]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/vm/vm_object.c
xnu-3789.31.2.tar.gz
[apple/xnu.git] / osfmk / vm / vm_object.c
index f49238416cd85a016087cbe036b3ac9a7384d356..72b525fdd76d7b5f9cc6ee82f2242c58a7ce43aa 100644 (file)
 #include <mach/memory_object_control_server.h>
 #include <mach/vm_param.h>
 
 #include <mach/memory_object_control_server.h>
 #include <mach/vm_param.h>
 
+#include <mach/sdt.h>
+
 #include <ipc/ipc_types.h>
 #include <ipc/ipc_port.h>
 
 #include <kern/kern_types.h>
 #include <kern/assert.h>
 #include <ipc/ipc_types.h>
 #include <ipc/ipc_port.h>
 
 #include <kern/kern_types.h>
 #include <kern/assert.h>
-#include <kern/lock.h>
 #include <kern/queue.h>
 #include <kern/xpr.h>
 #include <kern/queue.h>
 #include <kern/xpr.h>
+#include <kern/kalloc.h>
 #include <kern/zalloc.h>
 #include <kern/host.h>
 #include <kern/host_statistics.h>
 #include <kern/processor.h>
 #include <kern/misc_protos.h>
 #include <kern/zalloc.h>
 #include <kern/host.h>
 #include <kern/host_statistics.h>
 #include <kern/processor.h>
 #include <kern/misc_protos.h>
+#include <kern/policy_internal.h>
 
 #include <vm/memory_object.h>
 
 #include <vm/memory_object.h>
+#include <vm/vm_compressor_pager.h>
 #include <vm/vm_fault.h>
 #include <vm/vm_map.h>
 #include <vm/vm_object.h>
 #include <vm/vm_fault.h>
 #include <vm/vm_map.h>
 #include <vm/vm_object.h>
 #include <vm/vm_protos.h>
 #include <vm/vm_purgeable_internal.h>
 
 #include <vm/vm_protos.h>
 #include <vm/vm_purgeable_internal.h>
 
-#if CONFIG_EMBEDDED
-#include <sys/kern_memorystatus.h>
+#include <vm/vm_compressor.h>
+
+#if CONFIG_PHANTOM_CACHE
+#include <vm/vm_phantom_cache.h>
 #endif
 
 #endif
 
+boolean_t vm_object_collapse_compressor_allowed = TRUE;
+
+struct vm_counters vm_counters;
+
+#if VM_OBJECT_TRACKING
+boolean_t vm_object_tracking_inited = FALSE;
+btlog_t *vm_object_tracking_btlog;
+
+void
+vm_object_tracking_init(void)
+{
+       int vm_object_tracking;
+
+       vm_object_tracking = 1;
+       PE_parse_boot_argn("vm_object_tracking", &vm_object_tracking, 
+                          sizeof (vm_object_tracking));
+
+       if (vm_object_tracking) {
+               vm_object_tracking_btlog = btlog_create(
+                       VM_OBJECT_TRACKING_NUM_RECORDS,
+                       VM_OBJECT_TRACKING_BTDEPTH,
+                       TRUE /* caller_will_remove_entries_for_element? */);
+               assert(vm_object_tracking_btlog);
+               vm_object_tracking_inited = TRUE;
+       }
+}
+#endif /* VM_OBJECT_TRACKING */
+
 /*
  *     Virtual memory objects maintain the actual data
  *     associated with allocated virtual memory.  A given
 /*
  *     Virtual memory objects maintain the actual data
  *     associated with allocated virtual memory.  A given
@@ -203,9 +237,11 @@ static zone_t              vm_object_zone;         /* vm backing store zone */
  *     All wired-down kernel memory belongs to a single virtual
  *     memory object (kernel_object) to avoid wasting data structures.
  */
  *     All wired-down kernel memory belongs to a single virtual
  *     memory object (kernel_object) to avoid wasting data structures.
  */
-static struct vm_object                        kernel_object_store;
-vm_object_t                                            kernel_object;
+static struct vm_object                        kernel_object_store __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT)));
+vm_object_t                            kernel_object;
 
 
+static struct vm_object                        compressor_object_store __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT)));
+vm_object_t                            compressor_object = &compressor_object_store;
 
 /*
  *     The submap object is used as a placeholder for vm_map_submap
 
 /*
  *     The submap object is used as a placeholder for vm_map_submap
@@ -213,7 +249,7 @@ vm_object_t                                         kernel_object;
  *     is exported by the vm_map module.  The storage is declared
  *     here because it must be initialized here.
  */
  *     is exported by the vm_map module.  The storage is declared
  *     here because it must be initialized here.
  */
-static struct vm_object                        vm_submap_object_store;
+static struct vm_object                        vm_submap_object_store __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT)));
 
 /*
  *     Virtual memory objects are initialized from
 
 /*
  *     Virtual memory objects are initialized from
@@ -261,35 +297,47 @@ unsigned int vm_page_purged_others = 0;
 static vm_object_t     vm_object_cache_trim(
                                boolean_t called_from_vm_object_deallocate);
 
 static vm_object_t     vm_object_cache_trim(
                                boolean_t called_from_vm_object_deallocate);
 
-static queue_head_t    vm_object_cached_list;
-static int             vm_object_cached_count=0;
+static void            vm_object_deactivate_all_pages(
+                               vm_object_t     object);
+
 static int             vm_object_cached_high;  /* highest # cached objects */
 static int             vm_object_cached_max = 512;     /* may be patched*/
 
 static int             vm_object_cached_high;  /* highest # cached objects */
 static int             vm_object_cached_max = 512;     /* may be patched*/
 
-static lck_mtx_t       vm_object_cached_lock_data;
-static lck_mtx_ext_t   vm_object_cached_lock_data_ext;
-
 #define vm_object_cache_lock()         \
                lck_mtx_lock(&vm_object_cached_lock_data)
 #define vm_object_cache_lock_try()             \
                lck_mtx_try_lock(&vm_object_cached_lock_data)
 #define vm_object_cache_lock()         \
                lck_mtx_lock(&vm_object_cached_lock_data)
 #define vm_object_cache_lock_try()             \
                lck_mtx_try_lock(&vm_object_cached_lock_data)
+
+#endif /* VM_OBJECT_CACHE */
+
+static queue_head_t    vm_object_cached_list;
+static uint32_t                vm_object_cache_pages_freed = 0;
+static uint32_t                vm_object_cache_pages_moved = 0;
+static uint32_t                vm_object_cache_pages_skipped = 0;
+static uint32_t                vm_object_cache_adds = 0;
+static uint32_t                vm_object_cached_count = 0;
+static lck_mtx_t       vm_object_cached_lock_data;
+static lck_mtx_ext_t   vm_object_cached_lock_data_ext;
+
+static uint32_t                vm_object_page_grab_failed = 0;
+static uint32_t                vm_object_page_grab_skipped = 0;
+static uint32_t                vm_object_page_grab_returned = 0;
+static uint32_t                vm_object_page_grab_pmapped = 0;
+static uint32_t                vm_object_page_grab_reactivations = 0;
+
 #define vm_object_cache_lock_spin()            \
                lck_mtx_lock_spin(&vm_object_cached_lock_data)
 #define vm_object_cache_unlock()       \
                lck_mtx_unlock(&vm_object_cached_lock_data)
 
 #define vm_object_cache_lock_spin()            \
                lck_mtx_lock_spin(&vm_object_cached_lock_data)
 #define vm_object_cache_unlock()       \
                lck_mtx_unlock(&vm_object_cached_lock_data)
 
-#endif /* VM_OBJECT_CACHE */
-
-
-static void            vm_object_deactivate_all_pages(
-                               vm_object_t     object);
+static void    vm_object_cache_remove_locked(vm_object_t);
 
 
 #define        VM_OBJECT_HASH_COUNT            1024
 #define        VM_OBJECT_HASH_LOCK_COUNT       512
 
 
 
 #define        VM_OBJECT_HASH_COUNT            1024
 #define        VM_OBJECT_HASH_LOCK_COUNT       512
 
-static lck_mtx_t       vm_object_hashed_lock_data[VM_OBJECT_HASH_COUNT];
-static lck_mtx_ext_t   vm_object_hashed_lock_data_ext[VM_OBJECT_HASH_COUNT];
+static lck_mtx_t       vm_object_hashed_lock_data[VM_OBJECT_HASH_LOCK_COUNT];
+static lck_mtx_ext_t   vm_object_hashed_lock_data_ext[VM_OBJECT_HASH_LOCK_COUNT];
 
 static queue_head_t    vm_object_hashtable[VM_OBJECT_HASH_COUNT];
 static struct zone     *vm_object_hash_zone;
 
 static queue_head_t    vm_object_hashtable[VM_OBJECT_HASH_COUNT];
 static struct zone     *vm_object_hash_zone;
@@ -333,6 +381,39 @@ unsigned int vm_object_reap_count_async = 0;
 #define vm_object_reaper_unlock()      \
                lck_mtx_unlock(&vm_object_reaper_lock_data)
 
 #define vm_object_reaper_unlock()      \
                lck_mtx_unlock(&vm_object_reaper_lock_data)
 
+#if CONFIG_IOSCHED
+/* I/O Re-prioritization request list */
+queue_head_t   io_reprioritize_list;
+lck_spin_t     io_reprioritize_list_lock;
+
+#define IO_REPRIORITIZE_LIST_LOCK()    \
+               lck_spin_lock(&io_reprioritize_list_lock)
+#define IO_REPRIORITIZE_LIST_UNLOCK()  \
+               lck_spin_unlock(&io_reprioritize_list_lock)
+
+#define MAX_IO_REPRIORITIZE_REQS       8192
+zone_t                 io_reprioritize_req_zone;
+
+/* I/O Re-prioritization thread */
+int io_reprioritize_wakeup = 0;
+static void io_reprioritize_thread(void *param __unused, wait_result_t wr __unused);
+
+#define IO_REPRIO_THREAD_WAKEUP()      thread_wakeup((event_t)&io_reprioritize_wakeup)
+#define IO_REPRIO_THREAD_CONTINUATION()                                \
+{                                                              \
+       assert_wait(&io_reprioritize_wakeup, THREAD_UNINT);     \
+       thread_block(io_reprioritize_thread);                   \
+}
+
+void vm_page_request_reprioritize(vm_object_t, uint64_t, uint32_t, int);
+void vm_page_handle_prio_inversion(vm_object_t, vm_page_t);
+void vm_decmp_upl_reprioritize(upl_t, int);
+#endif
+
+#if 0
+#undef KERNEL_DEBUG
+#define KERNEL_DEBUG KERNEL_DEBUG_CONSTANT
+#endif
 
 
 static lck_mtx_t *
 
 
 static lck_mtx_t *
@@ -395,12 +476,26 @@ vm_object_hash_insert(
 {
        queue_t         bucket;
 
 {
        queue_t         bucket;
 
+       assert(vm_object_hash_lookup(entry->pager, FALSE) == NULL);
+
        bucket = &vm_object_hashtable[vm_object_hash(entry->pager)];
 
        queue_enter(bucket, entry, vm_object_hash_entry_t, hash_link);
 
        bucket = &vm_object_hashtable[vm_object_hash(entry->pager)];
 
        queue_enter(bucket, entry, vm_object_hash_entry_t, hash_link);
 
+       if (object->hashed) {
+               /*
+                * "hashed" was pre-set on this (new) object to avoid
+                * locking issues in vm_object_enter() (can't attempt to
+                * grab the object lock while holding the hash lock as
+                * a spinlock), so no need to set it here (and no need to
+                * hold the object's lock).
+                */
+       } else {
+               vm_object_lock_assert_exclusive(object);
+               object->hashed = TRUE;
+       }
+
        entry->object = object;
        entry->object = object;
-       object->hashed = TRUE;
 }
 
 static vm_object_hash_entry_t
 }
 
 static vm_object_hash_entry_t
@@ -440,20 +535,34 @@ _vm_object_allocate(
                object, size, 0,0,0);
 
        *object = vm_object_template;
                object, size, 0,0,0);
 
        *object = vm_object_template;
-       queue_init(&object->memq);
+       vm_page_queue_init(&object->memq);
        queue_init(&object->msr_q);
        queue_init(&object->msr_q);
-#if UPL_DEBUG
+#if UPL_DEBUG || CONFIG_IOSCHED
        queue_init(&object->uplq);
        queue_init(&object->uplq);
-#endif /* UPL_DEBUG */
+#endif
        vm_object_lock_init(object);
        vm_object_lock_init(object);
-       object->size = size;
+       object->vo_size = size;
+
+#if VM_OBJECT_TRACKING_OP_CREATED
+       if (vm_object_tracking_inited) {
+               void    *bt[VM_OBJECT_TRACKING_BTDEPTH];
+               int     numsaved = 0;
+
+               numsaved = OSBacktrace(bt, VM_OBJECT_TRACKING_BTDEPTH);
+               btlog_add_entry(vm_object_tracking_btlog,
+                               object,
+                               VM_OBJECT_TRACKING_OP_CREATED,
+                               bt,
+                               numsaved);
+       }
+#endif /* VM_OBJECT_TRACKING_OP_CREATED */
 }
 
 __private_extern__ vm_object_t
 vm_object_allocate(
        vm_object_size_t        size)
 {
 }
 
 __private_extern__ vm_object_t
 vm_object_allocate(
        vm_object_size_t        size)
 {
-       register vm_object_t object;
+       vm_object_t object;
 
        object = (vm_object_t) zalloc(vm_object_zone);
        
 
        object = (vm_object_t) zalloc(vm_object_zone);
        
@@ -467,9 +576,11 @@ vm_object_allocate(
 
 
 lck_grp_t              vm_object_lck_grp;
 
 
 lck_grp_t              vm_object_lck_grp;
-lck_grp_attr_t vm_object_lck_grp_attr;
+lck_grp_t              vm_object_cache_lck_grp;
+lck_grp_attr_t         vm_object_lck_grp_attr;
 lck_attr_t             vm_object_lck_attr;
 lck_attr_t             kernel_object_lck_attr;
 lck_attr_t             vm_object_lck_attr;
 lck_attr_t             kernel_object_lck_attr;
+lck_attr_t             compressor_object_lck_attr;
 
 /*
  *     vm_object_bootstrap:
 
 /*
  *     vm_object_bootstrap:
@@ -479,23 +590,27 @@ lck_attr_t                kernel_object_lck_attr;
 __private_extern__ void
 vm_object_bootstrap(void)
 {
 __private_extern__ void
 vm_object_bootstrap(void)
 {
-       register int    i;
+       int     i;
+       vm_size_t       vm_object_size;
 
 
-       vm_object_zone = zinit((vm_size_t) sizeof(struct vm_object),
-                               round_page(512*1024),
-                               round_page(12*1024),
-                               "vm objects");
+       vm_object_size = (sizeof(struct vm_object) + (VM_PACKED_POINTER_ALIGNMENT-1)) & ~(VM_PACKED_POINTER_ALIGNMENT - 1);
+
+       vm_object_zone = zinit(vm_object_size,
+                              round_page(512*1024),
+                              round_page(12*1024),
+                              "vm objects");
+       zone_change(vm_object_zone, Z_CALLERACCT, FALSE); /* don't charge caller */
+       zone_change(vm_object_zone, Z_NOENCRYPT, TRUE);
 
        vm_object_init_lck_grp();
 
 
        vm_object_init_lck_grp();
 
-#if VM_OBJECT_CACHE
        queue_init(&vm_object_cached_list);
 
        lck_mtx_init_ext(&vm_object_cached_lock_data,
                &vm_object_cached_lock_data_ext,
        queue_init(&vm_object_cached_list);
 
        lck_mtx_init_ext(&vm_object_cached_lock_data,
                &vm_object_cached_lock_data_ext,
-               &vm_object_lck_grp,
+               &vm_object_cache_lck_grp,
                &vm_object_lck_attr);
                &vm_object_lck_attr);
-#endif
+
        queue_init(&vm_object_reaper_queue);
 
        for (i = 0; i < VM_OBJECT_HASH_LOCK_COUNT; i++) {
        queue_init(&vm_object_reaper_queue);
 
        for (i = 0; i < VM_OBJECT_HASH_LOCK_COUNT; i++) {
@@ -514,6 +629,8 @@ vm_object_bootstrap(void)
                              round_page(512*1024),
                              round_page(12*1024),
                              "vm object hash entries");
                              round_page(512*1024),
                              round_page(12*1024),
                              "vm object hash entries");
+       zone_change(vm_object_hash_zone, Z_CALLERACCT, FALSE);
+       zone_change(vm_object_hash_zone, Z_NOENCRYPT, TRUE);
 
        for (i = 0; i < VM_OBJECT_HASH_COUNT; i++)
                queue_init(&vm_object_hashtable[i]);
 
        for (i = 0; i < VM_OBJECT_HASH_COUNT; i++)
                queue_init(&vm_object_hashtable[i]);
@@ -524,8 +641,10 @@ vm_object_bootstrap(void)
         */
 
        /* memq; Lock; init after allocation */
         */
 
        /* memq; Lock; init after allocation */
-       vm_object_template.memq.prev = NULL;
-       vm_object_template.memq.next = NULL;
+       
+
+       vm_object_template.memq.prev = 0;
+       vm_object_template.memq.next = 0;
 #if 0
        /*
         * We can't call vm_object_lock_init() here because that will
 #if 0
        /*
         * We can't call vm_object_lock_init() here because that will
@@ -536,7 +655,10 @@ vm_object_bootstrap(void)
         */
        vm_object_lock_init(&vm_object_template);
 #endif
         */
        vm_object_lock_init(&vm_object_template);
 #endif
-       vm_object_template.size = 0;
+#if DEVELOPMENT || DEBUG
+       vm_object_template.Lock_owner = 0;
+#endif
+       vm_object_template.vo_size = 0;
        vm_object_template.memq_hint = VM_PAGE_NULL;
        vm_object_template.ref_count = 1;
 #if    TASK_SWAPPER
        vm_object_template.memq_hint = VM_PAGE_NULL;
        vm_object_template.ref_count = 1;
 #if    TASK_SWAPPER
@@ -547,12 +669,15 @@ vm_object_bootstrap(void)
        vm_object_template.reusable_page_count = 0;
        vm_object_template.copy = VM_OBJECT_NULL;
        vm_object_template.shadow = VM_OBJECT_NULL;
        vm_object_template.reusable_page_count = 0;
        vm_object_template.copy = VM_OBJECT_NULL;
        vm_object_template.shadow = VM_OBJECT_NULL;
-       vm_object_template.shadow_offset = (vm_object_offset_t) 0;
+       vm_object_template.vo_shadow_offset = (vm_object_offset_t) 0;
        vm_object_template.pager = MEMORY_OBJECT_NULL;
        vm_object_template.paging_offset = 0;
        vm_object_template.pager_control = MEMORY_OBJECT_CONTROL_NULL;
        vm_object_template.copy_strategy = MEMORY_OBJECT_COPY_SYMMETRIC;
        vm_object_template.paging_in_progress = 0;
        vm_object_template.pager = MEMORY_OBJECT_NULL;
        vm_object_template.paging_offset = 0;
        vm_object_template.pager_control = MEMORY_OBJECT_CONTROL_NULL;
        vm_object_template.copy_strategy = MEMORY_OBJECT_COPY_SYMMETRIC;
        vm_object_template.paging_in_progress = 0;
+#if __LP64__
+       vm_object_template.__object1_unused_bits = 0;
+#endif /* __LP64__ */
        vm_object_template.activity_in_progress = 0;
 
        /* Begin bitfields */
        vm_object_template.activity_in_progress = 0;
 
        /* Begin bitfields */
@@ -568,8 +693,8 @@ vm_object_bootstrap(void)
        vm_object_template.pageout = FALSE;
        vm_object_template.alive = TRUE;
        vm_object_template.purgable = VM_PURGABLE_DENY;
        vm_object_template.pageout = FALSE;
        vm_object_template.alive = TRUE;
        vm_object_template.purgable = VM_PURGABLE_DENY;
+       vm_object_template.purgeable_when_ripe = FALSE;
        vm_object_template.shadowed = FALSE;
        vm_object_template.shadowed = FALSE;
-       vm_object_template.silent_overwrite = FALSE;
        vm_object_template.advisory_pageout = FALSE;
        vm_object_template.true_share = FALSE;
        vm_object_template.terminating = FALSE;
        vm_object_template.advisory_pageout = FALSE;
        vm_object_template.true_share = FALSE;
        vm_object_template.terminating = FALSE;
@@ -588,27 +713,30 @@ vm_object_bootstrap(void)
        vm_object_template.sequential = (vm_object_offset_t) 0;
        vm_object_template.pages_created = 0;
        vm_object_template.pages_used = 0;
        vm_object_template.sequential = (vm_object_offset_t) 0;
        vm_object_template.pages_created = 0;
        vm_object_template.pages_used = 0;
-
-#if    MACH_PAGEMAP
-       vm_object_template.existence_map = VM_EXTERNAL_NULL;
-#endif /* MACH_PAGEMAP */
+       vm_object_template.scan_collisions = 0;
+#if CONFIG_PHANTOM_CACHE
+       vm_object_template.phantom_object_id = 0;
+#endif
        vm_object_template.cow_hint = ~(vm_offset_t)0;
 #if    MACH_ASSERT
        vm_object_template.paging_object = VM_OBJECT_NULL;
 #endif /* MACH_ASSERT */
 
        /* cache bitfields */
        vm_object_template.cow_hint = ~(vm_offset_t)0;
 #if    MACH_ASSERT
        vm_object_template.paging_object = VM_OBJECT_NULL;
 #endif /* MACH_ASSERT */
 
        /* cache bitfields */
-       vm_object_template.wimg_bits = VM_WIMG_DEFAULT;
+       vm_object_template.wimg_bits = VM_WIMG_USE_DEFAULT;
+       vm_object_template.set_cache_attr = FALSE;
+       vm_object_template.object_slid = FALSE;
        vm_object_template.code_signed = FALSE;
        vm_object_template.hashed = FALSE;
        vm_object_template.transposed = FALSE;
        vm_object_template.mapping_in_progress = FALSE;
        vm_object_template.code_signed = FALSE;
        vm_object_template.hashed = FALSE;
        vm_object_template.transposed = FALSE;
        vm_object_template.mapping_in_progress = FALSE;
+       vm_object_template.phantom_isssd = FALSE;
        vm_object_template.volatile_empty = FALSE;
        vm_object_template.volatile_fault = FALSE;
        vm_object_template.all_reusable = FALSE;
        vm_object_template.blocked_access = FALSE;
        vm_object_template.__object2_unused_bits = 0;
        vm_object_template.volatile_empty = FALSE;
        vm_object_template.volatile_fault = FALSE;
        vm_object_template.all_reusable = FALSE;
        vm_object_template.blocked_access = FALSE;
        vm_object_template.__object2_unused_bits = 0;
-#if UPL_DEBUG
+#if CONFIG_IOSCHED || UPL_DEBUG
        vm_object_template.uplq.prev = NULL;
        vm_object_template.uplq.next = NULL;
 #endif /* UPL_DEBUG */
        vm_object_template.uplq.prev = NULL;
        vm_object_template.uplq.next = NULL;
 #endif /* UPL_DEBUG */
@@ -617,10 +745,33 @@ vm_object_bootstrap(void)
              sizeof (vm_object_template.pip_holders));
 #endif /* VM_PIP_DEBUG */
 
              sizeof (vm_object_template.pip_holders));
 #endif /* VM_PIP_DEBUG */
 
-       vm_object_template.objq.next=NULL;
-       vm_object_template.objq.prev=NULL;
+       vm_object_template.objq.next = NULL;
+       vm_object_template.objq.prev = NULL;
+
+       vm_object_template.purgeable_queue_type = PURGEABLE_Q_TYPE_MAX;
+       vm_object_template.purgeable_queue_group = 0;
+
+       vm_object_template.vo_cache_ts = 0;
+
+       vm_object_template.wire_tag = VM_KERN_MEMORY_NONE;
 
 
+       vm_object_template.io_tracking = FALSE;
+
+#if CONFIG_SECLUDED_MEMORY
+       vm_object_template.eligible_for_secluded = FALSE;
+       vm_object_template.can_grab_secluded = FALSE;
+#else /* CONFIG_SECLUDED_MEMORY */
+       vm_object_template.__object3_unused_bits = 0;
+#endif /* CONFIG_SECLUDED_MEMORY */
        
        
+#if DEBUG
+       bzero(&vm_object_template.purgeable_owner_bt[0],
+             sizeof (vm_object_template.purgeable_owner_bt));
+       vm_object_template.vo_purgeable_volatilizer = NULL;
+       bzero(&vm_object_template.purgeable_volatilizer_bt[0],
+             sizeof (vm_object_template.purgeable_volatilizer_bt));
+#endif /* DEBUG */
+
        /*
         *      Initialize the "kernel object"
         */
        /*
         *      Initialize the "kernel object"
         */
@@ -632,14 +783,13 @@ vm_object_bootstrap(void)
  *     VM_MAX_KERNEL_ADDRESS (vm_last_addr) is a maximum address, not a size.
  */
 
  *     VM_MAX_KERNEL_ADDRESS (vm_last_addr) is a maximum address, not a size.
  */
 
-#ifdef ppc
-       _vm_object_allocate(vm_last_addr + 1,
-                           kernel_object);
-#else
        _vm_object_allocate(VM_MAX_KERNEL_ADDRESS + 1,
                            kernel_object);
        _vm_object_allocate(VM_MAX_KERNEL_ADDRESS + 1,
                            kernel_object);
-#endif
+
+       _vm_object_allocate(VM_MAX_KERNEL_ADDRESS + 1,
+                           compressor_object);
        kernel_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
        kernel_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
+       compressor_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
 
        /*
         *      Initialize the "submap object".  Make it as large as the
 
        /*
         *      Initialize the "submap object".  Make it as large as the
@@ -647,13 +797,8 @@ vm_object_bootstrap(void)
         */
 
        vm_submap_object = &vm_submap_object_store;
         */
 
        vm_submap_object = &vm_submap_object_store;
-#ifdef ppc
-       _vm_object_allocate(vm_last_addr + 1,
-                           vm_submap_object);
-#else
        _vm_object_allocate(VM_MAX_KERNEL_ADDRESS + 1,
                            vm_submap_object);
        _vm_object_allocate(VM_MAX_KERNEL_ADDRESS + 1,
                            vm_submap_object);
-#endif
        vm_submap_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
 
        /*
        vm_submap_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
 
        /*
@@ -662,11 +807,31 @@ vm_object_bootstrap(void)
         * non-zone memory.
         */
        vm_object_reference(vm_submap_object);
         * non-zone memory.
         */
        vm_object_reference(vm_submap_object);
+}
 
 
-#if    MACH_PAGEMAP
-       vm_external_module_initialize();
-#endif /* MACH_PAGEMAP */
+#if CONFIG_IOSCHED
+void
+vm_io_reprioritize_init(void)
+{
+       kern_return_t   result;
+       thread_t        thread = THREAD_NULL;
+
+       /* Initialze the I/O reprioritization subsystem */
+        lck_spin_init(&io_reprioritize_list_lock, &vm_object_lck_grp, &vm_object_lck_attr);
+        queue_init(&io_reprioritize_list);
+
+       io_reprioritize_req_zone = zinit(sizeof(struct io_reprioritize_req),
+                                        MAX_IO_REPRIORITIZE_REQS * sizeof(struct io_reprioritize_req),
+                                        4096, "io_reprioritize_req");  
+
+       result = kernel_thread_start_priority(io_reprioritize_thread, NULL, 95 /* MAXPRI_KERNEL */, &thread);
+        if (result == KERN_SUCCESS) {
+                thread_deallocate(thread);
+        } else {
+                panic("Could not create io_reprioritize_thread");
+        }
 }
 }
+#endif
 
 void
 vm_object_reaper_init(void)
 
 void
 vm_object_reaper_init(void)
@@ -702,9 +867,12 @@ vm_object_init_lck_grp(void)
         */
        lck_grp_attr_setdefault(&vm_object_lck_grp_attr);
        lck_grp_init(&vm_object_lck_grp, "vm_object", &vm_object_lck_grp_attr);
         */
        lck_grp_attr_setdefault(&vm_object_lck_grp_attr);
        lck_grp_init(&vm_object_lck_grp, "vm_object", &vm_object_lck_grp_attr);
+       lck_grp_init(&vm_object_cache_lck_grp, "vm_object_cache", &vm_object_lck_grp_attr);
        lck_attr_setdefault(&vm_object_lck_attr);
        lck_attr_setdefault(&kernel_object_lck_attr);
        lck_attr_cleardebug(&kernel_object_lck_attr);
        lck_attr_setdefault(&vm_object_lck_attr);
        lck_attr_setdefault(&kernel_object_lck_attr);
        lck_attr_cleardebug(&kernel_object_lck_attr);
+       lck_attr_setdefault(&compressor_object_lck_attr);
+       lck_attr_cleardebug(&compressor_object_lck_attr);
 }
 
 #if VM_OBJECT_CACHE
 }
 
 #if VM_OBJECT_CACHE
@@ -728,9 +896,10 @@ static int cache_shadows = TRUE;
 unsigned long vm_object_deallocate_shared_successes = 0;
 unsigned long vm_object_deallocate_shared_failures = 0;
 unsigned long vm_object_deallocate_shared_swap_failures = 0;
 unsigned long vm_object_deallocate_shared_successes = 0;
 unsigned long vm_object_deallocate_shared_failures = 0;
 unsigned long vm_object_deallocate_shared_swap_failures = 0;
+
 __private_extern__ void
 vm_object_deallocate(
 __private_extern__ void
 vm_object_deallocate(
-       register vm_object_t    object)
+       vm_object_t     object)
 {
 #if VM_OBJECT_CACHE
        boolean_t       retry_cache_trim = FALSE;
 {
 #if VM_OBJECT_CACHE
        boolean_t       retry_cache_trim = FALSE;
@@ -744,20 +913,38 @@ vm_object_deallocate(
        if (object == VM_OBJECT_NULL)
                return;
 
        if (object == VM_OBJECT_NULL)
                return;
 
-       if (object == kernel_object) {
+       if (object == kernel_object || object == compressor_object) {
                vm_object_lock_shared(object);
 
                OSAddAtomic(-1, &object->ref_count);
 
                if (object->ref_count == 0) {
                vm_object_lock_shared(object);
 
                OSAddAtomic(-1, &object->ref_count);
 
                if (object->ref_count == 0) {
-                       panic("vm_object_deallocate: losing kernel_object\n");
+                       if (object == kernel_object)
+                               panic("vm_object_deallocate: losing kernel_object\n");
+                       else
+                               panic("vm_object_deallocate: losing compressor_object\n");
                }
                vm_object_unlock(object);
                return;
        }
 
                }
                vm_object_unlock(object);
                return;
        }
 
-       if (object->ref_count > 2 ||
-           (!object->named && object->ref_count > 1)) {
+       if (object->ref_count == 2 &&
+           object->named) {
+               /*
+                * This "named" object's reference count is about to
+                * drop from 2 to 1:
+                * we'll need to call memory_object_last_unmap().
+                */
+       } else if (object->ref_count == 2 &&
+                  object->internal &&
+                  object->shadow != VM_OBJECT_NULL) {
+               /*
+                * This internal object's reference count is about to
+                * drop from 2 to 1 and it has a shadow object:
+                * we'll want to try and collapse this object with its
+                * shadow.
+                */
+       } else if (object->ref_count >= 2) { 
                UInt32          original_ref_count;
                volatile UInt32 *ref_count_p;
                Boolean         atomic_swap;
                UInt32          original_ref_count;
                volatile UInt32 *ref_count_p;
                Boolean         atomic_swap;
@@ -778,19 +965,30 @@ vm_object_deallocate(
                 * Test again as "ref_count" could have changed.
                 * "named" shouldn't change.
                 */
                 * Test again as "ref_count" could have changed.
                 * "named" shouldn't change.
                 */
-               if (original_ref_count > 2 ||
-                   (!object->named && original_ref_count > 1)) {
+               if (original_ref_count == 2 &&
+                   object->named) {
+                       /* need to take slow path for m_o_last_unmap() */
+                       atomic_swap = FALSE;
+               } else if (original_ref_count == 2 &&
+                          object->internal &&
+                          object->shadow != VM_OBJECT_NULL) {
+                       /* need to take slow path for vm_object_collapse() */
+                       atomic_swap = FALSE;
+               } else if (original_ref_count < 2) { 
+                       /* need to take slow path for vm_object_terminate() */
+                       atomic_swap = FALSE;
+               } else {
+                       /* try an atomic update with the shared lock */
                        atomic_swap = OSCompareAndSwap(
                                original_ref_count,
                                original_ref_count - 1,
                                (UInt32 *) &object->ref_count);
                        if (atomic_swap == FALSE) {
                                vm_object_deallocate_shared_swap_failures++;
                        atomic_swap = OSCompareAndSwap(
                                original_ref_count,
                                original_ref_count - 1,
                                (UInt32 *) &object->ref_count);
                        if (atomic_swap == FALSE) {
                                vm_object_deallocate_shared_swap_failures++;
+                               /* fall back to the slow path... */
                        }
                        }
-
-               } else {
-                       atomic_swap = FALSE;
                }
                }
+                       
                vm_object_unlock(object);
 
                if (atomic_swap) {
                vm_object_unlock(object);
 
                if (atomic_swap) {
@@ -836,17 +1034,6 @@ vm_object_deallocate(
                                vm_object_lock(object);
                                vm_object_mapping_end(object);
                        }
                                vm_object_lock(object);
                                vm_object_mapping_end(object);
                        }
-                       /*
-                        * recheck the ref_count since we dropped the object lock
-                        * to call 'memory_object_last_unmap'... it's possible
-                        * additional references got taken and we only want
-                        * to deactivate the pages if this 'named' object will only
-                        * referenced by the backing pager once we drop our reference
-                        * below
-                        */
-                       if (!object->terminating && object->ref_count == 2)
-                               vm_object_deactivate_all_pages(object);
-
                        assert(object->ref_count > 0);
                }
 
                        assert(object->ref_count > 0);
                }
 
@@ -1071,6335 +1258,7571 @@ vm_object_deallocate(
 }
 
 
 }
 
 
-#if VM_OBJECT_CACHE
-/*
- *     Check to see whether we really need to trim
- *     down the cache. If so, remove an object from
- *     the cache, terminate it, and repeat.
- *
- *     Called with, and returns with, cache lock unlocked.
- */
-vm_object_t
-vm_object_cache_trim(
-       boolean_t called_from_vm_object_deallocate)
+
+vm_page_t
+vm_object_page_grab(
+       vm_object_t     object)
 {
 {
-       register vm_object_t object = VM_OBJECT_NULL;
-       vm_object_t shadow;
+       vm_page_t       p, next_p;
+       int             p_limit = 0;
+       int             p_skipped = 0;
 
 
-       for (;;) {
+       vm_object_lock_assert_exclusive(object);
 
 
-               /*
-                *      If we no longer need to trim the cache,
-                *      then we are done.
-                */
-               if (vm_object_cached_count <= vm_object_cached_max)
-                       return VM_OBJECT_NULL;
+       next_p = (vm_page_t)vm_page_queue_first(&object->memq);
+       p_limit = MIN(50, object->resident_page_count);
 
 
-               vm_object_cache_lock();
-               if (vm_object_cached_count <= vm_object_cached_max) {
-                       vm_object_cache_unlock();
-                       return VM_OBJECT_NULL;
-               }
+       while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)next_p) && --p_limit > 0) {
 
 
-               /*
-                *      We must trim down the cache, so remove
-                *      the first object in the cache.
-                */
-               XPR(XPR_VM_OBJECT,
-               "vm_object_cache_trim: removing from front of cache (%x, %x)\n",
-                       vm_object_cached_list.next,
-                       vm_object_cached_list.prev, 0, 0, 0);
+               p = next_p;
+               next_p = (vm_page_t)vm_page_queue_next(&next_p->listq);
 
 
-               object = (vm_object_t) queue_first(&vm_object_cached_list);
-               if(object == (vm_object_t) &vm_object_cached_list) {
-                       /* something's wrong with the calling parameter or */
-                       /* the value of vm_object_cached_count, just fix   */
-                       /* and return */
-                       if(vm_object_cached_max < 0)
-                               vm_object_cached_max = 0;
-                       vm_object_cached_count = 0;
-                       vm_object_cache_unlock();
-                       return VM_OBJECT_NULL;
-               }
-               vm_object_lock(object);
-               queue_remove(&vm_object_cached_list, object, vm_object_t,
-                            cached_list);
-               vm_object_cached_count--;
+               if (VM_PAGE_WIRED(p) || p->busy || p->cleaning || p->laundry || p->fictitious)
+                       goto move_page_in_obj;
 
 
-               vm_object_cache_unlock();
-               /*
-                *      Since this object is in the cache, we know
-                *      that it is initialized and has no references.
-                *      Take a reference to avoid recursive deallocations.
-                */
+               if (p->pmapped || p->dirty || p->precious) {
+                       vm_page_lockspin_queues();
 
 
-               assert(object->pager_initialized);
-               assert(object->ref_count == 0);
-               vm_object_lock_assert_exclusive(object);
-               object->ref_count++;
+                       if (p->pmapped) {
+                               int refmod_state;
 
 
-               /*
-                *      Terminate the object.
-                *      If the object had a shadow, we let vm_object_deallocate
-                *      deallocate it. "pageout" objects have a shadow, but
-                *      maintain a "paging reference" rather than a normal
-                *      reference.
-                *      (We are careful here to limit recursion.)
-                */
-               shadow = object->pageout?VM_OBJECT_NULL:object->shadow;
+                               vm_object_page_grab_pmapped++;
 
 
-               if(vm_object_terminate(object) != KERN_SUCCESS)
-                       continue;
+                               if (p->reference == FALSE || p->dirty == FALSE) {
 
 
-               if (shadow != VM_OBJECT_NULL) {
-                       if (called_from_vm_object_deallocate) {
-                               return shadow;
-                       } else {
-                               vm_object_deallocate(shadow);
+                                       refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(p));
+
+                                       if (refmod_state & VM_MEM_REFERENCED)
+                                               p->reference = TRUE;
+                                       if (refmod_state & VM_MEM_MODIFIED) {
+                                               SET_PAGE_DIRTY(p, FALSE);
+                                       }
+                               }
+                               if (p->dirty == FALSE && p->precious == FALSE) {
+
+                                       refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(p));
+
+                                       if (refmod_state & VM_MEM_REFERENCED)
+                                               p->reference = TRUE;
+                                       if (refmod_state & VM_MEM_MODIFIED) {
+                                               SET_PAGE_DIRTY(p, FALSE);
+                                       }
+
+                                       if (p->dirty == FALSE)
+                                               goto take_page;
+                               }
+                       }
+                       if ((p->vm_page_q_state != VM_PAGE_ON_ACTIVE_Q) && p->reference == TRUE) {
+                               vm_page_activate(p);
+
+                               VM_STAT_INCR(reactivations);
+                               vm_object_page_grab_reactivations++;
                        }
                        }
+                       vm_page_unlock_queues();
+move_page_in_obj:
+                       vm_page_queue_remove(&object->memq, p, vm_page_t, listq);
+                       vm_page_queue_enter(&object->memq, p, vm_page_t, listq);
+
+                       p_skipped++;
+                       continue;
                }
                }
+               vm_page_lockspin_queues();
+take_page:
+               vm_page_free_prepare_queues(p);
+               vm_object_page_grab_returned++;
+               vm_object_page_grab_skipped += p_skipped;
+
+               vm_page_unlock_queues();
+
+               vm_page_free_prepare_object(p, TRUE);
+               
+               return (p);
        }
        }
+       vm_object_page_grab_skipped += p_skipped;
+       vm_object_page_grab_failed++;
+
+       return (NULL);
 }
 }
-#endif
 
 
 
 
-/*
- *     Routine:        vm_object_terminate
- *     Purpose:
- *             Free all resources associated with a vm_object.
- *     In/out conditions:
- *             Upon entry, the object must be locked,
- *             and the object must have exactly one reference.
- *
- *             The shadow object reference is left alone.
- *
- *             The object must be unlocked if its found that pages
- *             must be flushed to a backing object.  If someone
- *             manages to map the object while it is being flushed
- *             the object is returned unlocked and unchanged.  Otherwise,
- *             upon exit, the cache will be unlocked, and the
- *             object will cease to exist.
- */
-static kern_return_t
-vm_object_terminate(
+
+#define EVICT_PREPARE_LIMIT    64
+#define EVICT_AGE              10
+
+static clock_sec_t     vm_object_cache_aging_ts = 0;
+
+static void
+vm_object_cache_remove_locked(
        vm_object_t     object)
 {
        vm_object_t     object)
 {
-       vm_object_t     shadow_object;
+       assert(object->purgable == VM_PURGABLE_DENY);
+       assert(object->wired_page_count == 0);
 
 
-       XPR(XPR_VM_OBJECT, "vm_object_terminate, object 0x%X ref %d\n",
-               object, object->ref_count, 0, 0, 0);
+       queue_remove(&vm_object_cached_list, object, vm_object_t, objq);
+       object->objq.next = NULL;
+       object->objq.prev = NULL;
 
 
-       if (!object->pageout && (!object->temporary || object->can_persist) &&
-           (object->pager != NULL || object->shadow_severed)) {
-               /*
-                * Clear pager_trusted bit so that the pages get yanked
-                * out of the object instead of cleaned in place.  This
-                * prevents a deadlock in XMM and makes more sense anyway.
-                */
-               object->pager_trusted = FALSE;
+       vm_object_cached_count--;
+}
 
 
-               vm_object_reap_pages(object, REAP_TERMINATE);
-       }
-       /*
-        *      Make sure the object isn't already being terminated
-        */
-       if (object->terminating) {
-               vm_object_lock_assert_exclusive(object);
-               object->ref_count--;
-               assert(object->ref_count > 0);
-               vm_object_unlock(object);
-               return KERN_FAILURE;
-       }
+void
+vm_object_cache_remove(
+       vm_object_t     object)
+{
+       vm_object_cache_lock_spin();
 
 
-       /*
-        * Did somebody get a reference to the object while we were
-        * cleaning it?
-        */
-       if (object->ref_count != 1) {
-               vm_object_lock_assert_exclusive(object);
-               object->ref_count--;
-               assert(object->ref_count > 0);
-               vm_object_res_deallocate(object);
-               vm_object_unlock(object);
-               return KERN_FAILURE;
-       }
+       if (object->objq.next || object->objq.prev)
+               vm_object_cache_remove_locked(object);
 
 
-       /*
-        *      Make sure no one can look us up now.
-        */
+       vm_object_cache_unlock();
+}
 
 
-       object->terminating = TRUE;
-       object->alive = FALSE;
+void
+vm_object_cache_add(
+       vm_object_t     object)
+{
+       clock_sec_t sec;
+       clock_nsec_t nsec;
 
 
-       if (object->hashed) {
-               lck_mtx_t       *lck;
+       assert(object->purgable == VM_PURGABLE_DENY);
+       assert(object->wired_page_count == 0);
 
 
-               lck = vm_object_hash_lock_spin(object->pager);
-               vm_object_remove(object);
-               vm_object_hash_unlock(lck);
+       if (object->resident_page_count == 0)
+               return;
+       clock_get_system_nanotime(&sec, &nsec);
+
+       vm_object_cache_lock_spin();
+
+       if (object->objq.next == NULL && object->objq.prev == NULL) {
+               queue_enter(&vm_object_cached_list, object, vm_object_t, objq);
+               object->vo_cache_ts = sec + EVICT_AGE;
+               object->vo_cache_pages_to_scan = object->resident_page_count;
+
+               vm_object_cached_count++;
+               vm_object_cache_adds++;
        }
        }
+       vm_object_cache_unlock();
+}
+
+int
+vm_object_cache_evict(
+       int     num_to_evict,
+       int     max_objects_to_examine)
+{
+       vm_object_t     object = VM_OBJECT_NULL;
+       vm_object_t     next_obj = VM_OBJECT_NULL;
+       vm_page_t       local_free_q = VM_PAGE_NULL;
+       vm_page_t       p;
+       vm_page_t       next_p;
+       int             object_cnt = 0;
+       vm_page_t       ep_array[EVICT_PREPARE_LIMIT];
+       int             ep_count;
+       int             ep_limit;
+       int             ep_index;
+       int             ep_freed = 0;
+       int             ep_moved = 0;
+       uint32_t        ep_skipped = 0;
+       clock_sec_t     sec;
+       clock_nsec_t    nsec;
+
+       KERNEL_DEBUG(0x13001ec | DBG_FUNC_START, 0, 0, 0, 0, 0);
        /*
        /*
-        *      Detach the object from its shadow if we are the shadow's
-        *      copy. The reference we hold on the shadow must be dropped
-        *      by our caller.
+        * do a couple of quick checks to see if it's 
+        * worthwhile grabbing the lock
         */
         */
-       if (((shadow_object = object->shadow) != VM_OBJECT_NULL) &&
-           !(object->pageout)) {
-               vm_object_lock(shadow_object);
-               if (shadow_object->copy == object)
-                       shadow_object->copy = VM_OBJECT_NULL;
-               vm_object_unlock(shadow_object);
+       if (queue_empty(&vm_object_cached_list)) {
+               KERNEL_DEBUG(0x13001ec | DBG_FUNC_END, 0, 0, 0, 0, 0);
+               return (0);
        }
        }
+       clock_get_system_nanotime(&sec, &nsec);
 
 
-       if (object->paging_in_progress != 0 ||
-           object->activity_in_progress != 0) {
-               /*
-                * There are still some paging_in_progress references
-                * on this object, meaning that there are some paging
-                * or other I/O operations in progress for this VM object.
-                * Such operations take some paging_in_progress references
-                * up front to ensure that the object doesn't go away, but
-                * they may also need to acquire a reference on the VM object,
-                * to map it in kernel space, for example.  That means that
-                * they may end up releasing the last reference on the VM
-                * object, triggering its termination, while still holding
-                * paging_in_progress references.  Waiting for these
-                * pending paging_in_progress references to go away here would
-                * deadlock.
-                *
-                * To avoid deadlocking, we'll let the vm_object_reaper_thread
-                * complete the VM object termination if it still holds
-                * paging_in_progress references at this point.
-                *
-                * No new paging_in_progress should appear now that the
-                * VM object is "terminating" and not "alive".
-                */
-               vm_object_reap_async(object);
-               vm_object_unlock(object);
-               /*
-                * Return KERN_FAILURE to let the caller know that we
-                * haven't completed the termination and it can't drop this
-                * object's reference on its shadow object yet.
-                * The reaper thread will take care of that once it has
-                * completed this object's termination.
-                */
-               return KERN_FAILURE;
-       }
        /*
        /*
-        * complete the VM object termination
+        * the object on the head of the queue has not
+        * yet sufficiently aged
         */
         */
-       vm_object_reap(object);
-       object = VM_OBJECT_NULL;
-
+       if (sec < vm_object_cache_aging_ts) {
+               KERNEL_DEBUG(0x13001ec | DBG_FUNC_END, 0, 0, 0, 0, 0);
+               return (0);
+       }
        /*
        /*
-        * the object lock was released by vm_object_reap()
-        *
-        * KERN_SUCCESS means that this object has been terminated
-        * and no longer needs its shadow object but still holds a
-        * reference on it.
-        * The caller is responsible for dropping that reference.
-        * We can't call vm_object_deallocate() here because that
-        * would create a recursion.
+        * don't need the queue lock to find 
+        * and lock an object on the cached list
         */
         */
-       return KERN_SUCCESS;
-}
+       vm_page_unlock_queues();
 
 
+       vm_object_cache_lock_spin();
 
 
-/*
- * vm_object_reap():
- *
- * Complete the termination of a VM object after it's been marked
- * as "terminating" and "!alive" by vm_object_terminate().
- *
- * The VM object must be locked by caller.
- * The lock will be released on return and the VM object is no longer valid.
- */
-void
-vm_object_reap(
-       vm_object_t object)
-{
-       memory_object_t         pager;
+       for (;;) {
+               next_obj = (vm_object_t)queue_first(&vm_object_cached_list);
 
 
-       vm_object_lock_assert_exclusive(object);
-       assert(object->paging_in_progress == 0);
-       assert(object->activity_in_progress == 0);
+               while (!queue_end(&vm_object_cached_list, (queue_entry_t)next_obj) && object_cnt++ < max_objects_to_examine) {
 
 
-       vm_object_reap_count++;
+                       object = next_obj;
+                       next_obj = (vm_object_t)queue_next(&next_obj->objq);
 
 
-       pager = object->pager;
-       object->pager = MEMORY_OBJECT_NULL;
+                       assert(object->purgable == VM_PURGABLE_DENY);
+                       assert(object->wired_page_count == 0);
+                       
+                       if (sec < object->vo_cache_ts) {
+                               KERNEL_DEBUG(0x130020c, object, object->resident_page_count, object->vo_cache_ts, sec, 0);
 
 
-       if (pager != MEMORY_OBJECT_NULL)
-               memory_object_control_disable(object->pager_control);
+                               vm_object_cache_aging_ts = object->vo_cache_ts;
+                               object = VM_OBJECT_NULL;
+                               break;
+                       }
+                       if (!vm_object_lock_try_scan(object)) {
+                               /*
+                                * just skip over this guy for now... if we find
+                                * an object to steal pages from, we'll revist in a bit...
+                                * hopefully, the lock will have cleared
+                                */
+                               KERNEL_DEBUG(0x13001f8, object, object->resident_page_count, 0, 0, 0);
 
 
-       object->ref_count--;
-#if    TASK_SWAPPER
-       assert(object->res_count == 0);
-#endif /* TASK_SWAPPER */
+                               object = VM_OBJECT_NULL;
+                               continue;
+                       }
+                       if (vm_page_queue_empty(&object->memq) || object->vo_cache_pages_to_scan == 0) {
+                               /*
+                                * this case really shouldn't happen, but it's not fatal
+                                * so deal with it... if we don't remove the object from
+                                * the list, we'll never move past it.
+                                */
+                               KERNEL_DEBUG(0x13001fc, object, object->resident_page_count, ep_freed, ep_moved, 0);
+                               
+                               vm_object_cache_remove_locked(object);
+                               vm_object_unlock(object);
+                               object = VM_OBJECT_NULL;
+                               continue;
+                       }
+                       /*
+                        * we have a locked object with pages...
+                        * time to start harvesting
+                        */
+                       break;
+               }
+               vm_object_cache_unlock();
 
 
-       assert (object->ref_count == 0);
+               if (object == VM_OBJECT_NULL)
+                       break;
 
 
-       /*
-        * remove from purgeable queue if it's on
-        */
-       if (object->objq.next || object->objq.prev) {
-               purgeable_q_t queue = vm_purgeable_object_remove(object);
-               assert(queue);
+               /*
+                * object is locked at this point and
+                * has resident pages
+                */
+               next_p = (vm_page_t)vm_page_queue_first(&object->memq);
 
 
-               /* Must take page lock for this - using it to protect token queue */
-               vm_page_lock_queues();
-               vm_purgeable_token_delete_first(queue);
-        
-               assert(queue->debug_count_objects>=0);
-               vm_page_unlock_queues();
-       }
-    
-       /*
-        *      Clean or free the pages, as appropriate.
-        *      It is possible for us to find busy/absent pages,
-        *      if some faults on this object were aborted.
-        */
-       if (object->pageout) {
-               assert(object->shadow != VM_OBJECT_NULL);
+               /*
+                * break the page scan into 2 pieces to minimize the time spent
+                * behind the page queue lock...
+                * the list of pages on these unused objects is likely to be cold
+                * w/r to the cpu cache which increases the time to scan the list
+                * tenfold...  and we may have a 'run' of pages we can't utilize that
+                * needs to be skipped over...
+                */
+               if ((ep_limit = num_to_evict - (ep_freed + ep_moved)) > EVICT_PREPARE_LIMIT)
+                       ep_limit = EVICT_PREPARE_LIMIT;
+               ep_count = 0;
 
 
-               vm_pageout_object_terminate(object);
+               while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)next_p) && object->vo_cache_pages_to_scan && ep_count < ep_limit) {
 
 
-       } else if (((object->temporary && !object->can_persist) || (pager == MEMORY_OBJECT_NULL))) {
+                       p = next_p;
+                       next_p = (vm_page_t)vm_page_queue_next(&next_p->listq);
 
 
-               vm_object_reap_pages(object, REAP_REAP);
-       }
-       assert(queue_empty(&object->memq));
-       assert(object->paging_in_progress == 0);
-       assert(object->activity_in_progress == 0);
-       assert(object->ref_count == 0);
+                       object->vo_cache_pages_to_scan--;
 
 
-       /*
-        * If the pager has not already been released by
-        * vm_object_destroy, we need to terminate it and
-        * release our reference to it here.
-        */
-       if (pager != MEMORY_OBJECT_NULL) {
-               vm_object_unlock(object);
-               vm_object_release_pager(pager, object->hashed);
-               vm_object_lock(object);
-       }
+                       if (VM_PAGE_WIRED(p) || p->busy || p->cleaning || p->laundry) {
+                               vm_page_queue_remove(&object->memq, p, vm_page_t, listq);
+                               vm_page_queue_enter(&object->memq, p, vm_page_t, listq);
 
 
-       /* kick off anyone waiting on terminating */
-       object->terminating = FALSE;
-       vm_object_paging_begin(object);
-       vm_object_paging_end(object);
-       vm_object_unlock(object);
+                               ep_skipped++;
+                               continue;
+                       }
+                       if (p->wpmapped || p->dirty || p->precious) {
+                               vm_page_queue_remove(&object->memq, p, vm_page_t, listq);
+                               vm_page_queue_enter(&object->memq, p, vm_page_t, listq);
 
 
-#if    MACH_PAGEMAP
-       vm_external_destroy(object->existence_map, object->size);
-#endif /* MACH_PAGEMAP */
+                               pmap_clear_reference(VM_PAGE_GET_PHYS_PAGE(p));
+                       }
+                       ep_array[ep_count++] = p;
+               }
+               KERNEL_DEBUG(0x13001f4 | DBG_FUNC_START, object, object->resident_page_count, ep_freed, ep_moved, 0);
 
 
-       object->shadow = VM_OBJECT_NULL;
+               vm_page_lockspin_queues();
 
 
-       vm_object_lock_destroy(object);
-       /*
-        *      Free the space for the object.
-        */
-       zfree(vm_object_zone, object);
-       object = VM_OBJECT_NULL;
-}
+               for (ep_index = 0; ep_index < ep_count; ep_index++) {
 
 
+                       p = ep_array[ep_index];
 
 
+                       if (p->wpmapped || p->dirty || p->precious) {
+                               p->reference = FALSE;
+                               p->no_cache = FALSE;
 
 
-#define V_O_R_MAX_BATCH 128
+                               /*
+                                * we've already filtered out pages that are in the laundry
+                                * so if we get here, this page can't be on the pageout queue
+                                */
+                               vm_page_queues_remove(p, FALSE);
+                               vm_page_enqueue_inactive(p, TRUE);
 
 
+                               ep_moved++;
+                       } else {
+#if CONFIG_PHANTOM_CACHE
+                               vm_phantom_cache_add_ghost(p);
+#endif
+                               vm_page_free_prepare_queues(p);
 
 
-#define VM_OBJ_REAP_FREELIST(_local_free_q, do_disconnect)             \
-       MACRO_BEGIN                                                     \
-       if (_local_free_q) {                                            \
-               if (do_disconnect) {                                    \
-                       vm_page_t m;                                    \
-                       for (m = _local_free_q;                         \
-                            m != VM_PAGE_NULL;                         \
-                            m = (vm_page_t) m->pageq.next) {           \
-                               if (m->pmapped) {                       \
-                                       pmap_disconnect(m->phys_page);  \
-                               }                                       \
-                       }                                               \
-               }                                                       \
-               vm_page_free_list(_local_free_q, TRUE);                 \
-               _local_free_q = VM_PAGE_NULL;                           \
-       }                                                               \
-       MACRO_END
+                               assert(p->pageq.next == 0 && p->pageq.prev == 0);
+                               /*
+                                * Add this page to our list of reclaimed pages,
+                                * to be freed later.
+                                */
+                               p->snext = local_free_q;
+                               local_free_q = p;
 
 
+                               ep_freed++;
+                       }
+               }
+               vm_page_unlock_queues();
 
 
-void
-vm_object_reap_pages(
-       vm_object_t     object,
-       int             reap_type)
-{
-       vm_page_t       p;
-       vm_page_t       next;
-       vm_page_t       local_free_q = VM_PAGE_NULL;
-       int             loop_count;
-       boolean_t       disconnect_on_release;
+               KERNEL_DEBUG(0x13001f4 | DBG_FUNC_END, object, object->resident_page_count, ep_freed, ep_moved, 0);
 
 
-       if (reap_type == REAP_DATA_FLUSH) {
+               if (local_free_q) {
+                       vm_page_free_list(local_free_q, TRUE);
+                       local_free_q = VM_PAGE_NULL;
+               }
+               if (object->vo_cache_pages_to_scan == 0) {
+                       KERNEL_DEBUG(0x1300208, object, object->resident_page_count, ep_freed, ep_moved, 0);
+
+                       vm_object_cache_remove(object);
+
+                       KERNEL_DEBUG(0x13001fc, object, object->resident_page_count, ep_freed, ep_moved, 0);
+               }
                /*
                /*
-                * We need to disconnect pages from all pmaps before
-                * releasing them to the free list
+                * done with this object
                 */
                 */
-               disconnect_on_release = TRUE;
-       } else {
+               vm_object_unlock(object);
+               object = VM_OBJECT_NULL;
+
                /*
                /*
-                * Either the caller has already disconnected the pages
-                * from all pmaps, or we disconnect them here as we add
-                * them to out local list of pages to be released.
-                * No need to re-disconnect them when we release the pages
-                * to the free list.
+                * at this point, we are not holding any locks
                 */
                 */
-               disconnect_on_release = FALSE;
+               if ((ep_freed + ep_moved) >= num_to_evict) {
+                       /*
+                        * we've reached our target for the
+                        * number of pages to evict
+                        */
+                       break;
+               }
+               vm_object_cache_lock_spin();
        }
        }
-               
-restart_after_sleep:
-       if (queue_empty(&object->memq))
-               return;
-       loop_count = V_O_R_MAX_BATCH + 1;
-
-       vm_page_lockspin_queues();
+       /*
+        * put the page queues lock back to the caller's
+        * idea of it 
+        */
+       vm_page_lock_queues();
 
 
-       next = (vm_page_t)queue_first(&object->memq);
+       vm_object_cache_pages_freed += ep_freed;
+       vm_object_cache_pages_moved += ep_moved;
+       vm_object_cache_pages_skipped += ep_skipped;
 
 
-       while (!queue_end(&object->memq, (queue_entry_t)next)) {
+       KERNEL_DEBUG(0x13001ec | DBG_FUNC_END, ep_freed, 0, 0, 0, 0);
+       return (ep_freed);
+}
 
 
-               p = next;
-               next = (vm_page_t)queue_next(&next->listq);
 
 
-               if (--loop_count == 0) {
-                                       
-                       vm_page_unlock_queues();
+#if VM_OBJECT_CACHE
+/*
+ *     Check to see whether we really need to trim
+ *     down the cache. If so, remove an object from
+ *     the cache, terminate it, and repeat.
+ *
+ *     Called with, and returns with, cache lock unlocked.
+ */
+vm_object_t
+vm_object_cache_trim(
+       boolean_t called_from_vm_object_deallocate)
+{
+       vm_object_t object = VM_OBJECT_NULL;
+       vm_object_t shadow;
 
 
-                       if (local_free_q) {
-                               /*
-                                * Free the pages we reclaimed so far
-                                * and take a little break to avoid
-                                * hogging the page queue lock too long
-                                */
-                               VM_OBJ_REAP_FREELIST(local_free_q,
-                                                    disconnect_on_release);
-                       } else
-                               mutex_pause(0);
+       for (;;) {
 
 
-                       loop_count = V_O_R_MAX_BATCH + 1;
+               /*
+                *      If we no longer need to trim the cache,
+                *      then we are done.
+                */
+               if (vm_object_cached_count <= vm_object_cached_max)
+                       return VM_OBJECT_NULL;
 
 
-                       vm_page_lockspin_queues();
+               vm_object_cache_lock();
+               if (vm_object_cached_count <= vm_object_cached_max) {
+                       vm_object_cache_unlock();
+                       return VM_OBJECT_NULL;
                }
                }
-               if (reap_type == REAP_DATA_FLUSH || reap_type == REAP_TERMINATE) {
-
-                       if (reap_type == REAP_DATA_FLUSH && (p->pageout == TRUE && p->list_req_pending == TRUE)) {
-                               p->list_req_pending = FALSE;
-                               p->cleaning = FALSE;
-                               p->pageout = FALSE;
-                               /*
-                                * need to drop the laundry count...
-                                * we may also need to remove it
-                                * from the I/O paging queue...
-                                * vm_pageout_throttle_up handles both cases
-                                *
-                                * the laundry and pageout_queue flags are cleared...
-                                */
-                               vm_pageout_throttle_up(p);
-
-                               /*
-                                * toss the wire count we picked up
-                                * when we intially set this page up
-                                * to be cleaned...
-                                */
-                               vm_page_unwire(p);
-                               PAGE_WAKEUP(p);
-
-                       } else if (p->busy || p->cleaning) {
-
-                               vm_page_unlock_queues();
-                               /*
-                                * free the pages reclaimed so far
-                                */
-                               VM_OBJ_REAP_FREELIST(local_free_q,
-                                                    disconnect_on_release);
 
 
-                               PAGE_SLEEP(object, p, THREAD_UNINT);
+               /*
+                *      We must trim down the cache, so remove
+                *      the first object in the cache.
+                */
+               XPR(XPR_VM_OBJECT,
+               "vm_object_cache_trim: removing from front of cache (%x, %x)\n",
+                       vm_object_cached_list.next,
+                       vm_object_cached_list.prev, 0, 0, 0);
 
 
-                               goto restart_after_sleep;
-                       }
+               object = (vm_object_t) queue_first(&vm_object_cached_list);
+               if(object == (vm_object_t) &vm_object_cached_list) {
+                       /* something's wrong with the calling parameter or */
+                       /* the value of vm_object_cached_count, just fix   */
+                       /* and return */
+                       if(vm_object_cached_max < 0)
+                               vm_object_cached_max = 0;
+                       vm_object_cached_count = 0;
+                       vm_object_cache_unlock();
+                       return VM_OBJECT_NULL;
                }
                }
-               switch (reap_type) {
-
-               case REAP_DATA_FLUSH:
-                       if (VM_PAGE_WIRED(p)) {
-                               /*
-                                * this is an odd case... perhaps we should
-                                * zero-fill this page since we're conceptually
-                                * tossing its data at this point, but leaving
-                                * it on the object to honor the 'wire' contract
-                                */
-                               continue;
-                       }
-                       break;
-                       
-               case REAP_PURGEABLE:
-                       if (VM_PAGE_WIRED(p)) {
-                               /* can't purge a wired page */
-                               vm_page_purged_wired++;
-                               continue;
-                       }
-
-                       if (p->busy) {
-                               /*
-                                * We can't reclaim a busy page but we can
-                                * make it pageable (it's not wired) to make
-                                * sure that it gets considered by
-                                * vm_pageout_scan() later.
-                                */
-                               vm_page_deactivate(p);
-                               vm_page_purged_busy++;
-                               continue;
-                       }
-
-                       if (p->cleaning || p->laundry || p->list_req_pending) {
-                               /*
-                                * page is being acted upon,
-                                * so don't mess with it
-                                */
-                               vm_page_purged_others++;
-                               continue;
-                       }
-                       assert(p->object != kernel_object);
-
-                       /*
-                        * we can discard this page...
-                        */
-                       if (p->pmapped == TRUE) {
-                               int refmod_state;
-                               /*
-                                * unmap the page
-                                */
-                               refmod_state = pmap_disconnect(p->phys_page);
-                               if (refmod_state & VM_MEM_MODIFIED) {
-                                       p->dirty = TRUE;
-                               }
-                       }
-                       if (p->dirty || p->precious) {
-                               /*
-                                * we saved the cost of cleaning this page !
-                                */
-                               vm_page_purged_count++;
-                       }
-
-                       break;
-
-               case REAP_TERMINATE:
-                       if (p->absent || p->private) {
-                               /*
-                                *      For private pages, VM_PAGE_FREE just
-                                *      leaves the page structure around for
-                                *      its owner to clean up.  For absent
-                                *      pages, the structure is returned to
-                                *      the appropriate pool.
-                                */
-                               break;
-                       }
-                       if (p->fictitious) {
-                               assert (p->phys_page == vm_page_guard_addr);
-                               break;
-                       }
-                       if (!p->dirty && p->wpmapped)
-                               p->dirty = pmap_is_modified(p->phys_page);
-
-                       if ((p->dirty || p->precious) && !p->error && object->alive) {
+               vm_object_lock(object);
+               queue_remove(&vm_object_cached_list, object, vm_object_t,
+                            cached_list);
+               vm_object_cached_count--;
 
 
-                               p->busy = TRUE;
+               vm_object_cache_unlock();
+               /*
+                *      Since this object is in the cache, we know
+                *      that it is initialized and has no references.
+                *      Take a reference to avoid recursive deallocations.
+                */
 
 
-                               VM_PAGE_QUEUES_REMOVE(p);
+               assert(object->pager_initialized);
+               assert(object->ref_count == 0);
+               vm_object_lock_assert_exclusive(object);
+               object->ref_count++;
 
 
-                               vm_page_unlock_queues();
-                               /*
-                                * free the pages reclaimed so far
-                                */
-                               VM_OBJ_REAP_FREELIST(local_free_q,
-                                                    disconnect_on_release);
+               /*
+                *      Terminate the object.
+                *      If the object had a shadow, we let vm_object_deallocate
+                *      deallocate it. "pageout" objects have a shadow, but
+                *      maintain a "paging reference" rather than a normal
+                *      reference.
+                *      (We are careful here to limit recursion.)
+                */
+               shadow = object->pageout?VM_OBJECT_NULL:object->shadow;
 
 
-                               /*
-                                * flush page... page will be freed
-                                * upon completion of I/O
-                                */
-                               vm_pageout_cluster(p);
-                               vm_object_paging_wait(object, THREAD_UNINT);
+               if(vm_object_terminate(object) != KERN_SUCCESS)
+                       continue;
 
 
-                               goto restart_after_sleep;
+               if (shadow != VM_OBJECT_NULL) {
+                       if (called_from_vm_object_deallocate) {
+                               return shadow;
+                       } else {
+                               vm_object_deallocate(shadow);
                        }
                        }
-                       break;
-
-               case REAP_REAP:
-                       break;
                }
                }
-               vm_page_free_prepare_queues(p);
-               assert(p->pageq.next == NULL && p->pageq.prev == NULL);
-               /*
-                * Add this page to our list of reclaimed pages,
-                * to be freed later.
-                */
-               p->pageq.next = (queue_entry_t) local_free_q;
-               local_free_q = p;
        }
        }
-       vm_page_unlock_queues();
-
-       /*
-        * Free the remaining reclaimed pages
-        */
-       VM_OBJ_REAP_FREELIST(local_free_q,
-                            disconnect_on_release);
 }
 }
+#endif
 
 
 
 
-void
-vm_object_reap_async(
+/*
+ *     Routine:        vm_object_terminate
+ *     Purpose:
+ *             Free all resources associated with a vm_object.
+ *     In/out conditions:
+ *             Upon entry, the object must be locked,
+ *             and the object must have exactly one reference.
+ *
+ *             The shadow object reference is left alone.
+ *
+ *             The object must be unlocked if its found that pages
+ *             must be flushed to a backing object.  If someone
+ *             manages to map the object while it is being flushed
+ *             the object is returned unlocked and unchanged.  Otherwise,
+ *             upon exit, the cache will be unlocked, and the
+ *             object will cease to exist.
+ */
+static kern_return_t
+vm_object_terminate(
        vm_object_t     object)
 {
        vm_object_t     object)
 {
-       vm_object_lock_assert_exclusive(object);
-
-       vm_object_reaper_lock_spin();
-
-       vm_object_reap_count_async++;
+       vm_object_t     shadow_object;
 
 
-       /* enqueue the VM object... */
-       queue_enter(&vm_object_reaper_queue, object,
-                   vm_object_t, cached_list);
+       XPR(XPR_VM_OBJECT, "vm_object_terminate, object 0x%X ref %d\n",
+               object, object->ref_count, 0, 0, 0);
 
 
-       vm_object_reaper_unlock();
+       vm_object_lock_assert_exclusive(object);
 
 
-       /* ... and wake up the reaper thread */
-       thread_wakeup((event_t) &vm_object_reaper_queue);
-}
+       if (!object->pageout && (!object->temporary || object->can_persist) &&
+           (object->pager != NULL || object->shadow_severed)) {
+               /*
+                * Clear pager_trusted bit so that the pages get yanked
+                * out of the object instead of cleaned in place.  This
+                * prevents a deadlock in XMM and makes more sense anyway.
+                */
+               object->pager_trusted = FALSE;
 
 
-
-void
-vm_object_reaper_thread(void)
-{
-       vm_object_t     object, shadow_object;
-
-       vm_object_reaper_lock_spin();
-
-       while (!queue_empty(&vm_object_reaper_queue)) {
-               queue_remove_first(&vm_object_reaper_queue,
-                                  object,
-                                  vm_object_t,
-                                  cached_list);
-
-               vm_object_reaper_unlock();
-               vm_object_lock(object);
-
-               assert(object->terminating);
-               assert(!object->alive);
-               
-               /*
-                * The pageout daemon might be playing with our pages.
-                * Now that the object is dead, it won't touch any more
-                * pages, but some pages might already be on their way out.
-                * Hence, we wait until the active paging activities have
-                * ceased before we break the association with the pager
-                * itself.
-                */
-               while (object->paging_in_progress != 0 ||
-                       object->activity_in_progress != 0) {
-                       vm_object_wait(object,
-                                      VM_OBJECT_EVENT_PAGING_IN_PROGRESS,
-                                      THREAD_UNINT);
-                       vm_object_lock(object);
-               }
-
-               shadow_object =
-                       object->pageout ? VM_OBJECT_NULL : object->shadow;
-
-               vm_object_reap(object);
-               /* cache is unlocked and object is no longer valid */
-               object = VM_OBJECT_NULL;
-
-               if (shadow_object != VM_OBJECT_NULL) {
-                       /*
-                        * Drop the reference "object" was holding on
-                        * its shadow object.
-                        */
-                       vm_object_deallocate(shadow_object);
-                       shadow_object = VM_OBJECT_NULL;
-               }
-               vm_object_reaper_lock_spin();
+               vm_object_reap_pages(object, REAP_TERMINATE);
        }
        }
-
-       /* wait for more work... */
-       assert_wait((event_t) &vm_object_reaper_queue, THREAD_UNINT);
-
-       vm_object_reaper_unlock();
-
-       thread_block((thread_continue_t) vm_object_reaper_thread);
-       /*NOTREACHED*/
-}
-
-/*
- *     Routine:        vm_object_pager_wakeup
- *     Purpose:        Wake up anyone waiting for termination of a pager.
- */
-
-static void
-vm_object_pager_wakeup(
-       memory_object_t pager)
-{
-       vm_object_hash_entry_t  entry;
-       boolean_t               waiting = FALSE;
-       lck_mtx_t               *lck;
-
        /*
        /*
-        *      If anyone was waiting for the memory_object_terminate
-        *      to be queued, wake them up now.
+        *      Make sure the object isn't already being terminated
         */
         */
-       lck = vm_object_hash_lock_spin(pager);
-       entry = vm_object_hash_lookup(pager, TRUE);
-       if (entry != VM_OBJECT_HASH_ENTRY_NULL)
-               waiting = entry->waiting;
-       vm_object_hash_unlock(lck);
-
-       if (entry != VM_OBJECT_HASH_ENTRY_NULL) {
-               if (waiting)
-                       thread_wakeup((event_t) pager);
-               vm_object_hash_entry_free(entry);
+       if (object->terminating) {
+               vm_object_lock_assert_exclusive(object);
+               object->ref_count--;
+               assert(object->ref_count > 0);
+               vm_object_unlock(object);
+               return KERN_FAILURE;
        }
        }
-}
-
-/*
- *     Routine:        vm_object_release_pager
- *     Purpose:        Terminate the pager and, upon completion,
- *                     release our last reference to it.
- *                     just like memory_object_terminate, except
- *                     that we wake up anyone blocked in vm_object_enter
- *                     waiting for termination message to be queued
- *                     before calling memory_object_init.
- */
-static void
-vm_object_release_pager(
-       memory_object_t pager,
-       boolean_t       hashed)
-{
 
        /*
 
        /*
-        *      Terminate the pager.
+        * Did somebody get a reference to the object while we were
+        * cleaning it?
         */
         */
-
-       (void) memory_object_terminate(pager);
-
-       if (hashed == TRUE) {
-               /*
-                *      Wakeup anyone waiting for this terminate
-                *      and remove the entry from the hash
-                */
-               vm_object_pager_wakeup(pager);
+       if (object->ref_count != 1) {
+               vm_object_lock_assert_exclusive(object);
+               object->ref_count--;
+               assert(object->ref_count > 0);
+               vm_object_res_deallocate(object);
+               vm_object_unlock(object);
+               return KERN_FAILURE;
        }
        }
-       /*
-        *      Release reference to pager.
-        */
-       memory_object_deallocate(pager);
-}
-
-/*
- *     Routine:        vm_object_destroy
- *     Purpose:
- *             Shut down a VM object, despite the
- *             presence of address map (or other) references
- *             to the vm_object.
- */
-kern_return_t
-vm_object_destroy(
-       vm_object_t             object,
-       __unused kern_return_t          reason)
-{
-       memory_object_t         old_pager;
-
-       if (object == VM_OBJECT_NULL)
-               return(KERN_SUCCESS);
 
        /*
 
        /*
-        *      Remove the pager association immediately.
-        *
-        *      This will prevent the memory manager from further
-        *      meddling.  [If it wanted to flush data or make
-        *      other changes, it should have done so before performing
-        *      the destroy call.]
+        *      Make sure no one can look us up now.
         */
 
         */
 
-       vm_object_lock(object);
-       object->can_persist = FALSE;
-       object->named = FALSE;
+       object->terminating = TRUE;
        object->alive = FALSE;
 
        object->alive = FALSE;
 
+       if ( !object->internal && (object->objq.next || object->objq.prev))
+               vm_object_cache_remove(object);
+
        if (object->hashed) {
                lck_mtx_t       *lck;
        if (object->hashed) {
                lck_mtx_t       *lck;
-               /*
-                *      Rip out the pager from the vm_object now...
-                */
+
                lck = vm_object_hash_lock_spin(object->pager);
                vm_object_remove(object);
                vm_object_hash_unlock(lck);
        }
                lck = vm_object_hash_lock_spin(object->pager);
                vm_object_remove(object);
                vm_object_hash_unlock(lck);
        }
-       old_pager = object->pager;
-       object->pager = MEMORY_OBJECT_NULL;
-       if (old_pager != MEMORY_OBJECT_NULL)
-               memory_object_control_disable(object->pager_control);
-
        /*
        /*
-        * Wait for the existing paging activity (that got
-        * through before we nulled out the pager) to subside.
+        *      Detach the object from its shadow if we are the shadow's
+        *      copy. The reference we hold on the shadow must be dropped
+        *      by our caller.
         */
         */
+       if (((shadow_object = object->shadow) != VM_OBJECT_NULL) &&
+           !(object->pageout)) {
+               vm_object_lock(shadow_object);
+               if (shadow_object->copy == object)
+                       shadow_object->copy = VM_OBJECT_NULL;
+               vm_object_unlock(shadow_object);
+       }
 
 
-       vm_object_paging_wait(object, THREAD_UNINT);
-       vm_object_unlock(object);
-
+       if (object->paging_in_progress != 0 ||
+           object->activity_in_progress != 0) {
+               /*
+                * There are still some paging_in_progress references
+                * on this object, meaning that there are some paging
+                * or other I/O operations in progress for this VM object.
+                * Such operations take some paging_in_progress references
+                * up front to ensure that the object doesn't go away, but
+                * they may also need to acquire a reference on the VM object,
+                * to map it in kernel space, for example.  That means that
+                * they may end up releasing the last reference on the VM
+                * object, triggering its termination, while still holding
+                * paging_in_progress references.  Waiting for these
+                * pending paging_in_progress references to go away here would
+                * deadlock.
+                *
+                * To avoid deadlocking, we'll let the vm_object_reaper_thread
+                * complete the VM object termination if it still holds
+                * paging_in_progress references at this point.
+                *
+                * No new paging_in_progress should appear now that the
+                * VM object is "terminating" and not "alive".
+                */
+               vm_object_reap_async(object);
+               vm_object_unlock(object);
+               /*
+                * Return KERN_FAILURE to let the caller know that we
+                * haven't completed the termination and it can't drop this
+                * object's reference on its shadow object yet.
+                * The reaper thread will take care of that once it has
+                * completed this object's termination.
+                */
+               return KERN_FAILURE;
+       }
        /*
        /*
-        *      Terminate the object now.
+        * complete the VM object termination
         */
         */
-       if (old_pager != MEMORY_OBJECT_NULL) {
-               vm_object_release_pager(old_pager, object->hashed);
-
-               /* 
-                * JMM - Release the caller's reference.  This assumes the
-                * caller had a reference to release, which is a big (but
-                * currently valid) assumption if this is driven from the
-                * vnode pager (it is holding a named reference when making
-                * this call)..
-                */
-               vm_object_deallocate(object);
+       vm_object_reap(object);
+       object = VM_OBJECT_NULL;
 
 
-       }
-       return(KERN_SUCCESS);
+       /*
+        * the object lock was released by vm_object_reap()
+        *
+        * KERN_SUCCESS means that this object has been terminated
+        * and no longer needs its shadow object but still holds a
+        * reference on it.
+        * The caller is responsible for dropping that reference.
+        * We can't call vm_object_deallocate() here because that
+        * would create a recursion.
+        */
+       return KERN_SUCCESS;
 }
 
 
 }
 
 
-#define VM_OBJ_DEACT_ALL_STATS DEBUG
-#if VM_OBJ_DEACT_ALL_STATS
-uint32_t vm_object_deactivate_all_pages_batches = 0;
-uint32_t vm_object_deactivate_all_pages_pages = 0;
-#endif /* VM_OBJ_DEACT_ALL_STATS */
 /*
 /*
- *     vm_object_deactivate_all_pages
+ * vm_object_reap():
  *
  *
- *     Deactivate all pages in the specified object.  (Keep its pages
- *     in memory even though it is no longer referenced.)
+ * Complete the termination of a VM object after it's been marked
+ * as "terminating" and "!alive" by vm_object_terminate().
  *
  *
- *     The object must be locked.
+ * The VM object must be locked by caller.
+ * The lock will be released on return and the VM object is no longer valid.
  */
  */
-static void
-vm_object_deactivate_all_pages(
-       register vm_object_t    object)
-{
-       register vm_page_t      p;
-       int                     loop_count;
-#if VM_OBJ_DEACT_ALL_STATS
-       int                     pages_count;
-#endif /* VM_OBJ_DEACT_ALL_STATS */
-#define V_O_D_A_P_MAX_BATCH    256
-
-       loop_count = V_O_D_A_P_MAX_BATCH;
-#if VM_OBJ_DEACT_ALL_STATS
-       pages_count = 0;
-#endif /* VM_OBJ_DEACT_ALL_STATS */
-       vm_page_lock_queues();
-       queue_iterate(&object->memq, p, vm_page_t, listq) {
-               if (--loop_count == 0) {
-#if VM_OBJ_DEACT_ALL_STATS
-                       hw_atomic_add(&vm_object_deactivate_all_pages_batches,
-                                     1);
-                       hw_atomic_add(&vm_object_deactivate_all_pages_pages,
-                                     pages_count);
-                       pages_count = 0;
-#endif /* VM_OBJ_DEACT_ALL_STATS */
-                       lck_mtx_yield(&vm_page_queue_lock);
-                       loop_count = V_O_D_A_P_MAX_BATCH;
-               }
-               if (!p->busy && !p->throttled) {
-#if VM_OBJ_DEACT_ALL_STATS
-                       pages_count++;
-#endif /* VM_OBJ_DEACT_ALL_STATS */
-                       vm_page_deactivate(p);
-               }
-       }
-#if VM_OBJ_DEACT_ALL_STATS
-       if (pages_count) {
-               hw_atomic_add(&vm_object_deactivate_all_pages_batches, 1);
-               hw_atomic_add(&vm_object_deactivate_all_pages_pages,
-                             pages_count);
-               pages_count = 0;
-       }
-#endif /* VM_OBJ_DEACT_ALL_STATS */
-       vm_page_unlock_queues();
-}
-
 
 
+void
+vm_object_reap(
+       vm_object_t object)
+{
+       memory_object_t         pager;
 
 
-/*
- * when deallocating pages it is necessary to hold 
- * the vm_page_queue_lock (a hot global lock) for certain operations
- * on the page... however, the majority of the work can be done
- * while merely holding the object lock... to mitigate the time spent behind the
- * global lock, go to a 2 pass algorithm... collect pages up to DELAYED_WORK_LIMIT
- * while doing all of the work that doesn't require the vm_page_queue_lock...
- * them call dw_do_work to acquire the vm_page_queue_lock and do the
- * necessary work for each page... we will grab the busy bit on the page
- * so that dw_do_work can drop the object lock if it can't immediately take the
- * vm_page_queue_lock in order to compete for the locks in the same order that
- * vm_pageout_scan takes them.
- */
-
-#define DELAYED_WORK_LIMIT     32
+       vm_object_lock_assert_exclusive(object);
+       assert(object->paging_in_progress == 0);
+       assert(object->activity_in_progress == 0);
 
 
-#define DW_clear_reference     0x01
-#define DW_move_page           0x02
-#define DW_clear_busy          0x04
-#define DW_PAGE_WAKEUP         0x08
+       vm_object_reap_count++;
 
 
+       /*
+        * Disown this purgeable object to cleanup its owner's purgeable
+        * ledgers.  We need to do this before disconnecting the object
+        * from its pager, to properly account for compressed pages.
+        */
+       if (object->internal &&
+           object->purgable != VM_PURGABLE_DENY) {
+               vm_purgeable_accounting(object,
+                                       object->purgable,
+                                       TRUE); /* disown */
+       }
 
 
-struct dw {
-       vm_page_t       dw_m;
-       int             dw_mask;
-};
+       pager = object->pager;
+       object->pager = MEMORY_OBJECT_NULL;
 
 
-static void dw_do_work(vm_object_t object, struct dw *dwp, int dw_count);
+       if (pager != MEMORY_OBJECT_NULL)
+               memory_object_control_disable(object->pager_control);
 
 
+       object->ref_count--;
+#if    TASK_SWAPPER
+       assert(object->res_count == 0);
+#endif /* TASK_SWAPPER */
 
 
-static void
-dw_do_work(
-       vm_object_t     object,
-       struct dw       *dwp,
-       int             dw_count)
-{
-       vm_page_t       m;
-       int             j;
+       assert (object->ref_count == 0);
 
        /*
 
        /*
-        * pageout_scan takes the vm_page_lock_queues first
-        * then tries for the object lock... to avoid what
-        * is effectively a lock inversion, we'll go to the
-        * trouble of taking them in that same order... otherwise
-        * if this object contains the majority of the pages resident
-        * in the UBC (or a small set of large objects actively being
-        * worked on contain the majority of the pages), we could
-        * cause the pageout_scan thread to 'starve' in its attempt
-        * to find pages to move to the free queue, since it has to
-        * successfully acquire the object lock of any candidate page
-        * before it can steal/clean it.
+        * remove from purgeable queue if it's on
         */
         */
-       if (!vm_page_trylockspin_queues()) {
-               vm_object_unlock(object);
+       if (object->internal) {
+               task_t owner;
 
 
-               vm_page_lockspin_queues();
+               owner = object->vo_purgeable_owner;
 
 
-               for (j = 0; ; j++) {
-                       if (!vm_object_lock_avoid(object) &&
-                           _vm_object_lock_try(object))
-                               break;
-                       vm_page_unlock_queues();
-                       mutex_pause(j);
-                       vm_page_lockspin_queues();
-               }
-       }
-       for (j = 0; j < dw_count; j++, dwp++) {
+               VM_OBJECT_UNWIRED(object);
 
 
-               m = dwp->dw_m;
+               if (object->purgable == VM_PURGABLE_DENY) {
+                       /* not purgeable: nothing to do */
+               } else if (object->purgable == VM_PURGABLE_VOLATILE) {
+                       purgeable_q_t queue;
 
 
-               if (dwp->dw_mask & DW_clear_reference)
-                       m->reference = FALSE;
+                       assert(object->vo_purgeable_owner == NULL);
 
 
-               if (dwp->dw_mask & DW_move_page) {
-                       VM_PAGE_QUEUES_REMOVE(m);
+                       queue = vm_purgeable_object_remove(object);
+                       assert(queue);
 
 
-                       assert(!m->laundry);
-                       assert(m->object != kernel_object);
-                       assert(m->pageq.next == NULL &&
-                              m->pageq.prev == NULL);
-                                       
-                       if (m->zero_fill) {
-                               queue_enter_first(&vm_page_queue_zf, m, vm_page_t, pageq);
-                               vm_zf_queue_count++;
-                       } else {
-                               queue_enter_first(&vm_page_queue_inactive, m, vm_page_t, pageq);
+                       if (object->purgeable_when_ripe) {
+                               /*
+                                * Must take page lock for this -
+                                * using it to protect token queue
+                                */
+                               vm_page_lock_queues();
+                               vm_purgeable_token_delete_first(queue);
+        
+                               assert(queue->debug_count_objects>=0);
+                               vm_page_unlock_queues();
                        }
                        }
-                       m->inactive = TRUE;
 
 
-                       if (!m->fictitious) {
-                               vm_page_inactive_count++;
-                               token_new_pagecount++;
-                       } else {
-                               assert(m->phys_page == vm_page_fictitious_addr);
+                       /*
+                        * Update "vm_page_purgeable_count" in bulk and mark
+                        * object as VM_PURGABLE_EMPTY to avoid updating 
+                        * "vm_page_purgeable_count" again in vm_page_remove()
+                        * when reaping the pages.
+                        */
+                       unsigned int delta;
+                       assert(object->resident_page_count >=
+                              object->wired_page_count);
+                       delta = (object->resident_page_count -
+                                object->wired_page_count);
+                       if (delta != 0) {
+                               assert(vm_page_purgeable_count >= delta);
+                               OSAddAtomic(-delta,
+                                           (SInt32 *)&vm_page_purgeable_count);
                        }
                        }
+                       if (object->wired_page_count != 0) {
+                               assert(vm_page_purgeable_wired_count >=
+                                      object->wired_page_count);
+                               OSAddAtomic(-object->wired_page_count,
+                                           (SInt32 *)&vm_page_purgeable_wired_count);
+                       }
+                       object->purgable = VM_PURGABLE_EMPTY;
                }
                }
-               if (dwp->dw_mask & DW_clear_busy)
-                       dwp->dw_m->busy = FALSE;
-
-               if (dwp->dw_mask & DW_PAGE_WAKEUP)
-                       PAGE_WAKEUP(dwp->dw_m);
+               else if (object->purgable == VM_PURGABLE_NONVOLATILE ||
+                        object->purgable == VM_PURGABLE_EMPTY) {
+                       /* remove from nonvolatile queue */
+                       assert(object->vo_purgeable_owner == TASK_NULL);
+                       vm_purgeable_nonvolatile_dequeue(object);
+               } else {
+                       panic("object %p in unexpected purgeable state 0x%x\n",
+                             object, object->purgable);
+               }
+               assert(object->objq.next == NULL);
+               assert(object->objq.prev == NULL);
        }
        }
-       vm_page_unlock_queues();
+    
+       /*
+        *      Clean or free the pages, as appropriate.
+        *      It is possible for us to find busy/absent pages,
+        *      if some faults on this object were aborted.
+        */
+       if (object->pageout) {
+               assert(object->shadow != VM_OBJECT_NULL);
 
 
-#if CONFIG_EMBEDDED
-       {
-       int percent_avail;
+               vm_pageout_object_terminate(object);
+
+       } else if (((object->temporary && !object->can_persist) || (pager == MEMORY_OBJECT_NULL))) {
+
+               vm_object_reap_pages(object, REAP_REAP);
+       }
+       assert(vm_page_queue_empty(&object->memq));
+       assert(object->paging_in_progress == 0);
+       assert(object->activity_in_progress == 0);
+       assert(object->ref_count == 0);
 
        /*
 
        /*
-        * Decide if we need to send a memory status notification.
+        * If the pager has not already been released by
+        * vm_object_destroy, we need to terminate it and
+        * release our reference to it here.
         */
         */
-       percent_avail = 
-               (vm_page_active_count + vm_page_inactive_count + 
-                vm_page_speculative_count + vm_page_free_count +
-                (IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 /
-               atop_64(max_mem);
-       if (percent_avail >= (kern_memorystatus_level + 5) || 
-           percent_avail <= (kern_memorystatus_level - 5)) {
-               kern_memorystatus_level = percent_avail;
-               thread_wakeup((event_t)&kern_memorystatus_wakeup);
-       }
+       if (pager != MEMORY_OBJECT_NULL) {
+               vm_object_unlock(object);
+               vm_object_release_pager(pager, object->hashed);
+               vm_object_lock(object);
        }
        }
-#endif
-}
 
 
+       /* kick off anyone waiting on terminating */
+       object->terminating = FALSE;
+       vm_object_paging_begin(object);
+       vm_object_paging_end(object);
+       vm_object_unlock(object);
 
 
+       object->shadow = VM_OBJECT_NULL;
 
 
-/*
- * The "chunk" macros are used by routines below when looking for pages to deactivate.  These
- * exist because of the need to handle shadow chains.  When deactivating pages, we only
- * want to deactive the ones at the top most level in the object chain.  In order to do
- * this efficiently, the specified address range is divided up into "chunks" and we use
- * a bit map to keep track of which pages have already been processed as we descend down
- * the shadow chain.  These chunk macros hide the details of the bit map implementation
- * as much as we can.
- *
- * For convenience, we use a 64-bit data type as the bit map, and therefore a chunk is
- * set to 64 pages.  The bit map is indexed from the low-order end, so that the lowest
- * order bit represents page 0 in the current range and highest order bit represents
- * page 63.
- *
- * For further convenience, we also use negative logic for the page state in the bit map.
- * The bit is set to 1 to indicate it has not yet been seen, and to 0 to indicate it has
- * been processed.  This way we can simply test the 64-bit long word to see if it's zero
- * to easily tell if the whole range has been processed.  Therefore, the bit map starts
- * out with all the bits set.  The macros below hide all these details from the caller.
- */
+#if VM_OBJECT_TRACKING
+       if (vm_object_tracking_inited) {
+               btlog_remove_entries_for_element(vm_object_tracking_btlog,
+                                                object);
+       }
+#endif /* VM_OBJECT_TRACKING */
 
 
-#define PAGES_IN_A_CHUNK       64      /* The number of pages in the chunk must */
-                                       /* be the same as the number of bits in  */
-                                       /* the chunk_state_t type. We use 64     */
-                                       /* just for convenience.                 */
+       vm_object_lock_destroy(object);
+       /*
+        *      Free the space for the object.
+        */
+       zfree(vm_object_zone, object);
+       object = VM_OBJECT_NULL;
+}
 
 
-#define CHUNK_SIZE     (PAGES_IN_A_CHUNK * PAGE_SIZE_64)       /* Size of a chunk in bytes */
 
 
-typedef uint64_t       chunk_state_t;
+unsigned int vm_max_batch = 256;
 
 
-/*
- * The bit map uses negative logic, so we start out with all 64 bits set to indicate
- * that no pages have been processed yet.  Also, if len is less than the full CHUNK_SIZE,
- * then we mark pages beyond the len as having been "processed" so that we don't waste time
- * looking at pages in that range.  This can save us from unnecessarily chasing down the 
- * shadow chain.
- */
+#define V_O_R_MAX_BATCH 128
 
 
-#define CHUNK_INIT(c, len)                                             \
+#define BATCH_LIMIT(max)       (vm_max_batch >= max ? max : vm_max_batch)
+
+
+#define VM_OBJ_REAP_FREELIST(_local_free_q, do_disconnect)             \
        MACRO_BEGIN                                                     \
        MACRO_BEGIN                                                     \
-       uint64_t p;                                                     \
-                                                                       \
-       (c) = 0xffffffffffffffffLL;                                     \
-                                                                       \
-       for (p = (len) / PAGE_SIZE_64; p < PAGES_IN_A_CHUNK; p++)       \
-               MARK_PAGE_HANDLED(c, p);                                \
+       if (_local_free_q) {                                            \
+               if (do_disconnect) {                                    \
+                       vm_page_t m;                                    \
+                       for (m = _local_free_q;                         \
+                            m != VM_PAGE_NULL;                         \
+                            m = m->snext) {            \
+                               if (m->pmapped) {                       \
+                                       pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); \
+                               }                                       \
+                       }                                               \
+               }                                                       \
+               vm_page_free_list(_local_free_q, TRUE);                 \
+               _local_free_q = VM_PAGE_NULL;                           \
+       }                                                               \
        MACRO_END
 
        MACRO_END
 
-/*
- * Return true if all pages in the chunk have not yet been processed.
- */
 
 
-#define CHUNK_NOT_COMPLETE(c)  ((c) != 0)
+void
+vm_object_reap_pages(
+       vm_object_t     object,
+       int             reap_type)
+{
+       vm_page_t       p;
+       vm_page_t       next;
+       vm_page_t       local_free_q = VM_PAGE_NULL;
+       int             loop_count;
+       boolean_t       disconnect_on_release;
+       pmap_flush_context      pmap_flush_context_storage;
 
 
-/*
- * Return true if the page at offset 'p' in the bit map has already been handled
- * while processing a higher level object in the shadow chain.
- */
+       if (reap_type == REAP_DATA_FLUSH) {
+               /*
+                * We need to disconnect pages from all pmaps before
+                * releasing them to the free list
+                */
+               disconnect_on_release = TRUE;
+       } else {
+               /*
+                * Either the caller has already disconnected the pages
+                * from all pmaps, or we disconnect them here as we add
+                * them to out local list of pages to be released.
+                * No need to re-disconnect them when we release the pages
+                * to the free list.
+                */
+               disconnect_on_release = FALSE;
+       }
+               
+restart_after_sleep:
+       if (vm_page_queue_empty(&object->memq))
+               return;
+       loop_count = BATCH_LIMIT(V_O_R_MAX_BATCH);
 
 
-#define PAGE_ALREADY_HANDLED(c, p)     (((c) & (1LL << (p))) == 0)
+       if (reap_type == REAP_PURGEABLE)
+               pmap_flush_context_init(&pmap_flush_context_storage);
 
 
-/*
- * Mark the page at offset 'p' in the bit map as having been processed.
- */
+       vm_page_lockspin_queues();
 
 
-#define MARK_PAGE_HANDLED(c, p) \
-MACRO_BEGIN \
-       (c) = (c) & ~(1LL << (p)); \
-MACRO_END
+       next = (vm_page_t)vm_page_queue_first(&object->memq);
 
 
+       while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)next)) {
 
 
-/*
- * Return true if the page at the given offset has been paged out.  Object is
- * locked upon entry and returned locked.
- */
+               p = next;
+               next = (vm_page_t)vm_page_queue_next(&next->listq);
 
 
-static boolean_t
-page_is_paged_out(
-       vm_object_t             object,
-       vm_object_offset_t      offset)
-{
-       kern_return_t   kr;
-       memory_object_t pager;
+               if (--loop_count == 0) {
+                                       
+                       vm_page_unlock_queues();
 
 
-       /*
-        * Check the existence map for the page if we have one, otherwise
-        * ask the pager about this page.
-        */
+                       if (local_free_q) {
 
 
-#if MACH_PAGEMAP
-       if (object->existence_map) {
-               if (vm_external_state_get(object->existence_map, offset)
-                   == VM_EXTERNAL_STATE_EXISTS) {
-                       /*
-                        * We found the page
-                        */
+                               if (reap_type == REAP_PURGEABLE) {
+                                       pmap_flush(&pmap_flush_context_storage);
+                                       pmap_flush_context_init(&pmap_flush_context_storage);
+                               }
+                               /*
+                                * Free the pages we reclaimed so far
+                                * and take a little break to avoid
+                                * hogging the page queue lock too long
+                                */
+                               VM_OBJ_REAP_FREELIST(local_free_q,
+                                                    disconnect_on_release);
+                       } else
+                               mutex_pause(0);
 
 
-                       return TRUE;
+                       loop_count = BATCH_LIMIT(V_O_R_MAX_BATCH);
+
+                       vm_page_lockspin_queues();
                }
                }
-       } else
-#endif
-               if (object->internal &&
-                  object->alive &&
-                  !object->terminating &&
-                  object->pager_ready) {
+               if (reap_type == REAP_DATA_FLUSH || reap_type == REAP_TERMINATE) {
 
 
-               /*
-                * We're already holding a "paging in progress" reference
-                * so the object can't disappear when we release the lock.
-                */
+                       if (p->busy || p->cleaning) {
 
 
-               assert(object->paging_in_progress);
-               pager = object->pager;
-               vm_object_unlock(object);
+                               vm_page_unlock_queues();
+                               /*
+                                * free the pages reclaimed so far
+                                */
+                               VM_OBJ_REAP_FREELIST(local_free_q,
+                                                    disconnect_on_release);
+
+                               PAGE_SLEEP(object, p, THREAD_UNINT);
 
 
-               kr = memory_object_data_request(
-                       pager,
-                       offset + object->paging_offset,
-                       0,      /* just poke the pager */
-                       VM_PROT_READ,
-                       NULL);
+                               goto restart_after_sleep;
+                       }
+                       if (p->laundry)
+                               vm_pageout_steal_laundry(p, TRUE);
+               }
+               switch (reap_type) {
 
 
-               vm_object_lock(object);
+               case REAP_DATA_FLUSH:
+                       if (VM_PAGE_WIRED(p)) {
+                               /*
+                                * this is an odd case... perhaps we should
+                                * zero-fill this page since we're conceptually
+                                * tossing its data at this point, but leaving
+                                * it on the object to honor the 'wire' contract
+                                */
+                               continue;
+                       }
+                       break;
+                       
+               case REAP_PURGEABLE:
+                       if (VM_PAGE_WIRED(p)) {
+                               /*
+                                * can't purge a wired page
+                                */
+                               vm_page_purged_wired++;
+                               continue;
+                       }
+                       if (p->laundry && !p->busy && !p->cleaning)
+                               vm_pageout_steal_laundry(p, TRUE);
+
+                       if (p->cleaning || p->laundry || p->absent) {
+                               /*
+                                * page is being acted upon,
+                                * so don't mess with it
+                                */
+                               vm_page_purged_others++;
+                               continue;
+                       }
+                       if (p->busy) {
+                               /*
+                                * We can't reclaim a busy page but we can
+                                * make it more likely to be paged (it's not wired) to make
+                                * sure that it gets considered by
+                                * vm_pageout_scan() later.
+                                */
+                               if (VM_PAGE_PAGEABLE(p))
+                                       vm_page_deactivate(p);
+                               vm_page_purged_busy++;
+                               continue;
+                       }
 
 
-               if (kr == KERN_SUCCESS) {
+                       assert(VM_PAGE_OBJECT(p) != kernel_object);
 
                        /*
 
                        /*
-                        * We found the page
+                        * we can discard this page...
                         */
                         */
+                       if (p->pmapped == TRUE) {
+                               /*
+                                * unmap the page
+                                */
+                               pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(p), PMAP_OPTIONS_NOFLUSH | PMAP_OPTIONS_NOREFMOD, (void *)&pmap_flush_context_storage);
+                       }
+                       vm_page_purged_count++;
 
 
-                       return TRUE;
-               }
-       }
-
-       return FALSE;
-}
-
+                       break;
 
 
-/*
- * Deactivate the pages in the specified object and range.  If kill_page is set, also discard any
- * page modified state from the pmap.  Update the chunk_state as we go along.  The caller must specify
- * a size that is less than or equal to the CHUNK_SIZE.
- */
+               case REAP_TERMINATE:
+                       if (p->absent || p->private) {
+                               /*
+                                *      For private pages, VM_PAGE_FREE just
+                                *      leaves the page structure around for
+                                *      its owner to clean up.  For absent
+                                *      pages, the structure is returned to
+                                *      the appropriate pool.
+                                */
+                               break;
+                       }
+                       if (p->fictitious) {
+                               assert (VM_PAGE_GET_PHYS_PAGE(p) == vm_page_guard_addr);
+                               break;
+                       }
+                       if (!p->dirty && p->wpmapped)
+                               p->dirty = pmap_is_modified(VM_PAGE_GET_PHYS_PAGE(p));
 
 
-static void
-deactivate_pages_in_object(
-       vm_object_t             object,
-       vm_object_offset_t      offset,
-       vm_object_size_t        size,
-       boolean_t               kill_page,
-       boolean_t               reusable_page,
-#if !MACH_ASSERT
-       __unused
-#endif
-       boolean_t               all_reusable,
-       chunk_state_t           *chunk_state)
-{
-       vm_page_t       m;
-       int             p;
-       struct  dw      dw_array[DELAYED_WORK_LIMIT];
-       struct  dw      *dwp;
-       int             dw_count;
-       unsigned int    reusable = 0;
+                       if ((p->dirty || p->precious) && !p->error && object->alive) {
 
 
+                               assert(!object->internal);
+                               
+                               p->free_when_done = TRUE;
 
 
-       /*
-        * Examine each page in the chunk.  The variable 'p' is the page number relative to the start of the
-        * chunk.  Since this routine is called once for each level in the shadow chain, the chunk_state may
-        * have pages marked as having been processed already.  We stop the loop early if we find we've handled
-        * all the pages in the chunk.
-        */
+                               if (!p->laundry) {
+                                       vm_page_queues_remove(p, TRUE);
+                                       /*
+                                        * flush page... page will be freed
+                                        * upon completion of I/O
+                                        */
+                                       (void)vm_pageout_cluster(p, FALSE, FALSE);
+                               }
+                               vm_page_unlock_queues();
+                               /*
+                                * free the pages reclaimed so far
+                                */
+                               VM_OBJ_REAP_FREELIST(local_free_q,
+                                                    disconnect_on_release);
 
 
-       dwp = &dw_array[0];
-       dw_count = 0;
+                               vm_object_paging_wait(object, THREAD_UNINT);
 
 
-       for(p = 0; size && CHUNK_NOT_COMPLETE(*chunk_state); p++, size -= PAGE_SIZE_64, offset += PAGE_SIZE_64) {
+                               goto restart_after_sleep;
+                       }
+                       break;
 
 
+               case REAP_REAP:
+                       break;
+               }
+               vm_page_free_prepare_queues(p);
+               assert(p->pageq.next == 0 && p->pageq.prev == 0);
                /*
                /*
-                * If this offset has already been found and handled in a higher level object, then don't
-                * do anything with it in the current shadow object.
+                * Add this page to our list of reclaimed pages,
+                * to be freed later.
                 */
                 */
+               p->snext = local_free_q;
+               local_free_q = p;
+       }
+       vm_page_unlock_queues();
 
 
-               if (PAGE_ALREADY_HANDLED(*chunk_state, p))
-                       continue;
-       
-               /*
-                * See if the page at this offset is around.  First check to see if the page is resident,
-                * then if not, check the existence map or with the pager.
-                */
+       /*
+        * Free the remaining reclaimed pages
+        */
+       if (reap_type == REAP_PURGEABLE)
+               pmap_flush(&pmap_flush_context_storage);
 
 
-               if ((m = vm_page_lookup(object, offset)) != VM_PAGE_NULL) {
+       VM_OBJ_REAP_FREELIST(local_free_q,
+                            disconnect_on_release);
+}
 
 
-                       /*
-                        * We found a page we were looking for.  Mark it as "handled" now in the chunk_state
-                        * so that we won't bother looking for a page at this offset again if there are more
-                        * shadow objects.  Then deactivate the page.
-                        */
 
 
-                       MARK_PAGE_HANDLED(*chunk_state, p);
-       
-                       if (( !VM_PAGE_WIRED(m)) && (!m->private) && (!m->gobbled) && (!m->busy)) {
-                               int     clear_refmod;
-       
-                               assert(!m->laundry);
-       
-                               clear_refmod = VM_MEM_REFERENCED;
-                               dwp->dw_mask = DW_clear_reference;
+void
+vm_object_reap_async(
+       vm_object_t     object)
+{
+       vm_object_lock_assert_exclusive(object);
 
 
-                               if ((kill_page) && (object->internal)) {
-                                       m->precious = FALSE;
-                                       m->dirty = FALSE;
+       vm_object_reaper_lock_spin();
 
 
-                                       clear_refmod |= VM_MEM_MODIFIED;
-#if CONFIG_EMBEDDED
-                                       dwp->dw_mask |= DW_move_page;
-#endif
-#if    MACH_PAGEMAP
-                                       vm_external_state_clr(object->existence_map, offset);
-#endif /* MACH_PAGEMAP */
+       vm_object_reap_count_async++;
 
 
-                                       if (reusable_page && !m->reusable) {
-                                               assert(!all_reusable);
-                                               assert(!object->all_reusable);
-                                               m->reusable = TRUE;
-                                               object->reusable_page_count++;
-                                               assert(object->resident_page_count >= object->reusable_page_count);
-                                               reusable++;
-                                       }
-                               }
-                               pmap_clear_refmod(m->phys_page, clear_refmod);
+       /* enqueue the VM object... */
+       queue_enter(&vm_object_reaper_queue, object,
+                   vm_object_t, cached_list);
 
 
-                               if (!m->throttled && !(reusable_page || all_reusable))
-                                       dwp->dw_mask |= DW_move_page;
-                               /*
-                                * dw_do_work may need to drop the object lock
-                                * if it does, we need the pages its looking at to
-                                * be held stable via the busy bit.
-                                */
-                               m->busy = TRUE;
-                               dwp->dw_mask |= (DW_clear_busy | DW_PAGE_WAKEUP);
+       vm_object_reaper_unlock();
 
 
-                               dwp->dw_m = m;
-                               dwp++;
-                               dw_count++;
+       /* ... and wake up the reaper thread */
+       thread_wakeup((event_t) &vm_object_reaper_queue);
+}
 
 
-                               if (dw_count >= DELAYED_WORK_LIMIT) {
-                                       if (reusable) {
-                                               OSAddAtomic(reusable,
-                                                           &vm_page_stats_reusable.reusable_count);
-                                               vm_page_stats_reusable.reusable += reusable;
-                                               reusable = 0;
-                                       }
-                                       dw_do_work(object, &dw_array[0], dw_count);
 
 
-                                       dwp = &dw_array[0];
-                                       dw_count = 0;
-                               }
-                       }
+void
+vm_object_reaper_thread(void)
+{
+       vm_object_t     object, shadow_object;
 
 
-               } else {
+       vm_object_reaper_lock_spin();
 
 
-                       /*
-                        * The page at this offset isn't memory resident, check to see if it's
-                        * been paged out.  If so, mark it as handled so we don't bother looking
-                        * for it in the shadow chain.
-                        */
+       while (!queue_empty(&vm_object_reaper_queue)) {
+               queue_remove_first(&vm_object_reaper_queue,
+                                  object,
+                                  vm_object_t,
+                                  cached_list);
 
 
-                       if (page_is_paged_out(object, offset)) {
-                               MARK_PAGE_HANDLED(*chunk_state, p);
+               vm_object_reaper_unlock();
+               vm_object_lock(object);
 
 
-                               /*
-                                * If we're killing a non-resident page, then clear the page in the existence 
-                                * map so we don't bother paging it back in if it's touched again in the future.
-                                */
+               assert(object->terminating);
+               assert(!object->alive);
+               
+               /*
+                * The pageout daemon might be playing with our pages.
+                * Now that the object is dead, it won't touch any more
+                * pages, but some pages might already be on their way out.
+                * Hence, we wait until the active paging activities have
+                * ceased before we break the association with the pager
+                * itself.
+                */
+               while (object->paging_in_progress != 0 ||
+                       object->activity_in_progress != 0) {
+                       vm_object_wait(object,
+                                      VM_OBJECT_EVENT_PAGING_IN_PROGRESS,
+                                      THREAD_UNINT);
+                       vm_object_lock(object);
+               }
 
 
-                               if ((kill_page) && (object->internal)) {
-#if    MACH_PAGEMAP
-                                       vm_external_state_clr(object->existence_map, offset);
-#endif /* MACH_PAGEMAP */
-                               }
-                       }
+               shadow_object =
+                       object->pageout ? VM_OBJECT_NULL : object->shadow;
+
+               vm_object_reap(object);
+               /* cache is unlocked and object is no longer valid */
+               object = VM_OBJECT_NULL;
+
+               if (shadow_object != VM_OBJECT_NULL) {
+                       /*
+                        * Drop the reference "object" was holding on
+                        * its shadow object.
+                        */
+                       vm_object_deallocate(shadow_object);
+                       shadow_object = VM_OBJECT_NULL;
                }
                }
+               vm_object_reaper_lock_spin();
        }
 
        }
 
-       if (reusable) {
-               OSAddAtomic(reusable, &vm_page_stats_reusable.reusable_count);
-               vm_page_stats_reusable.reusable += reusable;    
-               reusable = 0;
-       }
-               
-       if (dw_count)
-               dw_do_work(object, &dw_array[0], dw_count);
-}
+       /* wait for more work... */
+       assert_wait((event_t) &vm_object_reaper_queue, THREAD_UNINT);
+
+       vm_object_reaper_unlock();
 
 
+       thread_block((thread_continue_t) vm_object_reaper_thread);
+       /*NOTREACHED*/
+}
 
 /*
 
 /*
- * Deactive a "chunk" of the given range of the object starting at offset.  A "chunk"
- * will always be less than or equal to the given size.  The total range is divided up
- * into chunks for efficiency and performance related to the locks and handling the shadow
- * chain.  This routine returns how much of the given "size" it actually processed.  It's
- * up to the caler to loop and keep calling this routine until the entire range they want
- * to process has been done.
+ *     Routine:        vm_object_pager_wakeup
+ *     Purpose:        Wake up anyone waiting for termination of a pager.
  */
 
  */
 
-static vm_object_size_t
-deactivate_a_chunk(
-       vm_object_t             orig_object,
-       vm_object_offset_t      offset,
-       vm_object_size_t        size,
-       boolean_t               kill_page,
-       boolean_t               reusable_page,
-       boolean_t               all_reusable)
+static void
+vm_object_pager_wakeup(
+       memory_object_t pager)
 {
 {
-       vm_object_t             object;
-       vm_object_t             tmp_object;
-       vm_object_size_t        length;
-       chunk_state_t           chunk_state;
-
+       vm_object_hash_entry_t  entry;
+       boolean_t               waiting = FALSE;
+       lck_mtx_t               *lck;
 
        /*
 
        /*
-        * Get set to do a chunk.  We'll do up to CHUNK_SIZE, but no more than the
-        * remaining size the caller asked for.
+        *      If anyone was waiting for the memory_object_terminate
+        *      to be queued, wake them up now.
         */
         */
+       lck = vm_object_hash_lock_spin(pager);
+       entry = vm_object_hash_lookup(pager, TRUE);
+       if (entry != VM_OBJECT_HASH_ENTRY_NULL)
+               waiting = entry->waiting;
+       vm_object_hash_unlock(lck);
 
 
-       length = MIN(size, CHUNK_SIZE);
-
-       /*
-        * The chunk_state keeps track of which pages we've already processed if there's
-        * a shadow chain on this object.  At this point, we haven't done anything with this
-        * range of pages yet, so initialize the state to indicate no pages processed yet.
-        */
+       if (entry != VM_OBJECT_HASH_ENTRY_NULL) {
+               if (waiting)
+                       thread_wakeup((event_t) pager);
+               vm_object_hash_entry_free(entry);
+       }
+}
 
 
-       CHUNK_INIT(chunk_state, length);
-       object = orig_object;
+/*
+ *     Routine:        vm_object_release_pager
+ *     Purpose:        Terminate the pager and, upon completion,
+ *                     release our last reference to it.
+ *                     just like memory_object_terminate, except
+ *                     that we wake up anyone blocked in vm_object_enter
+ *                     waiting for termination message to be queued
+ *                     before calling memory_object_init.
+ */
+static void
+vm_object_release_pager(
+       memory_object_t pager,
+       boolean_t       hashed)
+{
 
        /*
 
        /*
-        * Start at the top level object and iterate around the loop once for each object
-        * in the shadow chain.  We stop processing early if we've already found all the pages
-        * in the range.  Otherwise we stop when we run out of shadow objects.
+        *      Terminate the pager.
         */
 
         */
 
-       while (object && CHUNK_NOT_COMPLETE(chunk_state)) {
-               vm_object_paging_begin(object);
-
-               deactivate_pages_in_object(object, offset, length, kill_page, reusable_page, all_reusable, &chunk_state);
-
-               vm_object_paging_end(object);
+       (void) memory_object_terminate(pager);
 
 
+       if (hashed == TRUE) {
                /*
                /*
-                * We've finished with this object, see if there's a shadow object.  If
-                * there is, update the offset and lock the new object.  We also turn off
-                * kill_page at this point since we only kill pages in the top most object.
+                *      Wakeup anyone waiting for this terminate
+                *      and remove the entry from the hash
                 */
                 */
-
-               tmp_object = object->shadow;
-
-               if (tmp_object) {
-                       kill_page = FALSE;
-                       reusable_page = FALSE;
-                       all_reusable = FALSE;
-                       offset += object->shadow_offset;
-                       vm_object_lock(tmp_object);
-               }
-
-               if (object != orig_object)
-                       vm_object_unlock(object);
-
-               object = tmp_object;
+               vm_object_pager_wakeup(pager);
        }
        }
-
-       if (object && object != orig_object)
-               vm_object_unlock(object);
-
-       return length;
+       /*
+        *      Release reference to pager.
+        */
+       memory_object_deallocate(pager);
 }
 
 }
 
-
-
 /*
 /*
- * Move any resident pages in the specified range to the inactive queue.  If kill_page is set,
- * we also clear the modified status of the page and "forget" any changes that have been made
- * to the page.
+ *     Routine:        vm_object_destroy
+ *     Purpose:
+ *             Shut down a VM object, despite the
+ *             presence of address map (or other) references
+ *             to the vm_object.
  */
  */
-
-__private_extern__ void
-vm_object_deactivate_pages(
+kern_return_t
+vm_object_destroy(
        vm_object_t             object,
        vm_object_t             object,
-       vm_object_offset_t      offset,
-       vm_object_size_t        size,
-       boolean_t               kill_page,
-       boolean_t               reusable_page)
+       __unused kern_return_t          reason)
 {
 {
-       vm_object_size_t        length;
-       boolean_t               all_reusable;
+       memory_object_t         old_pager;
+
+       if (object == VM_OBJECT_NULL)
+               return(KERN_SUCCESS);
 
        /*
 
        /*
-        * We break the range up into chunks and do one chunk at a time.  This is for
-        * efficiency and performance while handling the shadow chains and the locks.   
-        * The deactivate_a_chunk() function returns how much of the range it processed.
-        * We keep calling this routine until the given size is exhausted.
+        *      Remove the pager association immediately.
+        *
+        *      This will prevent the memory manager from further
+        *      meddling.  [If it wanted to flush data or make
+        *      other changes, it should have done so before performing
+        *      the destroy call.]
         */
 
         */
 
+       vm_object_lock(object);
+       object->can_persist = FALSE;
+       object->named = FALSE;
+       object->alive = FALSE;
 
 
-       all_reusable = FALSE;
-       if (reusable_page &&
-           object->size != 0 &&
-           object->size == size &&
-           object->reusable_page_count == 0) {
-               all_reusable = TRUE;
-               reusable_page = FALSE;
+       if (object->hashed) {
+               lck_mtx_t       *lck;
+               /*
+                *      Rip out the pager from the vm_object now...
+                */
+               lck = vm_object_hash_lock_spin(object->pager);
+               vm_object_remove(object);
+               vm_object_hash_unlock(lck);
        }
        }
+       old_pager = object->pager;
+       object->pager = MEMORY_OBJECT_NULL;
+       if (old_pager != MEMORY_OBJECT_NULL)
+               memory_object_control_disable(object->pager_control);
 
 
-       while (size) {
-               length = deactivate_a_chunk(object, offset, size, kill_page, reusable_page, all_reusable);
+       /*
+        * Wait for the existing paging activity (that got
+        * through before we nulled out the pager) to subside.
+        */
 
 
-               size -= length;
-               offset += length;
-       }
+       vm_object_paging_wait(object, THREAD_UNINT);
+       vm_object_unlock(object);
 
 
-       if (all_reusable) {
-               if (!object->all_reusable) {
-                       unsigned int reusable;
+       /*
+        *      Terminate the object now.
+        */
+       if (old_pager != MEMORY_OBJECT_NULL) {
+               vm_object_release_pager(old_pager, object->hashed);
+
+               /* 
+                * JMM - Release the caller's reference.  This assumes the
+                * caller had a reference to release, which is a big (but
+                * currently valid) assumption if this is driven from the
+                * vnode pager (it is holding a named reference when making
+                * this call)..
+                */
+               vm_object_deallocate(object);
 
 
-                       object->all_reusable = TRUE;
-                       assert(object->reusable_page_count == 0);
-                       /* update global stats */
-                       reusable = object->resident_page_count;
-                       OSAddAtomic(reusable,
-                                   &vm_page_stats_reusable.reusable_count);
-                       vm_page_stats_reusable.reusable += reusable;
-                       vm_page_stats_reusable.all_reusable_calls++;
-               }
-       } else if (reusable_page) {
-               vm_page_stats_reusable.partial_reusable_calls++;
        }
        }
+       return(KERN_SUCCESS);
 }
 
 }
 
-void
-vm_object_reuse_pages(
-       vm_object_t             object,
-       vm_object_offset_t      start_offset,
-       vm_object_offset_t      end_offset,
-       boolean_t               allow_partial_reuse)
-{
-       vm_object_offset_t      cur_offset;
-       vm_page_t               m;
-       unsigned int            reused, reusable;
-
-#define VM_OBJECT_REUSE_PAGE(object, m, reused)                                \
-       MACRO_BEGIN                                                     \
-               if ((m) != VM_PAGE_NULL &&                              \
-                   (m)->reusable) {                                    \
-                       assert((object)->reusable_page_count <=         \
-                              (object)->resident_page_count);          \
-                       assert((object)->reusable_page_count > 0);      \
-                       (object)->reusable_page_count--;                \
-                       (m)->reusable = FALSE;                          \
-                       (reused)++;                                     \
-               }                                                       \
-       MACRO_END
 
 
-       reused = 0;
-       reusable = 0;
+#if VM_OBJECT_CACHE
 
 
-       vm_object_lock_assert_exclusive(object);
+#define VM_OBJ_DEACT_ALL_STATS DEBUG
+#if VM_OBJ_DEACT_ALL_STATS
+uint32_t vm_object_deactivate_all_pages_batches = 0;
+uint32_t vm_object_deactivate_all_pages_pages = 0;
+#endif /* VM_OBJ_DEACT_ALL_STATS */
+/*
+ *     vm_object_deactivate_all_pages
+ *
+ *     Deactivate all pages in the specified object.  (Keep its pages
+ *     in memory even though it is no longer referenced.)
+ *
+ *     The object must be locked.
+ */
+static void
+vm_object_deactivate_all_pages(
+       vm_object_t     object)
+{
+       vm_page_t               p;
+       int                     loop_count;
+#if VM_OBJ_DEACT_ALL_STATS
+       int                     pages_count;
+#endif /* VM_OBJ_DEACT_ALL_STATS */
+#define V_O_D_A_P_MAX_BATCH    256
 
 
-       if (object->all_reusable) {
-               assert(object->reusable_page_count == 0);
-               object->all_reusable = FALSE;
-               if (end_offset - start_offset == object->size ||
-                   !allow_partial_reuse) {
-                       vm_page_stats_reusable.all_reuse_calls++;
-                       reused = object->resident_page_count;
-               } else {
-                       vm_page_stats_reusable.partial_reuse_calls++;
-                       queue_iterate(&object->memq, m, vm_page_t, listq) {
-                               if (m->offset < start_offset ||
-                                   m->offset >= end_offset) {
-                                       m->reusable = TRUE;
-                                       object->reusable_page_count++;
-                                       assert(object->resident_page_count >= object->reusable_page_count);
-                                       continue;
-                               } else {
-                                       assert(!m->reusable);
-                                       reused++;
-                               }
-                       }
-               }
-       } else if (object->resident_page_count >
-                  ((end_offset - start_offset) >> PAGE_SHIFT)) {
-               vm_page_stats_reusable.partial_reuse_calls++;
-               for (cur_offset = start_offset;
-                    cur_offset < end_offset;
-                    cur_offset += PAGE_SIZE_64) {
-                       if (object->reusable_page_count == 0) {
-                               break;
-                       }
-                       m = vm_page_lookup(object, cur_offset);
-                       VM_OBJECT_REUSE_PAGE(object, m, reused);
+       loop_count = BATCH_LIMIT(V_O_D_A_P_MAX_BATCH);
+#if VM_OBJ_DEACT_ALL_STATS
+       pages_count = 0;
+#endif /* VM_OBJ_DEACT_ALL_STATS */
+       vm_page_lock_queues();
+       vm_page_queue_iterate(&object->memq, p, vm_page_t, listq) {
+               if (--loop_count == 0) {
+#if VM_OBJ_DEACT_ALL_STATS
+                       hw_atomic_add(&vm_object_deactivate_all_pages_batches,
+                                     1);
+                       hw_atomic_add(&vm_object_deactivate_all_pages_pages,
+                                     pages_count);
+                       pages_count = 0;
+#endif /* VM_OBJ_DEACT_ALL_STATS */
+                       lck_mtx_yield(&vm_page_queue_lock);
+                       loop_count = BATCH_LIMIT(V_O_D_A_P_MAX_BATCH);
                }
                }
-       } else {
-               vm_page_stats_reusable.partial_reuse_calls++;
-               queue_iterate(&object->memq, m, vm_page_t, listq) {
-                       if (object->reusable_page_count == 0) {
-                               break;
-                       }
-                       if (m->offset < start_offset ||
-                           m->offset >= end_offset) {
-                               continue;
-                       }
-                       VM_OBJECT_REUSE_PAGE(object, m, reused);
+               if (!p->busy && (p->vm_page_q_state != VM_PAGE_ON_THROTTLED_Q)) {
+#if VM_OBJ_DEACT_ALL_STATS
+                       pages_count++;
+#endif /* VM_OBJ_DEACT_ALL_STATS */
+                       vm_page_deactivate(p);
                }
        }
                }
        }
-
-       /* update global stats */
-       OSAddAtomic(reusable-reused, &vm_page_stats_reusable.reusable_count);
-       vm_page_stats_reusable.reused += reused;
-       vm_page_stats_reusable.reusable += reusable;
+#if VM_OBJ_DEACT_ALL_STATS
+       if (pages_count) {
+               hw_atomic_add(&vm_object_deactivate_all_pages_batches, 1);
+               hw_atomic_add(&vm_object_deactivate_all_pages_pages,
+                             pages_count);
+               pages_count = 0;
+       }
+#endif /* VM_OBJ_DEACT_ALL_STATS */
+       vm_page_unlock_queues();
 }
 }
+#endif /* VM_OBJECT_CACHE */
+
+
 
 /*
 
 /*
- *     Routine:        vm_object_pmap_protect
- *
- *     Purpose:
- *             Reduces the permission for all physical
- *             pages in the specified object range.
- *
- *             If removing write permission only, it is
- *             sufficient to protect only the pages in
- *             the top-level object; only those pages may
- *             have write permission.
- *
- *             If removing all access, we must follow the
- *             shadow chain from the top-level object to
- *             remove access to all pages in shadowed objects.
+ * The "chunk" macros are used by routines below when looking for pages to deactivate.  These
+ * exist because of the need to handle shadow chains.  When deactivating pages, we only
+ * want to deactive the ones at the top most level in the object chain.  In order to do
+ * this efficiently, the specified address range is divided up into "chunks" and we use
+ * a bit map to keep track of which pages have already been processed as we descend down
+ * the shadow chain.  These chunk macros hide the details of the bit map implementation
+ * as much as we can.
  *
  *
- *             The object must *not* be locked.  The object must
- *             be temporary/internal.  
+ * For convenience, we use a 64-bit data type as the bit map, and therefore a chunk is
+ * set to 64 pages.  The bit map is indexed from the low-order end, so that the lowest
+ * order bit represents page 0 in the current range and highest order bit represents
+ * page 63.
  *
  *
- *              If pmap is not NULL, this routine assumes that
- *              the only mappings for the pages are in that
- *              pmap.
+ * For further convenience, we also use negative logic for the page state in the bit map.
+ * The bit is set to 1 to indicate it has not yet been seen, and to 0 to indicate it has
+ * been processed.  This way we can simply test the 64-bit long word to see if it's zero
+ * to easily tell if the whole range has been processed.  Therefore, the bit map starts
+ * out with all the bits set.  The macros below hide all these details from the caller.
  */
 
  */
 
-__private_extern__ void
-vm_object_pmap_protect(
-       register vm_object_t            object,
-       register vm_object_offset_t     offset,
-       vm_object_size_t                size,
-       pmap_t                          pmap,
-       vm_map_offset_t                 pmap_start,
-       vm_prot_t                       prot)
-{
-       if (object == VM_OBJECT_NULL)
-           return;
-       size = vm_object_round_page(size);
-       offset = vm_object_trunc_page(offset);
+#define PAGES_IN_A_CHUNK       64      /* The number of pages in the chunk must */
+                                       /* be the same as the number of bits in  */
+                                       /* the chunk_state_t type. We use 64     */
+                                       /* just for convenience.                 */
 
 
-       vm_object_lock(object);
+#define CHUNK_SIZE     (PAGES_IN_A_CHUNK * PAGE_SIZE_64)       /* Size of a chunk in bytes */
 
 
-       if (object->phys_contiguous) {
-               if (pmap != NULL) {
-                       vm_object_unlock(object);
-                       pmap_protect(pmap, pmap_start, pmap_start + size, prot);
-               } else {
-                       vm_object_offset_t phys_start, phys_end, phys_addr;
+typedef uint64_t       chunk_state_t;
 
 
-                       phys_start = object->shadow_offset + offset;
-                       phys_end = phys_start + size;
-                       assert(phys_start <= phys_end);
-                       assert(phys_end <= object->shadow_offset + object->size);
-                       vm_object_unlock(object);
+/*
+ * The bit map uses negative logic, so we start out with all 64 bits set to indicate
+ * that no pages have been processed yet.  Also, if len is less than the full CHUNK_SIZE,
+ * then we mark pages beyond the len as having been "processed" so that we don't waste time
+ * looking at pages in that range.  This can save us from unnecessarily chasing down the 
+ * shadow chain.
+ */
 
 
-                       for (phys_addr = phys_start;
-                            phys_addr < phys_end;
-                            phys_addr += PAGE_SIZE_64) {
-                               pmap_page_protect((ppnum_t) (phys_addr >> PAGE_SHIFT), prot);
-                       }
-               }
-               return;
-       }
+#define CHUNK_INIT(c, len)                                             \
+       MACRO_BEGIN                                                     \
+       uint64_t p;                                                     \
+                                                                       \
+       (c) = 0xffffffffffffffffLL;                                     \
+                                                                       \
+       for (p = (len) / PAGE_SIZE_64; p < PAGES_IN_A_CHUNK; p++)       \
+               MARK_PAGE_HANDLED(c, p);                                \
+       MACRO_END
 
 
-       assert(object->internal);
 
 
-       while (TRUE) {
-          if (ptoa_64(object->resident_page_count) > size/2 && pmap != PMAP_NULL) {
-               vm_object_unlock(object);
-               pmap_protect(pmap, pmap_start, pmap_start + size, prot);
-               return;
-           }
+/*
+ * Return true if all pages in the chunk have not yet been processed.
+ */
 
 
-           /* if we are doing large ranges with respect to resident */
-           /* page count then we should interate over pages otherwise */
-           /* inverse page look-up will be faster */
-           if (ptoa_64(object->resident_page_count / 4) <  size) {
-               vm_page_t               p;
-               vm_object_offset_t      end;
+#define CHUNK_NOT_COMPLETE(c)  ((c) != 0)
 
 
-               end = offset + size;
+/*
+ * Return true if the page at offset 'p' in the bit map has already been handled
+ * while processing a higher level object in the shadow chain.
+ */
 
 
-               if (pmap != PMAP_NULL) {
-                 queue_iterate(&object->memq, p, vm_page_t, listq) {
-                   if (!p->fictitious &&
-                       (offset <= p->offset) && (p->offset < end)) {
-                       vm_map_offset_t start;
+#define PAGE_ALREADY_HANDLED(c, p)     (((c) & (1LL << (p))) == 0)
 
 
-                       start = pmap_start + p->offset - offset;
-                       pmap_protect(pmap, start, start + PAGE_SIZE_64, prot);
-                   }
-                 }
-               } else {
-                 queue_iterate(&object->memq, p, vm_page_t, listq) {
-                   if (!p->fictitious &&
-                       (offset <= p->offset) && (p->offset < end)) {
+/*
+ * Mark the page at offset 'p' in the bit map as having been processed.
+ */
 
 
-                       pmap_page_protect(p->phys_page, prot);
-                   }
-                 }
-               }
-          } else {
-               vm_page_t               p;
-               vm_object_offset_t      end;
-               vm_object_offset_t      target_off;
+#define MARK_PAGE_HANDLED(c, p) \
+MACRO_BEGIN \
+       (c) = (c) & ~(1LL << (p)); \
+MACRO_END
 
 
-               end = offset + size;
 
 
-               if (pmap != PMAP_NULL) {
-                       for(target_off = offset; 
-                           target_off < end;
-                           target_off += PAGE_SIZE) {
-                               p = vm_page_lookup(object, target_off);
-                               if (p != VM_PAGE_NULL) {
-                                       vm_object_offset_t start;
-                                       start = pmap_start + 
-                                               (p->offset - offset);
-                                       pmap_protect(pmap, start, 
-                                                    start + PAGE_SIZE, prot);
-                               }
-                       }
-               } else {
-                       for(target_off = offset; 
-                               target_off < end; target_off += PAGE_SIZE) {
-                               p = vm_page_lookup(object, target_off);
-                               if (p != VM_PAGE_NULL) {
-                                       pmap_page_protect(p->phys_page, prot);
-                               }
-                       }
-               }
-         }
+/*
+ * Return true if the page at the given offset has been paged out.  Object is
+ * locked upon entry and returned locked.
+ */
 
 
-           if (prot == VM_PROT_NONE) {
-               /*
-                * Must follow shadow chain to remove access
-                * to pages in shadowed objects.
-                */
-               register vm_object_t    next_object;
+static boolean_t
+page_is_paged_out(
+       vm_object_t             object,
+       vm_object_offset_t      offset)
+{
+       if (object->internal &&
+          object->alive &&
+          !object->terminating &&
+          object->pager_ready) {
 
 
-               next_object = object->shadow;
-               if (next_object != VM_OBJECT_NULL) {
-                   offset += object->shadow_offset;
-                   vm_object_lock(next_object);
-                   vm_object_unlock(object);
-                   object = next_object;
-               }
-               else {
-                   /*
-                    * End of chain - we are done.
-                    */
-                   break;
+               if (VM_COMPRESSOR_PAGER_STATE_GET(object, offset) 
+                   == VM_EXTERNAL_STATE_EXISTS) {
+                       return TRUE;
                }
                }
-           }
-           else {
-               /*
-                * Pages in shadowed objects may never have
-                * write permission - we may stop here.
-                */
-               break;
-           }
        }
        }
-
-       vm_object_unlock(object);
+       return FALSE;
 }
 
 }
 
+
+
 /*
 /*
- *     Routine:        vm_object_copy_slowly
- *
- *     Description:
- *             Copy the specified range of the source
- *             virtual memory object without using
- *             protection-based optimizations (such
- *             as copy-on-write).  The pages in the
- *             region are actually copied.
- *
- *     In/out conditions:
- *             The caller must hold a reference and a lock
- *             for the source virtual memory object.  The source
- *             object will be returned *unlocked*.
- *
- *     Results:
- *             If the copy is completed successfully, KERN_SUCCESS is
- *             returned.  If the caller asserted the interruptible
- *             argument, and an interruption occurred while waiting
- *             for a user-generated event, MACH_SEND_INTERRUPTED is
- *             returned.  Other values may be returned to indicate
- *             hard errors during the copy operation.
+ * madvise_free_debug
  *
  *
- *             A new virtual memory object is returned in a
- *             parameter (_result_object).  The contents of this
- *             new object, starting at a zero offset, are a copy
- *             of the source memory region.  In the event of
- *             an error, this parameter will contain the value
- *             VM_OBJECT_NULL.
+ * To help debug madvise(MADV_FREE*) mis-usage, this triggers a
+ * zero-fill as soon as a page is affected by a madvise(MADV_FREE*), to
+ * simulate the loss of the page's contents as if the page had been
+ * reclaimed and then re-faulted.
  */
  */
-__private_extern__ kern_return_t
-vm_object_copy_slowly(
-       register vm_object_t    src_object,
-       vm_object_offset_t      src_offset,
-       vm_object_size_t        size,
-       boolean_t               interruptible,
-       vm_object_t             *_result_object)        /* OUT */
-{
-       vm_object_t             new_object;
-       vm_object_offset_t      new_offset;
-
-       struct vm_object_fault_info fault_info;
+#if DEVELOPMENT || DEBUG
+int madvise_free_debug = 1;
+#else /* DEBUG */
+int madvise_free_debug = 0;
+#endif /* DEBUG */
 
 
-       XPR(XPR_VM_OBJECT, "v_o_c_slowly obj 0x%x off 0x%x size 0x%x\n",
-           src_object, src_offset, size, 0, 0);
+/*
+ * Deactivate the pages in the specified object and range.  If kill_page is set, also discard any
+ * page modified state from the pmap.  Update the chunk_state as we go along.  The caller must specify
+ * a size that is less than or equal to the CHUNK_SIZE.
+ */
 
 
-       if (size == 0) {
-               vm_object_unlock(src_object);
-               *_result_object = VM_OBJECT_NULL;
-               return(KERN_INVALID_ARGUMENT);
-       }
+static void
+deactivate_pages_in_object(
+       vm_object_t             object,
+       vm_object_offset_t      offset,
+       vm_object_size_t        size,
+       boolean_t               kill_page,
+       boolean_t               reusable_page,
+       boolean_t               all_reusable,
+       chunk_state_t           *chunk_state,
+       pmap_flush_context      *pfc,
+       struct pmap             *pmap,
+       vm_map_offset_t         pmap_offset)
+{
+       vm_page_t       m;
+       int             p;
+       struct vm_page_delayed_work     dw_array[DEFAULT_DELAYED_WORK_LIMIT];
+       struct vm_page_delayed_work     *dwp;
+       int             dw_count;
+       int             dw_limit;
+       unsigned int    reusable = 0;
 
        /*
 
        /*
-        *      Prevent destruction of the source object while we copy.
+        * Examine each page in the chunk.  The variable 'p' is the page number relative to the start of the
+        * chunk.  Since this routine is called once for each level in the shadow chain, the chunk_state may
+        * have pages marked as having been processed already.  We stop the loop early if we find we've handled
+        * all the pages in the chunk.
         */
 
         */
 
-       vm_object_reference_locked(src_object);
-       vm_object_unlock(src_object);
+       dwp = &dw_array[0];
+       dw_count = 0;
+       dw_limit = DELAYED_WORK_LIMIT(DEFAULT_DELAYED_WORK_LIMIT);
 
 
-       /*
-        *      Create a new object to hold the copied pages.
-        *      A few notes:
-        *              We fill the new object starting at offset 0,
-        *               regardless of the input offset.
-        *              We don't bother to lock the new object within
-        *               this routine, since we have the only reference.
-        */
+       for(p = 0; size && CHUNK_NOT_COMPLETE(*chunk_state); p++, size -= PAGE_SIZE_64, offset += PAGE_SIZE_64, pmap_offset += PAGE_SIZE_64) {
 
 
-       new_object = vm_object_allocate(size);
-       new_offset = 0;
+               /*
+                * If this offset has already been found and handled in a higher level object, then don't
+                * do anything with it in the current shadow object.
+                */
 
 
-       assert(size == trunc_page_64(size));    /* Will the loop terminate? */
+               if (PAGE_ALREADY_HANDLED(*chunk_state, p))
+                       continue;
+       
+               /*
+                * See if the page at this offset is around.  First check to see if the page is resident,
+                * then if not, check the existence map or with the pager.
+                */
 
 
-       fault_info.interruptible = interruptible;
-       fault_info.behavior  = VM_BEHAVIOR_SEQUENTIAL;
-       fault_info.user_tag  = 0;
-       fault_info.lo_offset = src_offset;
-       fault_info.hi_offset = src_offset + size;
-       fault_info.no_cache  = FALSE;
-       fault_info.stealth = TRUE;
+               if ((m = vm_page_lookup(object, offset)) != VM_PAGE_NULL) {
 
 
-       for ( ;
-           size != 0 ;
-           src_offset += PAGE_SIZE_64, 
-                       new_offset += PAGE_SIZE_64, size -= PAGE_SIZE_64
-           ) {
-               vm_page_t       new_page;
-               vm_fault_return_t result;
+                       /*
+                        * We found a page we were looking for.  Mark it as "handled" now in the chunk_state
+                        * so that we won't bother looking for a page at this offset again if there are more
+                        * shadow objects.  Then deactivate the page.
+                        */
 
 
-               vm_object_lock(new_object);
+                       MARK_PAGE_HANDLED(*chunk_state, p);
+       
+                       if (( !VM_PAGE_WIRED(m)) && (!m->private) && (!m->gobbled) && (!m->busy) && (!m->laundry)) {
+                               int     clear_refmod;
+                               int     pmap_options;
+       
+                               dwp->dw_mask = 0;
 
 
-               while ((new_page = vm_page_alloc(new_object, new_offset))
-                               == VM_PAGE_NULL) {
+                               pmap_options = 0;
+                               clear_refmod = VM_MEM_REFERENCED;
+                               dwp->dw_mask |= DW_clear_reference;
 
 
-                       vm_object_unlock(new_object);
+                               if ((kill_page) && (object->internal)) {
+                                       if (madvise_free_debug) {
+                                               /*
+                                                * zero-fill the page now
+                                                * to simulate it being
+                                                * reclaimed and re-faulted.
+                                                */
+                                               pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(m));
+                                       }
+                                       m->precious = FALSE;
+                                       m->dirty = FALSE;
 
 
-                       if (!vm_page_wait(interruptible)) {
-                               vm_object_deallocate(new_object);
-                               vm_object_deallocate(src_object);
-                               *_result_object = VM_OBJECT_NULL;
-                               return(MACH_SEND_INTERRUPTED);
-                       }
-                       vm_object_lock(new_object);
-               }
-               vm_object_unlock(new_object);
+                                       clear_refmod |= VM_MEM_MODIFIED;
+                                       if (m->vm_page_q_state == VM_PAGE_ON_THROTTLED_Q) {
+                                               /*
+                                                * This page is now clean and
+                                                * reclaimable.  Move it out
+                                                * of the throttled queue, so
+                                                * that vm_pageout_scan() can
+                                                * find it.
+                                                */
+                                               dwp->dw_mask |= DW_move_page;
+                                       }
 
 
-               do {
-                       vm_prot_t       prot = VM_PROT_READ;
-                       vm_page_t       _result_page;
-                       vm_page_t       top_page;
-                       register
-                       vm_page_t       result_page;
-                       kern_return_t   error_code;
+                                       VM_COMPRESSOR_PAGER_STATE_CLR(object, offset);
 
 
-                       vm_object_lock(src_object);
-                       vm_object_paging_begin(src_object);
+                                       if (reusable_page && !m->reusable) {
+                                               assert(!all_reusable);
+                                               assert(!object->all_reusable);
+                                               m->reusable = TRUE;
+                                               object->reusable_page_count++;
+                                               assert(object->resident_page_count >= object->reusable_page_count);
+                                               reusable++;
+                                               /*
+                                                * Tell pmap this page is now
+                                                * "reusable" (to update pmap
+                                                * stats for all mappings).
+                                                */
+                                               pmap_options |= PMAP_OPTIONS_SET_REUSABLE;
+                                       }
+                               }
+                               pmap_options |= PMAP_OPTIONS_NOFLUSH;
+                               pmap_clear_refmod_options(VM_PAGE_GET_PHYS_PAGE(m),
+                                                         clear_refmod,
+                                                         pmap_options,
+                                                         (void *)pfc);
 
 
-                       if (size > (vm_size_t) -1) {
-                               /* 32-bit overflow */
-                               fault_info.cluster_size = (vm_size_t) (0 - PAGE_SIZE);
-                       } else {
-                               fault_info.cluster_size = (vm_size_t) size;
-                               assert(fault_info.cluster_size == size);
-                       }
+                               if ((m->vm_page_q_state != VM_PAGE_ON_THROTTLED_Q) && !(reusable_page || all_reusable))
+                                       dwp->dw_mask |= DW_move_page;
+                               
+                               if (dwp->dw_mask)
+                                       VM_PAGE_ADD_DELAYED_WORK(dwp, m,
+                                                                dw_count);
 
 
-                       XPR(XPR_VM_FAULT,"vm_object_copy_slowly -> vm_fault_page",0,0,0,0,0);
-                       result = vm_fault_page(src_object, src_offset,
-                               VM_PROT_READ, FALSE,
-                               &prot, &_result_page, &top_page,
-                               (int *)0,
-                               &error_code, FALSE, FALSE, &fault_info);
+                               if (dw_count >= dw_limit) {
+                                       if (reusable) {
+                                               OSAddAtomic(reusable,
+                                                           &vm_page_stats_reusable.reusable_count);
+                                               vm_page_stats_reusable.reusable += reusable;
+                                               reusable = 0;
+                                       }
+                                       vm_page_do_delayed_work(object, VM_KERN_MEMORY_NONE, &dw_array[0], dw_count);
 
 
-                       switch(result) {
-                       case VM_FAULT_SUCCESS:
-                               result_page = _result_page;
+                                       dwp = &dw_array[0];
+                                       dw_count = 0;
+                               }
+                       }
 
 
-                               /*
-                                *      We don't need to hold the object
-                                *      lock -- the busy page will be enough.
-                                *      [We don't care about picking up any
-                                *      new modifications.]
-                                *
-                                *      Copy the page to the new object.
-                                *
-                                *      POLICY DECISION:
-                                *              If result_page is clean,
-                                *              we could steal it instead
-                                *              of copying.
-                                */
+               } else {
 
 
-                               vm_object_unlock(result_page->object);
-                               vm_page_copy(result_page, new_page);
+                       /*
+                        * The page at this offset isn't memory resident, check to see if it's
+                        * been paged out.  If so, mark it as handled so we don't bother looking
+                        * for it in the shadow chain.
+                        */
+
+                       if (page_is_paged_out(object, offset)) {
+                               MARK_PAGE_HANDLED(*chunk_state, p);
 
                                /*
 
                                /*
-                                *      Let go of both pages (make them
-                                *      not busy, perform wakeup, activate).
+                                * If we're killing a non-resident page, then clear the page in the existence 
+                                * map so we don't bother paging it back in if it's touched again in the future.
                                 */
                                 */
-                               vm_object_lock(new_object);
-                               new_page->dirty = TRUE;
-                               PAGE_WAKEUP_DONE(new_page);
-                               vm_object_unlock(new_object);
-
-                               vm_object_lock(result_page->object);
-                               PAGE_WAKEUP_DONE(result_page);
 
 
-                               vm_page_lockspin_queues();
-                               if (!result_page->active &&
-                                   !result_page->inactive &&
-                                   !result_page->throttled)
-                                       vm_page_activate(result_page);
-                               vm_page_activate(new_page);
-                               vm_page_unlock_queues();
+                               if ((kill_page) && (object->internal)) {
 
 
-                               /*
-                                *      Release paging references and
-                                *      top-level placeholder page, if any.
-                                */
+                                       VM_COMPRESSOR_PAGER_STATE_CLR(object, offset);
 
 
-                               vm_fault_cleanup(result_page->object,
-                                                top_page);
-
-                               break;
-                               
-                       case VM_FAULT_RETRY:
-                               break;
+                                       if (pmap != PMAP_NULL) {
+                                               /*
+                                                * Tell pmap that this page
+                                                * is no longer mapped, to
+                                                * adjust the footprint ledger
+                                                * because this page is no
+                                                * longer compressed.
+                                                */
+                                               pmap_remove_options(
+                                                       pmap,
+                                                       pmap_offset,
+                                                       (pmap_offset +
+                                                        PAGE_SIZE),
+                                                       PMAP_OPTIONS_REMOVE);
+                                       }
+                               }
+                       }
+               }
+       }
 
 
-                       case VM_FAULT_FICTITIOUS_SHORTAGE:
-                               vm_page_more_fictitious();
-                               break;
+       if (reusable) {
+               OSAddAtomic(reusable, &vm_page_stats_reusable.reusable_count);
+               vm_page_stats_reusable.reusable += reusable;    
+               reusable = 0;
+       }
+               
+       if (dw_count)
+               vm_page_do_delayed_work(object, VM_KERN_MEMORY_NONE, &dw_array[0], dw_count);
+}
 
 
-                       case VM_FAULT_MEMORY_SHORTAGE:
-                               if (vm_page_wait(interruptible))
-                                       break;
-                               /* fall thru */
 
 
-                       case VM_FAULT_INTERRUPTED:
-                               vm_object_lock(new_object);
-                               VM_PAGE_FREE(new_page);
-                               vm_object_unlock(new_object);
-                                       
-                               vm_object_deallocate(new_object);
-                               vm_object_deallocate(src_object);
-                               *_result_object = VM_OBJECT_NULL;
-                               return(MACH_SEND_INTERRUPTED);
+/*
+ * Deactive a "chunk" of the given range of the object starting at offset.  A "chunk"
+ * will always be less than or equal to the given size.  The total range is divided up
+ * into chunks for efficiency and performance related to the locks and handling the shadow
+ * chain.  This routine returns how much of the given "size" it actually processed.  It's
+ * up to the caler to loop and keep calling this routine until the entire range they want
+ * to process has been done.
+ */
 
 
-                       case VM_FAULT_SUCCESS_NO_VM_PAGE:
-                               /* success but no VM page: fail */
-                               vm_object_paging_end(src_object);
-                               vm_object_unlock(src_object);
-                               /*FALLTHROUGH*/
-                       case VM_FAULT_MEMORY_ERROR:
-                               /*
-                                * A policy choice:
-                                *      (a) ignore pages that we can't
-                                *          copy
-                                *      (b) return the null object if
-                                *          any page fails [chosen]
-                                */
+static vm_object_size_t
+deactivate_a_chunk(
+       vm_object_t             orig_object,
+       vm_object_offset_t      offset,
+       vm_object_size_t        size,
+       boolean_t               kill_page,
+       boolean_t               reusable_page,
+       boolean_t               all_reusable,
+       pmap_flush_context      *pfc,
+       struct pmap             *pmap,
+       vm_map_offset_t         pmap_offset)
+{
+       vm_object_t             object;
+       vm_object_t             tmp_object;
+       vm_object_size_t        length;
+       chunk_state_t           chunk_state;
 
 
-                               vm_object_lock(new_object);
-                               VM_PAGE_FREE(new_page);
-                               vm_object_unlock(new_object);
 
 
-                               vm_object_deallocate(new_object);
-                               vm_object_deallocate(src_object);
-                               *_result_object = VM_OBJECT_NULL;
-                               return(error_code ? error_code:
-                                      KERN_MEMORY_ERROR);
+       /*
+        * Get set to do a chunk.  We'll do up to CHUNK_SIZE, but no more than the
+        * remaining size the caller asked for.
+        */
 
 
-                       default:
-                               panic("vm_object_copy_slowly: unexpected error"
-                                     " 0x%x from vm_fault_page()\n", result);
-                       }
-               } while (result != VM_FAULT_SUCCESS);
-       }
+       length = MIN(size, CHUNK_SIZE);
 
        /*
 
        /*
-        *      Lose the extra reference, and return our object.
+        * The chunk_state keeps track of which pages we've already processed if there's
+        * a shadow chain on this object.  At this point, we haven't done anything with this
+        * range of pages yet, so initialize the state to indicate no pages processed yet.
         */
         */
-       vm_object_deallocate(src_object);
-       *_result_object = new_object;
-       return(KERN_SUCCESS);
-}
-
-/*
- *     Routine:        vm_object_copy_quickly
- *
- *     Purpose:
- *             Copy the specified range of the source virtual
- *             memory object, if it can be done without waiting
- *             for user-generated events.
- *
- *     Results:
- *             If the copy is successful, the copy is returned in
- *             the arguments; otherwise, the arguments are not
- *             affected.
- *
- *     In/out conditions:
- *             The object should be unlocked on entry and exit.
- */
 
 
-/*ARGSUSED*/
-__private_extern__ boolean_t
-vm_object_copy_quickly(
-       vm_object_t             *_object,               /* INOUT */
-       __unused vm_object_offset_t     offset, /* IN */
-       __unused vm_object_size_t       size,   /* IN */
-       boolean_t               *_src_needs_copy,       /* OUT */
-       boolean_t               *_dst_needs_copy)       /* OUT */
-{
-       vm_object_t     object = *_object;
-       memory_object_copy_strategy_t copy_strategy;
+       CHUNK_INIT(chunk_state, length);
+       object = orig_object;
 
 
-       XPR(XPR_VM_OBJECT, "v_o_c_quickly obj 0x%x off 0x%x size 0x%x\n",
-           *_object, offset, size, 0, 0);
-       if (object == VM_OBJECT_NULL) {
-               *_src_needs_copy = FALSE;
-               *_dst_needs_copy = FALSE;
-               return(TRUE);
-       }
+       /*
+        * Start at the top level object and iterate around the loop once for each object
+        * in the shadow chain.  We stop processing early if we've already found all the pages
+        * in the range.  Otherwise we stop when we run out of shadow objects.
+        */
 
 
-       vm_object_lock(object);
+       while (object && CHUNK_NOT_COMPLETE(chunk_state)) {
+               vm_object_paging_begin(object);
 
 
-       copy_strategy = object->copy_strategy;
+               deactivate_pages_in_object(object, offset, length, kill_page, reusable_page, all_reusable, &chunk_state, pfc, pmap, pmap_offset);
 
 
-       switch (copy_strategy) {
-       case MEMORY_OBJECT_COPY_SYMMETRIC:
+               vm_object_paging_end(object);
 
                /*
 
                /*
-                *      Symmetric copy strategy.
-                *      Make another reference to the object.
-                *      Leave object/offset unchanged.
+                * We've finished with this object, see if there's a shadow object.  If
+                * there is, update the offset and lock the new object.  We also turn off
+                * kill_page at this point since we only kill pages in the top most object.
                 */
 
                 */
 
-               vm_object_reference_locked(object);
-               object->shadowed = TRUE;
-               vm_object_unlock(object);
+               tmp_object = object->shadow;
 
 
-               /*
-                *      Both source and destination must make
-                *      shadows, and the source must be made
-                *      read-only if not already.
-                */
+               if (tmp_object) {
+                       kill_page = FALSE;
+                       reusable_page = FALSE;
+                       all_reusable = FALSE;
+                       offset += object->vo_shadow_offset;
+                       vm_object_lock(tmp_object);
+               }
 
 
-               *_src_needs_copy = TRUE;
-               *_dst_needs_copy = TRUE;
+               if (object != orig_object)
+                       vm_object_unlock(object);
 
 
-               break;
+               object = tmp_object;
+       }
 
 
-       case MEMORY_OBJECT_COPY_DELAY:
-               vm_object_unlock(object);
-               return(FALSE);
+       if (object && object != orig_object)
+               vm_object_unlock(object);
 
 
-       default:
-               vm_object_unlock(object);
-               return(FALSE);
-       }
-       return(TRUE);
+       return length;
 }
 
 }
 
-static int copy_call_count = 0;
-static int copy_call_sleep_count = 0;
-static int copy_call_restart_count = 0;
+
 
 /*
 
 /*
- *     Routine:        vm_object_copy_call [internal]
- *
- *     Description:
- *             Copy the source object (src_object), using the
- *             user-managed copy algorithm.
- *
- *     In/out conditions:
- *             The source object must be locked on entry.  It
- *             will be *unlocked* on exit.
- *
- *     Results:
- *             If the copy is successful, KERN_SUCCESS is returned.
- *             A new object that represents the copied virtual
- *             memory is returned in a parameter (*_result_object).
- *             If the return value indicates an error, this parameter
- *             is not valid.
+ * Move any resident pages in the specified range to the inactive queue.  If kill_page is set,
+ * we also clear the modified status of the page and "forget" any changes that have been made
+ * to the page.
  */
  */
-static kern_return_t
-vm_object_copy_call(
-       vm_object_t             src_object,
-       vm_object_offset_t      src_offset,
+
+__private_extern__ void
+vm_object_deactivate_pages(
+       vm_object_t             object,
+       vm_object_offset_t      offset,
        vm_object_size_t        size,
        vm_object_size_t        size,
-       vm_object_t             *_result_object)        /* OUT */
+       boolean_t               kill_page,
+       boolean_t               reusable_page,
+       struct pmap             *pmap,
+       vm_map_offset_t         pmap_offset)
 {
 {
-       kern_return_t   kr;
-       vm_object_t     copy;
-       boolean_t       check_ready = FALSE;
-       uint32_t        try_failed_count = 0;
-
-       /*
-        *      If a copy is already in progress, wait and retry.
-        *
-        *      XXX
-        *      Consider making this call interruptable, as Mike
-        *      intended it to be.
-        *
-        *      XXXO
-        *      Need a counter or version or something to allow
-        *      us to use the copy that the currently requesting
-        *      thread is obtaining -- is it worth adding to the
-        *      vm object structure? Depends how common this case it.
-        */
-       copy_call_count++;
-       while (vm_object_wanted(src_object, VM_OBJECT_EVENT_COPY_CALL)) {
-               vm_object_sleep(src_object, VM_OBJECT_EVENT_COPY_CALL,
-                              THREAD_UNINT);
-               copy_call_restart_count++;
-       }
+       vm_object_size_t        length;
+       boolean_t               all_reusable;
+       pmap_flush_context      pmap_flush_context_storage;
 
        /*
 
        /*
-        *      Indicate (for the benefit of memory_object_create_copy)
-        *      that we want a copy for src_object. (Note that we cannot
-        *      do a real assert_wait before calling memory_object_copy,
-        *      so we simply set the flag.)
+        * We break the range up into chunks and do one chunk at a time.  This is for
+        * efficiency and performance while handling the shadow chains and the locks.   
+        * The deactivate_a_chunk() function returns how much of the range it processed.
+        * We keep calling this routine until the given size is exhausted.
         */
 
         */
 
-       vm_object_set_wanted(src_object, VM_OBJECT_EVENT_COPY_CALL);
-       vm_object_unlock(src_object);
 
 
+       all_reusable = FALSE;
+#if 11
        /*
        /*
-        *      Ask the memory manager to give us a memory object
-        *      which represents a copy of the src object.
-        *      The memory manager may give us a memory object
-        *      which we already have, or it may give us a
-        *      new memory object. This memory object will arrive
-        *      via memory_object_create_copy.
+        * For the sake of accurate "reusable" pmap stats, we need 
+        * to tell pmap about each page that is no longer "reusable",
+        * so we can't do the "all_reusable" optimization.
         */
         */
+#else
+       if (reusable_page &&
+           object->internal &&
+           object->vo_size != 0 &&
+           object->vo_size == size &&
+           object->reusable_page_count == 0) {
+               all_reusable = TRUE;
+               reusable_page = FALSE;
+       }
+#endif
 
 
-       kr = KERN_FAILURE;      /* XXX need to change memory_object.defs */
-       if (kr != KERN_SUCCESS) {
-               return kr;
+       if ((reusable_page || all_reusable) && object->all_reusable) {
+               /* This means MADV_FREE_REUSABLE has been called twice, which 
+                * is probably illegal. */
+               return;
        }
 
        }
 
-       /*
-        *      Wait for the copy to arrive.
-        */
-       vm_object_lock(src_object);
-       while (vm_object_wanted(src_object, VM_OBJECT_EVENT_COPY_CALL)) {
-               vm_object_sleep(src_object, VM_OBJECT_EVENT_COPY_CALL,
-                              THREAD_UNINT);
-               copy_call_sleep_count++;
+       pmap_flush_context_init(&pmap_flush_context_storage);
+
+       while (size) {
+               length = deactivate_a_chunk(object, offset, size, kill_page, reusable_page, all_reusable, &pmap_flush_context_storage, pmap, pmap_offset);
+
+               size -= length;
+               offset += length;
+               pmap_offset += length;
        }
        }
-Retry:
-       assert(src_object->copy != VM_OBJECT_NULL);
-       copy = src_object->copy;
-       if (!vm_object_lock_try(copy)) {
-               vm_object_unlock(src_object);
+       pmap_flush(&pmap_flush_context_storage);
 
 
-               try_failed_count++;
-               mutex_pause(try_failed_count);  /* wait a bit */
+       if (all_reusable) {
+               if (!object->all_reusable) {
+                       unsigned int reusable;
 
 
-               vm_object_lock(src_object);
-               goto Retry;
+                       object->all_reusable = TRUE;
+                       assert(object->reusable_page_count == 0);
+                       /* update global stats */
+                       reusable = object->resident_page_count;
+                       OSAddAtomic(reusable,
+                                   &vm_page_stats_reusable.reusable_count);
+                       vm_page_stats_reusable.reusable += reusable;
+                       vm_page_stats_reusable.all_reusable_calls++;
+               }
+       } else if (reusable_page) {
+               vm_page_stats_reusable.partial_reusable_calls++;
        }
        }
-       if (copy->size < src_offset+size)
-               copy->size = src_offset+size;
+}
 
 
-       if (!copy->pager_ready)
-               check_ready = TRUE;
+void
+vm_object_reuse_pages(
+       vm_object_t             object,
+       vm_object_offset_t      start_offset,
+       vm_object_offset_t      end_offset,
+       boolean_t               allow_partial_reuse)
+{
+       vm_object_offset_t      cur_offset;
+       vm_page_t               m;
+       unsigned int            reused, reusable;
 
 
-       /*
-        *      Return the copy.
-        */
-       *_result_object = copy;
-       vm_object_unlock(copy);
-       vm_object_unlock(src_object);
+#define VM_OBJECT_REUSE_PAGE(object, m, reused)                                \
+       MACRO_BEGIN                                                     \
+               if ((m) != VM_PAGE_NULL &&                              \
+                   (m)->reusable) {                                    \
+                       assert((object)->reusable_page_count <=         \
+                              (object)->resident_page_count);          \
+                       assert((object)->reusable_page_count > 0);      \
+                       (object)->reusable_page_count--;                \
+                       (m)->reusable = FALSE;                          \
+                       (reused)++;                                     \
+                       /*                                              \
+                        * Tell pmap that this page is no longer        \
+                        * "reusable", to update the "reusable" stats   \
+                        * for all the pmaps that have mapped this      \
+                        * page.                                        \
+                        */                                             \
+                       pmap_clear_refmod_options(VM_PAGE_GET_PHYS_PAGE((m)), \
+                                                 0, /* refmod */       \
+                                                 (PMAP_OPTIONS_CLEAR_REUSABLE \
+                                                  | PMAP_OPTIONS_NOFLUSH), \
+                                                 NULL);                \
+               }                                                       \
+       MACRO_END
 
 
-       /* Wait for the copy to be ready. */
-       if (check_ready == TRUE) {
-               vm_object_lock(copy);
-               while (!copy->pager_ready) {
-                       vm_object_sleep(copy, VM_OBJECT_EVENT_PAGER_READY, THREAD_UNINT);
+       reused = 0;
+       reusable = 0;
+
+       vm_object_lock_assert_exclusive(object);
+
+       if (object->all_reusable) {
+               panic("object %p all_reusable: can't update pmap stats\n",
+                     object);
+               assert(object->reusable_page_count == 0);
+               object->all_reusable = FALSE;
+               if (end_offset - start_offset == object->vo_size ||
+                   !allow_partial_reuse) {
+                       vm_page_stats_reusable.all_reuse_calls++;
+                       reused = object->resident_page_count;
+               } else {
+                       vm_page_stats_reusable.partial_reuse_calls++;
+                       vm_page_queue_iterate(&object->memq, m, vm_page_t, listq) {
+                               if (m->offset < start_offset ||
+                                   m->offset >= end_offset) {
+                                       m->reusable = TRUE;
+                                       object->reusable_page_count++;
+                                       assert(object->resident_page_count >= object->reusable_page_count);
+                                       continue;
+                               } else {
+                                       assert(!m->reusable);
+                                       reused++;
+                               }
+                       }
+               }
+       } else if (object->resident_page_count >
+                  ((end_offset - start_offset) >> PAGE_SHIFT)) {
+               vm_page_stats_reusable.partial_reuse_calls++;
+               for (cur_offset = start_offset;
+                    cur_offset < end_offset;
+                    cur_offset += PAGE_SIZE_64) {
+                       if (object->reusable_page_count == 0) {
+                               break;
+                       }
+                       m = vm_page_lookup(object, cur_offset);
+                       VM_OBJECT_REUSE_PAGE(object, m, reused);
+               }
+       } else {
+               vm_page_stats_reusable.partial_reuse_calls++;
+               vm_page_queue_iterate(&object->memq, m, vm_page_t, listq) {
+                       if (object->reusable_page_count == 0) {
+                               break;
+                       }
+                       if (m->offset < start_offset ||
+                           m->offset >= end_offset) {
+                               continue;
+                       }
+                       VM_OBJECT_REUSE_PAGE(object, m, reused);
                }
                }
-               vm_object_unlock(copy);
        }
 
        }
 
-       return KERN_SUCCESS;
+       /* update global stats */
+       OSAddAtomic(reusable-reused, &vm_page_stats_reusable.reusable_count);
+       vm_page_stats_reusable.reused += reused;
+       vm_page_stats_reusable.reusable += reusable;
 }
 
 }
 
-static int copy_delayed_lock_collisions = 0;
-static int copy_delayed_max_collisions = 0;
-static int copy_delayed_lock_contention = 0;
-static int copy_delayed_protect_iterate = 0;
-
 /*
 /*
- *     Routine:        vm_object_copy_delayed [internal]
+ *     Routine:        vm_object_pmap_protect
  *
  *
- *     Description:
- *             Copy the specified virtual memory object, using
- *             the asymmetric copy-on-write algorithm.
+ *     Purpose:
+ *             Reduces the permission for all physical
+ *             pages in the specified object range.
  *
  *
- *     In/out conditions:
- *             The src_object must be locked on entry.  It will be unlocked
- *             on exit - so the caller must also hold a reference to it.
+ *             If removing write permission only, it is
+ *             sufficient to protect only the pages in
+ *             the top-level object; only those pages may
+ *             have write permission.
  *
  *
- *             This routine will not block waiting for user-generated
- *             events.  It is not interruptible.
+ *             If removing all access, we must follow the
+ *             shadow chain from the top-level object to
+ *             remove access to all pages in shadowed objects.
+ *
+ *             The object must *not* be locked.  The object must
+ *             be temporary/internal.  
+ *
+ *              If pmap is not NULL, this routine assumes that
+ *              the only mappings for the pages are in that
+ *              pmap.
  */
  */
-__private_extern__ vm_object_t
-vm_object_copy_delayed(
-       vm_object_t             src_object,
-       vm_object_offset_t      src_offset,
-       vm_object_size_t        size,
-       boolean_t               src_object_shared)
+
+__private_extern__ void
+vm_object_pmap_protect(
+       vm_object_t                     object,
+       vm_object_offset_t              offset,
+       vm_object_size_t                size,
+       pmap_t                          pmap,
+       vm_map_offset_t                 pmap_start,
+       vm_prot_t                       prot)
 {
 {
-       vm_object_t             new_copy = VM_OBJECT_NULL;
-       vm_object_t             old_copy;
-       vm_page_t               p;
-       vm_object_size_t        copy_size = src_offset + size;
+       vm_object_pmap_protect_options(object, offset, size,
+                                      pmap, pmap_start, prot, 0);
+}
 
 
+__private_extern__ void
+vm_object_pmap_protect_options(
+       vm_object_t                     object,
+       vm_object_offset_t              offset,
+       vm_object_size_t                size,
+       pmap_t                          pmap,
+       vm_map_offset_t                 pmap_start,
+       vm_prot_t                       prot,
+       int                             options)
+{
+       pmap_flush_context      pmap_flush_context_storage;
+       boolean_t               delayed_pmap_flush = FALSE;
 
 
-       int collisions = 0;
-       /*
-        *      The user-level memory manager wants to see all of the changes
-        *      to this object, but it has promised not to make any changes on
-        *      its own.
-        *
-        *      Perform an asymmetric copy-on-write, as follows:
-        *              Create a new object, called a "copy object" to hold
-        *               pages modified by the new mapping  (i.e., the copy,
-        *               not the original mapping).
-        *              Record the original object as the backing object for
-        *               the copy object.  If the original mapping does not
-        *               change a page, it may be used read-only by the copy.
-        *              Record the copy object in the original object.
-        *               When the original mapping causes a page to be modified,
-        *               it must be copied to a new page that is "pushed" to
-        *               the copy object.
-        *              Mark the new mapping (the copy object) copy-on-write.
-        *               This makes the copy object itself read-only, allowing
-        *               it to be reused if the original mapping makes no
-        *               changes, and simplifying the synchronization required
-        *               in the "push" operation described above.
-        *
-        *      The copy-on-write is said to be assymetric because the original
-        *      object is *not* marked copy-on-write. A copied page is pushed
-        *      to the copy object, regardless which party attempted to modify
-        *      the page.
-        *
-        *      Repeated asymmetric copy operations may be done. If the
-        *      original object has not been changed since the last copy, its
-        *      copy object can be reused. Otherwise, a new copy object can be
-        *      inserted between the original object and its previous copy
-        *      object.  Since any copy object is read-only, this cannot affect
-        *      affect the contents of the previous copy object.
-        *
-        *      Note that a copy object is higher in the object tree than the
-        *      original object; therefore, use of the copy object recorded in
-        *      the original object must be done carefully, to avoid deadlock.
-        */
+       if (object == VM_OBJECT_NULL)
+               return;
+       size = vm_object_round_page(size);
+       offset = vm_object_trunc_page(offset);
 
 
- Retry:
-       /*
-        * Wait for paging in progress.
-        */
-       if (!src_object->true_share &&
-           (src_object->paging_in_progress != 0 ||
-            src_object->activity_in_progress != 0)) {
-               if (src_object_shared == TRUE) {
-                       vm_object_unlock(src_object);
-                       vm_object_lock(src_object);
-                       src_object_shared = FALSE;
-                       goto Retry;
-               }
-               vm_object_paging_wait(src_object, THREAD_UNINT);
-       }
-       /*
-        *      See whether we can reuse the result of a previous
-        *      copy operation.
-        */
+       vm_object_lock(object);
 
 
-       old_copy = src_object->copy;
-       if (old_copy != VM_OBJECT_NULL) {
-               int lock_granted;
+       if (object->phys_contiguous) {
+               if (pmap != NULL) {
+                       vm_object_unlock(object);
+                       pmap_protect_options(pmap,
+                                            pmap_start,
+                                            pmap_start + size,
+                                            prot,
+                                            options & ~PMAP_OPTIONS_NOFLUSH,
+                                            NULL);
+               } else {
+                       vm_object_offset_t phys_start, phys_end, phys_addr;
 
 
-               /*
-                *      Try to get the locks (out of order)
-                */
-               if (src_object_shared == TRUE)
-                       lock_granted = vm_object_lock_try_shared(old_copy);
-               else
-                       lock_granted = vm_object_lock_try(old_copy);
+                       phys_start = object->vo_shadow_offset + offset;
+                       phys_end = phys_start + size;
+                       assert(phys_start <= phys_end);
+                       assert(phys_end <= object->vo_shadow_offset + object->vo_size);
+                       vm_object_unlock(object);
 
 
-               if (!lock_granted) {
-                       vm_object_unlock(src_object);
+                       pmap_flush_context_init(&pmap_flush_context_storage);
+                       delayed_pmap_flush = FALSE;
 
 
-                       if (collisions++ == 0)
-                               copy_delayed_lock_contention++;
-                       mutex_pause(collisions);
+                       for (phys_addr = phys_start;
+                            phys_addr < phys_end;
+                            phys_addr += PAGE_SIZE_64) {
+                               pmap_page_protect_options(
+                                       (ppnum_t) (phys_addr >> PAGE_SHIFT),
+                                       prot,
+                                       options | PMAP_OPTIONS_NOFLUSH,
+                                       (void *)&pmap_flush_context_storage);
+                               delayed_pmap_flush = TRUE;
+                       }
+                       if (delayed_pmap_flush == TRUE)
+                               pmap_flush(&pmap_flush_context_storage);
+               }
+               return;
+       }
 
 
-                       /* Heisenberg Rules */
-                       copy_delayed_lock_collisions++;
+       assert(object->internal);
 
 
-                       if (collisions > copy_delayed_max_collisions)
-                               copy_delayed_max_collisions = collisions;
+       while (TRUE) {
+          if (ptoa_64(object->resident_page_count) > size/2 && pmap != PMAP_NULL) {
+               vm_object_unlock(object);
+               pmap_protect_options(pmap, pmap_start, pmap_start + size, prot,
+                                    options & ~PMAP_OPTIONS_NOFLUSH, NULL);
+               return;
+           }
 
 
-                       if (src_object_shared == TRUE)
-                               vm_object_lock_shared(src_object);
-                       else
-                               vm_object_lock(src_object);
+          pmap_flush_context_init(&pmap_flush_context_storage);
+          delayed_pmap_flush = FALSE;
 
 
-                       goto Retry;
-               }
+           /*
+            * if we are doing large ranges with respect to resident
+            * page count then we should interate over pages otherwise
+            * inverse page look-up will be faster
+            */
+           if (ptoa_64(object->resident_page_count / 4) <  size) {
+               vm_page_t               p;
+               vm_object_offset_t      end;
 
 
-               /*
-                *      Determine whether the old copy object has
-                *      been modified.
-                */
+               end = offset + size;
 
 
-               if (old_copy->resident_page_count == 0 &&
-                   !old_copy->pager_created) {
-                       /*
-                        *      It has not been modified.
-                        *
-                        *      Return another reference to
-                        *      the existing copy-object if
-                        *      we can safely grow it (if
-                        *      needed).
-                        */
+               vm_page_queue_iterate(&object->memq, p, vm_page_t, listq) {
+                       if (!p->fictitious && (offset <= p->offset) && (p->offset < end)) {
+                               vm_map_offset_t start;
+
+                               start = pmap_start + p->offset - offset;
+
+                               if (pmap != PMAP_NULL)
+                                       pmap_protect_options(
+                                               pmap,
+                                               start,
+                                               start + PAGE_SIZE_64,
+                                               prot,
+                                               options | PMAP_OPTIONS_NOFLUSH,
+                                               &pmap_flush_context_storage);
+                               else
+                                       pmap_page_protect_options(
+                                               VM_PAGE_GET_PHYS_PAGE(p),
+                                               prot,
+                                               options | PMAP_OPTIONS_NOFLUSH,
+                                               &pmap_flush_context_storage);
+                                       delayed_pmap_flush = TRUE;
+                       }
+               }
 
 
-                       if (old_copy->size < copy_size) {
-                               if (src_object_shared == TRUE) {
-                                       vm_object_unlock(old_copy);
-                                       vm_object_unlock(src_object);
-                               
-                                       vm_object_lock(src_object);
-                                       src_object_shared = FALSE;
-                                       goto Retry;
-                               }
-                               /*
-                                * We can't perform a delayed copy if any of the
-                                * pages in the extended range are wired (because
-                                * we can't safely take write permission away from
-                                * wired pages).  If the pages aren't wired, then
-                                * go ahead and protect them.
-                                */
-                               copy_delayed_protect_iterate++;
+          } else {
+               vm_page_t               p;
+               vm_object_offset_t      end;
+               vm_object_offset_t      target_off;
 
 
-                               queue_iterate(&src_object->memq, p, vm_page_t, listq) {
-                                       if (!p->fictitious && 
-                                           p->offset >= old_copy->size && 
-                                           p->offset < copy_size) {
-                                               if (VM_PAGE_WIRED(p)) {
-                                                       vm_object_unlock(old_copy);
-                                                       vm_object_unlock(src_object);
+               end = offset + size;
 
 
-                                                       if (new_copy != VM_OBJECT_NULL) {
-                                                               vm_object_unlock(new_copy);
-                                                               vm_object_deallocate(new_copy);
-                                                       }
+               for (target_off = offset; 
+                    target_off < end; target_off += PAGE_SIZE) {
 
 
-                                                       return VM_OBJECT_NULL;
-                                               } else {
-                                                       pmap_page_protect(p->phys_page, 
-                                                                         (VM_PROT_ALL & ~VM_PROT_WRITE));
-                                               }
-                                       }
-                               }
-                               old_copy->size = copy_size;
-                       }
-                       if (src_object_shared == TRUE)
-                               vm_object_reference_shared(old_copy);
-                       else
-                               vm_object_reference_locked(old_copy);
-                       vm_object_unlock(old_copy);
-                       vm_object_unlock(src_object);
+                       p = vm_page_lookup(object, target_off);
 
 
-                       if (new_copy != VM_OBJECT_NULL) {
-                               vm_object_unlock(new_copy);
-                               vm_object_deallocate(new_copy);
-                       }
-                       return(old_copy);
+                       if (p != VM_PAGE_NULL) {
+                               vm_object_offset_t start;
+
+                               start = pmap_start + (p->offset - offset);
+
+                               if (pmap != PMAP_NULL)
+                                       pmap_protect_options(
+                                               pmap,
+                                               start,
+                                               start + PAGE_SIZE_64,
+                                               prot,
+                                               options | PMAP_OPTIONS_NOFLUSH,
+                                               &pmap_flush_context_storage);
+                               else
+                                       pmap_page_protect_options(
+                                               VM_PAGE_GET_PHYS_PAGE(p),
+                                               prot,
+                                               options | PMAP_OPTIONS_NOFLUSH,
+                                               &pmap_flush_context_storage);
+                                       delayed_pmap_flush = TRUE;
+                       }
                }
                }
-               
-               
+           }
+           if (delayed_pmap_flush == TRUE)
+                   pmap_flush(&pmap_flush_context_storage);
 
 
+           if (prot == VM_PROT_NONE) {
                /*
                /*
-                * Adjust the size argument so that the newly-created 
-                * copy object will be large enough to back either the
-                * old copy object or the new mapping.
+                * Must follow shadow chain to remove access
+                * to pages in shadowed objects.
                 */
                 */
-               if (old_copy->size > copy_size)
-                       copy_size = old_copy->size;
-
-               if (new_copy == VM_OBJECT_NULL) {
-                       vm_object_unlock(old_copy);
-                       vm_object_unlock(src_object);
-                       new_copy = vm_object_allocate(copy_size);
-                       vm_object_lock(src_object);
-                       vm_object_lock(new_copy);
+               vm_object_t     next_object;
 
 
-                       src_object_shared = FALSE;
-                       goto Retry;
+               next_object = object->shadow;
+               if (next_object != VM_OBJECT_NULL) {
+                   offset += object->vo_shadow_offset;
+                   vm_object_lock(next_object);
+                   vm_object_unlock(object);
+                   object = next_object;
                }
                }
-               new_copy->size = copy_size;     
-
+               else {
+                   /*
+                    * End of chain - we are done.
+                    */
+                   break;
+               }
+           }
+           else {
                /*
                /*
-                *      The copy-object is always made large enough to
-                *      completely shadow the original object, since
-                *      it may have several users who want to shadow
-                *      the original object at different points.
+                * Pages in shadowed objects may never have
+                * write permission - we may stop here.
                 */
                 */
+               break;
+           }
+       }
 
 
-               assert((old_copy->shadow == src_object) &&
-                   (old_copy->shadow_offset == (vm_object_offset_t) 0));
+       vm_object_unlock(object);
+}
 
 
-       } else if (new_copy == VM_OBJECT_NULL) {
-               vm_object_unlock(src_object);
-               new_copy = vm_object_allocate(copy_size);
-               vm_object_lock(src_object);
-               vm_object_lock(new_copy);
+/*
+ *     Routine:        vm_object_copy_slowly
+ *
+ *     Description:
+ *             Copy the specified range of the source
+ *             virtual memory object without using
+ *             protection-based optimizations (such
+ *             as copy-on-write).  The pages in the
+ *             region are actually copied.
+ *
+ *     In/out conditions:
+ *             The caller must hold a reference and a lock
+ *             for the source virtual memory object.  The source
+ *             object will be returned *unlocked*.
+ *
+ *     Results:
+ *             If the copy is completed successfully, KERN_SUCCESS is
+ *             returned.  If the caller asserted the interruptible
+ *             argument, and an interruption occurred while waiting
+ *             for a user-generated event, MACH_SEND_INTERRUPTED is
+ *             returned.  Other values may be returned to indicate
+ *             hard errors during the copy operation.
+ *
+ *             A new virtual memory object is returned in a
+ *             parameter (_result_object).  The contents of this
+ *             new object, starting at a zero offset, are a copy
+ *             of the source memory region.  In the event of
+ *             an error, this parameter will contain the value
+ *             VM_OBJECT_NULL.
+ */
+__private_extern__ kern_return_t
+vm_object_copy_slowly(
+       vm_object_t             src_object,
+       vm_object_offset_t      src_offset,
+       vm_object_size_t        size,
+       boolean_t               interruptible,
+       vm_object_t             *_result_object)        /* OUT */
+{
+       vm_object_t             new_object;
+       vm_object_offset_t      new_offset;
 
 
-               src_object_shared = FALSE;
-               goto Retry;
+       struct vm_object_fault_info fault_info;
+
+       XPR(XPR_VM_OBJECT, "v_o_c_slowly obj 0x%x off 0x%x size 0x%x\n",
+           src_object, src_offset, size, 0, 0);
+
+       if (size == 0) {
+               vm_object_unlock(src_object);
+               *_result_object = VM_OBJECT_NULL;
+               return(KERN_INVALID_ARGUMENT);
        }
 
        /*
        }
 
        /*
-        * We now have the src object locked, and the new copy object
-        * allocated and locked (and potentially the old copy locked).
-        * Before we go any further, make sure we can still perform
-        * a delayed copy, as the situation may have changed.
-        *
-        * Specifically, we can't perform a delayed copy if any of the
-        * pages in the range are wired (because we can't safely take
-        * write permission away from wired pages).  If the pages aren't
-        * wired, then go ahead and protect them.
+        *      Prevent destruction of the source object while we copy.
         */
         */
-       copy_delayed_protect_iterate++;
 
 
-       queue_iterate(&src_object->memq, p, vm_page_t, listq) {
-               if (!p->fictitious && p->offset < copy_size) {
-                       if (VM_PAGE_WIRED(p)) {
-                               if (old_copy)
-                                       vm_object_unlock(old_copy);
-                               vm_object_unlock(src_object);
-                               vm_object_unlock(new_copy);
-                               vm_object_deallocate(new_copy);
-                               return VM_OBJECT_NULL;
-                       } else {
-                               pmap_page_protect(p->phys_page, 
-                                                 (VM_PROT_ALL & ~VM_PROT_WRITE));
-                       }
-               }
-       }
-       if (old_copy != VM_OBJECT_NULL) {
-               /*
-                *      Make the old copy-object shadow the new one.
-                *      It will receive no more pages from the original
-                *      object.
-                */
-
-               /* remove ref. from old_copy */
-               vm_object_lock_assert_exclusive(src_object);
-               src_object->ref_count--;
-               assert(src_object->ref_count > 0);
-               vm_object_lock_assert_exclusive(old_copy);
-               old_copy->shadow = new_copy;
-               vm_object_lock_assert_exclusive(new_copy);
-               assert(new_copy->ref_count > 0);
-               new_copy->ref_count++;          /* for old_copy->shadow ref. */
-
-#if TASK_SWAPPER
-               if (old_copy->res_count) {
-                       VM_OBJ_RES_INCR(new_copy);
-                       VM_OBJ_RES_DECR(src_object);
-               }
-#endif
-
-               vm_object_unlock(old_copy);     /* done with old_copy */
-       }
+       vm_object_reference_locked(src_object);
+       vm_object_unlock(src_object);
 
        /*
 
        /*
-        *      Point the new copy at the existing object.
+        *      Create a new object to hold the copied pages.
+        *      A few notes:
+        *              We fill the new object starting at offset 0,
+        *               regardless of the input offset.
+        *              We don't bother to lock the new object within
+        *               this routine, since we have the only reference.
         */
         */
-       vm_object_lock_assert_exclusive(new_copy);
-       new_copy->shadow = src_object;
-       new_copy->shadow_offset = 0;
-       new_copy->shadowed = TRUE;      /* caller must set needs_copy */
-
-       vm_object_lock_assert_exclusive(src_object);
-       vm_object_reference_locked(src_object);
-       src_object->copy = new_copy;
-       vm_object_unlock(src_object);
-       vm_object_unlock(new_copy);
-
-       XPR(XPR_VM_OBJECT,
-               "vm_object_copy_delayed: used copy object %X for source %X\n",
-               new_copy, src_object, 0, 0, 0);
 
 
-       return new_copy;
-}
+       new_object = vm_object_allocate(size);
+       new_offset = 0;
 
 
-/*
- *     Routine:        vm_object_copy_strategically
- *
- *     Purpose:
- *             Perform a copy according to the source object's
- *             declared strategy.  This operation may block,
- *             and may be interrupted.
- */
-__private_extern__ kern_return_t
-vm_object_copy_strategically(
-       register vm_object_t    src_object,
-       vm_object_offset_t      src_offset,
-       vm_object_size_t        size,
-       vm_object_t             *dst_object,    /* OUT */
-       vm_object_offset_t      *dst_offset,    /* OUT */
-       boolean_t               *dst_needs_copy) /* OUT */
-{
-       boolean_t       result;
-       boolean_t       interruptible = THREAD_ABORTSAFE; /* XXX */
-       boolean_t       object_lock_shared = FALSE;
-       memory_object_copy_strategy_t copy_strategy;
+       assert(size == trunc_page_64(size));    /* Will the loop terminate? */
 
 
-       assert(src_object != VM_OBJECT_NULL);
+       fault_info.interruptible = interruptible;
+       fault_info.behavior  = VM_BEHAVIOR_SEQUENTIAL;
+       fault_info.user_tag = 0;
+       fault_info.pmap_options = 0;
+       fault_info.lo_offset = src_offset;
+       fault_info.hi_offset = src_offset + size;
+       fault_info.no_cache  = FALSE;
+       fault_info.stealth = TRUE;
+       fault_info.io_sync = FALSE;
+       fault_info.cs_bypass = FALSE;
+       fault_info.mark_zf_absent = FALSE;
+       fault_info.batch_pmap_op = FALSE;
 
 
-       copy_strategy = src_object->copy_strategy;
+       for ( ;
+           size != 0 ;
+           src_offset += PAGE_SIZE_64, 
+                       new_offset += PAGE_SIZE_64, size -= PAGE_SIZE_64
+           ) {
+               vm_page_t       new_page;
+               vm_fault_return_t result;
 
 
-       if (copy_strategy == MEMORY_OBJECT_COPY_DELAY) {
-               vm_object_lock_shared(src_object);
-               object_lock_shared = TRUE;
-       } else
-               vm_object_lock(src_object);
+               vm_object_lock(new_object);
 
 
-       /*
-        *      The copy strategy is only valid if the memory manager
-        *      is "ready". Internal objects are always ready.
-        */
+               while ((new_page = vm_page_alloc(new_object, new_offset))
+                               == VM_PAGE_NULL) {
 
 
-       while (!src_object->internal && !src_object->pager_ready) {
-               wait_result_t wait_result;
+                       vm_object_unlock(new_object);
 
 
-               if (object_lock_shared == TRUE) {
-                       vm_object_unlock(src_object);
-                       vm_object_lock(src_object);
-                       object_lock_shared = FALSE;
-                       continue;
-               }
-               wait_result = vm_object_sleep(  src_object,
-                                               VM_OBJECT_EVENT_PAGER_READY,
-                                               interruptible);
-               if (wait_result != THREAD_AWAKENED) {
-                       vm_object_unlock(src_object);
-                       *dst_object = VM_OBJECT_NULL;
-                       *dst_offset = 0;
-                       *dst_needs_copy = FALSE;
-                       return(MACH_SEND_INTERRUPTED);
+                       if (!vm_page_wait(interruptible)) {
+                               vm_object_deallocate(new_object);
+                               vm_object_deallocate(src_object);
+                               *_result_object = VM_OBJECT_NULL;
+                               return(MACH_SEND_INTERRUPTED);
+                       }
+                       vm_object_lock(new_object);
                }
                }
-       }
-
-       /*
-        *      Use the appropriate copy strategy.
-        */
+               vm_object_unlock(new_object);
 
 
-       switch (copy_strategy) {
-           case MEMORY_OBJECT_COPY_DELAY:
-               *dst_object = vm_object_copy_delayed(src_object,
-                                                    src_offset, size, object_lock_shared);
-               if (*dst_object != VM_OBJECT_NULL) {
-                       *dst_offset = src_offset;
-                       *dst_needs_copy = TRUE;
-                       result = KERN_SUCCESS;
-                       break;
-               }
-               vm_object_lock(src_object);
-               /* fall thru when delayed copy not allowed */
+               do {
+                       vm_prot_t       prot = VM_PROT_READ;
+                       vm_page_t       _result_page;
+                       vm_page_t       top_page;
+                       vm_page_t       result_page;
+                       kern_return_t   error_code;
+                       vm_object_t     result_page_object;
 
 
-           case MEMORY_OBJECT_COPY_NONE:
-               result = vm_object_copy_slowly(src_object, src_offset, size,
-                                              interruptible, dst_object);
-               if (result == KERN_SUCCESS) {
-                       *dst_offset = 0;
-                       *dst_needs_copy = FALSE;
-               }
-               break;
 
 
-           case MEMORY_OBJECT_COPY_CALL:
-               result = vm_object_copy_call(src_object, src_offset, size,
-                               dst_object);
-               if (result == KERN_SUCCESS) {
-                       *dst_offset = src_offset;
-                       *dst_needs_copy = TRUE;
-               }
-               break;
+                       vm_object_lock(src_object);
 
 
-           case MEMORY_OBJECT_COPY_SYMMETRIC:
-               XPR(XPR_VM_OBJECT, "v_o_c_strategically obj 0x%x off 0x%x size 0x%x\n", src_object, src_offset, size, 0, 0);
-               vm_object_unlock(src_object);
-               result = KERN_MEMORY_RESTART_COPY;
-               break;
+                       if (src_object->internal &&
+                           src_object->shadow == VM_OBJECT_NULL &&
+                           (vm_page_lookup(src_object,
+                                           src_offset) == VM_PAGE_NULL) &&
+                           (src_object->pager == NULL ||
+                            (VM_COMPRESSOR_PAGER_STATE_GET(src_object,
+                                                           src_offset) ==
+                             VM_EXTERNAL_STATE_ABSENT))) {
+                               /*
+                                * This page is neither resident nor compressed
+                                * and there's no shadow object below 
+                                * "src_object", so this page is really missing.
+                                * There's no need to zero-fill it just to copy
+                                * it:  let's leave it missing in "new_object"
+                                * and get zero-filled on demand.
+                                */
+                               vm_object_unlock(src_object);
+                               /* free the unused "new_page"... */
+                               vm_object_lock(new_object);
+                               VM_PAGE_FREE(new_page);
+                               new_page = VM_PAGE_NULL;
+                               vm_object_unlock(new_object);
+                               /* ...and go to next page in "src_object" */
+                               result = VM_FAULT_SUCCESS;
+                               break;
+                       }
 
 
-           default:
-               panic("copy_strategically: bad strategy");
-               result = KERN_INVALID_ARGUMENT;
-       }
-       return(result);
-}
+                       vm_object_paging_begin(src_object);
 
 
-/*
- *     vm_object_shadow:
- *
- *     Create a new object which is backed by the
- *     specified existing object range.  The source
- *     object reference is deallocated.
- *
- *     The new object and offset into that object
- *     are returned in the source parameters.
- */
-boolean_t vm_object_shadow_check = FALSE;
+                       if (size > (vm_size_t) -1) {
+                               /* 32-bit overflow */
+                               fault_info.cluster_size = (vm_size_t) (0 - PAGE_SIZE);
+                       } else {
+                               fault_info.cluster_size = (vm_size_t) size;
+                               assert(fault_info.cluster_size == size);
+                       }
 
 
-__private_extern__ boolean_t
-vm_object_shadow(
-       vm_object_t             *object,        /* IN/OUT */
-       vm_object_offset_t      *offset,        /* IN/OUT */
-       vm_object_size_t        length)
-{
-       register vm_object_t    source;
-       register vm_object_t    result;
+                       XPR(XPR_VM_FAULT,"vm_object_copy_slowly -> vm_fault_page",0,0,0,0,0);
+                       _result_page = VM_PAGE_NULL;
+                       result = vm_fault_page(src_object, src_offset,
+                               VM_PROT_READ, FALSE,
+                               FALSE, /* page not looked up */
+                               &prot, &_result_page, &top_page,
+                               (int *)0,
+                               &error_code, FALSE, FALSE, &fault_info);
 
 
-       source = *object;
-#if 0
-       /*
-        * XXX FBDP
-        * This assertion is valid but it gets triggered by Rosetta for example
-        * due to a combination of vm_remap() that changes a VM object's
-        * copy_strategy from SYMMETRIC to DELAY and vm_protect(VM_PROT_COPY)
-        * that then sets "needs_copy" on its map entry.  This creates a
-        * mapping situation that VM should never see and doesn't know how to
-        * handle.
-        * It's not clear if this can create any real problem but we should
-        * look into fixing this, probably by having vm_protect(VM_PROT_COPY)
-        * do more than just set "needs_copy" to handle the copy-on-write...
-        * In the meantime, let's disable the assertion.
-        */
-       assert(source->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC);
-#endif
+                       switch(result) {
+                       case VM_FAULT_SUCCESS:
+                               result_page = _result_page;
+                               result_page_object = VM_PAGE_OBJECT(result_page);
 
 
-       /*
-        *      Determine if we really need a shadow.
-        */
+                               /*
+                                *      Copy the page to the new object.
+                                *
+                                *      POLICY DECISION:
+                                *              If result_page is clean,
+                                *              we could steal it instead
+                                *              of copying.
+                                */
 
 
-       if (vm_object_shadow_check && source->ref_count == 1 &&
-           (source->shadow == VM_OBJECT_NULL ||
-            source->shadow->copy == VM_OBJECT_NULL))
-       {
-               source->shadowed = FALSE;
-               return FALSE;
-       }
+                               vm_page_copy(result_page, new_page);
+                               vm_object_unlock(result_page_object);
 
 
-       /*
-        *      Allocate a new object with the given length
-        */
+                               /*
+                                *      Let go of both pages (make them
+                                *      not busy, perform wakeup, activate).
+                                */
+                               vm_object_lock(new_object);
+                               SET_PAGE_DIRTY(new_page, FALSE);
+                               PAGE_WAKEUP_DONE(new_page);
+                               vm_object_unlock(new_object);
 
 
-       if ((result = vm_object_allocate(length)) == VM_OBJECT_NULL)
-               panic("vm_object_shadow: no object for shadowing");
+                               vm_object_lock(result_page_object);
+                               PAGE_WAKEUP_DONE(result_page);
 
 
-       /*
-        *      The new object shadows the source object, adding
-        *      a reference to it.  Our caller changes his reference
-        *      to point to the new object, removing a reference to
-        *      the source object.  Net result: no change of reference
-        *      count.
-        */
-       result->shadow = source;
-       
-       /*
-        *      Store the offset into the source object,
-        *      and fix up the offset into the new object.
-        */
+                               vm_page_lockspin_queues();
+                               if ((result_page->vm_page_q_state == VM_PAGE_ON_SPECULATIVE_Q) ||
+                                   (result_page->vm_page_q_state == VM_PAGE_NOT_ON_Q)) {
+                                       vm_page_activate(result_page);
+                               }
+                               vm_page_activate(new_page);
+                               vm_page_unlock_queues();
+
+                               /*
+                                *      Release paging references and
+                                *      top-level placeholder page, if any.
+                                */
+
+                               vm_fault_cleanup(result_page_object,
+                                                top_page);
+
+                               break;
+                               
+                       case VM_FAULT_RETRY:
+                               break;
+
+                       case VM_FAULT_MEMORY_SHORTAGE:
+                               if (vm_page_wait(interruptible))
+                                       break;
+                               /* fall thru */
+
+                       case VM_FAULT_INTERRUPTED:
+                               vm_object_lock(new_object);
+                               VM_PAGE_FREE(new_page);
+                               vm_object_unlock(new_object);
+                                       
+                               vm_object_deallocate(new_object);
+                               vm_object_deallocate(src_object);
+                               *_result_object = VM_OBJECT_NULL;
+                               return(MACH_SEND_INTERRUPTED);
+
+                       case VM_FAULT_SUCCESS_NO_VM_PAGE:
+                               /* success but no VM page: fail */
+                               vm_object_paging_end(src_object);
+                               vm_object_unlock(src_object);
+                               /*FALLTHROUGH*/
+                       case VM_FAULT_MEMORY_ERROR:
+                               /*
+                                * A policy choice:
+                                *      (a) ignore pages that we can't
+                                *          copy
+                                *      (b) return the null object if
+                                *          any page fails [chosen]
+                                */
+
+                               vm_object_lock(new_object);
+                               VM_PAGE_FREE(new_page);
+                               vm_object_unlock(new_object);
+
+                               vm_object_deallocate(new_object);
+                               vm_object_deallocate(src_object);
+                               *_result_object = VM_OBJECT_NULL;
+                               return(error_code ? error_code:
+                                      KERN_MEMORY_ERROR);
 
 
-       result->shadow_offset = *offset;
+                       default:
+                               panic("vm_object_copy_slowly: unexpected error"
+                                     " 0x%x from vm_fault_page()\n", result);
+                       }
+               } while (result != VM_FAULT_SUCCESS);
+       }
 
        /*
 
        /*
-        *      Return the new things
+        *      Lose the extra reference, and return our object.
         */
         */
-
-       *offset = 0;
-       *object = result;
-       return TRUE;
+       vm_object_deallocate(src_object);
+       *_result_object = new_object;
+       return(KERN_SUCCESS);
 }
 
 /*
 }
 
 /*
- *     The relationship between vm_object structures and
- *     the memory_object requires careful synchronization.
- *
- *     All associations are created by memory_object_create_named
- *  for external pagers and vm_object_pager_create for internal
- *  objects as follows:
- *
- *             pager:  the memory_object itself, supplied by
- *                     the user requesting a mapping (or the kernel,
- *                     when initializing internal objects); the
- *                     kernel simulates holding send rights by keeping
- *                     a port reference;
- *
- *             pager_request:
- *                     the memory object control port,
- *                     created by the kernel; the kernel holds
- *                     receive (and ownership) rights to this
- *                     port, but no other references.
- *
- *     When initialization is complete, the "initialized" field
- *     is asserted.  Other mappings using a particular memory object,
- *     and any references to the vm_object gained through the
- *     port association must wait for this initialization to occur.
- *
- *     In order to allow the memory manager to set attributes before
- *     requests (notably virtual copy operations, but also data or
- *     unlock requests) are made, a "ready" attribute is made available.
- *     Only the memory manager may affect the value of this attribute.
- *     Its value does not affect critical kernel functions, such as
- *     internal object initialization or destruction.  [Furthermore,
- *     memory objects created by the kernel are assumed to be ready
- *     immediately; the default memory manager need not explicitly
- *     set the "ready" attribute.]
- *
- *     [Both the "initialized" and "ready" attribute wait conditions
- *     use the "pager" field as the wait event.]
- *
- *     The port associations can be broken down by any of the
- *     following routines:
- *             vm_object_terminate:
- *                     No references to the vm_object remain, and
- *                     the object cannot (or will not) be cached.
- *                     This is the normal case, and is done even
- *                     though one of the other cases has already been
- *                     done.
- *             memory_object_destroy:
- *                     The memory manager has requested that the
- *                     kernel relinquish references to the memory
- *                     object. [The memory manager may not want to
- *                     destroy the memory object, but may wish to
- *                     refuse or tear down existing memory mappings.]
- *
- *     Each routine that breaks an association must break all of
- *     them at once.  At some later time, that routine must clear
- *     the pager field and release the memory object references.
- *     [Furthermore, each routine must cope with the simultaneous
- *     or previous operations of the others.]
+ *     Routine:        vm_object_copy_quickly
  *
  *
- *     In addition to the lock on the object, the vm_object_hash_lock
- *     governs the associations.  References gained through the
- *     association require use of the hash lock.
+ *     Purpose:
+ *             Copy the specified range of the source virtual
+ *             memory object, if it can be done without waiting
+ *             for user-generated events.
  *
  *
- *     Because the pager field may be cleared spontaneously, it
- *     cannot be used to determine whether a memory object has
- *     ever been associated with a particular vm_object.  [This
- *     knowledge is important to the shadow object mechanism.]
- *     For this reason, an additional "created" attribute is
- *     provided.
+ *     Results:
+ *             If the copy is successful, the copy is returned in
+ *             the arguments; otherwise, the arguments are not
+ *             affected.
  *
  *
- *     During various paging operations, the pager reference found in the
- *     vm_object must be valid.  To prevent this from being released,
- *     (other than being removed, i.e., made null), routines may use
- *     the vm_object_paging_begin/end routines [actually, macros].
- *     The implementation uses the "paging_in_progress" and "wanted" fields.
- *     [Operations that alter the validity of the pager values include the
- *     termination routines and vm_object_collapse.]
+ *     In/out conditions:
+ *             The object should be unlocked on entry and exit.
  */
 
  */
 
-
-/*
- *     Routine:        vm_object_enter
- *     Purpose:
- *             Find a VM object corresponding to the given
- *             pager; if no such object exists, create one,
- *             and initialize the pager.
- */
-vm_object_t
-vm_object_enter(
-       memory_object_t         pager,
-       vm_object_size_t        size,
-       boolean_t               internal,
-       boolean_t               init,
-       boolean_t               named)
+/*ARGSUSED*/
+__private_extern__ boolean_t
+vm_object_copy_quickly(
+       vm_object_t             *_object,               /* INOUT */
+       __unused vm_object_offset_t     offset, /* IN */
+       __unused vm_object_size_t       size,   /* IN */
+       boolean_t               *_src_needs_copy,       /* OUT */
+       boolean_t               *_dst_needs_copy)       /* OUT */
 {
 {
-       register vm_object_t    object;
-       vm_object_t             new_object;
-       boolean_t               must_init;
-       vm_object_hash_entry_t  entry, new_entry;
-       uint32_t        try_failed_count = 0;
-       lck_mtx_t       *lck;
+       vm_object_t     object = *_object;
+       memory_object_copy_strategy_t copy_strategy;
 
 
-       if (pager == MEMORY_OBJECT_NULL)
-               return(vm_object_allocate(size));
+       XPR(XPR_VM_OBJECT, "v_o_c_quickly obj 0x%x off 0x%x size 0x%x\n",
+           *_object, offset, size, 0, 0);
+       if (object == VM_OBJECT_NULL) {
+               *_src_needs_copy = FALSE;
+               *_dst_needs_copy = FALSE;
+               return(TRUE);
+       }
 
 
-       new_object = VM_OBJECT_NULL;
-       new_entry = VM_OBJECT_HASH_ENTRY_NULL;
-       must_init = init;
+       vm_object_lock(object);
 
 
-       /*
-        *      Look for an object associated with this port.
-        */
-Retry:
-       lck = vm_object_hash_lock_spin(pager);
-       do {
-               entry = vm_object_hash_lookup(pager, FALSE);
+       copy_strategy = object->copy_strategy;
 
 
-               if (entry == VM_OBJECT_HASH_ENTRY_NULL) {
-                       if (new_object == VM_OBJECT_NULL) {
-                               /*
-                                *      We must unlock to create a new object;
-                                *      if we do so, we must try the lookup again.
-                                */
-                               vm_object_hash_unlock(lck);
-                               assert(new_entry == VM_OBJECT_HASH_ENTRY_NULL);
-                               new_entry = vm_object_hash_entry_alloc(pager);
-                               new_object = vm_object_allocate(size);
-                               lck = vm_object_hash_lock_spin(pager);
-                       } else {
-                               /*
-                                *      Lookup failed twice, and we have something
-                                *      to insert; set the object.
-                                */
-                               vm_object_hash_insert(new_entry, new_object);
-                               entry = new_entry;
-                               new_entry = VM_OBJECT_HASH_ENTRY_NULL;
-                               new_object = VM_OBJECT_NULL;
-                               must_init = TRUE;
-                       }
-               } else if (entry->object == VM_OBJECT_NULL) {
-                       /*
-                        *      If a previous object is being terminated,
-                        *      we must wait for the termination message
-                        *      to be queued (and lookup the entry again).
-                        */
-                       entry->waiting = TRUE;
-                       entry = VM_OBJECT_HASH_ENTRY_NULL;
-                       assert_wait((event_t) pager, THREAD_UNINT);
-                       vm_object_hash_unlock(lck);
-
-                       thread_block(THREAD_CONTINUE_NULL);
-                       lck = vm_object_hash_lock_spin(pager);
-               }
-       } while (entry == VM_OBJECT_HASH_ENTRY_NULL);
-
-       object = entry->object;
-       assert(object != VM_OBJECT_NULL);
-
-       if (!must_init) {
-               if ( !vm_object_lock_try(object)) {
-
-                       vm_object_hash_unlock(lck);
-
-                       try_failed_count++;
-                       mutex_pause(try_failed_count);  /* wait a bit */
-                       goto Retry;
-               }
-               assert(!internal || object->internal);
-#if VM_OBJECT_CACHE
-               if (object->ref_count == 0) {
-                       if ( !vm_object_cache_lock_try()) {
-
-                               vm_object_hash_unlock(lck);
-                               vm_object_unlock(object);
-
-                               try_failed_count++;
-                               mutex_pause(try_failed_count);  /* wait a bit */
-                               goto Retry;
-                       }
-                       XPR(XPR_VM_OBJECT_CACHE,
-                           "vm_object_enter: removing %x from cache, head (%x, %x)\n",
-                               object,
-                               vm_object_cached_list.next,
-                               vm_object_cached_list.prev, 0,0);
-                       queue_remove(&vm_object_cached_list, object,
-                                    vm_object_t, cached_list);
-                       vm_object_cached_count--;
-
-                       vm_object_cache_unlock();
-               }
-#endif
-               if (named) {
-                       assert(!object->named);
-                       object->named = TRUE;
-               }
-               vm_object_lock_assert_exclusive(object);
-               object->ref_count++;
-               vm_object_res_reference(object);
-
-               vm_object_hash_unlock(lck);
-               vm_object_unlock(object);
-
-               VM_STAT_INCR(hits);
-       } else
-               vm_object_hash_unlock(lck);
-
-       assert(object->ref_count > 0);
-
-       VM_STAT_INCR(lookups);
-
-       XPR(XPR_VM_OBJECT,
-               "vm_o_enter: pager 0x%x obj 0x%x must_init %d\n",
-               pager, object, must_init, 0, 0);
-
-       /*
-        *      If we raced to create a vm_object but lost, let's
-        *      throw away ours.
-        */
-
-       if (new_object != VM_OBJECT_NULL)
-               vm_object_deallocate(new_object);
-
-       if (new_entry != VM_OBJECT_HASH_ENTRY_NULL)
-               vm_object_hash_entry_free(new_entry);
-
-       if (must_init) {
-               memory_object_control_t control;
-
-               /*
-                *      Allocate request port.
-                */
-
-               control = memory_object_control_allocate(object);
-               assert (control != MEMORY_OBJECT_CONTROL_NULL);
-
-               vm_object_lock(object);
-               assert(object != kernel_object);
+       switch (copy_strategy) {
+       case MEMORY_OBJECT_COPY_SYMMETRIC:
 
                /*
 
                /*
-                *      Copy the reference we were given.
+                *      Symmetric copy strategy.
+                *      Make another reference to the object.
+                *      Leave object/offset unchanged.
                 */
 
                 */
 
-               memory_object_reference(pager);
-               object->pager_created = TRUE;
-               object->pager = pager;
-               object->internal = internal;
-               object->pager_trusted = internal;
-               if (!internal) {
-                       /* copy strategy invalid until set by memory manager */
-                       object->copy_strategy = MEMORY_OBJECT_COPY_INVALID;
-               }
-               object->pager_control = control;
-               object->pager_ready = FALSE;
-
+               vm_object_reference_locked(object);
+               object->shadowed = TRUE;
                vm_object_unlock(object);
 
                /*
                vm_object_unlock(object);
 
                /*
-                *      Let the pager know we're using it.
+                *      Both source and destination must make
+                *      shadows, and the source must be made
+                *      read-only if not already.
                 */
 
                 */
 
-               (void) memory_object_init(pager,
-                       object->pager_control,
-                       PAGE_SIZE);
-
-               vm_object_lock(object);
-               if (named)
-                       object->named = TRUE;
-               if (internal) {
-                       object->pager_ready = TRUE;
-                       vm_object_wakeup(object, VM_OBJECT_EVENT_PAGER_READY);
-               }
-
-               object->pager_initialized = TRUE;
-               vm_object_wakeup(object, VM_OBJECT_EVENT_INITIALIZED);
-       } else {
-               vm_object_lock(object);
-       }
+               *_src_needs_copy = TRUE;
+               *_dst_needs_copy = TRUE;
 
 
-       /*
-        *      [At this point, the object must be locked]
-        */
+               break;
 
 
-       /*
-        *      Wait for the work above to be done by the first
-        *      thread to map this object.
-        */
+       case MEMORY_OBJECT_COPY_DELAY:
+               vm_object_unlock(object);
+               return(FALSE);
 
 
-       while (!object->pager_initialized) {
-               vm_object_sleep(object,
-                               VM_OBJECT_EVENT_INITIALIZED,
-                               THREAD_UNINT);
+       default:
+               vm_object_unlock(object);
+               return(FALSE);
        }
        }
-       vm_object_unlock(object);
-
-       XPR(XPR_VM_OBJECT,
-           "vm_object_enter: vm_object %x, memory_object %x, internal %d\n",
-           object, object->pager, internal, 0,0);
-       return(object);
+       return(TRUE);
 }
 
 }
 
+static int copy_call_count = 0;
+static int copy_call_sleep_count = 0;
+static int copy_call_restart_count = 0;
+
 /*
 /*
- *     Routine:        vm_object_pager_create
- *     Purpose:
- *             Create a memory object for an internal object.
+ *     Routine:        vm_object_copy_call [internal]
+ *
+ *     Description:
+ *             Copy the source object (src_object), using the
+ *             user-managed copy algorithm.
+ *
  *     In/out conditions:
  *     In/out conditions:
- *             The object is locked on entry and exit;
- *             it may be unlocked within this call.
- *     Limitations:
- *             Only one thread may be performing a
- *             vm_object_pager_create on an object at
- *             a time.  Presumably, only the pageout
- *             daemon will be using this routine.
+ *             The source object must be locked on entry.  It
+ *             will be *unlocked* on exit.
+ *
+ *     Results:
+ *             If the copy is successful, KERN_SUCCESS is returned.
+ *             A new object that represents the copied virtual
+ *             memory is returned in a parameter (*_result_object).
+ *             If the return value indicates an error, this parameter
+ *             is not valid.
  */
  */
-
-void
-vm_object_pager_create(
-       register vm_object_t    object)
+static kern_return_t
+vm_object_copy_call(
+       vm_object_t             src_object,
+       vm_object_offset_t      src_offset,
+       vm_object_size_t        size,
+       vm_object_t             *_result_object)        /* OUT */
 {
 {
-       memory_object_t         pager;
-       vm_object_hash_entry_t  entry;
-       lck_mtx_t               *lck;
-#if    MACH_PAGEMAP
-       vm_object_size_t        size;
-       vm_external_map_t       map;
-#endif /* MACH_PAGEMAP */
-
-       XPR(XPR_VM_OBJECT, "vm_object_pager_create, object 0x%X\n",
-               object, 0,0,0,0);
-
-       assert(object != kernel_object);
-
-       if (memory_manager_default_check() != KERN_SUCCESS)
-               return;
+       kern_return_t   kr;
+       vm_object_t     copy;
+       boolean_t       check_ready = FALSE;
+       uint32_t        try_failed_count = 0;
 
        /*
 
        /*
-        *      Prevent collapse or termination by holding a paging reference
+        *      If a copy is already in progress, wait and retry.
+        *
+        *      XXX
+        *      Consider making this call interruptable, as Mike
+        *      intended it to be.
+        *
+        *      XXXO
+        *      Need a counter or version or something to allow
+        *      us to use the copy that the currently requesting
+        *      thread is obtaining -- is it worth adding to the
+        *      vm object structure? Depends how common this case it.
         */
         */
-
-       vm_object_paging_begin(object);
-       if (object->pager_created) {
-               /*
-                *      Someone else got to it first...
-                *      wait for them to finish initializing the ports
-                */
-               while (!object->pager_initialized) {
-                       vm_object_sleep(object,
-                                       VM_OBJECT_EVENT_INITIALIZED,
-                                       THREAD_UNINT);
-               }
-               vm_object_paging_end(object);
-               return;
+       copy_call_count++;
+       while (vm_object_wanted(src_object, VM_OBJECT_EVENT_COPY_CALL)) {
+               vm_object_sleep(src_object, VM_OBJECT_EVENT_COPY_CALL,
+                              THREAD_UNINT);
+               copy_call_restart_count++;
        }
 
        /*
        }
 
        /*
-        *      Indicate that a memory object has been assigned
-        *      before dropping the lock, to prevent a race.
+        *      Indicate (for the benefit of memory_object_create_copy)
+        *      that we want a copy for src_object. (Note that we cannot
+        *      do a real assert_wait before calling memory_object_copy,
+        *      so we simply set the flag.)
         */
 
         */
 
-       object->pager_created = TRUE;
-       object->paging_offset = 0;
-               
-#if    MACH_PAGEMAP
-       size = object->size;
-#endif /* MACH_PAGEMAP */
-       vm_object_unlock(object);
+       vm_object_set_wanted(src_object, VM_OBJECT_EVENT_COPY_CALL);
+       vm_object_unlock(src_object);
 
 
-#if    MACH_PAGEMAP
-       map = vm_external_create(size);
-       vm_object_lock(object);
-       assert(object->size == size);
-       object->existence_map = map;
-       vm_object_unlock(object);
-#endif /* MACH_PAGEMAP */
+       /*
+        *      Ask the memory manager to give us a memory object
+        *      which represents a copy of the src object.
+        *      The memory manager may give us a memory object
+        *      which we already have, or it may give us a
+        *      new memory object. This memory object will arrive
+        *      via memory_object_create_copy.
+        */
 
 
-       if ((uint32_t) object->size != object->size) {
-               panic("vm_object_pager_create(): object size 0x%llx >= 4GB\n",
-                     (uint64_t) object->size);
+       kr = KERN_FAILURE;      /* XXX need to change memory_object.defs */
+       if (kr != KERN_SUCCESS) {
+               return kr;
        }
 
        /*
        }
 
        /*
-        *      Create the [internal] pager, and associate it with this object.
-        *
-        *      We make the association here so that vm_object_enter()
-        *      can look up the object to complete initializing it.  No
-        *      user will ever map this object.
-        */
-       {
-               memory_object_default_t         dmm;
-
-               /* acquire a reference for the default memory manager */
-               dmm = memory_manager_default_reference();
-
-               assert(object->temporary);
-
-               /* create our new memory object */
-               assert((vm_size_t) object->size == object->size);
-               (void) memory_object_create(dmm, (vm_size_t) object->size,
-                                           &pager);
-
-               memory_object_default_deallocate(dmm);
-       }
-
-       entry = vm_object_hash_entry_alloc(pager);
-
-       lck = vm_object_hash_lock_spin(pager);
-       vm_object_hash_insert(entry, object);
-       vm_object_hash_unlock(lck);
-
-       /*
-        *      A reference was returned by
-        *      memory_object_create(), and it is
-        *      copied by vm_object_enter().
+        *      Wait for the copy to arrive.
         */
         */
+       vm_object_lock(src_object);
+       while (vm_object_wanted(src_object, VM_OBJECT_EVENT_COPY_CALL)) {
+               vm_object_sleep(src_object, VM_OBJECT_EVENT_COPY_CALL,
+                              THREAD_UNINT);
+               copy_call_sleep_count++;
+       }
+Retry:
+       assert(src_object->copy != VM_OBJECT_NULL);
+       copy = src_object->copy;
+       if (!vm_object_lock_try(copy)) {
+               vm_object_unlock(src_object);
 
 
-       if (vm_object_enter(pager, object->size, TRUE, TRUE, FALSE) != object)
-               panic("vm_object_pager_create: mismatch");
+               try_failed_count++;
+               mutex_pause(try_failed_count);  /* wait a bit */
 
 
-       /*
-        *      Drop the reference we were passed.
-        */
-       memory_object_deallocate(pager);
+               vm_object_lock(src_object);
+               goto Retry;
+       }
+       if (copy->vo_size < src_offset+size)
+               copy->vo_size = src_offset+size;
 
 
-       vm_object_lock(object);
+       if (!copy->pager_ready)
+               check_ready = TRUE;
 
        /*
 
        /*
-        *      Release the paging reference
+        *      Return the copy.
         */
         */
-       vm_object_paging_end(object);
-}
-
-/*
- *     Routine:        vm_object_remove
- *     Purpose:
- *             Eliminate the pager/object association
- *             for this pager.
- *     Conditions:
- *             The object cache must be locked.
- */
-__private_extern__ void
-vm_object_remove(
-       vm_object_t     object)
-{
-       memory_object_t pager;
-
-       if ((pager = object->pager) != MEMORY_OBJECT_NULL) {
-               vm_object_hash_entry_t  entry;
+       *_result_object = copy;
+       vm_object_unlock(copy);
+       vm_object_unlock(src_object);
 
 
-               entry = vm_object_hash_lookup(pager, FALSE);
-               if (entry != VM_OBJECT_HASH_ENTRY_NULL)
-                       entry->object = VM_OBJECT_NULL;
+       /* Wait for the copy to be ready. */
+       if (check_ready == TRUE) {
+               vm_object_lock(copy);
+               while (!copy->pager_ready) {
+                       vm_object_sleep(copy, VM_OBJECT_EVENT_PAGER_READY, THREAD_UNINT);
+               }
+               vm_object_unlock(copy);
        }
 
        }
 
+       return KERN_SUCCESS;
 }
 
 }
 
-/*
- *     Global variables for vm_object_collapse():
- *
- *             Counts for normal collapses and bypasses.
- *             Debugging variables, to watch or disable collapse.
- */
-static long    object_collapses = 0;
-static long    object_bypasses  = 0;
-
-static boolean_t       vm_object_collapse_allowed = TRUE;
-static boolean_t       vm_object_bypass_allowed = TRUE;
-
-#if MACH_PAGEMAP
-static int     vm_external_discarded;
-static int     vm_external_collapsed;
-#endif
-
-unsigned long vm_object_collapse_encrypted = 0;
+static int copy_delayed_lock_collisions = 0;
+static int copy_delayed_max_collisions = 0;
+static int copy_delayed_lock_contention = 0;
+static int copy_delayed_protect_iterate = 0;
 
 /*
 
 /*
- *     Routine:        vm_object_do_collapse
- *     Purpose:
- *             Collapse an object with the object backing it.
- *             Pages in the backing object are moved into the
- *             parent, and the backing object is deallocated.
- *     Conditions:
- *             Both objects and the cache are locked; the page
- *             queues are unlocked.
+ *     Routine:        vm_object_copy_delayed [internal]
+ *
+ *     Description:
+ *             Copy the specified virtual memory object, using
+ *             the asymmetric copy-on-write algorithm.
+ *
+ *     In/out conditions:
+ *             The src_object must be locked on entry.  It will be unlocked
+ *             on exit - so the caller must also hold a reference to it.
  *
  *
+ *             This routine will not block waiting for user-generated
+ *             events.  It is not interruptible.
  */
  */
-static void
-vm_object_do_collapse(
-       vm_object_t object,
-       vm_object_t backing_object)
+__private_extern__ vm_object_t
+vm_object_copy_delayed(
+       vm_object_t             src_object,
+       vm_object_offset_t      src_offset,
+       vm_object_size_t        size,
+       boolean_t               src_object_shared)
 {
 {
-       vm_page_t p, pp;
-       vm_object_offset_t new_offset, backing_offset;
-       vm_object_size_t size;
+       vm_object_t             new_copy = VM_OBJECT_NULL;
+       vm_object_t             old_copy;
+       vm_page_t               p;
+       vm_object_size_t        copy_size = src_offset + size;
+       pmap_flush_context      pmap_flush_context_storage;
+       boolean_t               delayed_pmap_flush = FALSE;
 
 
-       vm_object_lock_assert_exclusive(object);
-       vm_object_lock_assert_exclusive(backing_object);
 
 
-       backing_offset = object->shadow_offset;
-       size = object->size;
+       int collisions = 0;
+       /*
+        *      The user-level memory manager wants to see all of the changes
+        *      to this object, but it has promised not to make any changes on
+        *      its own.
+        *
+        *      Perform an asymmetric copy-on-write, as follows:
+        *              Create a new object, called a "copy object" to hold
+        *               pages modified by the new mapping  (i.e., the copy,
+        *               not the original mapping).
+        *              Record the original object as the backing object for
+        *               the copy object.  If the original mapping does not
+        *               change a page, it may be used read-only by the copy.
+        *              Record the copy object in the original object.
+        *               When the original mapping causes a page to be modified,
+        *               it must be copied to a new page that is "pushed" to
+        *               the copy object.
+        *              Mark the new mapping (the copy object) copy-on-write.
+        *               This makes the copy object itself read-only, allowing
+        *               it to be reused if the original mapping makes no
+        *               changes, and simplifying the synchronization required
+        *               in the "push" operation described above.
+        *
+        *      The copy-on-write is said to be assymetric because the original
+        *      object is *not* marked copy-on-write. A copied page is pushed
+        *      to the copy object, regardless which party attempted to modify
+        *      the page.
+        *
+        *      Repeated asymmetric copy operations may be done. If the
+        *      original object has not been changed since the last copy, its
+        *      copy object can be reused. Otherwise, a new copy object can be
+        *      inserted between the original object and its previous copy
+        *      object.  Since any copy object is read-only, this cannot affect
+        *      affect the contents of the previous copy object.
+        *
+        *      Note that a copy object is higher in the object tree than the
+        *      original object; therefore, use of the copy object recorded in
+        *      the original object must be done carefully, to avoid deadlock.
+        */
 
 
+       copy_size = vm_object_round_page(copy_size);
+ Retry:
        /*
        /*
-        *      Move all in-memory pages from backing_object
-        *      to the parent.  Pages that have been paged out
-        *      will be overwritten by any of the parent's
-        *      pages that shadow them.
+        * Wait for paging in progress.
         */
         */
-       
-       while (!queue_empty(&backing_object->memq)) {
-               
-               p = (vm_page_t) queue_first(&backing_object->memq);
-               
-               new_offset = (p->offset - backing_offset);
-               
-               assert(!p->busy || p->absent);
+       if (!src_object->true_share &&
+           (src_object->paging_in_progress != 0 ||
+            src_object->activity_in_progress != 0)) {
+               if (src_object_shared == TRUE) {
+                       vm_object_unlock(src_object);
+                       vm_object_lock(src_object);
+                       src_object_shared = FALSE;
+                       goto Retry;
+               }
+               vm_object_paging_wait(src_object, THREAD_UNINT);
+       }
+       /*
+        *      See whether we can reuse the result of a previous
+        *      copy operation.
+        */
+
+       old_copy = src_object->copy;
+       if (old_copy != VM_OBJECT_NULL) {
+               int lock_granted;
 
                /*
 
                /*
-                *      If the parent has a page here, or if
-                *      this page falls outside the parent,
-                *      dispose of it.
-                *
-                *      Otherwise, move it as planned.
+                *      Try to get the locks (out of order)
                 */
                 */
-               
-               if (p->offset < backing_offset || new_offset >= size) {
-                       VM_PAGE_FREE(p);
-               } else {
-                       /*
-                        * ENCRYPTED SWAP:
-                        * The encryption key includes the "pager" and the
-                        * "paging_offset".  These will not change during the 
-                        * object collapse, so we can just move an encrypted
-                        * page from one object to the other in this case.
-                        * We can't decrypt the page here, since we can't drop
-                        * the object lock.
-                        */
-                       if (p->encrypted) {
-                               vm_object_collapse_encrypted++;
-                       }
-                       pp = vm_page_lookup(object, new_offset);
-                       if (pp == VM_PAGE_NULL) {
+               if (src_object_shared == TRUE)
+                       lock_granted = vm_object_lock_try_shared(old_copy);
+               else
+                       lock_granted = vm_object_lock_try(old_copy);
 
 
-                               /*
-                                *      Parent now has no page.
-                                *      Move the backing object's page up.
-                                */
+               if (!lock_granted) {
+                       vm_object_unlock(src_object);
 
 
-                               vm_page_rename(p, object, new_offset, TRUE);
-#if    MACH_PAGEMAP
-                       } else if (pp->absent) {
+                       if (collisions++ == 0)
+                               copy_delayed_lock_contention++;
+                       mutex_pause(collisions);
 
 
-                               /*
-                                *      Parent has an absent page...
-                                *      it's not being paged in, so
-                                *      it must really be missing from
-                                *      the parent.
-                                *
-                                *      Throw out the absent page...
-                                *      any faults looking for that
-                                *      page will restart with the new
-                                *      one.
-                                */
+                       /* Heisenberg Rules */
+                       copy_delayed_lock_collisions++;
 
 
-                               VM_PAGE_FREE(pp);
-                               vm_page_rename(p, object, new_offset, TRUE);
-#endif /* MACH_PAGEMAP */
-                       } else {
-                               assert(! pp->absent);
+                       if (collisions > copy_delayed_max_collisions)
+                               copy_delayed_max_collisions = collisions;
 
 
-                               /*
-                                *      Parent object has a real page.
-                                *      Throw away the backing object's
-                                *      page.
-                                */
-                               VM_PAGE_FREE(p);
-                       }
-               }
-       }
-       
-#if    !MACH_PAGEMAP
-       assert((!object->pager_created && (object->pager == MEMORY_OBJECT_NULL))
-               || (!backing_object->pager_created
-               &&  (backing_object->pager == MEMORY_OBJECT_NULL)));
-#else 
-        assert(!object->pager_created && object->pager == MEMORY_OBJECT_NULL);
-#endif /* !MACH_PAGEMAP */
+                       if (src_object_shared == TRUE)
+                               vm_object_lock_shared(src_object);
+                       else
+                               vm_object_lock(src_object);
 
 
-       if (backing_object->pager != MEMORY_OBJECT_NULL) {
-               vm_object_hash_entry_t  entry;
+                       goto Retry;
+               }
 
                /*
 
                /*
-                *      Move the pager from backing_object to object.
-                *
-                *      XXX We're only using part of the paging space
-                *      for keeps now... we ought to discard the
-                *      unused portion.
+                *      Determine whether the old copy object has
+                *      been modified.
                 */
 
                 */
 
-               assert(!object->paging_in_progress);
-               assert(!object->activity_in_progress);
-               object->pager = backing_object->pager;
+               if (old_copy->resident_page_count == 0 &&
+                   !old_copy->pager_created) {
+                       /*
+                        *      It has not been modified.
+                        *
+                        *      Return another reference to
+                        *      the existing copy-object if
+                        *      we can safely grow it (if
+                        *      needed).
+                        */
 
 
-               if (backing_object->hashed) {
-                       lck_mtx_t       *lck;
+                       if (old_copy->vo_size < copy_size) {
+                               if (src_object_shared == TRUE) {
+                                       vm_object_unlock(old_copy);
+                                       vm_object_unlock(src_object);
+                               
+                                       vm_object_lock(src_object);
+                                       src_object_shared = FALSE;
+                                       goto Retry;
+                               }
+                               /*
+                                * We can't perform a delayed copy if any of the
+                                * pages in the extended range are wired (because
+                                * we can't safely take write permission away from
+                                * wired pages).  If the pages aren't wired, then
+                                * go ahead and protect them.
+                                */
+                               copy_delayed_protect_iterate++;
 
 
-                       lck = vm_object_hash_lock_spin(backing_object->pager);
-                       entry = vm_object_hash_lookup(object->pager, FALSE);
-                       assert(entry != VM_OBJECT_HASH_ENTRY_NULL);
-                       entry->object = object;
-                       vm_object_hash_unlock(lck);
+                               pmap_flush_context_init(&pmap_flush_context_storage);
+                               delayed_pmap_flush = FALSE;
 
 
-                       object->hashed = TRUE;
+                               vm_page_queue_iterate(&src_object->memq, p, vm_page_t, listq) {
+                                       if (!p->fictitious && 
+                                           p->offset >= old_copy->vo_size && 
+                                           p->offset < copy_size) {
+                                               if (VM_PAGE_WIRED(p)) {
+                                                       vm_object_unlock(old_copy);
+                                                       vm_object_unlock(src_object);
+
+                                                       if (new_copy != VM_OBJECT_NULL) {
+                                                               vm_object_unlock(new_copy);
+                                                               vm_object_deallocate(new_copy);
+                                                       }
+                                                       if (delayed_pmap_flush == TRUE)
+                                                               pmap_flush(&pmap_flush_context_storage);
+
+                                                       return VM_OBJECT_NULL;
+                                               } else {
+                                                       pmap_page_protect_options(VM_PAGE_GET_PHYS_PAGE(p), (VM_PROT_ALL & ~VM_PROT_WRITE),
+                                                                                 PMAP_OPTIONS_NOFLUSH, (void *)&pmap_flush_context_storage);
+                                                       delayed_pmap_flush = TRUE;
+                                               }
+                                       }
+                               }
+                               if (delayed_pmap_flush == TRUE)
+                                       pmap_flush(&pmap_flush_context_storage);
+
+                               old_copy->vo_size = copy_size;
+                       }
+                       if (src_object_shared == TRUE)
+                               vm_object_reference_shared(old_copy);
+                       else
+                               vm_object_reference_locked(old_copy);
+                       vm_object_unlock(old_copy);
+                       vm_object_unlock(src_object);
+
+                       if (new_copy != VM_OBJECT_NULL) {
+                               vm_object_unlock(new_copy);
+                               vm_object_deallocate(new_copy);
+                       }
+                       return(old_copy);
                }
                }
-               object->pager_created = backing_object->pager_created;
-               object->pager_control = backing_object->pager_control;
-               object->pager_ready = backing_object->pager_ready;
-               object->pager_initialized = backing_object->pager_initialized;
-               object->paging_offset =
-                   backing_object->paging_offset + backing_offset;
-               if (object->pager_control != MEMORY_OBJECT_CONTROL_NULL) {
-                       memory_object_control_collapse(object->pager_control,
-                                                      object);
+               
+               
+
+               /*
+                * Adjust the size argument so that the newly-created 
+                * copy object will be large enough to back either the
+                * old copy object or the new mapping.
+                */
+               if (old_copy->vo_size > copy_size)
+                       copy_size = old_copy->vo_size;
+
+               if (new_copy == VM_OBJECT_NULL) {
+                       vm_object_unlock(old_copy);
+                       vm_object_unlock(src_object);
+                       new_copy = vm_object_allocate(copy_size);
+                       vm_object_lock(src_object);
+                       vm_object_lock(new_copy);
+
+                       src_object_shared = FALSE;
+                       goto Retry;
                }
                }
-       }
+               new_copy->vo_size = copy_size;  
 
 
-#if    MACH_PAGEMAP
-       /*
-        *      If the shadow offset is 0, the use the existence map from
-        *      the backing object if there is one. If the shadow offset is
-        *      not zero, toss it.
-        *
-        *      XXX - If the shadow offset is not 0 then a bit copy is needed
-        *      if the map is to be salvaged.  For now, we just just toss the
-        *      old map, giving the collapsed object no map. This means that
-        *      the pager is invoked for zero fill pages.  If analysis shows
-        *      that this happens frequently and is a performance hit, then
-        *      this code should be fixed to salvage the map.
-        */
-       assert(object->existence_map == VM_EXTERNAL_NULL);
-       if (backing_offset || (size != backing_object->size)) {
-               vm_external_discarded++;
-               vm_external_destroy(backing_object->existence_map,
-                       backing_object->size);
-       }
-       else {
-               vm_external_collapsed++;
-               object->existence_map = backing_object->existence_map;
-       }
-       backing_object->existence_map = VM_EXTERNAL_NULL;
-#endif /* MACH_PAGEMAP */
+               /*
+                *      The copy-object is always made large enough to
+                *      completely shadow the original object, since
+                *      it may have several users who want to shadow
+                *      the original object at different points.
+                */
 
 
-       /*
-        *      Object now shadows whatever backing_object did.
-        *      Note that the reference to backing_object->shadow
-        *      moves from within backing_object to within object.
-        */
-       
-       assert(!object->phys_contiguous);
-       assert(!backing_object->phys_contiguous);
-       object->shadow = backing_object->shadow;
-       if (object->shadow) {
-               object->shadow_offset += backing_object->shadow_offset;
-       } else {
-               /* no shadow, therefore no shadow offset... */
-               object->shadow_offset = 0;
+               assert((old_copy->shadow == src_object) &&
+                   (old_copy->vo_shadow_offset == (vm_object_offset_t) 0));
+
+       } else if (new_copy == VM_OBJECT_NULL) {
+               vm_object_unlock(src_object);
+               new_copy = vm_object_allocate(copy_size);
+               vm_object_lock(src_object);
+               vm_object_lock(new_copy);
+
+               src_object_shared = FALSE;
+               goto Retry;
        }
        }
-       assert((object->shadow == VM_OBJECT_NULL) ||
-              (object->shadow->copy != backing_object));
 
        /*
 
        /*
-        *      Discard backing_object.
+        * We now have the src object locked, and the new copy object
+        * allocated and locked (and potentially the old copy locked).
+        * Before we go any further, make sure we can still perform
+        * a delayed copy, as the situation may have changed.
         *
         *
-        *      Since the backing object has no pages, no
-        *      pager left, and no object references within it,
-        *      all that is necessary is to dispose of it.
+        * Specifically, we can't perform a delayed copy if any of the
+        * pages in the range are wired (because we can't safely take
+        * write permission away from wired pages).  If the pages aren't
+        * wired, then go ahead and protect them.
         */
         */
-       
-       assert((backing_object->ref_count == 1) &&
-              (backing_object->resident_page_count == 0) &&
-              (backing_object->paging_in_progress == 0) &&
-              (backing_object->activity_in_progress == 0));
+       copy_delayed_protect_iterate++;
 
 
-       backing_object->alive = FALSE;
-       vm_object_unlock(backing_object);
+       pmap_flush_context_init(&pmap_flush_context_storage);
+       delayed_pmap_flush = FALSE;
 
 
-       XPR(XPR_VM_OBJECT, "vm_object_collapse, collapsed 0x%X\n",
-               backing_object, 0,0,0,0);
+       vm_page_queue_iterate(&src_object->memq, p, vm_page_t, listq) {
+               if (!p->fictitious && p->offset < copy_size) {
+                       if (VM_PAGE_WIRED(p)) {
+                               if (old_copy)
+                                       vm_object_unlock(old_copy);
+                               vm_object_unlock(src_object);
+                               vm_object_unlock(new_copy);
+                               vm_object_deallocate(new_copy);
 
 
-       vm_object_lock_destroy(backing_object);
+                               if (delayed_pmap_flush == TRUE)
+                                       pmap_flush(&pmap_flush_context_storage);
 
 
-       zfree(vm_object_zone, backing_object);
-       
-       object_collapses++;
-}
+                               return VM_OBJECT_NULL;
+                       } else {
+                               pmap_page_protect_options(VM_PAGE_GET_PHYS_PAGE(p), (VM_PROT_ALL & ~VM_PROT_WRITE),
+                                                         PMAP_OPTIONS_NOFLUSH, (void *)&pmap_flush_context_storage);
+                               delayed_pmap_flush = TRUE;
+                       }
+               }
+       }
+       if (delayed_pmap_flush == TRUE)
+               pmap_flush(&pmap_flush_context_storage);
 
 
-static void
-vm_object_do_bypass(
-       vm_object_t object,
-       vm_object_t backing_object)
-{
-       /*
-        *      Make the parent shadow the next object
-        *      in the chain.
-        */
-       
-       vm_object_lock_assert_exclusive(object);
-       vm_object_lock_assert_exclusive(backing_object);
+       if (old_copy != VM_OBJECT_NULL) {
+               /*
+                *      Make the old copy-object shadow the new one.
+                *      It will receive no more pages from the original
+                *      object.
+                */
 
 
-#if    TASK_SWAPPER
-       /*
-        *      Do object reference in-line to 
-        *      conditionally increment shadow's
-        *      residence count.  If object is not
-        *      resident, leave residence count
-        *      on shadow alone.
-        */
-       if (backing_object->shadow != VM_OBJECT_NULL) {
-               vm_object_lock(backing_object->shadow);
-               vm_object_lock_assert_exclusive(backing_object->shadow);
-               backing_object->shadow->ref_count++;
-               if (object->res_count != 0)
-                       vm_object_res_reference(backing_object->shadow);
-               vm_object_unlock(backing_object->shadow);
-       }
-#else  /* TASK_SWAPPER */
-       vm_object_reference(backing_object->shadow);
-#endif /* TASK_SWAPPER */
+               /* remove ref. from old_copy */
+               vm_object_lock_assert_exclusive(src_object);
+               src_object->ref_count--;
+               assert(src_object->ref_count > 0);
+               vm_object_lock_assert_exclusive(old_copy);
+               old_copy->shadow = new_copy;
+               vm_object_lock_assert_exclusive(new_copy);
+               assert(new_copy->ref_count > 0);
+               new_copy->ref_count++;          /* for old_copy->shadow ref. */
 
 
-       assert(!object->phys_contiguous);
-       assert(!backing_object->phys_contiguous);
-       object->shadow = backing_object->shadow;
-       if (object->shadow) {
-               object->shadow_offset += backing_object->shadow_offset;
-       } else {
-               /* no shadow, therefore no shadow offset... */
-               object->shadow_offset = 0;
+#if TASK_SWAPPER
+               if (old_copy->res_count) {
+                       VM_OBJ_RES_INCR(new_copy);
+                       VM_OBJ_RES_DECR(src_object);
+               }
+#endif
+
+               vm_object_unlock(old_copy);     /* done with old_copy */
        }
        }
-       
+
        /*
        /*
-        *      Backing object might have had a copy pointer
-        *      to us.  If it did, clear it. 
+        *      Point the new copy at the existing object.
         */
         */
-       if (backing_object->copy == object) {
-               backing_object->copy = VM_OBJECT_NULL;
-       }
-       
-       /*
-        *      Drop the reference count on backing_object.
-#if    TASK_SWAPPER
-        *      Since its ref_count was at least 2, it
-        *      will not vanish; so we don't need to call
-        *      vm_object_deallocate.
-        *      [with a caveat for "named" objects]
-        * 
-        *      The res_count on the backing object is
-        *      conditionally decremented.  It's possible
-        *      (via vm_pageout_scan) to get here with
-        *      a "swapped" object, which has a 0 res_count,
-        *      in which case, the backing object res_count
-        *      is already down by one.
-#else
-        *      Don't call vm_object_deallocate unless
-        *      ref_count drops to zero.
-        *
-        *      The ref_count can drop to zero here if the
-        *      backing object could be bypassed but not
-        *      collapsed, such as when the backing object
-        *      is temporary and cachable.
-#endif
-        */
-       if (backing_object->ref_count > 2 ||
-           (!backing_object->named && backing_object->ref_count > 1)) {
-               vm_object_lock_assert_exclusive(backing_object);
-               backing_object->ref_count--;
-#if    TASK_SWAPPER
-               if (object->res_count != 0)
-                       vm_object_res_deallocate(backing_object);
-               assert(backing_object->ref_count > 0);
-#endif /* TASK_SWAPPER */
-               vm_object_unlock(backing_object);
-       } else {
-
-               /*
-                *      Drop locks so that we can deallocate
-                *      the backing object.
-                */
+       vm_object_lock_assert_exclusive(new_copy);
+       new_copy->shadow = src_object;
+       new_copy->vo_shadow_offset = 0;
+       new_copy->shadowed = TRUE;      /* caller must set needs_copy */
 
 
-#if    TASK_SWAPPER
-               if (object->res_count == 0) {
-                       /* XXX get a reference for the deallocate below */
-                       vm_object_res_reference(backing_object);
-               }
-#endif /* TASK_SWAPPER */
-               vm_object_unlock(object);
-               vm_object_unlock(backing_object);
-               vm_object_deallocate(backing_object);
+       vm_object_lock_assert_exclusive(src_object);
+       vm_object_reference_locked(src_object);
+       src_object->copy = new_copy;
+       vm_object_unlock(src_object);
+       vm_object_unlock(new_copy);
 
 
-               /*
-                *      Relock object. We don't have to reverify
-                *      its state since vm_object_collapse will
-                *      do that for us as it starts at the
-                *      top of its loop.
-                */
+       XPR(XPR_VM_OBJECT,
+               "vm_object_copy_delayed: used copy object %X for source %X\n",
+               new_copy, src_object, 0, 0, 0);
 
 
-               vm_object_lock(object);
-       }
-       
-       object_bypasses++;
+       return new_copy;
 }
 
 }
 
-               
 /*
 /*
- *     vm_object_collapse:
- *
- *     Perform an object collapse or an object bypass if appropriate.
- *     The real work of collapsing and bypassing is performed in
- *     the routines vm_object_do_collapse and vm_object_do_bypass.
- *
- *     Requires that the object be locked and the page queues be unlocked.
+ *     Routine:        vm_object_copy_strategically
  *
  *
+ *     Purpose:
+ *             Perform a copy according to the source object's
+ *             declared strategy.  This operation may block,
+ *             and may be interrupted.
  */
  */
-static unsigned long vm_object_collapse_calls = 0;
-static unsigned long vm_object_collapse_objects = 0;
-static unsigned long vm_object_collapse_do_collapse = 0;
-static unsigned long vm_object_collapse_do_bypass = 0;
-static unsigned long vm_object_collapse_delays = 0;
-__private_extern__ void
-vm_object_collapse(
-       register vm_object_t                    object,
-       register vm_object_offset_t             hint_offset,
-       boolean_t                               can_bypass)
+__private_extern__ kern_return_t
+vm_object_copy_strategically(
+       vm_object_t             src_object,
+       vm_object_offset_t      src_offset,
+       vm_object_size_t        size,
+       vm_object_t             *dst_object,    /* OUT */
+       vm_object_offset_t      *dst_offset,    /* OUT */
+       boolean_t               *dst_needs_copy) /* OUT */
 {
 {
-       register vm_object_t                    backing_object;
-       register unsigned int                   rcount;
-       register unsigned int                   size;
-       vm_object_t                             original_object;
-       int                                     object_lock_type;
-       int                                     backing_object_lock_type;
-
-       vm_object_collapse_calls++;
-
-       if (! vm_object_collapse_allowed &&
-           ! (can_bypass && vm_object_bypass_allowed)) {
-               return;
-       }
+       boolean_t       result;
+       boolean_t       interruptible = THREAD_ABORTSAFE; /* XXX */
+       boolean_t       object_lock_shared = FALSE;
+       memory_object_copy_strategy_t copy_strategy;
 
 
-       XPR(XPR_VM_OBJECT, "vm_object_collapse, obj 0x%X\n", 
-               object, 0,0,0,0);
+       assert(src_object != VM_OBJECT_NULL);
 
 
-       if (object == VM_OBJECT_NULL)
-               return;
+       copy_strategy = src_object->copy_strategy;
 
 
-       original_object = object;
+       if (copy_strategy == MEMORY_OBJECT_COPY_DELAY) {
+               vm_object_lock_shared(src_object);
+               object_lock_shared = TRUE;
+       } else
+               vm_object_lock(src_object);
 
        /*
 
        /*
-        * The top object was locked "exclusive" by the caller.
-        * In the first pass, to determine if we can collapse the shadow chain,
-        * take a "shared" lock on the shadow objects.  If we can collapse,
-        * we'll have to go down the chain again with exclusive locks.
+        *      The copy strategy is only valid if the memory manager
+        *      is "ready". Internal objects are always ready.
         */
         */
-       object_lock_type = OBJECT_LOCK_EXCLUSIVE;
-       backing_object_lock_type = OBJECT_LOCK_SHARED;
-
-retry:
-       object = original_object;
-       vm_object_lock_assert_exclusive(object);
 
 
-       while (TRUE) {
-               vm_object_collapse_objects++;
-               /*
-                *      Verify that the conditions are right for either
-                *      collapse or bypass:
-                */
+       while (!src_object->internal && !src_object->pager_ready) {
+               wait_result_t wait_result;
 
 
-               /*
-                *      There is a backing object, and
-                */
-       
-               backing_object = object->shadow;
-               if (backing_object == VM_OBJECT_NULL) {
-                       if (object != original_object) {
-                               vm_object_unlock(object);
-                       }
-                       return;
+               if (object_lock_shared == TRUE) {
+                       vm_object_unlock(src_object);
+                       vm_object_lock(src_object);
+                       object_lock_shared = FALSE;
+                       continue;
                }
                }
-               if (backing_object_lock_type == OBJECT_LOCK_SHARED) {
-                       vm_object_lock_shared(backing_object);
-               } else {
-                       vm_object_lock(backing_object);
+               wait_result = vm_object_sleep(  src_object,
+                                               VM_OBJECT_EVENT_PAGER_READY,
+                                               interruptible);
+               if (wait_result != THREAD_AWAKENED) {
+                       vm_object_unlock(src_object);
+                       *dst_object = VM_OBJECT_NULL;
+                       *dst_offset = 0;
+                       *dst_needs_copy = FALSE;
+                       return(MACH_SEND_INTERRUPTED);
                }
                }
+       }
 
 
-               /*
-                *      No pages in the object are currently
-                *      being paged out, and
-                */
-               if (object->paging_in_progress != 0 ||
-                   object->activity_in_progress != 0) {
-                       /* try and collapse the rest of the shadow chain */
-                       if (object != original_object) {
-                               vm_object_unlock(object);
-                       }
-                       object = backing_object;
-                       object_lock_type = backing_object_lock_type;
-                       continue;
+       /*
+        *      Use the appropriate copy strategy.
+        */
+
+       switch (copy_strategy) {
+           case MEMORY_OBJECT_COPY_DELAY:
+               *dst_object = vm_object_copy_delayed(src_object,
+                                                    src_offset, size, object_lock_shared);
+               if (*dst_object != VM_OBJECT_NULL) {
+                       *dst_offset = src_offset;
+                       *dst_needs_copy = TRUE;
+                       result = KERN_SUCCESS;
+                       break;
                }
                }
+               vm_object_lock(src_object);
+               /* fall thru when delayed copy not allowed */
 
 
-               /*
-                *      ...
-                *              The backing object is not read_only,
-                *              and no pages in the backing object are
-                *              currently being paged out.
-                *              The backing object is internal.
-                *
-                */
-       
-               if (!backing_object->internal ||
-                   backing_object->paging_in_progress != 0 ||
-                   backing_object->activity_in_progress != 0) {
-                       /* try and collapse the rest of the shadow chain */
-                       if (object != original_object) {
-                               vm_object_unlock(object);
-                       }
-                       object = backing_object;
-                       object_lock_type = backing_object_lock_type;
-                       continue;
+           case MEMORY_OBJECT_COPY_NONE:
+               result = vm_object_copy_slowly(src_object, src_offset, size,
+                                              interruptible, dst_object);
+               if (result == KERN_SUCCESS) {
+                       *dst_offset = 0;
+                       *dst_needs_copy = FALSE;
                }
                }
-       
-               /*
-                *      The backing object can't be a copy-object:
-                *      the shadow_offset for the copy-object must stay
-                *      as 0.  Furthermore (for the 'we have all the
-                *      pages' case), if we bypass backing_object and
-                *      just shadow the next object in the chain, old
-                *      pages from that object would then have to be copied
-                *      BOTH into the (former) backing_object and into the
-                *      parent object.
-                */
-               if (backing_object->shadow != VM_OBJECT_NULL &&
-                   backing_object->shadow->copy == backing_object) {
-                       /* try and collapse the rest of the shadow chain */
-                       if (object != original_object) {
-                               vm_object_unlock(object);
-                       }
-                       object = backing_object;
-                       object_lock_type = backing_object_lock_type;
-                       continue;
+               break;
+
+           case MEMORY_OBJECT_COPY_CALL:
+               result = vm_object_copy_call(src_object, src_offset, size,
+                               dst_object);
+               if (result == KERN_SUCCESS) {
+                       *dst_offset = src_offset;
+                       *dst_needs_copy = TRUE;
                }
                }
+               break;
 
 
-               /*
-                *      We can now try to either collapse the backing
-                *      object (if the parent is the only reference to
-                *      it) or (perhaps) remove the parent's reference
-                *      to it.
-                *
-                *      If there is exactly one reference to the backing
-                *      object, we may be able to collapse it into the
-                *      parent.
-                *
-                *      If MACH_PAGEMAP is defined:
-                *      The parent must not have a pager created for it,
-                *      since collapsing a backing_object dumps new pages
-                *      into the parent that its pager doesn't know about
-                *      (and the collapse code can't merge the existence
-                *      maps).
-                *      Otherwise:
-                *      As long as one of the objects is still not known
-                *      to the pager, we can collapse them.
-                */
-               if (backing_object->ref_count == 1 &&
-                   (!object->pager_created 
-#if    !MACH_PAGEMAP
-                    || !backing_object->pager_created
-#endif /*!MACH_PAGEMAP */
-                   ) && vm_object_collapse_allowed) {
-
-                       /*
-                        * We need the exclusive lock on the VM objects.
-                        */
-                       if (backing_object_lock_type != OBJECT_LOCK_EXCLUSIVE) {
-                               /*
-                                * We have an object and its shadow locked 
-                                * "shared".  We can't just upgrade the locks
-                                * to "exclusive", as some other thread might
-                                * also have these objects locked "shared" and
-                                * attempt to upgrade one or the other to 
-                                * "exclusive".  The upgrades would block
-                                * forever waiting for the other "shared" locks
-                                * to get released.
-                                * So we have to release the locks and go
-                                * down the shadow chain again (since it could
-                                * have changed) with "exclusive" locking.
-                                */
-                               vm_object_unlock(backing_object);
-                               if (object != original_object)
-                                       vm_object_unlock(object);
-                               object_lock_type = OBJECT_LOCK_EXCLUSIVE;
-                               backing_object_lock_type = OBJECT_LOCK_EXCLUSIVE;
-                               goto retry;
-                       }
-
-                       XPR(XPR_VM_OBJECT, 
-                  "vm_object_collapse: %x to %x, pager %x, pager_control %x\n",
-                               backing_object, object,
-                               backing_object->pager, 
-                               backing_object->pager_control, 0);
-
-                       /*
-                        *      Collapse the object with its backing
-                        *      object, and try again with the object's
-                        *      new backing object.
-                        */
-
-                       vm_object_do_collapse(object, backing_object);
-                       vm_object_collapse_do_collapse++;
-                       continue;
-               }
-
-               /*
-                *      Collapsing the backing object was not possible
-                *      or permitted, so let's try bypassing it.
-                */
+           case MEMORY_OBJECT_COPY_SYMMETRIC:
+               XPR(XPR_VM_OBJECT, "v_o_c_strategically obj 0x%x off 0x%x size 0x%x\n", src_object, src_offset, size, 0, 0);
+               vm_object_unlock(src_object);
+               result = KERN_MEMORY_RESTART_COPY;
+               break;
 
 
-               if (! (can_bypass && vm_object_bypass_allowed)) {
-                       /* try and collapse the rest of the shadow chain */
-                       if (object != original_object) {
-                               vm_object_unlock(object);
-                       }
-                       object = backing_object;
-                       object_lock_type = backing_object_lock_type;
-                       continue;
-               }
+           default:
+               panic("copy_strategically: bad strategy");
+               result = KERN_INVALID_ARGUMENT;
+       }
+       return(result);
+}
 
 
+/*
+ *     vm_object_shadow:
+ *
+ *     Create a new object which is backed by the
+ *     specified existing object range.  The source
+ *     object reference is deallocated.
+ *
+ *     The new object and offset into that object
+ *     are returned in the source parameters.
+ */
+boolean_t vm_object_shadow_check = TRUE;
 
 
-               /*
-                *      If the object doesn't have all its pages present,
-                *      we have to make sure no pages in the backing object
-                *      "show through" before bypassing it.
-                */
-               size = atop(object->size);
-               rcount = object->resident_page_count;
-               if (rcount != size) {
-                       vm_object_offset_t      offset;
-                       vm_object_offset_t      backing_offset;
-                       unsigned int            backing_rcount;
-                       unsigned int            lookups = 0;
+__private_extern__ boolean_t
+vm_object_shadow(
+       vm_object_t             *object,        /* IN/OUT */
+       vm_object_offset_t      *offset,        /* IN/OUT */
+       vm_object_size_t        length)
+{
+       vm_object_t     source;
+       vm_object_t     result;
 
 
-                       /*
-                        *      If the backing object has a pager but no pagemap,
-                        *      then we cannot bypass it, because we don't know
-                        *      what pages it has.
-                        */
-                       if (backing_object->pager_created
-#if    MACH_PAGEMAP
-                           && (backing_object->existence_map == VM_EXTERNAL_NULL)
-#endif /* MACH_PAGEMAP */
-                               ) {
-                               /* try and collapse the rest of the shadow chain */
-                               if (object != original_object) {
-                                       vm_object_unlock(object);
-                               }
-                               object = backing_object;
-                               object_lock_type = backing_object_lock_type;
-                               continue;
-                       }
+       source = *object;
+       assert(source != VM_OBJECT_NULL);
+       if (source == VM_OBJECT_NULL)
+               return FALSE;
 
 
-                       /*
-                        *      If the object has a pager but no pagemap,
-                        *      then we cannot bypass it, because we don't know
-                        *      what pages it has.
-                        */
-                       if (object->pager_created
-#if    MACH_PAGEMAP
-                           && (object->existence_map == VM_EXTERNAL_NULL)
-#endif /* MACH_PAGEMAP */
-                               ) {
-                               /* try and collapse the rest of the shadow chain */
-                               if (object != original_object) {
-                                       vm_object_unlock(object);
-                               }
-                               object = backing_object;
-                               object_lock_type = backing_object_lock_type;
-                               continue;
-                       }
+#if 0
+       /*
+        * XXX FBDP
+        * This assertion is valid but it gets triggered by Rosetta for example
+        * due to a combination of vm_remap() that changes a VM object's
+        * copy_strategy from SYMMETRIC to DELAY and vm_protect(VM_PROT_COPY)
+        * that then sets "needs_copy" on its map entry.  This creates a
+        * mapping situation that VM should never see and doesn't know how to
+        * handle.
+        * It's not clear if this can create any real problem but we should
+        * look into fixing this, probably by having vm_protect(VM_PROT_COPY)
+        * do more than just set "needs_copy" to handle the copy-on-write...
+        * In the meantime, let's disable the assertion.
+        */
+       assert(source->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC);
+#endif
 
 
-                       /*
-                        *      If all of the pages in the backing object are
-                        *      shadowed by the parent object, the parent
-                        *      object no longer has to shadow the backing
-                        *      object; it can shadow the next one in the
-                        *      chain.
-                        *
-                        *      If the backing object has existence info,
-                        *      we must check examine its existence info
-                        *      as well.
-                        *
-                        */
+       /*
+        *      Determine if we really need a shadow.
+        *
+        *      If the source object is larger than what we are trying
+        *      to create, then force the shadow creation even if the
+        *      ref count is 1.  This will allow us to [potentially]
+        *      collapse the underlying object away in the future
+        *      (freeing up the extra data it might contain and that
+        *      we don't need).
+        */
 
 
-                       backing_offset = object->shadow_offset;
-                       backing_rcount = backing_object->resident_page_count;
+       assert(source->copy_strategy != MEMORY_OBJECT_COPY_NONE); /* Purgeable objects shouldn't have shadow objects. */
 
 
-#if    MACH_PAGEMAP
-#define EXISTS_IN_OBJECT(obj, off, rc) \
-       (vm_external_state_get((obj)->existence_map, \
-        (vm_offset_t)(off)) == VM_EXTERNAL_STATE_EXISTS || \
-        ((rc) && ++lookups && vm_page_lookup((obj), (off)) != VM_PAGE_NULL && (rc)--))
-#else
-#define EXISTS_IN_OBJECT(obj, off, rc) \
-       (((rc) && ++lookups && vm_page_lookup((obj), (off)) != VM_PAGE_NULL && (rc)--))
-#endif /* MACH_PAGEMAP */
+       if (vm_object_shadow_check &&
+           source->vo_size == length &&
+           source->ref_count == 1 &&
+           (source->shadow == VM_OBJECT_NULL ||
+            source->shadow->copy == VM_OBJECT_NULL) )
+       {
+               /* lock the object and check again */
+               vm_object_lock(source);
+               if (source->vo_size == length &&
+                   source->ref_count == 1 &&
+                   (source->shadow == VM_OBJECT_NULL ||
+                    source->shadow->copy == VM_OBJECT_NULL))
+               {
+                       source->shadowed = FALSE;
+                       vm_object_unlock(source);
+                       return FALSE;
+               }
+               /* things changed while we were locking "source"... */
+               vm_object_unlock(source);
+       }
 
 
-                       /*
-                        * Check the hint location first
-                        * (since it is often the quickest way out of here).
-                        */
-                       if (object->cow_hint != ~(vm_offset_t)0)
-                               hint_offset = (vm_object_offset_t)object->cow_hint;
-                       else
-                               hint_offset = (hint_offset > 8 * PAGE_SIZE_64) ?
-                                             (hint_offset - 8 * PAGE_SIZE_64) : 0;
+       /*
+        *      Allocate a new object with the given length
+        */
 
 
-                       if (EXISTS_IN_OBJECT(backing_object, hint_offset +
-                                            backing_offset, backing_rcount) &&
-                           !EXISTS_IN_OBJECT(object, hint_offset, rcount)) {
-                               /* dependency right at the hint */
-                               object->cow_hint = (vm_offset_t) hint_offset; /* atomic */
-                               /* try and collapse the rest of the shadow chain */
-                               if (object != original_object) {
-                                       vm_object_unlock(object);
-                               }
-                               object = backing_object;
-                               object_lock_type = backing_object_lock_type;
-                               continue;
-                       }
+       if ((result = vm_object_allocate(length)) == VM_OBJECT_NULL)
+               panic("vm_object_shadow: no object for shadowing");
 
 
-                       /*
-                        * If the object's window onto the backing_object
-                        * is large compared to the number of resident
-                        * pages in the backing object, it makes sense to
-                        * walk the backing_object's resident pages first.
-                        *
-                        * NOTE: Pages may be in both the existence map and 
-                        * resident.  So, we can't permanently decrement
-                        * the rcount here because the second loop may
-                        * find the same pages in the backing object'
-                        * existence map that we found here and we would
-                        * double-decrement the rcount.  We also may or
-                        * may not have found the 
-                        */
-                       if (backing_rcount && 
-#if    MACH_PAGEMAP
-                           size > ((backing_object->existence_map) ?
-                            backing_rcount : (backing_rcount >> 1))
-#else
-                           size > (backing_rcount >> 1)
-#endif /* MACH_PAGEMAP */
-                               ) {
-                               unsigned int rc = rcount;
-                               vm_page_t p;
+       /*
+        *      The new object shadows the source object, adding
+        *      a reference to it.  Our caller changes his reference
+        *      to point to the new object, removing a reference to
+        *      the source object.  Net result: no change of reference
+        *      count.
+        */
+       result->shadow = source;
+       
+       /*
+        *      Store the offset into the source object,
+        *      and fix up the offset into the new object.
+        */
 
 
-                               backing_rcount = backing_object->resident_page_count;
-                               p = (vm_page_t)queue_first(&backing_object->memq);
-                               do {
-                                       /* Until we get more than one lookup lock */
-                                       if (lookups > 256) {
-                                               vm_object_collapse_delays++;
-                                               lookups = 0;
-                                               mutex_pause(0);
-                                       }
+       result->vo_shadow_offset = *offset;
 
 
-                                       offset = (p->offset - backing_offset);
-                                       if (offset < object->size &&
-                                           offset != hint_offset &&
-                                           !EXISTS_IN_OBJECT(object, offset, rc)) {
-                                               /* found a dependency */
-                                               object->cow_hint = (vm_offset_t) offset; /* atomic */
-                                               
-                                               break;
-                                       }
-                                       p = (vm_page_t) queue_next(&p->listq);
+       /*
+        *      Return the new things
+        */
 
 
-                               } while (--backing_rcount);
-                               if (backing_rcount != 0 ) {
-                                       /* try and collapse the rest of the shadow chain */
-                                       if (object != original_object) {
-                                               vm_object_unlock(object);
-                                       }
-                                       object = backing_object;
-                                       object_lock_type = backing_object_lock_type;
-                                       continue;
-                               }
-                       }
-
-                       /*
-                        * Walk through the offsets looking for pages in the
-                        * backing object that show through to the object.
-                        */
-                       if (backing_rcount
-#if MACH_PAGEMAP
-                           || backing_object->existence_map
-#endif /* MACH_PAGEMAP */
-                               ) {
-                               offset = hint_offset;
-                               
-                               while((offset =
-                                     (offset + PAGE_SIZE_64 < object->size) ?
-                                     (offset + PAGE_SIZE_64) : 0) != hint_offset) {
-
-                                       /* Until we get more than one lookup lock */
-                                       if (lookups > 256) {
-                                               vm_object_collapse_delays++;
-                                               lookups = 0;
-                                               mutex_pause(0);
-                                       }
-
-                                       if (EXISTS_IN_OBJECT(backing_object, offset +
-                                           backing_offset, backing_rcount) &&
-                                           !EXISTS_IN_OBJECT(object, offset, rcount)) {
-                                               /* found a dependency */
-                                               object->cow_hint = (vm_offset_t) offset; /* atomic */
-                                               break;
-                                       }
-                               }
-                               if (offset != hint_offset) {
-                                       /* try and collapse the rest of the shadow chain */
-                                       if (object != original_object) {
-                                               vm_object_unlock(object);
-                                       }
-                                       object = backing_object;
-                                       object_lock_type = backing_object_lock_type;
-                                       continue;
-                               }
-                       }
-               }
-
-               /*
-                * We need "exclusive" locks on the 2 VM objects.
-                */
-               if (backing_object_lock_type != OBJECT_LOCK_EXCLUSIVE) {
-                       vm_object_unlock(backing_object);
-                       if (object != original_object)
-                               vm_object_unlock(object);
-                       object_lock_type = OBJECT_LOCK_EXCLUSIVE;
-                       backing_object_lock_type = OBJECT_LOCK_EXCLUSIVE;
-                       goto retry;
-               }
-
-               /* reset the offset hint for any objects deeper in the chain */
-               object->cow_hint = (vm_offset_t)0;
-
-               /*
-                *      All interesting pages in the backing object
-                *      already live in the parent or its pager.
-                *      Thus we can bypass the backing object.
-                */
-
-               vm_object_do_bypass(object, backing_object);
-               vm_object_collapse_do_bypass++;
-
-               /*
-                *      Try again with this object's new backing object.
-                */
-
-               continue;
-       }
-
-       if (object != original_object) {
-               vm_object_unlock(object);
-       }
-}
+       *offset = 0;
+       *object = result;
+       return TRUE;
+}
 
 /*
 
 /*
- *     Routine:        vm_object_page_remove: [internal]
- *     Purpose:
- *             Removes all physical pages in the specified
- *             object range from the object's list of pages.
+ *     The relationship between vm_object structures and
+ *     the memory_object requires careful synchronization.
  *
  *
- *     In/out conditions:
- *             The object must be locked.
- *             The object must not have paging_in_progress, usually
- *             guaranteed by not having a pager.
- */
-unsigned int vm_object_page_remove_lookup = 0;
-unsigned int vm_object_page_remove_iterate = 0;
-
-__private_extern__ void
-vm_object_page_remove(
-       register vm_object_t            object,
-       register vm_object_offset_t     start,
-       register vm_object_offset_t     end)
-{
-       register vm_page_t      p, next;
-
-       /*
-        *      One and two page removals are most popular.
-        *      The factor of 16 here is somewhat arbitrary.
-        *      It balances vm_object_lookup vs iteration.
-        */
-
-       if (atop_64(end - start) < (unsigned)object->resident_page_count/16) {
-               vm_object_page_remove_lookup++;
-
-               for (; start < end; start += PAGE_SIZE_64) {
-                       p = vm_page_lookup(object, start);
-                       if (p != VM_PAGE_NULL) {
-                               assert(!p->cleaning && !p->pageout);
-                               if (!p->fictitious && p->pmapped)
-                                       pmap_disconnect(p->phys_page);
-                               VM_PAGE_FREE(p);
-                       }
-               }
-       } else {
-               vm_object_page_remove_iterate++;
-
-               p = (vm_page_t) queue_first(&object->memq);
-               while (!queue_end(&object->memq, (queue_entry_t) p)) {
-                       next = (vm_page_t) queue_next(&p->listq);
-                       if ((start <= p->offset) && (p->offset < end)) {
-                               assert(!p->cleaning && !p->pageout);
-                               if (!p->fictitious && p->pmapped)
-                                       pmap_disconnect(p->phys_page);
-                               VM_PAGE_FREE(p);
-                       }
-                       p = next;
-               }
-       }
-}
-
-
-/*
- *     Routine:        vm_object_coalesce
- *     Function:       Coalesces two objects backing up adjoining
- *                     regions of memory into a single object.
+ *     All associations are created by memory_object_create_named
+ *  for external pagers and vm_object_compressor_pager_create for internal
+ *  objects as follows:
  *
  *
- *     returns TRUE if objects were combined.
+ *             pager:  the memory_object itself, supplied by
+ *                     the user requesting a mapping (or the kernel,
+ *                     when initializing internal objects); the
+ *                     kernel simulates holding send rights by keeping
+ *                     a port reference;
  *
  *
- *     NOTE:   Only works at the moment if the second object is NULL -
- *             if it's not, which object do we lock first?
+ *             pager_request:
+ *                     the memory object control port,
+ *                     created by the kernel; the kernel holds
+ *                     receive (and ownership) rights to this
+ *                     port, but no other references.
  *
  *
- *     Parameters:
- *             prev_object     First object to coalesce
- *             prev_offset     Offset into prev_object
- *             next_object     Second object into coalesce
- *             next_offset     Offset into next_object
+ *     When initialization is complete, the "initialized" field
+ *     is asserted.  Other mappings using a particular memory object,
+ *     and any references to the vm_object gained through the
+ *     port association must wait for this initialization to occur.
  *
  *
- *             prev_size       Size of reference to prev_object
- *             next_size       Size of reference to next_object
+ *     In order to allow the memory manager to set attributes before
+ *     requests (notably virtual copy operations, but also data or
+ *     unlock requests) are made, a "ready" attribute is made available.
+ *     Only the memory manager may affect the value of this attribute.
+ *     Its value does not affect critical kernel functions, such as
+ *     internal object initialization or destruction.  [Furthermore,
+ *     memory objects created by the kernel are assumed to be ready
+ *     immediately; the default memory manager need not explicitly
+ *     set the "ready" attribute.]
  *
  *
- *     Conditions:
- *     The object(s) must *not* be locked. The map must be locked
- *     to preserve the reference to the object(s).
+ *     [Both the "initialized" and "ready" attribute wait conditions
+ *     use the "pager" field as the wait event.]
+ *
+ *     The port associations can be broken down by any of the
+ *     following routines:
+ *             vm_object_terminate:
+ *                     No references to the vm_object remain, and
+ *                     the object cannot (or will not) be cached.
+ *                     This is the normal case, and is done even
+ *                     though one of the other cases has already been
+ *                     done.
+ *             memory_object_destroy:
+ *                     The memory manager has requested that the
+ *                     kernel relinquish references to the memory
+ *                     object. [The memory manager may not want to
+ *                     destroy the memory object, but may wish to
+ *                     refuse or tear down existing memory mappings.]
+ *
+ *     Each routine that breaks an association must break all of
+ *     them at once.  At some later time, that routine must clear
+ *     the pager field and release the memory object references.
+ *     [Furthermore, each routine must cope with the simultaneous
+ *     or previous operations of the others.]
+ *
+ *     In addition to the lock on the object, the vm_object_hash_lock
+ *     governs the associations.  References gained through the
+ *     association require use of the hash lock.
+ *
+ *     Because the pager field may be cleared spontaneously, it
+ *     cannot be used to determine whether a memory object has
+ *     ever been associated with a particular vm_object.  [This
+ *     knowledge is important to the shadow object mechanism.]
+ *     For this reason, an additional "created" attribute is
+ *     provided.
+ *
+ *     During various paging operations, the pager reference found in the
+ *     vm_object must be valid.  To prevent this from being released,
+ *     (other than being removed, i.e., made null), routines may use
+ *     the vm_object_paging_begin/end routines [actually, macros].
+ *     The implementation uses the "paging_in_progress" and "wanted" fields.
+ *     [Operations that alter the validity of the pager values include the
+ *     termination routines and vm_object_collapse.]
  */
  */
-static int vm_object_coalesce_count = 0;
-
-__private_extern__ boolean_t
-vm_object_coalesce(
-       register vm_object_t            prev_object,
-       vm_object_t                     next_object,
-       vm_object_offset_t              prev_offset,
-       __unused vm_object_offset_t next_offset,
-       vm_object_size_t                prev_size,
-       vm_object_size_t                next_size)
-{
-       vm_object_size_t        newsize;
-
-#ifdef lint
-       next_offset++;
-#endif /* lint */
 
 
-       if (next_object != VM_OBJECT_NULL) {
-               return(FALSE);
-       }
 
 
-       if (prev_object == VM_OBJECT_NULL) {
-               return(TRUE);
-       }
+/*
+ *     Routine:        vm_object_enter
+ *     Purpose:
+ *             Find a VM object corresponding to the given
+ *             pager; if no such object exists, create one,
+ *             and initialize the pager.
+ */
+vm_object_t
+vm_object_enter(
+       memory_object_t         pager,
+       vm_object_size_t        size,
+       boolean_t               internal,
+       boolean_t               init,
+       boolean_t               named)
+{
+       vm_object_t             object;
+       vm_object_t             new_object;
+       boolean_t               must_init;
+       vm_object_hash_entry_t  entry, new_entry;
+       uint32_t        try_failed_count = 0;
+       lck_mtx_t       *lck;
 
 
-       XPR(XPR_VM_OBJECT,
-       "vm_object_coalesce: 0x%X prev_off 0x%X prev_size 0x%X next_size 0x%X\n",
-               prev_object, prev_offset, prev_size, next_size, 0);
+       if (pager == MEMORY_OBJECT_NULL)
+               return(vm_object_allocate(size));
 
 
-       vm_object_lock(prev_object);
+       new_object = VM_OBJECT_NULL;
+       new_entry = VM_OBJECT_HASH_ENTRY_NULL;
+       must_init = init;
 
        /*
 
        /*
-        *      Try to collapse the object first
+        *      Look for an object associated with this port.
         */
         */
-       vm_object_collapse(prev_object, prev_offset, TRUE);
+Retry:
+       lck = vm_object_hash_lock_spin(pager);
+       do {
+               entry = vm_object_hash_lookup(pager, FALSE);
 
 
-       /*
-        *      Can't coalesce if pages not mapped to
-        *      prev_entry may be in use any way:
-        *      . more than one reference
-        *      . paged out
-        *      . shadows another object
-        *      . has a copy elsewhere
-        *      . is purgeable
-        *      . paging references (pages might be in page-list)
-        */
+               if (entry == VM_OBJECT_HASH_ENTRY_NULL) {
+                       if (new_object == VM_OBJECT_NULL) {
+                               /*
+                                *      We must unlock to create a new object;
+                                *      if we do so, we must try the lookup again.
+                                */
+                               vm_object_hash_unlock(lck);
+                               assert(new_entry == VM_OBJECT_HASH_ENTRY_NULL);
+                               new_entry = vm_object_hash_entry_alloc(pager);
+                               new_object = vm_object_allocate(size);
+                               /*
+                                * Set new_object->hashed now, while noone
+                                * knows about this object yet and we
+                                * don't need to lock it.  Once it's in
+                                * the hash table, we would have to lock
+                                * the object to set its "hashed" bit and
+                                * we can't lock the object while holding
+                                * the hash lock as a spinlock...
+                                */
+                               new_object->hashed = TRUE;
+                               lck = vm_object_hash_lock_spin(pager);
+                       } else {
+                               /*
+                                *      Lookup failed twice, and we have something
+                                *      to insert; set the object.
+                                */
+                               /*
+                                * We can't lock the object here since we're
+                                * holding the hash lock as a spin lock.
+                                * We've already pre-set "new_object->hashed"
+                                * when we created "new_object" above, so we
+                                * won't need to modify the object in
+                                * vm_object_hash_insert().
+                                */
+                               assert(new_object->hashed);
+                               vm_object_hash_insert(new_entry, new_object);
+                               entry = new_entry;
+                               new_entry = VM_OBJECT_HASH_ENTRY_NULL;
+                               new_object = VM_OBJECT_NULL;
+                               must_init = TRUE;
+                       }
+               } else if (entry->object == VM_OBJECT_NULL) {
+                       /*
+                        *      If a previous object is being terminated,
+                        *      we must wait for the termination message
+                        *      to be queued (and lookup the entry again).
+                        */
+                       entry->waiting = TRUE;
+                       entry = VM_OBJECT_HASH_ENTRY_NULL;
+                       assert_wait((event_t) pager, THREAD_UNINT);
+                       vm_object_hash_unlock(lck);
 
 
-       if ((prev_object->ref_count > 1) ||
-           prev_object->pager_created ||
-           (prev_object->shadow != VM_OBJECT_NULL) ||
-           (prev_object->copy != VM_OBJECT_NULL) ||
-           (prev_object->true_share != FALSE) ||
-           (prev_object->purgable != VM_PURGABLE_DENY) ||
-           (prev_object->paging_in_progress != 0) ||
-           (prev_object->activity_in_progress != 0)) {
-               vm_object_unlock(prev_object);
-               return(FALSE);
-       }
+                       thread_block(THREAD_CONTINUE_NULL);
+                       lck = vm_object_hash_lock_spin(pager);
+               }
+       } while (entry == VM_OBJECT_HASH_ENTRY_NULL);
 
 
-       vm_object_coalesce_count++;
+       object = entry->object;
+       assert(object != VM_OBJECT_NULL);
 
 
-       /*
-        *      Remove any pages that may still be in the object from
-        *      a previous deallocation.
-        */
-       vm_object_page_remove(prev_object,
-               prev_offset + prev_size,
-               prev_offset + prev_size + next_size);
+       if (!must_init) {
+               if ( !vm_object_lock_try(object)) {
 
 
-       /*
-        *      Extend the object if necessary.
-        */
-       newsize = prev_offset + prev_size + next_size;
-       if (newsize > prev_object->size) {
-#if    MACH_PAGEMAP
-               /*
-                *      We cannot extend an object that has existence info,
-                *      since the existence info might then fail to cover
-                *      the entire object.
-                *
-                *      This assertion must be true because the object
-                *      has no pager, and we only create existence info
-                *      for objects with pagers.
-                */
-               assert(prev_object->existence_map == VM_EXTERNAL_NULL);
-#endif /* MACH_PAGEMAP */
-               prev_object->size = newsize;
-       }
+                       vm_object_hash_unlock(lck);
 
 
-       vm_object_unlock(prev_object);
-       return(TRUE);
-}
+                       try_failed_count++;
+                       mutex_pause(try_failed_count);  /* wait a bit */
+                       goto Retry;
+               }
+               assert(!internal || object->internal);
+#if VM_OBJECT_CACHE
+               if (object->ref_count == 0) {
+                       if ( !vm_object_cache_lock_try()) {
 
 
-/*
- *     Attach a set of physical pages to an object, so that they can
- *     be mapped by mapping the object.  Typically used to map IO memory.
- *
- *     The mapping function and its private data are used to obtain the
- *     physical addresses for each page to be mapped.
- */
-void
-vm_object_page_map(
-       vm_object_t             object,
-       vm_object_offset_t      offset,
-       vm_object_size_t        size,
-       vm_object_offset_t      (*map_fn)(void *map_fn_data, 
-               vm_object_offset_t offset),
-               void            *map_fn_data)   /* private to map_fn */
-{
-       int64_t num_pages;
-       int     i;
-       vm_page_t       m;
-       vm_page_t       old_page;
-       vm_object_offset_t      addr;
+                               vm_object_hash_unlock(lck);
+                               vm_object_unlock(object);
+
+                               try_failed_count++;
+                               mutex_pause(try_failed_count);  /* wait a bit */
+                               goto Retry;
+                       }
+                       XPR(XPR_VM_OBJECT_CACHE,
+                           "vm_object_enter: removing %x from cache, head (%x, %x)\n",
+                               object,
+                               vm_object_cached_list.next,
+                               vm_object_cached_list.prev, 0,0);
+                       queue_remove(&vm_object_cached_list, object,
+                                    vm_object_t, cached_list);
+                       vm_object_cached_count--;
+
+                       vm_object_cache_unlock();
+               }
+#endif
+               if (named) {
+                       assert(!object->named);
+                       object->named = TRUE;
+               }
+               vm_object_lock_assert_exclusive(object);
+               object->ref_count++;
+               vm_object_res_reference(object);
 
 
-       num_pages = atop_64(size);
+               vm_object_hash_unlock(lck);
+               vm_object_unlock(object);
 
 
-       for (i = 0; i < num_pages; i++, offset += PAGE_SIZE_64) {
+               VM_STAT_INCR(hits);
+       } else
+               vm_object_hash_unlock(lck);
 
 
-           addr = (*map_fn)(map_fn_data, offset);
+       assert(object->ref_count > 0);
 
 
-           while ((m = vm_page_grab_fictitious()) == VM_PAGE_NULL)
-               vm_page_more_fictitious();
+       VM_STAT_INCR(lookups);
 
 
-           vm_object_lock(object);
-           if ((old_page = vm_page_lookup(object, offset))
-                       != VM_PAGE_NULL)
-           {
-                   VM_PAGE_FREE(old_page);
-           }
+       XPR(XPR_VM_OBJECT,
+               "vm_o_enter: pager 0x%x obj 0x%x must_init %d\n",
+               pager, object, must_init, 0, 0);
 
 
-           assert((ppnum_t) addr == addr);
-           vm_page_init(m, (ppnum_t) addr);
-           /*
-            * private normally requires lock_queues but since we
-            * are initializing the page, its not necessary here
-            */
-           m->private = TRUE;          /* don`t free page */
-           m->wire_count = 1;
-           vm_page_insert(m, object, offset);
+       /*
+        *      If we raced to create a vm_object but lost, let's
+        *      throw away ours.
+        */
 
 
-           PAGE_WAKEUP_DONE(m);
-           vm_object_unlock(object);
+       if (new_object != VM_OBJECT_NULL) {
+               /*
+                * Undo the pre-setting of "new_object->hashed" before
+                * deallocating "new_object", since we did not insert it
+                * into the hash table after all.
+                */
+               assert(new_object->hashed);
+               new_object->hashed = FALSE;
+               vm_object_deallocate(new_object);
        }
        }
-}
 
 
-#include <mach_kdb.h>
+       if (new_entry != VM_OBJECT_HASH_ENTRY_NULL)
+               vm_object_hash_entry_free(new_entry);
+
+       if (must_init) {
+               memory_object_control_t control;
+
+               /*
+                *      Allocate request port.
+                */
 
 
-#if    MACH_KDB
-#include <ddb/db_output.h>
-#include <vm/vm_print.h>
+               control = memory_object_control_allocate(object);
+               assert (control != MEMORY_OBJECT_CONTROL_NULL);
 
 
-#define printf kdbprintf
+               vm_object_lock(object);
+               assert(object != kernel_object);
 
 
-extern boolean_t       vm_object_cached(
-                               vm_object_t object);
+               /*
+                *      Copy the reference we were given.
+                */
 
 
-extern void            print_bitstring(
-                               char byte);
+               memory_object_reference(pager);
+               object->pager_created = TRUE;
+               object->pager = pager;
+               object->internal = internal;
+               object->pager_trusted = internal;
+               if (!internal) {
+                       /* copy strategy invalid until set by memory manager */
+                       object->copy_strategy = MEMORY_OBJECT_COPY_INVALID;
+               }
+               object->pager_control = control;
+               object->pager_ready = FALSE;
 
 
-boolean_t      vm_object_print_pages = FALSE;
+               vm_object_unlock(object);
 
 
-void
-print_bitstring(
-       char byte)
-{
-       printf("%c%c%c%c%c%c%c%c",
-              ((byte & (1 << 0)) ? '1' : '0'),
-              ((byte & (1 << 1)) ? '1' : '0'),
-              ((byte & (1 << 2)) ? '1' : '0'),
-              ((byte & (1 << 3)) ? '1' : '0'),
-              ((byte & (1 << 4)) ? '1' : '0'),
-              ((byte & (1 << 5)) ? '1' : '0'),
-              ((byte & (1 << 6)) ? '1' : '0'),
-              ((byte & (1 << 7)) ? '1' : '0'));
-}
+               /*
+                *      Let the pager know we're using it.
+                */
 
 
-boolean_t
-vm_object_cached(
-       __unused register vm_object_t object)
-{
-#if VM_OBJECT_CACHE
-       register vm_object_t o;
+               (void) memory_object_init(pager,
+                       object->pager_control,
+                       PAGE_SIZE);
 
 
-       queue_iterate(&vm_object_cached_list, o, vm_object_t, cached_list) {
-               if (object == o) {
-                       return TRUE;
+               vm_object_lock(object);
+               if (named)
+                       object->named = TRUE;
+               if (internal) {
+                       vm_object_lock_assert_exclusive(object);
+                       object->pager_ready = TRUE;
+                       vm_object_wakeup(object, VM_OBJECT_EVENT_PAGER_READY);
                }
                }
-       }
-#endif
-       return FALSE;
-}
 
 
-#if    MACH_PAGEMAP
-/*
- *     vm_external_print:      [ debug ]
- */
-void
-vm_external_print(
-       vm_external_map_t       emap,
-       vm_object_size_t        size)
-{
-       if (emap == VM_EXTERNAL_NULL) {
-               printf("0  ");
+               object->pager_initialized = TRUE;
+               vm_object_wakeup(object, VM_OBJECT_EVENT_INITIALIZED);
        } else {
        } else {
-               vm_object_size_t existence_size = stob(size);
-               printf("{ size=%lld, map=[", (uint64_t) existence_size);
-               if (existence_size > 0) {
-                       print_bitstring(emap[0]);
-               }
-               if (existence_size > 1) {
-                       print_bitstring(emap[1]);
-               }
-               if (existence_size > 2) {
-                       printf("...");
-                       print_bitstring(emap[existence_size-1]);
-               }
-               printf("] }\n");
+               vm_object_lock(object);
        }
        }
-       return;
-}
-#endif /* MACH_PAGEMAP */
-
-int
-vm_follow_object(
-       vm_object_t object)
-{
-       int count = 0;
-       int orig_db_indent = db_indent;
 
 
-       while (TRUE) {
-               if (object == VM_OBJECT_NULL) {
-                       db_indent = orig_db_indent;
-                       return count;
-               }
-
-               count += 1;
+       /*
+        *      [At this point, the object must be locked]
+        */
 
 
-               iprintf("object 0x%x", object);
-               printf(", shadow=0x%x", object->shadow);
-               printf(", copy=0x%x", object->copy);
-               printf(", pager=0x%x", object->pager);
-               printf(", ref=%d\n", object->ref_count);
+       /*
+        *      Wait for the work above to be done by the first
+        *      thread to map this object.
+        */
 
 
-               db_indent += 2;
-               object = object->shadow;
+       while (!object->pager_initialized) {
+               vm_object_sleep(object,
+                               VM_OBJECT_EVENT_INITIALIZED,
+                               THREAD_UNINT);
        }
        }
+       vm_object_unlock(object);
 
 
+       XPR(XPR_VM_OBJECT,
+           "vm_object_enter: vm_object %x, memory_object %x, internal %d\n",
+           object, object->pager, internal, 0,0);
+       return(object);
 }
 
 /*
 }
 
 /*
- *     vm_object_print:        [ debug ]
+ *     Routine:        vm_object_compressor_pager_create
+ *     Purpose:
+ *             Create a memory object for an internal object.
+ *     In/out conditions:
+ *             The object is locked on entry and exit;
+ *             it may be unlocked within this call.
+ *     Limitations:
+ *             Only one thread may be performing a
+ *             vm_object_compressor_pager_create on an object at
+ *             a time.  Presumably, only the pageout
+ *             daemon will be using this routine.
  */
  */
+
 void
 void
-vm_object_print(db_expr_t db_addr, __unused boolean_t have_addr,
-               __unused db_expr_t arg_count, __unused char *modif)
+vm_object_compressor_pager_create(
+       vm_object_t     object)
 {
 {
-       vm_object_t     object;
-       register vm_page_t p;
-       const char *s;
-
-       register int count;
+       memory_object_t         pager;
+       vm_object_hash_entry_t  entry;
+       lck_mtx_t               *lck;
+       vm_object_t             pager_object = VM_OBJECT_NULL;
 
 
-       object = (vm_object_t) (long) db_addr;
-       if (object == VM_OBJECT_NULL)
-               return;
+       assert(object != kernel_object);
 
 
-       iprintf("object 0x%x\n", object);
+       /*
+        *      Prevent collapse or termination by holding a paging reference
+        */
 
 
-       db_indent += 2;
+       vm_object_paging_begin(object);
+       if (object->pager_created) {
+               /*
+                *      Someone else got to it first...
+                *      wait for them to finish initializing the ports
+                */
+               while (!object->pager_initialized) {
+                       vm_object_sleep(object,
+                                       VM_OBJECT_EVENT_INITIALIZED,
+                                       THREAD_UNINT);
+               }
+               vm_object_paging_end(object);
+               return;
+       }
 
 
-       iprintf("size=0x%x", object->size);
-       printf(", memq_hint=%p", object->memq_hint);
-       printf(", ref_count=%d\n", object->ref_count);
-       iprintf("");
-#if    TASK_SWAPPER
-       printf("res_count=%d, ", object->res_count);
-#endif /* TASK_SWAPPER */
-       printf("resident_page_count=%d\n", object->resident_page_count);
+       /*
+        *      Indicate that a memory object has been assigned
+        *      before dropping the lock, to prevent a race.
+        */
 
 
-       iprintf("shadow=0x%x", object->shadow);
-       if (object->shadow) {
-               register int i = 0;
-               vm_object_t shadow = object;
-               while((shadow = shadow->shadow))
-                       i++;
-               printf(" (depth %d)", i);
-       }
-       printf(", copy=0x%x", object->copy);
-       printf(", shadow_offset=0x%x", object->shadow_offset);
-       printf(", last_alloc=0x%x\n", object->last_alloc);
-
-       iprintf("pager=0x%x", object->pager);
-       printf(", paging_offset=0x%x", object->paging_offset);
-       printf(", pager_control=0x%x\n", object->pager_control);
-
-       iprintf("copy_strategy=%d[", object->copy_strategy);
-       switch (object->copy_strategy) {
-               case MEMORY_OBJECT_COPY_NONE:
-               printf("copy_none");
-               break;
+       object->pager_created = TRUE;
+       object->paging_offset = 0;
+               
+       vm_object_unlock(object);
 
 
-               case MEMORY_OBJECT_COPY_CALL:
-               printf("copy_call");
-               break;
+       if ((uint32_t) (object->vo_size/PAGE_SIZE) !=
+           (object->vo_size/PAGE_SIZE)) {
+               panic("vm_object_compressor_pager_create(%p): "
+                     "object size 0x%llx >= 0x%llx\n",
+                     object,
+                     (uint64_t) object->vo_size,
+                     0x0FFFFFFFFULL*PAGE_SIZE);
+       }
 
 
-               case MEMORY_OBJECT_COPY_DELAY:
-               printf("copy_delay");
-               break;
+       /*
+        *      Create the [internal] pager, and associate it with this object.
+        *
+        *      We make the association here so that vm_object_enter()
+        *      can look up the object to complete initializing it.  No
+        *      user will ever map this object.
+        */
+       {
+               assert(object->temporary);
 
 
-               case MEMORY_OBJECT_COPY_SYMMETRIC:
-               printf("copy_symmetric");
-               break;
+               /* create our new memory object */
+               assert((uint32_t) (object->vo_size/PAGE_SIZE) ==
+                      (object->vo_size/PAGE_SIZE));
+               (void) compressor_memory_object_create(
+                       (memory_object_size_t) object->vo_size,
+                       &pager);
+               if (pager == NULL) {
+                       panic("vm_object_compressor_pager_create(): "
+                             "no pager for object %p size 0x%llx\n",
+                             object, (uint64_t) object->vo_size);
+               }
+       }
 
 
-               case MEMORY_OBJECT_COPY_INVALID:
-               printf("copy_invalid");
-               break;
+       entry = vm_object_hash_entry_alloc(pager);
 
 
-               default:
-               printf("?");
-       }
-       printf("]");
-
-       iprintf("all_wanted=0x%x<", object->all_wanted);
-       s = "";
-       if (vm_object_wanted(object, VM_OBJECT_EVENT_INITIALIZED)) {
-               printf("%sinit", s);
-               s = ",";
-       }
-       if (vm_object_wanted(object, VM_OBJECT_EVENT_PAGER_READY)) {
-               printf("%sready", s);
-               s = ",";
-       }
-       if (vm_object_wanted(object, VM_OBJECT_EVENT_PAGING_IN_PROGRESS)) {
-               printf("%spaging", s);
-               s = ",";
-       }
-       if (vm_object_wanted(object, VM_OBJECT_EVENT_LOCK_IN_PROGRESS)) {
-               printf("%slock", s);
-               s = ",";
-       }
-       if (vm_object_wanted(object, VM_OBJECT_EVENT_UNCACHING)) {
-               printf("%suncaching", s);
-               s = ",";
-       }
-       if (vm_object_wanted(object, VM_OBJECT_EVENT_COPY_CALL)) {
-               printf("%scopy_call", s);
-               s = ",";
-       }
-       if (vm_object_wanted(object, VM_OBJECT_EVENT_CACHING)) {
-               printf("%scaching", s);
-               s = ",";
-       }
-       printf(">");
-       printf(", paging_in_progress=%d\n", object->paging_in_progress);
-       printf(", activity_in_progress=%d\n", object->activity_in_progress);
-
-       iprintf("%screated, %sinit, %sready, %spersist, %strusted, %spageout, %s, %s\n",
-               (object->pager_created ? "" : "!"),
-               (object->pager_initialized ? "" : "!"),
-               (object->pager_ready ? "" : "!"),
-               (object->can_persist ? "" : "!"),
-               (object->pager_trusted ? "" : "!"),
-               (object->pageout ? "" : "!"),
-               (object->internal ? "internal" : "external"),
-               (object->temporary ? "temporary" : "permanent"));
-       iprintf("%salive, %spurgeable, %spurgeable_volatile, %spurgeable_empty, %sshadowed, %scached, %sprivate\n",
-               (object->alive ? "" : "!"),
-               ((object->purgable != VM_PURGABLE_DENY) ? "" : "!"),
-               ((object->purgable == VM_PURGABLE_VOLATILE) ? "" : "!"),
-               ((object->purgable == VM_PURGABLE_EMPTY) ? "" : "!"),
-               (object->shadowed ? "" : "!"),
-               (vm_object_cached(object) ? "" : "!"),
-               (object->private ? "" : "!"));
-       iprintf("%sadvisory_pageout, %ssilent_overwrite\n",
-               (object->advisory_pageout ? "" : "!"),
-               (object->silent_overwrite ? "" : "!"));
-
-#if    MACH_PAGEMAP
-       iprintf("existence_map=");
-       vm_external_print(object->existence_map, object->size);
-#endif /* MACH_PAGEMAP */
-#if    MACH_ASSERT
-       iprintf("paging_object=0x%x\n", object->paging_object);
-#endif /* MACH_ASSERT */
+       vm_object_lock(object);
+       lck = vm_object_hash_lock_spin(pager);
+       vm_object_hash_insert(entry, object);
+       vm_object_hash_unlock(lck);
+       vm_object_unlock(object);
 
 
-       if (vm_object_print_pages) {
-               count = 0;
-               p = (vm_page_t) queue_first(&object->memq);
-               while (!queue_end(&object->memq, (queue_entry_t) p)) {
-                       if (count == 0) {
-                               iprintf("memory:=");
-                       } else if (count == 2) {
-                               printf("\n");
-                               iprintf(" ...");
-                               count = 0;
-                       } else {
-                               printf(",");
-                       }
-                       count++;
+       /*
+        *      A reference was returned by
+        *      memory_object_create(), and it is
+        *      copied by vm_object_enter().
+        */
 
 
-                       printf("(off=0x%llX,page=%p)", p->offset, p);
-                       p = (vm_page_t) queue_next(&p->listq);
-               }
-               if (count != 0) {
-                       printf("\n");
-               }
+       pager_object = vm_object_enter(pager, object->vo_size, TRUE, TRUE, FALSE);
+
+       if (pager_object != object) {
+               panic("vm_object_compressor_pager_create: mismatch (pager: %p, pager_object: %p, orig_object: %p, orig_object size: 0x%llx)\n", pager, pager_object, object, (uint64_t) object->vo_size);
        }
        }
-       db_indent -= 2;
-}
 
 
+       /*
+        *      Drop the reference we were passed.
+        */
+       memory_object_deallocate(pager);
+
+       vm_object_lock(object);
+
+       /*
+        *      Release the paging reference
+        */
+       vm_object_paging_end(object);
+}
 
 /*
 
 /*
- *     vm_object_find          [ debug ]
- *
- *     Find all tasks which reference the given vm_object.
+ *     Routine:        vm_object_remove
+ *     Purpose:
+ *             Eliminate the pager/object association
+ *             for this pager.
+ *     Conditions:
+ *             The object cache must be locked.
  */
  */
-
-boolean_t vm_object_find(vm_object_t object);
-boolean_t vm_object_print_verbose = FALSE;
-
-boolean_t
-vm_object_find(
-       vm_object_t     object)
+__private_extern__ void
+vm_object_remove(
+       vm_object_t     object)
 {
 {
-        task_t task;
-       vm_map_t map;
-       vm_map_entry_t entry;
-       boolean_t found = FALSE;
-
-       queue_iterate(&tasks, task, task_t, tasks) {
-               map = task->map;
-               for (entry = vm_map_first_entry(map);
-                        entry && entry != vm_map_to_entry(map);
-                        entry = entry->vme_next) {
-
-                       vm_object_t obj;
-
-                       /* 
-                        * For the time being skip submaps,
-                        * only the kernel can have submaps,
-                        * and unless we are interested in 
-                        * kernel objects, we can simply skip 
-                        * submaps. See sb/dejan/nmk18b7/src/mach_kernel/vm
-                        * for a full solution.
-                        */
-                       if (entry->is_sub_map)
-                               continue;
-                       if (entry) 
-                               obj = entry->object.vm_object;
-                       else 
-                               continue;
+       memory_object_t pager;
 
 
-                       while (obj != VM_OBJECT_NULL) {
-                               if (obj == object) {
-                                       if (!found) {
-                                               printf("TASK\t\tMAP\t\tENTRY\n");
-                                               found = TRUE;
-                                       }
-                                       printf("0x%x\t0x%x\t0x%x\n", 
-                                                  task, map, entry);
-                               }
-                               obj = obj->shadow;
-                       }
-               }
+       if ((pager = object->pager) != MEMORY_OBJECT_NULL) {
+               vm_object_hash_entry_t  entry;
+
+               entry = vm_object_hash_lookup(pager, FALSE);
+               if (entry != VM_OBJECT_HASH_ENTRY_NULL)
+                       entry->object = VM_OBJECT_NULL;
        }
 
        }
 
-       return(found);
 }
 
 }
 
-#endif /* MACH_KDB */
+/*
+ *     Global variables for vm_object_collapse():
+ *
+ *             Counts for normal collapses and bypasses.
+ *             Debugging variables, to watch or disable collapse.
+ */
+static long    object_collapses = 0;
+static long    object_bypasses  = 0;
 
 
-kern_return_t
-vm_object_populate_with_private(
-               vm_object_t             object,
-               vm_object_offset_t      offset,
-               ppnum_t                 phys_page,
-               vm_size_t               size)
-{
-       ppnum_t                 base_page;
-       vm_object_offset_t      base_offset;
+static boolean_t       vm_object_collapse_allowed = TRUE;
+static boolean_t       vm_object_bypass_allowed = TRUE;
 
 
+unsigned long vm_object_collapse_encrypted = 0;
 
 
-       if(!object->private)
-               return KERN_FAILURE;
+void vm_object_do_collapse_compressor(vm_object_t object,
+                                     vm_object_t backing_object);
+void
+vm_object_do_collapse_compressor(
+       vm_object_t object,
+       vm_object_t backing_object)
+{
+       vm_object_offset_t new_offset, backing_offset;
+       vm_object_size_t size;
 
 
-       base_page = phys_page;
+       vm_counters.do_collapse_compressor++;
 
 
-       vm_object_lock(object);
-       if(!object->phys_contiguous) {
-               vm_page_t       m;
-               if((base_offset = trunc_page_64(offset)) != offset) {
-                       vm_object_unlock(object);
-                       return KERN_FAILURE;
-               }
-               base_offset += object->paging_offset;
-               while(size) {
-                       m = vm_page_lookup(object, base_offset);
-                       if(m != VM_PAGE_NULL) {
-                               if(m->fictitious) {
-                                       if (m->phys_page != vm_page_guard_addr) {
+       vm_object_lock_assert_exclusive(object);
+       vm_object_lock_assert_exclusive(backing_object);
 
 
-                                               vm_page_lockspin_queues();
-                                               m->private = TRUE;
-                                               vm_page_unlock_queues();
+       size = object->vo_size;
 
 
-                                               m->fictitious = FALSE;
-                                               m->phys_page = base_page;
-                                               if(!m->busy) {
-                                                       m->busy = TRUE;
-                                               }
-                                               if(!m->absent) {
-                                                       m->absent = TRUE;
-                                               }
-                                               m->list_req_pending = TRUE;
-                                       }
-                               } else if (m->phys_page != base_page) {
-                                       if (m->pmapped) {
-                                               /*
-                                                * pmap call to clear old mapping
-                                                */
-                                               pmap_disconnect(m->phys_page);
-                                       }
-                                       m->phys_page = base_page;
-                               }
+       /*
+        *      Move all compressed pages from backing_object
+        *      to the parent.
+        */
 
 
-                               /*
-                                * ENCRYPTED SWAP:
-                                * We're not pointing to the same
-                                * physical page any longer and the
-                                * contents of the new one are not
-                                * supposed to be encrypted.
-                                * XXX What happens to the original
-                                * physical page. Is it lost ?
-                                */
-                               m->encrypted = FALSE;
+       for (backing_offset = object->vo_shadow_offset;
+            backing_offset < object->vo_shadow_offset + object->vo_size;
+            backing_offset += PAGE_SIZE) {
+               memory_object_offset_t backing_pager_offset;
+
+               /* find the next compressed page at or after this offset */
+               backing_pager_offset = (backing_offset +
+                                       backing_object->paging_offset);
+               backing_pager_offset = vm_compressor_pager_next_compressed(
+                       backing_object->pager,
+                       backing_pager_offset);
+               if (backing_pager_offset == (memory_object_offset_t) -1) {
+                       /* no more compressed pages */
+                       break;
+               }
+               backing_offset = (backing_pager_offset -
+                                 backing_object->paging_offset);
 
 
-                       } else {
-                               while ((m = vm_page_grab_fictitious()) == VM_PAGE_NULL)
-                                       vm_page_more_fictitious();      
+               new_offset = backing_offset - object->vo_shadow_offset;
 
 
-                               /*
-                                * private normally requires lock_queues but since we
-                                * are initializing the page, its not necessary here
-                                */
-                               m->private = TRUE;
-                               m->fictitious = FALSE;
-                               m->phys_page = base_page;
-                               m->list_req_pending = TRUE;
-                               m->absent = TRUE;
-                               m->unusual = TRUE;
+               if (new_offset >= object->vo_size) {
+                       /* we're out of the scope of "object": done */
+                       break;
+               }
 
 
-                               vm_page_insert(m, object, base_offset);
-                       }
-                       base_page++;                                                                    /* Go to the next physical page */
-                       base_offset += PAGE_SIZE;
-                       size -= PAGE_SIZE;
+               if ((vm_page_lookup(object, new_offset) != VM_PAGE_NULL) ||
+                   (vm_compressor_pager_state_get(object->pager,
+                                                  (new_offset +
+                                                   object->paging_offset)) ==
+                    VM_EXTERNAL_STATE_EXISTS)) {
+                       /*
+                        * This page already exists in object, resident or
+                        * compressed.
+                        * We don't need this compressed page in backing_object
+                        * and it will be reclaimed when we release
+                        * backing_object.
+                        */
+                       continue;
                }
                }
-       } else {
-               /* NOTE: we should check the original settings here */
-               /* if we have a size > zero a pmap call should be made */
-               /* to disable the range */      
 
 
-               /* pmap_? */
-               
-               /* shadows on contiguous memory are not allowed */
-               /* we therefore can use the offset field */
-               object->shadow_offset = (vm_object_offset_t)phys_page << PAGE_SHIFT;
-               object->size = size;
+               /*
+                * backing_object has this page in the VM compressor and
+                * we need to transfer it to object.
+                */
+               vm_counters.do_collapse_compressor_pages++;
+               vm_compressor_pager_transfer(
+                       /* destination: */
+                       object->pager,
+                       (new_offset + object->paging_offset),
+                       /* source: */
+                       backing_object->pager,
+                       (backing_offset + backing_object->paging_offset));
        }
        }
-       vm_object_unlock(object);
-       return KERN_SUCCESS;
 }
 
 /*
 }
 
 /*
- *     memory_object_free_from_cache:
- *
- *     Walk the vm_object cache list, removing and freeing vm_objects 
- *     which are backed by the pager identified by the caller, (pager_ops).  
- *     Remove up to "count" objects, if there are that may available
- *     in the cache.
+ *     Routine:        vm_object_do_collapse
+ *     Purpose:
+ *             Collapse an object with the object backing it.
+ *             Pages in the backing object are moved into the
+ *             parent, and the backing object is deallocated.
+ *     Conditions:
+ *             Both objects and the cache are locked; the page
+ *             queues are unlocked.
  *
  *
- *     Walk the list at most once, return the number of vm_objects
- *     actually freed.
  */
  */
-
-__private_extern__ kern_return_t
-memory_object_free_from_cache(
-       __unused host_t         host,
-       __unused memory_object_pager_ops_t pager_ops,
-       int             *count)
+static void
+vm_object_do_collapse(
+       vm_object_t object,
+       vm_object_t backing_object)
 {
 {
-#if VM_OBJECT_CACHE
-       int     object_released = 0;
-
-       register vm_object_t object = VM_OBJECT_NULL;
-       vm_object_t shadow;
+       vm_page_t p, pp;
+       vm_object_offset_t new_offset, backing_offset;
+       vm_object_size_t size;
 
 
-/*
-       if(host == HOST_NULL)
-               return(KERN_INVALID_ARGUMENT);
-*/
+       vm_object_lock_assert_exclusive(object);
+       vm_object_lock_assert_exclusive(backing_object);
 
 
- try_again:
-       vm_object_cache_lock();
+       assert(object->purgable == VM_PURGABLE_DENY);
+       assert(backing_object->purgable == VM_PURGABLE_DENY);
 
 
-       queue_iterate(&vm_object_cached_list, object, 
-                                       vm_object_t, cached_list) {
-               if (object->pager &&
-                   (pager_ops == object->pager->mo_pager_ops)) {
-                       vm_object_lock(object);
-                       queue_remove(&vm_object_cached_list, object, 
-                                       vm_object_t, cached_list);
-                       vm_object_cached_count--;
+       backing_offset = object->vo_shadow_offset;
+       size = object->vo_size;
 
 
-                       vm_object_cache_unlock();
-                       /*
-                       *       Since this object is in the cache, we know
-                       *       that it is initialized and has only a pager's
-                       *       (implicit) reference. Take a reference to avoid
-                       *       recursive deallocations.
-                       */
-
-                       assert(object->pager_initialized);
-                       assert(object->ref_count == 0);
-                       vm_object_lock_assert_exclusive(object);
-                       object->ref_count++;
+       /*
+        *      Move all in-memory pages from backing_object
+        *      to the parent.  Pages that have been paged out
+        *      will be overwritten by any of the parent's
+        *      pages that shadow them.
+        */
+       
+       while (!vm_page_queue_empty(&backing_object->memq)) {
+               
+               p = (vm_page_t) vm_page_queue_first(&backing_object->memq);
+               
+               new_offset = (p->offset - backing_offset);
+               
+               assert(!p->busy || p->absent);
 
 
+               /*
+                *      If the parent has a page here, or if
+                *      this page falls outside the parent,
+                *      dispose of it.
+                *
+                *      Otherwise, move it as planned.
+                */
+               
+               if (p->offset < backing_offset || new_offset >= size) {
+                       VM_PAGE_FREE(p);
+               } else {
                        /*
                        /*
-                       *       Terminate the object.
-                       *       If the object had a shadow, we let 
-                       *       vm_object_deallocate deallocate it. 
-                       *       "pageout" objects have a shadow, but
-                       *       maintain a "paging reference" rather 
-                       *       than a normal reference.
-                       *       (We are careful here to limit recursion.)
-                       */
-                       shadow = object->pageout?VM_OBJECT_NULL:object->shadow;
+                        * ENCRYPTED SWAP:
+                        * The encryption key includes the "pager" and the
+                        * "paging_offset".  These will not change during the 
+                        * object collapse, so we can just move an encrypted
+                        * page from one object to the other in this case.
+                        * We can't decrypt the page here, since we can't drop
+                        * the object lock.
+                        */
+                       if (p->encrypted) {
+                               vm_object_collapse_encrypted++;
+                       }
+                       pp = vm_page_lookup(object, new_offset);
+                       if (pp == VM_PAGE_NULL) {
 
 
-                       if ((vm_object_terminate(object) == KERN_SUCCESS)
-                                       && (shadow != VM_OBJECT_NULL)) {
-                               vm_object_deallocate(shadow);
+                               if (VM_COMPRESSOR_PAGER_STATE_GET(object,
+                                                                 new_offset)
+                                   == VM_EXTERNAL_STATE_EXISTS) {
+                                       /*
+                                        * Parent object has this page
+                                        * in the VM compressor.
+                                        * Throw away the backing
+                                        * object's page.
+                                        */
+                                       VM_PAGE_FREE(p);
+                               } else {
+                                       /*
+                                        *      Parent now has no page.
+                                        *      Move the backing object's page
+                                        *      up.
+                                        */
+                                       vm_page_rename(p, object, new_offset,
+                                                      TRUE);
+                               }
+                       } else {
+                               assert(! pp->absent);
+
+                               /*
+                                *      Parent object has a real page.
+                                *      Throw away the backing object's
+                                *      page.
+                                */
+                               VM_PAGE_FREE(p);
                        }
                        }
-               
-                       if(object_released++ == *count)
-                               return KERN_SUCCESS;
-                       goto try_again;
                }
        }
                }
        }
-       vm_object_cache_unlock();
-       *count  = object_released;
-#else
-       *count = 0;
-#endif
-       return KERN_SUCCESS;
-}
 
 
+       if (vm_object_collapse_compressor_allowed &&
+           object->pager != MEMORY_OBJECT_NULL &&
+           backing_object->pager != MEMORY_OBJECT_NULL) {
 
 
+               /* move compressed pages from backing_object to object */
+               vm_object_do_collapse_compressor(object, backing_object);
 
 
-kern_return_t
-memory_object_create_named(
-       memory_object_t pager,
-       memory_object_offset_t  size,
-       memory_object_control_t         *control)
-{
-       vm_object_t             object;
-       vm_object_hash_entry_t  entry;
-       lck_mtx_t               *lck;
+       } else if (backing_object->pager != MEMORY_OBJECT_NULL) {
+               vm_object_hash_entry_t  entry;
 
 
-       *control = MEMORY_OBJECT_CONTROL_NULL;
-       if (pager == MEMORY_OBJECT_NULL)
-               return KERN_INVALID_ARGUMENT;
+               assert((!object->pager_created &&
+                       (object->pager == MEMORY_OBJECT_NULL)) ||
+                      (!backing_object->pager_created &&
+                       (backing_object->pager == MEMORY_OBJECT_NULL)));
+               /*
+                *      Move the pager from backing_object to object.
+                *
+                *      XXX We're only using part of the paging space
+                *      for keeps now... we ought to discard the
+                *      unused portion.
+                */
 
 
-       lck = vm_object_hash_lock_spin(pager);
-       entry = vm_object_hash_lookup(pager, FALSE);
+               assert(!object->paging_in_progress);
+               assert(!object->activity_in_progress);
+               assert(!object->pager_created);
+               assert(object->pager == NULL);
+               object->pager = backing_object->pager;
 
 
-       if ((entry != VM_OBJECT_HASH_ENTRY_NULL) &&
-                       (entry->object != VM_OBJECT_NULL)) {
-               if (entry->object->named == TRUE)
-                       panic("memory_object_create_named: caller already holds the right");    }
-       vm_object_hash_unlock(lck);
+               if (backing_object->hashed) {
+                       lck_mtx_t       *lck;
 
 
-       if ((object = vm_object_enter(pager, size, FALSE, FALSE, TRUE)) == VM_OBJECT_NULL) {
-               return(KERN_INVALID_OBJECT);
+                       lck = vm_object_hash_lock_spin(backing_object->pager);
+                       entry = vm_object_hash_lookup(object->pager, FALSE);
+                       assert(entry != VM_OBJECT_HASH_ENTRY_NULL);
+                       entry->object = object;
+                       vm_object_hash_unlock(lck);
+
+                       object->hashed = TRUE;
+               }
+               object->pager_created = backing_object->pager_created;
+               object->pager_control = backing_object->pager_control;
+               object->pager_ready = backing_object->pager_ready;
+               object->pager_initialized = backing_object->pager_initialized;
+               object->paging_offset =
+                   backing_object->paging_offset + backing_offset;
+               if (object->pager_control != MEMORY_OBJECT_CONTROL_NULL) {
+                       memory_object_control_collapse(object->pager_control,
+                                                      object);
+               }
+               /* the backing_object has lost its pager: reset all fields */
+               backing_object->pager_created = FALSE;
+               backing_object->pager_control = NULL;
+               backing_object->pager_ready = FALSE;
+               backing_object->paging_offset = 0;
+               backing_object->pager = NULL;
        }
        }
+       /*
+        *      Object now shadows whatever backing_object did.
+        *      Note that the reference to backing_object->shadow
+        *      moves from within backing_object to within object.
+        */
        
        
-       /* wait for object (if any) to be ready */
-       if (object != VM_OBJECT_NULL) {
-               vm_object_lock(object);
-               object->named = TRUE;
-               while (!object->pager_ready) {
-                       vm_object_sleep(object,
-                                       VM_OBJECT_EVENT_PAGER_READY,
-                                       THREAD_UNINT);
+       assert(!object->phys_contiguous);
+       assert(!backing_object->phys_contiguous);
+       object->shadow = backing_object->shadow;
+       if (object->shadow) {
+               object->vo_shadow_offset += backing_object->vo_shadow_offset;
+               /* "backing_object" gave its shadow to "object" */
+               backing_object->shadow = VM_OBJECT_NULL;
+               backing_object->vo_shadow_offset = 0;
+       } else {
+               /* no shadow, therefore no shadow offset... */
+               object->vo_shadow_offset = 0;
+       }
+       assert((object->shadow == VM_OBJECT_NULL) ||
+              (object->shadow->copy != backing_object));
+
+       /*
+        *      Discard backing_object.
+        *
+        *      Since the backing object has no pages, no
+        *      pager left, and no object references within it,
+        *      all that is necessary is to dispose of it.
+        */
+       object_collapses++;
+       
+       assert(backing_object->ref_count == 1);
+       assert(backing_object->resident_page_count == 0);
+       assert(backing_object->paging_in_progress == 0);
+       assert(backing_object->activity_in_progress == 0);
+       assert(backing_object->shadow == VM_OBJECT_NULL);
+       assert(backing_object->vo_shadow_offset == 0);
+
+       if (backing_object->pager != MEMORY_OBJECT_NULL) {
+               /* ... unless it has a pager; need to terminate pager too */
+               vm_counters.do_collapse_terminate++;
+               if (vm_object_terminate(backing_object) != KERN_SUCCESS) {
+                       vm_counters.do_collapse_terminate_failure++;
                }
                }
-               *control = object->pager_control;
-               vm_object_unlock(object);
+               return;
        }
        }
-       return (KERN_SUCCESS);
-}
 
 
+       assert(backing_object->pager == NULL);
 
 
-/*
- *     Routine:        memory_object_recover_named [user interface]
- *     Purpose:
- *             Attempt to recover a named reference for a VM object.
- *             VM will verify that the object has not already started
- *             down the termination path, and if it has, will optionally
- *             wait for that to finish.
- *     Returns:
- *             KERN_SUCCESS - we recovered a named reference on the object
- *             KERN_FAILURE - we could not recover a reference (object dead)
- *             KERN_INVALID_ARGUMENT - bad memory object control
- */
-kern_return_t
-memory_object_recover_named(
-       memory_object_control_t control,
-       boolean_t               wait_on_terminating)
-{
-       vm_object_t             object;
+       backing_object->alive = FALSE;
+       vm_object_unlock(backing_object);
 
 
-       object = memory_object_control_to_vm_object(control);
-       if (object == VM_OBJECT_NULL) {
-               return (KERN_INVALID_ARGUMENT);
-       }
-restart:
-       vm_object_lock(object);
+       XPR(XPR_VM_OBJECT, "vm_object_collapse, collapsed 0x%X\n",
+               backing_object, 0,0,0,0);
 
 
-       if (object->terminating && wait_on_terminating) {
-               vm_object_wait(object, 
-                       VM_OBJECT_EVENT_PAGING_IN_PROGRESS, 
-                       THREAD_UNINT);
-               goto restart;
+#if VM_OBJECT_TRACKING
+       if (vm_object_tracking_inited) {
+               btlog_remove_entries_for_element(vm_object_tracking_btlog,
+                                                backing_object);
        }
        }
+#endif /* VM_OBJECT_TRACKING */
 
 
-       if (!object->alive) {
-               vm_object_unlock(object);
-               return KERN_FAILURE;
-       }
+       vm_object_lock_destroy(backing_object);
 
 
-       if (object->named == TRUE) {
-               vm_object_unlock(object);
-               return KERN_SUCCESS;
-       }
-#if VM_OBJECT_CACHE
-       if ((object->ref_count == 0) && (!object->terminating)) {
-               if (!vm_object_cache_lock_try()) {
-                       vm_object_unlock(object);
-                       goto restart;
-               }
-               queue_remove(&vm_object_cached_list, object,
-                                    vm_object_t, cached_list);
-               vm_object_cached_count--;
-               XPR(XPR_VM_OBJECT_CACHE,
-                   "memory_object_recover_named: removing %X, head (%X, %X)\n",
-                   object, 
-                   vm_object_cached_list.next,
-                   vm_object_cached_list.prev, 0,0);
-               
-               vm_object_cache_unlock();
-       }
-#endif
-       object->named = TRUE;
-       vm_object_lock_assert_exclusive(object);
-       object->ref_count++;
-       vm_object_res_reference(object);
-       while (!object->pager_ready) {
-               vm_object_sleep(object,
-                               VM_OBJECT_EVENT_PAGER_READY,
-                               THREAD_UNINT);
-       }
-       vm_object_unlock(object);
-       return (KERN_SUCCESS);
+       zfree(vm_object_zone, backing_object);
+       
 }
 
 }
 
+static void
+vm_object_do_bypass(
+       vm_object_t object,
+       vm_object_t backing_object)
+{
+       /*
+        *      Make the parent shadow the next object
+        *      in the chain.
+        */
+       
+       vm_object_lock_assert_exclusive(object);
+       vm_object_lock_assert_exclusive(backing_object);
 
 
-/*
- *     vm_object_release_name:  
- *
- *     Enforces name semantic on memory_object reference count decrement
- *     This routine should not be called unless the caller holds a name
- *     reference gained through the memory_object_create_named.
- *
- *     If the TERMINATE_IDLE flag is set, the call will return if the
- *     reference count is not 1. i.e. idle with the only remaining reference
- *     being the name.
- *     If the decision is made to proceed the name field flag is set to
- *     false and the reference count is decremented.  If the RESPECT_CACHE
- *     flag is set and the reference count has gone to zero, the 
- *     memory_object is checked to see if it is cacheable otherwise when
- *     the reference count is zero, it is simply terminated.
- */
-
-__private_extern__ kern_return_t
-vm_object_release_name(
-       vm_object_t     object,
-       int             flags)
-{
-       vm_object_t     shadow;
-       boolean_t       original_object = TRUE;
-
-       while (object != VM_OBJECT_NULL) {
-
-               vm_object_lock(object);
+#if    TASK_SWAPPER
+       /*
+        *      Do object reference in-line to 
+        *      conditionally increment shadow's
+        *      residence count.  If object is not
+        *      resident, leave residence count
+        *      on shadow alone.
+        */
+       if (backing_object->shadow != VM_OBJECT_NULL) {
+               vm_object_lock(backing_object->shadow);
+               vm_object_lock_assert_exclusive(backing_object->shadow);
+               backing_object->shadow->ref_count++;
+               if (object->res_count != 0)
+                       vm_object_res_reference(backing_object->shadow);
+               vm_object_unlock(backing_object->shadow);
+       }
+#else  /* TASK_SWAPPER */
+       vm_object_reference(backing_object->shadow);
+#endif /* TASK_SWAPPER */
 
 
-               assert(object->alive);
-               if (original_object)
-                       assert(object->named);
-               assert(object->ref_count > 0);
+       assert(!object->phys_contiguous);
+       assert(!backing_object->phys_contiguous);
+       object->shadow = backing_object->shadow;
+       if (object->shadow) {
+               object->vo_shadow_offset += backing_object->vo_shadow_offset;
+       } else {
+               /* no shadow, therefore no shadow offset... */
+               object->vo_shadow_offset = 0;
+       }
+       
+       /*
+        *      Backing object might have had a copy pointer
+        *      to us.  If it did, clear it. 
+        */
+       if (backing_object->copy == object) {
+               backing_object->copy = VM_OBJECT_NULL;
+       }
+       
+       /*
+        *      Drop the reference count on backing_object.
+#if    TASK_SWAPPER
+        *      Since its ref_count was at least 2, it
+        *      will not vanish; so we don't need to call
+        *      vm_object_deallocate.
+        *      [with a caveat for "named" objects]
+        * 
+        *      The res_count on the backing object is
+        *      conditionally decremented.  It's possible
+        *      (via vm_pageout_scan) to get here with
+        *      a "swapped" object, which has a 0 res_count,
+        *      in which case, the backing object res_count
+        *      is already down by one.
+#else
+        *      Don't call vm_object_deallocate unless
+        *      ref_count drops to zero.
+        *
+        *      The ref_count can drop to zero here if the
+        *      backing object could be bypassed but not
+        *      collapsed, such as when the backing object
+        *      is temporary and cachable.
+#endif
+        */
+       if (backing_object->ref_count > 2 ||
+           (!backing_object->named && backing_object->ref_count > 1)) {
+               vm_object_lock_assert_exclusive(backing_object);
+               backing_object->ref_count--;
+#if    TASK_SWAPPER
+               if (object->res_count != 0)
+                       vm_object_res_deallocate(backing_object);
+               assert(backing_object->ref_count > 0);
+#endif /* TASK_SWAPPER */
+               vm_object_unlock(backing_object);
+       } else {
 
                /*
 
                /*
-                *      We have to wait for initialization before
-                *      destroying or caching the object.
+                *      Drop locks so that we can deallocate
+                *      the backing object.
                 */
 
                 */
 
-               if (object->pager_created && !object->pager_initialized) {
-                       assert(!object->can_persist);
-                       vm_object_assert_wait(object,
-                                       VM_OBJECT_EVENT_INITIALIZED,
-                                       THREAD_UNINT);
-                       vm_object_unlock(object);
-                       thread_block(THREAD_CONTINUE_NULL);
-                       continue;
+#if    TASK_SWAPPER
+               if (object->res_count == 0) {
+                       /* XXX get a reference for the deallocate below */
+                       vm_object_res_reference(backing_object);
                }
                }
+#endif /* TASK_SWAPPER */
+               /*
+                * vm_object_collapse (the caller of this function) is
+                * now called from contexts that may not guarantee that a
+                * valid reference is held on the object... w/o a valid
+                * reference, it is unsafe and unwise (you will definitely
+                * regret it) to unlock the object and then retake the lock
+                * since the object may be terminated and recycled in between.
+                * The "activity_in_progress" reference will keep the object
+                * 'stable'.
+                */
+               vm_object_activity_begin(object);
+               vm_object_unlock(object);
 
 
-               if (((object->ref_count > 1)
-                       && (flags & MEMORY_OBJECT_TERMINATE_IDLE))
-                       || (object->terminating)) {
-                       vm_object_unlock(object);
-                       return KERN_FAILURE;
-               } else {
-                       if (flags & MEMORY_OBJECT_RELEASE_NO_OP) {
-                               vm_object_unlock(object);
-                               return KERN_SUCCESS;
-                       }
-               }
-               
-               if ((flags & MEMORY_OBJECT_RESPECT_CACHE) &&
-                                       (object->ref_count == 1)) {
-                       if (original_object)
-                               object->named = FALSE;
-                       vm_object_unlock(object);
-                       /* let vm_object_deallocate push this thing into */
-                       /* the cache, if that it is where it is bound */
-                       vm_object_deallocate(object);
-                       return KERN_SUCCESS;
-               }
-               VM_OBJ_RES_DECR(object);
-               shadow = object->pageout?VM_OBJECT_NULL:object->shadow;
+               vm_object_unlock(backing_object);
+               vm_object_deallocate(backing_object);
 
 
-               if (object->ref_count == 1) {
-                       if (vm_object_terminate(object) != KERN_SUCCESS) {
-                               if (original_object) {
-                                       return KERN_FAILURE;
-                               } else {
-                                       return KERN_SUCCESS;
-                               }
-                       }
-                       if (shadow != VM_OBJECT_NULL) {
-                               original_object = FALSE;
-                               object = shadow;
-                               continue;
-                       }
-                       return KERN_SUCCESS;
-               } else {
-                       vm_object_lock_assert_exclusive(object);
-                       object->ref_count--;
-                       assert(object->ref_count > 0);
-                       if(original_object)
-                               object->named = FALSE;
-                       vm_object_unlock(object);
-                       return KERN_SUCCESS;
-               }
+               /*
+                *      Relock object. We don't have to reverify
+                *      its state since vm_object_collapse will
+                *      do that for us as it starts at the
+                *      top of its loop.
+                */
+
+               vm_object_lock(object);
+               vm_object_activity_end(object);
        }
        }
-       /*NOTREACHED*/
-       assert(0);
-       return KERN_FAILURE;
+       
+       object_bypasses++;
 }
 
 }
 
+               
+/*
+ *     vm_object_collapse:
+ *
+ *     Perform an object collapse or an object bypass if appropriate.
+ *     The real work of collapsing and bypassing is performed in
+ *     the routines vm_object_do_collapse and vm_object_do_bypass.
+ *
+ *     Requires that the object be locked and the page queues be unlocked.
+ *
+ */
+static unsigned long vm_object_collapse_calls = 0;
+static unsigned long vm_object_collapse_objects = 0;
+static unsigned long vm_object_collapse_do_collapse = 0;
+static unsigned long vm_object_collapse_do_bypass = 0;
 
 
-__private_extern__ kern_return_t
-vm_object_lock_request(
-       vm_object_t                     object,
-       vm_object_offset_t              offset,
-       vm_object_size_t                size,
-       memory_object_return_t          should_return,
-       int                             flags,
-       vm_prot_t                       prot)
+__private_extern__ void
+vm_object_collapse(
+       vm_object_t                             object,
+       vm_object_offset_t                      hint_offset,
+       boolean_t                               can_bypass)
 {
 {
-       __unused boolean_t      should_flush;
+       vm_object_t                             backing_object;
+       unsigned int                            rcount;
+       unsigned int                            size;
+       vm_object_t                             original_object;
+       int                                     object_lock_type;
+       int                                     backing_object_lock_type;
 
 
-       should_flush = flags & MEMORY_OBJECT_DATA_FLUSH;
+       vm_object_collapse_calls++;
 
 
-        XPR(XPR_MEMORY_OBJECT,
-           "vm_o_lock_request, obj 0x%X off 0x%X size 0x%X flags %X prot %X\n",
-           object, offset, size, 
-           (((should_return&1)<<1)|should_flush), prot);
+       if (! vm_object_collapse_allowed &&
+           ! (can_bypass && vm_object_bypass_allowed)) {
+               return;
+       }
 
 
-       /*
-        *      Check for bogus arguments.
-        */
-       if (object == VM_OBJECT_NULL)
-               return (KERN_INVALID_ARGUMENT);
+       XPR(XPR_VM_OBJECT, "vm_object_collapse, obj 0x%X\n", 
+               object, 0,0,0,0);
 
 
-       if ((prot & ~VM_PROT_ALL) != 0 && prot != VM_PROT_NO_CHANGE)
-               return (KERN_INVALID_ARGUMENT);
+       if (object == VM_OBJECT_NULL)
+               return;
 
 
-       size = round_page_64(size);
+       original_object = object;
 
        /*
 
        /*
-        *      Lock the object, and acquire a paging reference to
-        *      prevent the memory_object reference from being released.
+        * The top object was locked "exclusive" by the caller.
+        * In the first pass, to determine if we can collapse the shadow chain,
+        * take a "shared" lock on the shadow objects.  If we can collapse,
+        * we'll have to go down the chain again with exclusive locks.
         */
         */
-       vm_object_lock(object);
-       vm_object_paging_begin(object);
-
-       (void)vm_object_update(object,
-               offset, size, NULL, NULL, should_return, flags, prot);
+       object_lock_type = OBJECT_LOCK_EXCLUSIVE;
+       backing_object_lock_type = OBJECT_LOCK_SHARED;
 
 
-       vm_object_paging_end(object);
-       vm_object_unlock(object);
+retry:
+       object = original_object;
+       vm_object_lock_assert_exclusive(object);
 
 
-       return (KERN_SUCCESS);
-}
+       while (TRUE) {
+               vm_object_collapse_objects++;
+               /*
+                *      Verify that the conditions are right for either
+                *      collapse or bypass:
+                */
 
 
-/*
- * Empty a purgeable object by grabbing the physical pages assigned to it and
- * putting them on the free queue without writing them to backing store, etc.
- * When the pages are next touched they will be demand zero-fill pages.  We
- * skip pages which are busy, being paged in/out, wired, etc.  We do _not_
- * skip referenced/dirty pages, pages on the active queue, etc.  We're more
+               /*
+                *      There is a backing object, and
+                */
+       
+               backing_object = object->shadow;
+               if (backing_object == VM_OBJECT_NULL) {
+                       if (object != original_object) {
+                               vm_object_unlock(object);
+                       }
+                       return;
+               }
+               if (backing_object_lock_type == OBJECT_LOCK_SHARED) {
+                       vm_object_lock_shared(backing_object);
+               } else {
+                       vm_object_lock(backing_object);
+               }
+
+               /*
+                *      No pages in the object are currently
+                *      being paged out, and
+                */
+               if (object->paging_in_progress != 0 ||
+                   object->activity_in_progress != 0) {
+                       /* try and collapse the rest of the shadow chain */
+                       if (object != original_object) {
+                               vm_object_unlock(object);
+                       }
+                       object = backing_object;
+                       object_lock_type = backing_object_lock_type;
+                       continue;
+               }
+
+               /*
+                *      ...
+                *              The backing object is not read_only,
+                *              and no pages in the backing object are
+                *              currently being paged out.
+                *              The backing object is internal.
+                *
+                */
+       
+               if (!backing_object->internal ||
+                   backing_object->paging_in_progress != 0 ||
+                   backing_object->activity_in_progress != 0) {
+                       /* try and collapse the rest of the shadow chain */
+                       if (object != original_object) {
+                               vm_object_unlock(object);
+                       }
+                       object = backing_object;
+                       object_lock_type = backing_object_lock_type;
+                       continue;
+               }
+
+               /*
+                * Purgeable objects are not supposed to engage in
+                * copy-on-write activities, so should not have
+                * any shadow objects or be a shadow object to another
+                * object.
+                * Collapsing a purgeable object would require some
+                * updates to the purgeable compressed ledgers.
+                */
+               if (object->purgable != VM_PURGABLE_DENY ||
+                   backing_object->purgable != VM_PURGABLE_DENY) {
+                       panic("vm_object_collapse() attempting to collapse "
+                             "purgeable object: %p(%d) %p(%d)\n",
+                             object, object->purgable,
+                             backing_object, backing_object->purgable);
+                       /* try and collapse the rest of the shadow chain */
+                       if (object != original_object) {
+                               vm_object_unlock(object);
+                       }
+                       object = backing_object;
+                       object_lock_type = backing_object_lock_type;
+                       continue;
+               }
+       
+               /*
+                *      The backing object can't be a copy-object:
+                *      the shadow_offset for the copy-object must stay
+                *      as 0.  Furthermore (for the 'we have all the
+                *      pages' case), if we bypass backing_object and
+                *      just shadow the next object in the chain, old
+                *      pages from that object would then have to be copied
+                *      BOTH into the (former) backing_object and into the
+                *      parent object.
+                */
+               if (backing_object->shadow != VM_OBJECT_NULL &&
+                   backing_object->shadow->copy == backing_object) {
+                       /* try and collapse the rest of the shadow chain */
+                       if (object != original_object) {
+                               vm_object_unlock(object);
+                       }
+                       object = backing_object;
+                       object_lock_type = backing_object_lock_type;
+                       continue;
+               }
+
+               /*
+                *      We can now try to either collapse the backing
+                *      object (if the parent is the only reference to
+                *      it) or (perhaps) remove the parent's reference
+                *      to it.
+                *
+                *      If there is exactly one reference to the backing
+                *      object, we may be able to collapse it into the
+                *      parent.
+                *
+                *      As long as one of the objects is still not known
+                *      to the pager, we can collapse them.
+                */
+               if (backing_object->ref_count == 1 &&
+                   (vm_object_collapse_compressor_allowed ||
+                    !object->pager_created 
+                    || (!backing_object->pager_created)
+                   ) && vm_object_collapse_allowed) {
+
+                       /*
+                        * We need the exclusive lock on the VM objects.
+                        */
+                       if (backing_object_lock_type != OBJECT_LOCK_EXCLUSIVE) {
+                               /*
+                                * We have an object and its shadow locked 
+                                * "shared".  We can't just upgrade the locks
+                                * to "exclusive", as some other thread might
+                                * also have these objects locked "shared" and
+                                * attempt to upgrade one or the other to 
+                                * "exclusive".  The upgrades would block
+                                * forever waiting for the other "shared" locks
+                                * to get released.
+                                * So we have to release the locks and go
+                                * down the shadow chain again (since it could
+                                * have changed) with "exclusive" locking.
+                                */
+                               vm_object_unlock(backing_object);
+                               if (object != original_object)
+                                       vm_object_unlock(object);
+                               object_lock_type = OBJECT_LOCK_EXCLUSIVE;
+                               backing_object_lock_type = OBJECT_LOCK_EXCLUSIVE;
+                               goto retry;
+                       }
+
+                       XPR(XPR_VM_OBJECT, 
+                  "vm_object_collapse: %x to %x, pager %x, pager_control %x\n",
+                               backing_object, object,
+                               backing_object->pager, 
+                               backing_object->pager_control, 0);
+
+                       /*
+                        *      Collapse the object with its backing
+                        *      object, and try again with the object's
+                        *      new backing object.
+                        */
+
+                       vm_object_do_collapse(object, backing_object);
+                       vm_object_collapse_do_collapse++;
+                       continue;
+               }
+
+               /*
+                *      Collapsing the backing object was not possible
+                *      or permitted, so let's try bypassing it.
+                */
+
+               if (! (can_bypass && vm_object_bypass_allowed)) {
+                       /* try and collapse the rest of the shadow chain */
+                       if (object != original_object) {
+                               vm_object_unlock(object);
+                       }
+                       object = backing_object;
+                       object_lock_type = backing_object_lock_type;
+                       continue;
+               }
+
+
+               /*
+                *      If the object doesn't have all its pages present,
+                *      we have to make sure no pages in the backing object
+                *      "show through" before bypassing it.
+                */
+               size = (unsigned int)atop(object->vo_size);
+               rcount = object->resident_page_count;
+
+               if (rcount != size) {
+                       vm_object_offset_t      offset;
+                       vm_object_offset_t      backing_offset;
+                       unsigned int            backing_rcount;
+
+                       /*
+                        *      If the backing object has a pager but no pagemap,
+                        *      then we cannot bypass it, because we don't know
+                        *      what pages it has.
+                        */
+                       if (backing_object->pager_created) {
+                               /* try and collapse the rest of the shadow chain */
+                               if (object != original_object) {
+                                       vm_object_unlock(object);
+                               }
+                               object = backing_object;
+                               object_lock_type = backing_object_lock_type;
+                               continue;
+                       }
+
+                       /*
+                        *      If the object has a pager but no pagemap,
+                        *      then we cannot bypass it, because we don't know
+                        *      what pages it has.
+                        */
+                       if (object->pager_created) {
+                               /* try and collapse the rest of the shadow chain */
+                               if (object != original_object) {
+                                       vm_object_unlock(object);
+                               }
+                               object = backing_object;
+                               object_lock_type = backing_object_lock_type;
+                               continue;
+                       }
+
+                       backing_offset = object->vo_shadow_offset;
+                       backing_rcount = backing_object->resident_page_count;
+
+                       if ( (int)backing_rcount - (int)(atop(backing_object->vo_size) - size) > (int)rcount) {
+                                /*
+                                * we have enough pages in the backing object to guarantee that
+                                * at least 1 of them must be 'uncovered' by a resident page
+                                * in the object we're evaluating, so move on and
+                                * try to collapse the rest of the shadow chain
+                                */
+                               if (object != original_object) {
+                                       vm_object_unlock(object);
+                               }
+                               object = backing_object;
+                               object_lock_type = backing_object_lock_type;
+                               continue;
+                       }
+
+                       /*
+                        *      If all of the pages in the backing object are
+                        *      shadowed by the parent object, the parent
+                        *      object no longer has to shadow the backing
+                        *      object; it can shadow the next one in the
+                        *      chain.
+                        *
+                        *      If the backing object has existence info,
+                        *      we must check examine its existence info
+                        *      as well.
+                        *
+                        */
+
+#define EXISTS_IN_OBJECT(obj, off, rc)                 \
+       ((VM_COMPRESSOR_PAGER_STATE_GET((obj), (off))   \
+         == VM_EXTERNAL_STATE_EXISTS) ||               \
+        ((rc) && vm_page_lookup((obj), (off)) != VM_PAGE_NULL && (rc)--))
+
+                       /*
+                        * Check the hint location first
+                        * (since it is often the quickest way out of here).
+                        */
+                       if (object->cow_hint != ~(vm_offset_t)0)
+                               hint_offset = (vm_object_offset_t)object->cow_hint;
+                       else
+                               hint_offset = (hint_offset > 8 * PAGE_SIZE_64) ?
+                                             (hint_offset - 8 * PAGE_SIZE_64) : 0;
+
+                       if (EXISTS_IN_OBJECT(backing_object, hint_offset +
+                                            backing_offset, backing_rcount) &&
+                           !EXISTS_IN_OBJECT(object, hint_offset, rcount)) {
+                               /* dependency right at the hint */
+                               object->cow_hint = (vm_offset_t) hint_offset; /* atomic */
+                               /* try and collapse the rest of the shadow chain */
+                               if (object != original_object) {
+                                       vm_object_unlock(object);
+                               }
+                               object = backing_object;
+                               object_lock_type = backing_object_lock_type;
+                               continue;
+                       }
+
+                       /*
+                        * If the object's window onto the backing_object
+                        * is large compared to the number of resident
+                        * pages in the backing object, it makes sense to
+                        * walk the backing_object's resident pages first.
+                        *
+                        * NOTE: Pages may be in both the existence map and/or
+                         * resident, so if we don't find a dependency while
+                        * walking the backing object's resident page list
+                        * directly, and there is an existence map, we'll have
+                        * to run the offset based 2nd pass.  Because we may
+                        * have to run both passes, we need to be careful
+                        * not to decrement 'rcount' in the 1st pass
+                        */
+                       if (backing_rcount && backing_rcount < (size / 8)) {
+                               unsigned int rc = rcount;
+                               vm_page_t p;
+
+                               backing_rcount = backing_object->resident_page_count;
+                               p = (vm_page_t)vm_page_queue_first(&backing_object->memq);
+                               do {
+                                       offset = (p->offset - backing_offset);
+
+                                       if (offset < object->vo_size &&
+                                           offset != hint_offset &&
+                                           !EXISTS_IN_OBJECT(object, offset, rc)) {
+                                               /* found a dependency */
+                                               object->cow_hint = (vm_offset_t) offset; /* atomic */
+                                               
+                                               break;
+                                       }
+                                       p = (vm_page_t) vm_page_queue_next(&p->listq);
+
+                               } while (--backing_rcount);
+                               if (backing_rcount != 0 ) {
+                                       /* try and collapse the rest of the shadow chain */
+                                       if (object != original_object) {
+                                               vm_object_unlock(object);
+                                       }
+                                       object = backing_object;
+                                       object_lock_type = backing_object_lock_type;
+                                       continue;
+                               }
+                       }
+
+                       /*
+                        * Walk through the offsets looking for pages in the
+                        * backing object that show through to the object.
+                        */
+                       if (backing_rcount) {
+                               offset = hint_offset;
+                               
+                               while((offset =
+                                     (offset + PAGE_SIZE_64 < object->vo_size) ?
+                                     (offset + PAGE_SIZE_64) : 0) != hint_offset) {
+
+                                       if (EXISTS_IN_OBJECT(backing_object, offset +
+                                           backing_offset, backing_rcount) &&
+                                           !EXISTS_IN_OBJECT(object, offset, rcount)) {
+                                               /* found a dependency */
+                                               object->cow_hint = (vm_offset_t) offset; /* atomic */
+                                               break;
+                                       }
+                               }
+                               if (offset != hint_offset) {
+                                       /* try and collapse the rest of the shadow chain */
+                                       if (object != original_object) {
+                                               vm_object_unlock(object);
+                                       }
+                                       object = backing_object;
+                                       object_lock_type = backing_object_lock_type;
+                                       continue;
+                               }
+                       }
+               }
+
+               /*
+                * We need "exclusive" locks on the 2 VM objects.
+                */
+               if (backing_object_lock_type != OBJECT_LOCK_EXCLUSIVE) {
+                       vm_object_unlock(backing_object);
+                       if (object != original_object)
+                               vm_object_unlock(object);
+                       object_lock_type = OBJECT_LOCK_EXCLUSIVE;
+                       backing_object_lock_type = OBJECT_LOCK_EXCLUSIVE;
+                       goto retry;
+               }
+
+               /* reset the offset hint for any objects deeper in the chain */
+               object->cow_hint = (vm_offset_t)0;
+
+               /*
+                *      All interesting pages in the backing object
+                *      already live in the parent or its pager.
+                *      Thus we can bypass the backing object.
+                */
+
+               vm_object_do_bypass(object, backing_object);
+               vm_object_collapse_do_bypass++;
+
+               /*
+                *      Try again with this object's new backing object.
+                */
+
+               continue;
+       }
+
+       /* NOT REACHED */
+       /*
+       if (object != original_object) {
+               vm_object_unlock(object);
+       }
+       */
+}
+
+/*
+ *     Routine:        vm_object_page_remove: [internal]
+ *     Purpose:
+ *             Removes all physical pages in the specified
+ *             object range from the object's list of pages.
+ *
+ *     In/out conditions:
+ *             The object must be locked.
+ *             The object must not have paging_in_progress, usually
+ *             guaranteed by not having a pager.
+ */
+unsigned int vm_object_page_remove_lookup = 0;
+unsigned int vm_object_page_remove_iterate = 0;
+
+__private_extern__ void
+vm_object_page_remove(
+       vm_object_t             object,
+       vm_object_offset_t      start,
+       vm_object_offset_t      end)
+{
+       vm_page_t       p, next;
+
+       /*
+        *      One and two page removals are most popular.
+        *      The factor of 16 here is somewhat arbitrary.
+        *      It balances vm_object_lookup vs iteration.
+        */
+
+       if (atop_64(end - start) < (unsigned)object->resident_page_count/16) {
+               vm_object_page_remove_lookup++;
+
+               for (; start < end; start += PAGE_SIZE_64) {
+                       p = vm_page_lookup(object, start);
+                       if (p != VM_PAGE_NULL) {
+                               assert(!p->cleaning && !p->laundry);
+                               if (!p->fictitious && p->pmapped)
+                                       pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(p));
+                               VM_PAGE_FREE(p);
+                       }
+               }
+       } else {
+               vm_object_page_remove_iterate++;
+
+               p = (vm_page_t) vm_page_queue_first(&object->memq);
+               while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t) p)) {
+                       next = (vm_page_t) vm_page_queue_next(&p->listq);
+                       if ((start <= p->offset) && (p->offset < end)) {
+                               assert(!p->cleaning && !p->laundry);
+                               if (!p->fictitious && p->pmapped)
+                                       pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(p));
+                               VM_PAGE_FREE(p);
+                       }
+                       p = next;
+               }
+       }
+}
+
+
+/*
+ *     Routine:        vm_object_coalesce
+ *     Function:       Coalesces two objects backing up adjoining
+ *                     regions of memory into a single object.
+ *
+ *     returns TRUE if objects were combined.
+ *
+ *     NOTE:   Only works at the moment if the second object is NULL -
+ *             if it's not, which object do we lock first?
+ *
+ *     Parameters:
+ *             prev_object     First object to coalesce
+ *             prev_offset     Offset into prev_object
+ *             next_object     Second object into coalesce
+ *             next_offset     Offset into next_object
+ *
+ *             prev_size       Size of reference to prev_object
+ *             next_size       Size of reference to next_object
+ *
+ *     Conditions:
+ *     The object(s) must *not* be locked. The map must be locked
+ *     to preserve the reference to the object(s).
+ */
+static int vm_object_coalesce_count = 0;
+
+__private_extern__ boolean_t
+vm_object_coalesce(
+       vm_object_t                     prev_object,
+       vm_object_t                     next_object,
+       vm_object_offset_t              prev_offset,
+       __unused vm_object_offset_t next_offset,
+       vm_object_size_t                prev_size,
+       vm_object_size_t                next_size)
+{
+       vm_object_size_t        newsize;
+
+#ifdef lint
+       next_offset++;
+#endif /* lint */
+
+       if (next_object != VM_OBJECT_NULL) {
+               return(FALSE);
+       }
+
+       if (prev_object == VM_OBJECT_NULL) {
+               return(TRUE);
+       }
+
+       XPR(XPR_VM_OBJECT,
+       "vm_object_coalesce: 0x%X prev_off 0x%X prev_size 0x%X next_size 0x%X\n",
+               prev_object, prev_offset, prev_size, next_size, 0);
+
+       vm_object_lock(prev_object);
+
+       /*
+        *      Try to collapse the object first
+        */
+       vm_object_collapse(prev_object, prev_offset, TRUE);
+
+       /*
+        *      Can't coalesce if pages not mapped to
+        *      prev_entry may be in use any way:
+        *      . more than one reference
+        *      . paged out
+        *      . shadows another object
+        *      . has a copy elsewhere
+        *      . is purgeable
+        *      . paging references (pages might be in page-list)
+        */
+
+       if ((prev_object->ref_count > 1) ||
+           prev_object->pager_created ||
+           (prev_object->shadow != VM_OBJECT_NULL) ||
+           (prev_object->copy != VM_OBJECT_NULL) ||
+           (prev_object->true_share != FALSE) ||
+           (prev_object->purgable != VM_PURGABLE_DENY) ||
+           (prev_object->paging_in_progress != 0) ||
+           (prev_object->activity_in_progress != 0)) {
+               vm_object_unlock(prev_object);
+               return(FALSE);
+       }
+
+       vm_object_coalesce_count++;
+
+       /*
+        *      Remove any pages that may still be in the object from
+        *      a previous deallocation.
+        */
+       vm_object_page_remove(prev_object,
+               prev_offset + prev_size,
+               prev_offset + prev_size + next_size);
+
+       /*
+        *      Extend the object if necessary.
+        */
+       newsize = prev_offset + prev_size + next_size;
+       if (newsize > prev_object->vo_size) {
+               prev_object->vo_size = newsize;
+       }
+
+       vm_object_unlock(prev_object);
+       return(TRUE);
+}
+
+kern_return_t
+vm_object_populate_with_private(
+               vm_object_t             object,
+               vm_object_offset_t      offset,
+               ppnum_t                 phys_page,
+               vm_size_t               size)
+{
+       ppnum_t                 base_page;
+       vm_object_offset_t      base_offset;
+
+
+       if (!object->private)
+               return KERN_FAILURE;
+
+       base_page = phys_page;
+
+       vm_object_lock(object);
+
+       if (!object->phys_contiguous) {
+               vm_page_t       m;
+
+               if ((base_offset = trunc_page_64(offset)) != offset) {
+                       vm_object_unlock(object);
+                       return KERN_FAILURE;
+               }
+               base_offset += object->paging_offset;
+
+               while (size) {
+                       m = vm_page_lookup(object, base_offset);
+
+                       if (m != VM_PAGE_NULL) {
+                               if (m->fictitious) {
+                                       if (VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr) {
+
+                                               vm_page_lockspin_queues();
+                                               m->private = TRUE;
+                                               vm_page_unlock_queues();
+
+                                               m->fictitious = FALSE;
+                                               VM_PAGE_SET_PHYS_PAGE(m, base_page);
+                                       }
+                               } else if (VM_PAGE_GET_PHYS_PAGE(m) != base_page) {
+
+                                       if ( !m->private) {
+                                               /*
+                                                * we'd leak a real page... that can't be right
+                                                */
+                                               panic("vm_object_populate_with_private - %p not private", m);
+                                       }
+                                       if (m->pmapped) {
+                                               /*
+                                                * pmap call to clear old mapping
+                                                */
+                                               pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
+                                       }
+                                       VM_PAGE_SET_PHYS_PAGE(m, base_page);
+                               }
+                               if (m->encrypted) {
+                                       /*
+                                        * we should never see this on a ficticious or private page
+                                        */
+                                       panic("vm_object_populate_with_private - %p encrypted", m);
+                               }
+
+                       } else {
+                               while ((m = vm_page_grab_fictitious()) == VM_PAGE_NULL)
+                                       vm_page_more_fictitious();      
+
+                               /*
+                                * private normally requires lock_queues but since we
+                                * are initializing the page, its not necessary here
+                                */
+                               m->private = TRUE;
+                               m->fictitious = FALSE;
+                               VM_PAGE_SET_PHYS_PAGE(m, base_page);
+                               m->unusual = TRUE;
+                               m->busy = FALSE;
+
+                               vm_page_insert(m, object, base_offset);
+                       }
+                       base_page++;                                                                    /* Go to the next physical page */
+                       base_offset += PAGE_SIZE;
+                       size -= PAGE_SIZE;
+               }
+       } else {
+               /* NOTE: we should check the original settings here */
+               /* if we have a size > zero a pmap call should be made */
+               /* to disable the range */      
+
+               /* pmap_? */
+               
+               /* shadows on contiguous memory are not allowed */
+               /* we therefore can use the offset field */
+               object->vo_shadow_offset = (vm_object_offset_t)phys_page << PAGE_SHIFT;
+               object->vo_size = size;
+       }
+       vm_object_unlock(object);
+
+       return KERN_SUCCESS;
+}
+
+/*
+ *     memory_object_free_from_cache:
+ *
+ *     Walk the vm_object cache list, removing and freeing vm_objects 
+ *     which are backed by the pager identified by the caller, (pager_ops).  
+ *     Remove up to "count" objects, if there are that may available
+ *     in the cache.
+ *
+ *     Walk the list at most once, return the number of vm_objects
+ *     actually freed.
+ */
+
+__private_extern__ kern_return_t
+memory_object_free_from_cache(
+       __unused host_t         host,
+       __unused memory_object_pager_ops_t pager_ops,
+       int             *count)
+{
+#if VM_OBJECT_CACHE
+       int     object_released = 0;
+
+       vm_object_t object = VM_OBJECT_NULL;
+       vm_object_t shadow;
+
+/*
+       if(host == HOST_NULL)
+               return(KERN_INVALID_ARGUMENT);
+*/
+
+ try_again:
+       vm_object_cache_lock();
+
+       queue_iterate(&vm_object_cached_list, object, 
+                                       vm_object_t, cached_list) {
+               if (object->pager &&
+                   (pager_ops == object->pager->mo_pager_ops)) {
+                       vm_object_lock(object);
+                       queue_remove(&vm_object_cached_list, object, 
+                                       vm_object_t, cached_list);
+                       vm_object_cached_count--;
+
+                       vm_object_cache_unlock();
+                       /*
+                       *       Since this object is in the cache, we know
+                       *       that it is initialized and has only a pager's
+                       *       (implicit) reference. Take a reference to avoid
+                       *       recursive deallocations.
+                       */
+
+                       assert(object->pager_initialized);
+                       assert(object->ref_count == 0);
+                       vm_object_lock_assert_exclusive(object);
+                       object->ref_count++;
+
+                       /*
+                       *       Terminate the object.
+                       *       If the object had a shadow, we let 
+                       *       vm_object_deallocate deallocate it. 
+                       *       "pageout" objects have a shadow, but
+                       *       maintain a "paging reference" rather 
+                       *       than a normal reference.
+                       *       (We are careful here to limit recursion.)
+                       */
+                       shadow = object->pageout?VM_OBJECT_NULL:object->shadow;
+
+                       if ((vm_object_terminate(object) == KERN_SUCCESS)
+                                       && (shadow != VM_OBJECT_NULL)) {
+                               vm_object_deallocate(shadow);
+                       }
+               
+                       if(object_released++ == *count)
+                               return KERN_SUCCESS;
+                       goto try_again;
+               }
+       }
+       vm_object_cache_unlock();
+       *count  = object_released;
+#else
+       *count = 0;
+#endif
+       return KERN_SUCCESS;
+}
+
+
+
+kern_return_t
+memory_object_create_named(
+       memory_object_t pager,
+       memory_object_offset_t  size,
+       memory_object_control_t         *control)
+{
+       vm_object_t             object;
+       vm_object_hash_entry_t  entry;
+       lck_mtx_t               *lck;
+
+       *control = MEMORY_OBJECT_CONTROL_NULL;
+       if (pager == MEMORY_OBJECT_NULL)
+               return KERN_INVALID_ARGUMENT;
+
+       lck = vm_object_hash_lock_spin(pager);
+       entry = vm_object_hash_lookup(pager, FALSE);
+
+       if ((entry != VM_OBJECT_HASH_ENTRY_NULL) &&
+                       (entry->object != VM_OBJECT_NULL)) {
+               if (entry->object->named == TRUE)
+                       panic("memory_object_create_named: caller already holds the right");    }
+       vm_object_hash_unlock(lck);
+
+       if ((object = vm_object_enter(pager, size, FALSE, FALSE, TRUE)) == VM_OBJECT_NULL) {
+               return(KERN_INVALID_OBJECT);
+       }
+       
+       /* wait for object (if any) to be ready */
+       if (object != VM_OBJECT_NULL) {
+               vm_object_lock(object);
+               object->named = TRUE;
+               while (!object->pager_ready) {
+                       vm_object_sleep(object,
+                                       VM_OBJECT_EVENT_PAGER_READY,
+                                       THREAD_UNINT);
+               }
+               *control = object->pager_control;
+               vm_object_unlock(object);
+       }
+       return (KERN_SUCCESS);
+}
+
+
+/*
+ *     Routine:        memory_object_recover_named [user interface]
+ *     Purpose:
+ *             Attempt to recover a named reference for a VM object.
+ *             VM will verify that the object has not already started
+ *             down the termination path, and if it has, will optionally
+ *             wait for that to finish.
+ *     Returns:
+ *             KERN_SUCCESS - we recovered a named reference on the object
+ *             KERN_FAILURE - we could not recover a reference (object dead)
+ *             KERN_INVALID_ARGUMENT - bad memory object control
+ */
+kern_return_t
+memory_object_recover_named(
+       memory_object_control_t control,
+       boolean_t               wait_on_terminating)
+{
+       vm_object_t             object;
+
+       object = memory_object_control_to_vm_object(control);
+       if (object == VM_OBJECT_NULL) {
+               return (KERN_INVALID_ARGUMENT);
+       }
+restart:
+       vm_object_lock(object);
+
+       if (object->terminating && wait_on_terminating) {
+               vm_object_wait(object, 
+                       VM_OBJECT_EVENT_PAGING_IN_PROGRESS, 
+                       THREAD_UNINT);
+               goto restart;
+       }
+
+       if (!object->alive) {
+               vm_object_unlock(object);
+               return KERN_FAILURE;
+       }
+
+       if (object->named == TRUE) {
+               vm_object_unlock(object);
+               return KERN_SUCCESS;
+       }
+#if VM_OBJECT_CACHE
+       if ((object->ref_count == 0) && (!object->terminating)) {
+               if (!vm_object_cache_lock_try()) {
+                       vm_object_unlock(object);
+                       goto restart;
+               }
+               queue_remove(&vm_object_cached_list, object,
+                                    vm_object_t, cached_list);
+               vm_object_cached_count--;
+               XPR(XPR_VM_OBJECT_CACHE,
+                   "memory_object_recover_named: removing %X, head (%X, %X)\n",
+                   object, 
+                   vm_object_cached_list.next,
+                   vm_object_cached_list.prev, 0,0);
+               
+               vm_object_cache_unlock();
+       }
+#endif
+       object->named = TRUE;
+       vm_object_lock_assert_exclusive(object);
+       object->ref_count++;
+       vm_object_res_reference(object);
+       while (!object->pager_ready) {
+               vm_object_sleep(object,
+                               VM_OBJECT_EVENT_PAGER_READY,
+                               THREAD_UNINT);
+       }
+       vm_object_unlock(object);
+       return (KERN_SUCCESS);
+}
+
+
+/*
+ *     vm_object_release_name:  
+ *
+ *     Enforces name semantic on memory_object reference count decrement
+ *     This routine should not be called unless the caller holds a name
+ *     reference gained through the memory_object_create_named.
+ *
+ *     If the TERMINATE_IDLE flag is set, the call will return if the
+ *     reference count is not 1. i.e. idle with the only remaining reference
+ *     being the name.
+ *     If the decision is made to proceed the name field flag is set to
+ *     false and the reference count is decremented.  If the RESPECT_CACHE
+ *     flag is set and the reference count has gone to zero, the 
+ *     memory_object is checked to see if it is cacheable otherwise when
+ *     the reference count is zero, it is simply terminated.
+ */
+
+__private_extern__ kern_return_t
+vm_object_release_name(
+       vm_object_t     object,
+       int             flags)
+{
+       vm_object_t     shadow;
+       boolean_t       original_object = TRUE;
+
+       while (object != VM_OBJECT_NULL) {
+
+               vm_object_lock(object);
+
+               assert(object->alive);
+               if (original_object)
+                       assert(object->named);
+               assert(object->ref_count > 0);
+
+               /*
+                *      We have to wait for initialization before
+                *      destroying or caching the object.
+                */
+
+               if (object->pager_created && !object->pager_initialized) {
+                       assert(!object->can_persist);
+                       vm_object_assert_wait(object,
+                                       VM_OBJECT_EVENT_INITIALIZED,
+                                       THREAD_UNINT);
+                       vm_object_unlock(object);
+                       thread_block(THREAD_CONTINUE_NULL);
+                       continue;
+               }
+
+               if (((object->ref_count > 1)
+                       && (flags & MEMORY_OBJECT_TERMINATE_IDLE))
+                       || (object->terminating)) {
+                       vm_object_unlock(object);
+                       return KERN_FAILURE;
+               } else {
+                       if (flags & MEMORY_OBJECT_RELEASE_NO_OP) {
+                               vm_object_unlock(object);
+                               return KERN_SUCCESS;
+                       }
+               }
+               
+               if ((flags & MEMORY_OBJECT_RESPECT_CACHE) &&
+                                       (object->ref_count == 1)) {
+                       if (original_object)
+                               object->named = FALSE;
+                       vm_object_unlock(object);
+                       /* let vm_object_deallocate push this thing into */
+                       /* the cache, if that it is where it is bound */
+                       vm_object_deallocate(object);
+                       return KERN_SUCCESS;
+               }
+               VM_OBJ_RES_DECR(object);
+               shadow = object->pageout?VM_OBJECT_NULL:object->shadow;
+
+               if (object->ref_count == 1) {
+                       if (vm_object_terminate(object) != KERN_SUCCESS) {
+                               if (original_object) {
+                                       return KERN_FAILURE;
+                               } else {
+                                       return KERN_SUCCESS;
+                               }
+                       }
+                       if (shadow != VM_OBJECT_NULL) {
+                               original_object = FALSE;
+                               object = shadow;
+                               continue;
+                       }
+                       return KERN_SUCCESS;
+               } else {
+                       vm_object_lock_assert_exclusive(object);
+                       object->ref_count--;
+                       assert(object->ref_count > 0);
+                       if(original_object)
+                               object->named = FALSE;
+                       vm_object_unlock(object);
+                       return KERN_SUCCESS;
+               }
+       }
+       /*NOTREACHED*/
+       assert(0);
+       return KERN_FAILURE;
+}
+
+
+__private_extern__ kern_return_t
+vm_object_lock_request(
+       vm_object_t                     object,
+       vm_object_offset_t              offset,
+       vm_object_size_t                size,
+       memory_object_return_t          should_return,
+       int                             flags,
+       vm_prot_t                       prot)
+{
+       __unused boolean_t      should_flush;
+
+       should_flush = flags & MEMORY_OBJECT_DATA_FLUSH;
+
+        XPR(XPR_MEMORY_OBJECT,
+           "vm_o_lock_request, obj 0x%X off 0x%X size 0x%X flags %X prot %X\n",
+           object, offset, size, 
+           (((should_return&1)<<1)|should_flush), prot);
+
+       /*
+        *      Check for bogus arguments.
+        */
+       if (object == VM_OBJECT_NULL)
+               return (KERN_INVALID_ARGUMENT);
+
+       if ((prot & ~VM_PROT_ALL) != 0 && prot != VM_PROT_NO_CHANGE)
+               return (KERN_INVALID_ARGUMENT);
+
+       size = round_page_64(size);
+
+       /*
+        *      Lock the object, and acquire a paging reference to
+        *      prevent the memory_object reference from being released.
+        */
+       vm_object_lock(object);
+       vm_object_paging_begin(object);
+
+       (void)vm_object_update(object,
+               offset, size, NULL, NULL, should_return, flags, prot);
+
+       vm_object_paging_end(object);
+       vm_object_unlock(object);
+
+       return (KERN_SUCCESS);
+}
+
+/*
+ * Empty a purgeable object by grabbing the physical pages assigned to it and
+ * putting them on the free queue without writing them to backing store, etc.
+ * When the pages are next touched they will be demand zero-fill pages.  We
+ * skip pages which are busy, being paged in/out, wired, etc.  We do _not_
+ * skip referenced/dirty pages, pages on the active queue, etc.  We're more
  * than happy to grab these since this is a purgeable object.  We mark the
  * object as "empty" after reaping its pages.
  *
  * than happy to grab these since this is a purgeable object.  We mark the
  * object as "empty" after reaping its pages.
  *
- * On entry the object must be locked and it must be
- * purgeable with no delayed copies pending.
+ * On entry the object must be locked and it must be
+ * purgeable with no delayed copies pending.
+ */
+void
+vm_object_purge(vm_object_t object, int flags)
+{
+       unsigned int    object_page_count = 0;
+       unsigned int    pgcount = 0;
+       boolean_t       skipped_object = FALSE;
+
+        vm_object_lock_assert_exclusive(object);
+
+       if (object->purgable == VM_PURGABLE_DENY)
+               return;
+
+       assert(object->copy == VM_OBJECT_NULL);
+       assert(object->copy_strategy == MEMORY_OBJECT_COPY_NONE);
+
+       /*
+        * We need to set the object's state to VM_PURGABLE_EMPTY *before*
+        * reaping its pages.  We update vm_page_purgeable_count in bulk
+        * and we don't want vm_page_remove() to update it again for each
+        * page we reap later.
+        *
+        * For the purgeable ledgers, pages from VOLATILE and EMPTY objects
+        * are all accounted for in the "volatile" ledgers, so this does not
+        * make any difference.
+        * If we transitioned directly from NONVOLATILE to EMPTY,
+        * vm_page_purgeable_count must have been updated when the object
+        * was dequeued from its volatile queue and the purgeable ledgers
+        * must have also been updated accordingly at that time (in
+        * vm_object_purgable_control()).
+        */
+       if (object->purgable == VM_PURGABLE_VOLATILE) {
+               unsigned int delta;
+               assert(object->resident_page_count >=
+                      object->wired_page_count);
+               delta = (object->resident_page_count -
+                        object->wired_page_count);
+               if (delta != 0) {
+                       assert(vm_page_purgeable_count >=
+                              delta);
+                       OSAddAtomic(-delta,
+                                   (SInt32 *)&vm_page_purgeable_count);
+               }
+               if (object->wired_page_count != 0) {
+                       assert(vm_page_purgeable_wired_count >=
+                              object->wired_page_count);
+                       OSAddAtomic(-object->wired_page_count,
+                                   (SInt32 *)&vm_page_purgeable_wired_count);
+               }
+               object->purgable = VM_PURGABLE_EMPTY;
+       }
+       assert(object->purgable == VM_PURGABLE_EMPTY);
+       
+       object_page_count = object->resident_page_count;
+
+       vm_object_reap_pages(object, REAP_PURGEABLE);
+
+       if (object->pager != NULL) {
+
+               assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
+
+               if (object->activity_in_progress == 0 &&
+                   object->paging_in_progress == 0) {
+                       /*
+                        * Also reap any memory coming from this object
+                        * in the VM compressor.
+                        *
+                        * There are no operations in progress on the VM object
+                        * and no operation can start while we're holding the
+                        * VM object lock, so it's safe to reap the compressed
+                        * pages and update the page counts.
+                        */
+                       pgcount = vm_compressor_pager_get_count(object->pager);
+                       if (pgcount) {
+                               pgcount = vm_compressor_pager_reap_pages(object->pager, flags);
+                               vm_compressor_pager_count(object->pager,
+                                                         -pgcount,
+                                                         FALSE, /* shared */
+                                                         object);
+                               vm_purgeable_compressed_update(object,
+                                                              -pgcount);
+                       }
+                       if ( !(flags & C_DONT_BLOCK)) {
+                               assert(vm_compressor_pager_get_count(object->pager)
+                                      == 0);
+                       }
+               } else {
+                       /*
+                        * There's some kind of paging activity in progress
+                        * for this object, which could result in a page
+                        * being compressed or decompressed, possibly while
+                        * the VM object is not locked, so it could race
+                        * with us.
+                        *
+                        * We can't really synchronize this without possibly 
+                        * causing a deadlock when the compressor needs to
+                        * allocate or free memory while compressing or
+                        * decompressing a page from a purgeable object
+                        * mapped in the kernel_map...
+                        *
+                        * So let's not attempt to purge the compressor
+                        * pager if there's any kind of operation in
+                        * progress on the VM object.
+                        */
+                       skipped_object = TRUE;
+               }
+       }
+
+       vm_object_lock_assert_exclusive(object);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (MACHDBG_CODE(DBG_MACH_VM, OBJECT_PURGE_ONE)),
+                             VM_KERNEL_UNSLIDE_OR_PERM(object), /* purged object */
+                             object_page_count,
+                             pgcount,
+                             skipped_object,
+                             0);
+
+}
+                               
+
+/*
+ * vm_object_purgeable_control() allows the caller to control and investigate the
+ * state of a purgeable object.  A purgeable object is created via a call to
+ * vm_allocate() with VM_FLAGS_PURGABLE specified.  A purgeable object will
+ * never be coalesced with any other object -- even other purgeable objects --
+ * and will thus always remain a distinct object.  A purgeable object has
+ * special semantics when its reference count is exactly 1.  If its reference
+ * count is greater than 1, then a purgeable object will behave like a normal
+ * object and attempts to use this interface will result in an error return
+ * of KERN_INVALID_ARGUMENT.
+ *
+ * A purgeable object may be put into a "volatile" state which will make the
+ * object's pages elligable for being reclaimed without paging to backing
+ * store if the system runs low on memory.  If the pages in a volatile
+ * purgeable object are reclaimed, the purgeable object is said to have been
+ * "emptied."  When a purgeable object is emptied the system will reclaim as
+ * many pages from the object as it can in a convenient manner (pages already
+ * en route to backing store or busy for other reasons are left as is).  When
+ * a purgeable object is made volatile, its pages will generally be reclaimed
+ * before other pages in the application's working set.  This semantic is
+ * generally used by applications which can recreate the data in the object
+ * faster than it can be paged in.  One such example might be media assets
+ * which can be reread from a much faster RAID volume.
+ *
+ * A purgeable object may be designated as "non-volatile" which means it will
+ * behave like all other objects in the system with pages being written to and
+ * read from backing store as needed to satisfy system memory needs.  If the
+ * object was emptied before the object was made non-volatile, that fact will
+ * be returned as the old state of the purgeable object (see
+ * VM_PURGABLE_SET_STATE below).  In this case, any pages of the object which
+ * were reclaimed as part of emptying the object will be refaulted in as
+ * zero-fill on demand.  It is up to the application to note that an object
+ * was emptied and recreate the objects contents if necessary.  When a
+ * purgeable object is made non-volatile, its pages will generally not be paged
+ * out to backing store in the immediate future.  A purgeable object may also
+ * be manually emptied.
+ *
+ * Finally, the current state (non-volatile, volatile, volatile & empty) of a
+ * volatile purgeable object may be queried at any time.  This information may
+ * be used as a control input to let the application know when the system is
+ * experiencing memory pressure and is reclaiming memory.
+ *
+ * The specified address may be any address within the purgeable object.  If
+ * the specified address does not represent any object in the target task's
+ * virtual address space, then KERN_INVALID_ADDRESS will be returned.  If the
+ * object containing the specified address is not a purgeable object, then
+ * KERN_INVALID_ARGUMENT will be returned.  Otherwise, KERN_SUCCESS will be
+ * returned.
+ *
+ * The control parameter may be any one of VM_PURGABLE_SET_STATE or
+ * VM_PURGABLE_GET_STATE.  For VM_PURGABLE_SET_STATE, the in/out parameter
+ * state is used to set the new state of the purgeable object and return its
+ * old state.  For VM_PURGABLE_GET_STATE, the current state of the purgeable
+ * object is returned in the parameter state.
+ *
+ * The in/out parameter state may be one of VM_PURGABLE_NONVOLATILE,
+ * VM_PURGABLE_VOLATILE or VM_PURGABLE_EMPTY.  These, respectively, represent
+ * the non-volatile, volatile and volatile/empty states described above.
+ * Setting the state of a purgeable object to VM_PURGABLE_EMPTY will
+ * immediately reclaim as many pages in the object as can be conveniently
+ * collected (some may have already been written to backing store or be
+ * otherwise busy).
+ *
+ * The process of making a purgeable object non-volatile and determining its
+ * previous state is atomic.  Thus, if a purgeable object is made
+ * VM_PURGABLE_NONVOLATILE and the old state is returned as
+ * VM_PURGABLE_VOLATILE, then the purgeable object's previous contents are
+ * completely intact and will remain so until the object is made volatile
+ * again.  If the old state is returned as VM_PURGABLE_EMPTY then the object
+ * was reclaimed while it was in a volatile state and its previous contents
+ * have been lost.
+ */
+/*
+ * The object must be locked.
+ */
+kern_return_t
+vm_object_purgable_control(
+       vm_object_t     object,
+       vm_purgable_t   control,
+       int             *state)
+{
+       int             old_state;
+       int             new_state;
+
+       if (object == VM_OBJECT_NULL) {
+               /*
+                * Object must already be present or it can't be purgeable.
+                */
+               return KERN_INVALID_ARGUMENT;
+       }
+
+       vm_object_lock_assert_exclusive(object);
+
+       /*
+        * Get current state of the purgeable object.
+        */
+       old_state = object->purgable;
+       if (old_state == VM_PURGABLE_DENY)
+               return KERN_INVALID_ARGUMENT;
+    
+       /* purgeable cant have delayed copies - now or in the future */
+       assert(object->copy == VM_OBJECT_NULL); 
+       assert(object->copy_strategy == MEMORY_OBJECT_COPY_NONE);
+
+       /*
+        * Execute the desired operation.
+        */
+       if (control == VM_PURGABLE_GET_STATE) {
+               *state = old_state;
+               return KERN_SUCCESS;
+       }
+
+       if ((*state) & VM_PURGABLE_DEBUG_EMPTY) {
+               object->volatile_empty = TRUE;
+       }
+       if ((*state) & VM_PURGABLE_DEBUG_FAULT) {
+               object->volatile_fault = TRUE;
+       }
+
+       new_state = *state & VM_PURGABLE_STATE_MASK;
+       if (new_state == VM_PURGABLE_VOLATILE &&
+           object->volatile_empty) {
+               new_state = VM_PURGABLE_EMPTY;
+       }
+
+       switch (new_state) {
+       case VM_PURGABLE_DENY:
+       case VM_PURGABLE_NONVOLATILE:
+               object->purgable = new_state;
+
+               if (old_state == VM_PURGABLE_VOLATILE) {
+                       unsigned int delta;
+
+                       assert(object->resident_page_count >=
+                              object->wired_page_count);
+                       delta = (object->resident_page_count -
+                                object->wired_page_count);
+
+                       assert(vm_page_purgeable_count >= delta);
+
+                       if (delta != 0) {
+                               OSAddAtomic(-delta,
+                                           (SInt32 *)&vm_page_purgeable_count);
+                       }
+                       if (object->wired_page_count != 0) {
+                               assert(vm_page_purgeable_wired_count >=
+                                      object->wired_page_count);
+                               OSAddAtomic(-object->wired_page_count,
+                                           (SInt32 *)&vm_page_purgeable_wired_count);
+                       }
+
+                       vm_page_lock_queues();
+
+                       /* object should be on a queue */
+                       assert(object->objq.next != NULL &&
+                              object->objq.prev != NULL);
+                       purgeable_q_t queue;
+
+                       /*
+                        * Move object from its volatile queue to the
+                        * non-volatile queue...
+                        */
+                       queue = vm_purgeable_object_remove(object);
+                       assert(queue);
+
+                       if (object->purgeable_when_ripe) {
+                               vm_purgeable_token_delete_last(queue);
+                       }
+                       assert(queue->debug_count_objects>=0);
+
+                       vm_page_unlock_queues();
+               }
+               if (old_state == VM_PURGABLE_VOLATILE ||
+                   old_state == VM_PURGABLE_EMPTY) {
+                       /*
+                        * Transfer the object's pages from the volatile to
+                        * non-volatile ledgers.
+                        */
+                       vm_purgeable_accounting(object, VM_PURGABLE_VOLATILE,
+                                               FALSE);
+               }
+
+               break;
+
+       case VM_PURGABLE_VOLATILE:
+               if (object->volatile_fault) {
+                       vm_page_t       p;
+                       int             refmod;
+
+                       vm_page_queue_iterate(&object->memq, p, vm_page_t, listq) {
+                               if (p->busy ||
+                                   VM_PAGE_WIRED(p) ||
+                                   p->fictitious) {
+                                       continue;
+                               }
+                               refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(p));
+                               if ((refmod & VM_MEM_MODIFIED) &&
+                                   !p->dirty) {
+                                       SET_PAGE_DIRTY(p, FALSE);
+                               }
+                       }
+               }
+                                              
+               if (old_state == VM_PURGABLE_EMPTY &&
+                   object->resident_page_count == 0 &&
+                   object->pager == NULL)
+                       break;
+
+               purgeable_q_t queue;
+        
+               /* find the correct queue */
+               if ((*state&VM_PURGABLE_ORDERING_MASK) == VM_PURGABLE_ORDERING_OBSOLETE)
+                       queue = &purgeable_queues[PURGEABLE_Q_TYPE_OBSOLETE];
+               else {
+                       if ((*state&VM_PURGABLE_BEHAVIOR_MASK) == VM_PURGABLE_BEHAVIOR_FIFO)
+                               queue = &purgeable_queues[PURGEABLE_Q_TYPE_FIFO];
+                       else
+                               queue = &purgeable_queues[PURGEABLE_Q_TYPE_LIFO];
+               }
+        
+               if (old_state == VM_PURGABLE_NONVOLATILE ||
+                   old_state == VM_PURGABLE_EMPTY) {
+                       unsigned int delta;
+
+                       if ((*state & VM_PURGABLE_NO_AGING_MASK) ==
+                           VM_PURGABLE_NO_AGING) {
+                               object->purgeable_when_ripe = FALSE;
+                       } else {
+                               object->purgeable_when_ripe = TRUE;
+                       }
+                               
+                       if (object->purgeable_when_ripe) {
+                               kern_return_t result;
+
+                               /* try to add token... this can fail */
+                               vm_page_lock_queues();
+
+                               result = vm_purgeable_token_add(queue);
+                               if (result != KERN_SUCCESS) {
+                                       vm_page_unlock_queues();
+                                       return result;
+                               }
+                               vm_page_unlock_queues();
+                       }
+
+                       assert(object->resident_page_count >=
+                              object->wired_page_count);
+                       delta = (object->resident_page_count -
+                                object->wired_page_count);
+
+                       if (delta != 0) {
+                               OSAddAtomic(delta,
+                                           &vm_page_purgeable_count);
+                       }
+                       if (object->wired_page_count != 0) {
+                               OSAddAtomic(object->wired_page_count,
+                                           &vm_page_purgeable_wired_count);
+                       }
+
+                       object->purgable = new_state;
+
+                       /* object should be on "non-volatile" queue */
+                       assert(object->objq.next != NULL);
+                       assert(object->objq.prev != NULL);
+               }
+               else if (old_state == VM_PURGABLE_VOLATILE) {
+                       purgeable_q_t   old_queue;
+                       boolean_t       purgeable_when_ripe;
+
+                       /*
+                        * if reassigning priorities / purgeable groups, we don't change the
+                        * token queue. So moving priorities will not make pages stay around longer.
+                        * Reasoning is that the algorithm gives most priority to the most important
+                        * object. If a new token is added, the most important object' priority is boosted.
+                        * This biases the system already for purgeable queues that move a lot.
+                        * It doesn't seem more biasing is neccessary in this case, where no new object is added.
+                        */
+                       assert(object->objq.next != NULL && object->objq.prev != NULL); /* object should be on a queue */
+            
+                       old_queue = vm_purgeable_object_remove(object);
+                       assert(old_queue);
+            
+                       if ((*state & VM_PURGABLE_NO_AGING_MASK) ==
+                           VM_PURGABLE_NO_AGING) {
+                               purgeable_when_ripe = FALSE;
+                       } else {
+                               purgeable_when_ripe = TRUE;
+                       }
+                               
+                       if (old_queue != queue ||
+                           (purgeable_when_ripe !=
+                            object->purgeable_when_ripe)) {
+                               kern_return_t result;
+
+                               /* Changing queue. Have to move token. */
+                               vm_page_lock_queues();
+                               if (object->purgeable_when_ripe) {
+                                       vm_purgeable_token_delete_last(old_queue);
+                               }
+                               object->purgeable_when_ripe = purgeable_when_ripe;
+                               if (object->purgeable_when_ripe) {
+                                       result = vm_purgeable_token_add(queue);
+                                       assert(result==KERN_SUCCESS);   /* this should never fail since we just freed a token */
+                               }
+                               vm_page_unlock_queues();
+
+                       }
+               };
+               vm_purgeable_object_add(object, queue, (*state&VM_VOLATILE_GROUP_MASK)>>VM_VOLATILE_GROUP_SHIFT );
+               if (old_state == VM_PURGABLE_NONVOLATILE) {
+                       vm_purgeable_accounting(object, VM_PURGABLE_NONVOLATILE,
+                                               FALSE);
+               }
+
+               assert(queue->debug_count_objects>=0);
+        
+               break;
+
+
+       case VM_PURGABLE_EMPTY:
+               if (object->volatile_fault) {
+                       vm_page_t       p;
+                       int             refmod;
+
+                       vm_page_queue_iterate(&object->memq, p, vm_page_t, listq) {
+                               if (p->busy ||
+                                   VM_PAGE_WIRED(p) ||
+                                   p->fictitious) {
+                                       continue;
+                               }
+                               refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(p));
+                               if ((refmod & VM_MEM_MODIFIED) &&
+                                   !p->dirty) {
+                                       SET_PAGE_DIRTY(p, FALSE);
+                               }
+                       }
+               }
+
+               if (old_state == new_state) {
+                       /* nothing changes */
+                       break;
+               }
+
+               assert(old_state == VM_PURGABLE_NONVOLATILE ||
+                      old_state == VM_PURGABLE_VOLATILE);
+               if (old_state == VM_PURGABLE_VOLATILE) {
+                       purgeable_q_t old_queue;
+
+                       /* object should be on a queue */
+                       assert(object->objq.next != NULL &&
+                              object->objq.prev != NULL);
+
+                       old_queue = vm_purgeable_object_remove(object);
+                       assert(old_queue);
+                       if (object->purgeable_when_ripe) {
+                               vm_page_lock_queues();
+                               vm_purgeable_token_delete_first(old_queue);
+                               vm_page_unlock_queues();
+                       }
+               }
+
+               if (old_state == VM_PURGABLE_NONVOLATILE) {
+                       /*
+                        * This object's pages were previously accounted as
+                        * "non-volatile" and now need to be accounted as
+                        * "volatile".
+                        */
+                       vm_purgeable_accounting(object, VM_PURGABLE_NONVOLATILE,
+                                               FALSE);
+                       /*
+                        * Set to VM_PURGABLE_EMPTY because the pages are no
+                        * longer accounted in the "non-volatile" ledger
+                        * and are also not accounted for in
+                        * "vm_page_purgeable_count".
+                        */
+                       object->purgable = VM_PURGABLE_EMPTY;
+               }
+
+               (void) vm_object_purge(object, 0);
+               assert(object->purgable == VM_PURGABLE_EMPTY);
+
+               break;
+       }
+
+       *state = old_state;
+
+       vm_object_lock_assert_exclusive(object);
+
+       return KERN_SUCCESS;
+}
+
+kern_return_t
+vm_object_get_page_counts(
+       vm_object_t             object,
+       vm_object_offset_t      offset,
+       vm_object_size_t        size,
+       unsigned int            *resident_page_count,
+       unsigned int            *dirty_page_count)
+{
+
+       kern_return_t           kr = KERN_SUCCESS;
+       boolean_t               count_dirty_pages = FALSE;
+       vm_page_t               p = VM_PAGE_NULL;
+       unsigned int            local_resident_count = 0;
+       unsigned int            local_dirty_count = 0;
+       vm_object_offset_t      cur_offset = 0;
+       vm_object_offset_t      end_offset = 0;
+
+       if (object == VM_OBJECT_NULL)
+               return KERN_INVALID_ARGUMENT;
+
+
+       cur_offset = offset;
+       
+       end_offset = offset + size;
+
+       vm_object_lock_assert_exclusive(object);
+
+       if (dirty_page_count != NULL) {
+
+               count_dirty_pages = TRUE;
+       }
+
+       if (resident_page_count != NULL && count_dirty_pages == FALSE) {
+               /*
+                * Fast path when:
+                * - we only want the resident page count, and,
+                * - the entire object is exactly covered by the request.
+                */
+               if (offset == 0 && (object->vo_size == size)) {
+
+                       *resident_page_count = object->resident_page_count;
+                       goto out;
+               }
+       }
+
+       if (object->resident_page_count <= (size >> PAGE_SHIFT)) {
+
+               vm_page_queue_iterate(&object->memq, p, vm_page_t, listq) {
+               
+                       if (p->offset >= cur_offset && p->offset < end_offset) {
+
+                               local_resident_count++;
+
+                               if (count_dirty_pages) {
+                                       
+                                       if (p->dirty || (p->wpmapped && pmap_is_modified(VM_PAGE_GET_PHYS_PAGE(p)))) {
+                                               
+                                               local_dirty_count++;
+                                       }
+                               }
+                       }
+               }
+       } else {
+
+               for (cur_offset = offset; cur_offset < end_offset; cur_offset += PAGE_SIZE_64) {
+       
+                       p = vm_page_lookup(object, cur_offset);
+               
+                       if (p != VM_PAGE_NULL) {
+
+                               local_resident_count++;
+
+                               if (count_dirty_pages) {
+                                       
+                                       if (p->dirty || (p->wpmapped && pmap_is_modified(VM_PAGE_GET_PHYS_PAGE(p)))) {
+                               
+                                               local_dirty_count++;
+                                       }
+                               }
+                       }
+               }
+
+       }
+
+       if (resident_page_count != NULL) {
+               *resident_page_count = local_resident_count;
+       }
+
+       if (dirty_page_count != NULL) {
+               *dirty_page_count = local_dirty_count;
+       }
+
+out:
+       return kr;
+}
+
+
+#if    TASK_SWAPPER
+/*
+ * vm_object_res_deallocate
+ *
+ * (recursively) decrement residence counts on vm objects and their shadows.
+ * Called from vm_object_deallocate and when swapping out an object.
+ *
+ * The object is locked, and remains locked throughout the function,
+ * even as we iterate down the shadow chain.  Locks on intermediate objects
+ * will be dropped, but not the original object.
+ *
+ * NOTE: this function used to use recursion, rather than iteration.
+ */
+
+__private_extern__ void
+vm_object_res_deallocate(
+       vm_object_t     object)
+{
+       vm_object_t orig_object = object;
+       /*
+        * Object is locked so it can be called directly
+        * from vm_object_deallocate.  Original object is never
+        * unlocked.
+        */
+       assert(object->res_count > 0);
+       while  (--object->res_count == 0) {
+               assert(object->ref_count >= object->res_count);
+               vm_object_deactivate_all_pages(object);
+               /* iterate on shadow, if present */
+               if (object->shadow != VM_OBJECT_NULL) {
+                       vm_object_t tmp_object = object->shadow;
+                       vm_object_lock(tmp_object);
+                       if (object != orig_object)
+                               vm_object_unlock(object);
+                       object = tmp_object;
+                       assert(object->res_count > 0);
+               } else
+                       break;
+       }
+       if (object != orig_object)
+               vm_object_unlock(object);
+}
+
+/*
+ * vm_object_res_reference
+ *
+ * Internal function to increment residence count on a vm object
+ * and its shadows.  It is called only from vm_object_reference, and
+ * when swapping in a vm object, via vm_map_swap.
+ *
+ * The object is locked, and remains locked throughout the function,
+ * even as we iterate down the shadow chain.  Locks on intermediate objects
+ * will be dropped, but not the original object.
+ *
+ * NOTE: this function used to use recursion, rather than iteration.
+ */
+
+__private_extern__ void
+vm_object_res_reference(
+       vm_object_t     object)
+{
+       vm_object_t orig_object = object;
+       /* 
+        * Object is locked, so this can be called directly
+        * from vm_object_reference.  This lock is never released.
+        */
+       while  ((++object->res_count == 1)  && 
+               (object->shadow != VM_OBJECT_NULL)) {
+               vm_object_t tmp_object = object->shadow;
+
+               assert(object->ref_count >= object->res_count);
+               vm_object_lock(tmp_object);
+               if (object != orig_object)
+                       vm_object_unlock(object);
+               object = tmp_object;
+       }
+       if (object != orig_object)
+               vm_object_unlock(object);
+       assert(orig_object->ref_count >= orig_object->res_count);
+}
+#endif /* TASK_SWAPPER */
+
+/*
+ *     vm_object_reference:
+ *
+ *     Gets another reference to the given object.
+ */
+#ifdef vm_object_reference
+#undef vm_object_reference
+#endif
+__private_extern__ void
+vm_object_reference(
+       vm_object_t     object)
+{
+       if (object == VM_OBJECT_NULL)
+               return;
+
+       vm_object_lock(object);
+       assert(object->ref_count > 0);
+       vm_object_reference_locked(object);
+       vm_object_unlock(object);
+}
+
+#ifdef MACH_BSD
+/*
+ * Scale the vm_object_cache
+ * This is required to make sure that the vm_object_cache is big
+ * enough to effectively cache the mapped file.
+ * This is really important with UBC as all the regular file vnodes
+ * have memory object associated with them. Havving this cache too
+ * small results in rapid reclaim of vnodes and hurts performance a LOT!
+ *
+ * This is also needed as number of vnodes can be dynamically scaled.
+ */
+kern_return_t
+adjust_vm_object_cache(
+       __unused vm_size_t oval,
+       __unused vm_size_t nval)
+{
+#if VM_OBJECT_CACHE
+       vm_object_cached_max = nval;
+       vm_object_cache_trim(FALSE);
+#endif
+       return (KERN_SUCCESS);
+}
+#endif /* MACH_BSD */
+
+
+/*
+ * vm_object_transpose
+ *
+ * This routine takes two VM objects of the same size and exchanges
+ * their backing store.
+ * The objects should be "quiesced" via a UPL operation with UPL_SET_IO_WIRE
+ * and UPL_BLOCK_ACCESS if they are referenced anywhere.
+ *
+ * The VM objects must not be locked by caller.
  */
  */
-void
-vm_object_purge(vm_object_t object)
+unsigned int vm_object_transpose_count = 0;
+kern_return_t
+vm_object_transpose(
+       vm_object_t             object1,
+       vm_object_t             object2,
+       vm_object_size_t        transpose_size)
 {
 {
-        vm_object_lock_assert_exclusive(object);
+       vm_object_t             tmp_object;
+       kern_return_t           retval;
+       boolean_t               object1_locked, object2_locked;
+       vm_page_t               page;
+       vm_object_offset_t      page_offset;
+       lck_mtx_t               *hash_lck;
+       vm_object_hash_entry_t  hash_entry;
+
+       tmp_object = VM_OBJECT_NULL;
+       object1_locked = FALSE; object2_locked = FALSE;
+
+       if (object1 == object2 ||
+           object1 == VM_OBJECT_NULL ||
+           object2 == VM_OBJECT_NULL) {
+               /*
+                * If the 2 VM objects are the same, there's
+                * no point in exchanging their backing store.
+                */
+               retval = KERN_INVALID_VALUE;
+               goto done;
+       }
+
+       /*
+        * Since we need to lock both objects at the same time,
+        * make sure we always lock them in the same order to
+        * avoid deadlocks.
+        */
+       if (object1 >  object2) {
+               tmp_object = object1;
+               object1 = object2;
+               object2 = tmp_object;
+       }
+
+       /*
+        * Allocate a temporary VM object to hold object1's contents
+        * while we copy object2 to object1.
+        */
+       tmp_object = vm_object_allocate(transpose_size);
+       vm_object_lock(tmp_object);
+       tmp_object->can_persist = FALSE;
+
+
+       /*
+        * Grab control of the 1st VM object.
+        */
+       vm_object_lock(object1);
+       object1_locked = TRUE;
+       if (!object1->alive || object1->terminating ||
+           object1->copy || object1->shadow || object1->shadowed ||
+           object1->purgable != VM_PURGABLE_DENY) {
+               /*
+                * We don't deal with copy or shadow objects (yet).
+                */
+               retval = KERN_INVALID_VALUE;
+               goto done;
+       }
+       /*
+        * We're about to mess with the object's backing store and 
+        * taking a "paging_in_progress" reference wouldn't be enough
+        * to prevent any paging activity on this object, so the caller should
+        * have "quiesced" the objects beforehand, via a UPL operation with
+        * UPL_SET_IO_WIRE (to make sure all the pages are there and wired)
+        * and UPL_BLOCK_ACCESS (to mark the pages "busy").
+        * 
+        * Wait for any paging operation to complete (but only paging, not 
+        * other kind of activities not linked to the pager).  After we're
+        * statisfied that there's no more paging in progress, we keep the
+        * object locked, to guarantee that no one tries to access its pager.
+        */
+       vm_object_paging_only_wait(object1, THREAD_UNINT);
+
+       /*
+        * Same as above for the 2nd object...
+        */
+       vm_object_lock(object2);
+       object2_locked = TRUE;
+       if (! object2->alive || object2->terminating ||
+           object2->copy || object2->shadow || object2->shadowed ||
+           object2->purgable != VM_PURGABLE_DENY) {
+               retval = KERN_INVALID_VALUE;
+               goto done;
+       }
+       vm_object_paging_only_wait(object2, THREAD_UNINT);
+
+
+       if (object1->vo_size != object2->vo_size ||
+           object1->vo_size != transpose_size) {
+               /*
+                * If the 2 objects don't have the same size, we can't
+                * exchange their backing stores or one would overflow.
+                * If their size doesn't match the caller's
+                * "transpose_size", we can't do it either because the
+                * transpose operation will affect the entire span of 
+                * the objects.
+                */
+               retval = KERN_INVALID_VALUE;
+               goto done;
+       }
+
+
+       /*
+        * Transpose the lists of resident pages.
+        * This also updates the resident_page_count and the memq_hint.
+        */
+       if (object1->phys_contiguous || vm_page_queue_empty(&object1->memq)) {
+               /*
+                * No pages in object1, just transfer pages
+                * from object2 to object1.  No need to go through
+                * an intermediate object.
+                */
+               while (!vm_page_queue_empty(&object2->memq)) {
+                       page = (vm_page_t) vm_page_queue_first(&object2->memq);
+                       vm_page_rename(page, object1, page->offset, FALSE);
+               }
+               assert(vm_page_queue_empty(&object2->memq));
+       } else if (object2->phys_contiguous || vm_page_queue_empty(&object2->memq)) {
+               /*
+                * No pages in object2, just transfer pages
+                * from object1 to object2.  No need to go through
+                * an intermediate object.
+                */
+               while (!vm_page_queue_empty(&object1->memq)) {
+                       page = (vm_page_t) vm_page_queue_first(&object1->memq);
+                       vm_page_rename(page, object2, page->offset, FALSE);
+               }
+               assert(vm_page_queue_empty(&object1->memq));
+       } else {
+               /* transfer object1's pages to tmp_object */
+               while (!vm_page_queue_empty(&object1->memq)) {
+                       page = (vm_page_t) vm_page_queue_first(&object1->memq);
+                       page_offset = page->offset;
+                       vm_page_remove(page, TRUE);
+                       page->offset = page_offset;
+                       vm_page_queue_enter(&tmp_object->memq, page, vm_page_t, listq);
+               }
+               assert(vm_page_queue_empty(&object1->memq));
+               /* transfer object2's pages to object1 */
+               while (!vm_page_queue_empty(&object2->memq)) {
+                       page = (vm_page_t) vm_page_queue_first(&object2->memq);
+                       vm_page_rename(page, object1, page->offset, FALSE);
+               }
+               assert(vm_page_queue_empty(&object2->memq));
+               /* transfer tmp_object's pages to object2 */
+               while (!vm_page_queue_empty(&tmp_object->memq)) {
+                       page = (vm_page_t) vm_page_queue_first(&tmp_object->memq);
+                       vm_page_queue_remove(&tmp_object->memq, page,
+                                            vm_page_t, listq);
+                       vm_page_insert(page, object2, page->offset);
+               }
+               assert(vm_page_queue_empty(&tmp_object->memq));
+       }
+
+#define __TRANSPOSE_FIELD(field)                               \
+MACRO_BEGIN                                                    \
+       tmp_object->field = object1->field;                     \
+       object1->field = object2->field;                        \
+       object2->field = tmp_object->field;                     \
+MACRO_END
+
+       /* "Lock" refers to the object not its contents */
+       /* "size" should be identical */
+       assert(object1->vo_size == object2->vo_size);
+       /* "memq_hint" was updated above when transposing pages */
+       /* "ref_count" refers to the object not its contents */
+#if TASK_SWAPPER
+       /* "res_count" refers to the object not its contents */
+#endif
+       /* "resident_page_count" was updated above when transposing pages */
+       /* "wired_page_count" was updated above when transposing pages */
+       /* "reusable_page_count" was updated above when transposing pages */
+       /* there should be no "copy" */
+       assert(!object1->copy);
+       assert(!object2->copy);
+       /* there should be no "shadow" */
+       assert(!object1->shadow);
+       assert(!object2->shadow);
+       __TRANSPOSE_FIELD(vo_shadow_offset); /* used by phys_contiguous objects */
+       __TRANSPOSE_FIELD(pager);
+       __TRANSPOSE_FIELD(paging_offset);
+       __TRANSPOSE_FIELD(pager_control);
+       /* update the memory_objects' pointers back to the VM objects */
+       if (object1->pager_control != MEMORY_OBJECT_CONTROL_NULL) {
+               memory_object_control_collapse(object1->pager_control,
+                                              object1);
+       }
+       if (object2->pager_control != MEMORY_OBJECT_CONTROL_NULL) {
+               memory_object_control_collapse(object2->pager_control,
+                                              object2);
+       }
+       __TRANSPOSE_FIELD(copy_strategy);
+       /* "paging_in_progress" refers to the object not its contents */
+       assert(!object1->paging_in_progress);
+       assert(!object2->paging_in_progress);
+       assert(object1->activity_in_progress);
+       assert(object2->activity_in_progress);
+       /* "all_wanted" refers to the object not its contents */
+       __TRANSPOSE_FIELD(pager_created);
+       __TRANSPOSE_FIELD(pager_initialized);
+       __TRANSPOSE_FIELD(pager_ready);
+       __TRANSPOSE_FIELD(pager_trusted);
+       __TRANSPOSE_FIELD(can_persist);
+       __TRANSPOSE_FIELD(internal);
+       __TRANSPOSE_FIELD(temporary);
+       __TRANSPOSE_FIELD(private);
+       __TRANSPOSE_FIELD(pageout);
+       /* "alive" should be set */
+       assert(object1->alive);
+       assert(object2->alive);
+       /* "purgeable" should be non-purgeable */
+       assert(object1->purgable == VM_PURGABLE_DENY);
+       assert(object2->purgable == VM_PURGABLE_DENY);
+       /* "shadowed" refers to the the object not its contents */
+       __TRANSPOSE_FIELD(purgeable_when_ripe);
+       __TRANSPOSE_FIELD(advisory_pageout);
+       __TRANSPOSE_FIELD(true_share);
+       /* "terminating" should not be set */
+       assert(!object1->terminating);
+       assert(!object2->terminating);
+       __TRANSPOSE_FIELD(named);
+       /* "shadow_severed" refers to the object not its contents */
+       __TRANSPOSE_FIELD(phys_contiguous);
+       __TRANSPOSE_FIELD(nophyscache);
+       /* "cached_list.next" points to transposed object */
+       object1->cached_list.next = (queue_entry_t) object2;
+       object2->cached_list.next = (queue_entry_t) object1;
+       /* "cached_list.prev" should be NULL */
+       assert(object1->cached_list.prev == NULL);
+       assert(object2->cached_list.prev == NULL);
+       /* "msr_q" is linked to the object not its contents */
+       assert(queue_empty(&object1->msr_q));
+       assert(queue_empty(&object2->msr_q));
+       __TRANSPOSE_FIELD(last_alloc);
+       __TRANSPOSE_FIELD(sequential);
+       __TRANSPOSE_FIELD(pages_created);
+       __TRANSPOSE_FIELD(pages_used);
+       __TRANSPOSE_FIELD(scan_collisions);
+       __TRANSPOSE_FIELD(cow_hint);
+#if MACH_ASSERT
+       __TRANSPOSE_FIELD(paging_object);
+#endif
+       __TRANSPOSE_FIELD(wimg_bits);
+       __TRANSPOSE_FIELD(set_cache_attr);
+       __TRANSPOSE_FIELD(code_signed);
+       if (object1->hashed) {
+               hash_lck = vm_object_hash_lock_spin(object2->pager);
+               hash_entry = vm_object_hash_lookup(object2->pager, FALSE);
+               assert(hash_entry != VM_OBJECT_HASH_ENTRY_NULL);
+               hash_entry->object = object2;
+               vm_object_hash_unlock(hash_lck);
+       }
+       if (object2->hashed) {
+               hash_lck = vm_object_hash_lock_spin(object1->pager);
+               hash_entry = vm_object_hash_lookup(object1->pager, FALSE);
+               assert(hash_entry != VM_OBJECT_HASH_ENTRY_NULL);
+               hash_entry->object = object1;
+               vm_object_hash_unlock(hash_lck);
+       }
+       __TRANSPOSE_FIELD(hashed);
+       object1->transposed = TRUE;
+       object2->transposed = TRUE;
+       __TRANSPOSE_FIELD(mapping_in_progress);
+       __TRANSPOSE_FIELD(volatile_empty);
+       __TRANSPOSE_FIELD(volatile_fault);
+       __TRANSPOSE_FIELD(all_reusable);
+       assert(object1->blocked_access);
+       assert(object2->blocked_access);
+       assert(object1->__object2_unused_bits == 0);
+       assert(object2->__object2_unused_bits == 0);
+#if UPL_DEBUG
+       /* "uplq" refers to the object not its contents (see upl_transpose()) */
+#endif
+       assert((object1->purgable == VM_PURGABLE_DENY) || (object1->objq.next == NULL));
+       assert((object1->purgable == VM_PURGABLE_DENY) || (object1->objq.prev == NULL));
+       assert((object2->purgable == VM_PURGABLE_DENY) || (object2->objq.next == NULL));
+       assert((object2->purgable == VM_PURGABLE_DENY) || (object2->objq.prev == NULL));
 
 
-       if (object->purgable == VM_PURGABLE_DENY)
-               return;
+#undef __TRANSPOSE_FIELD
 
 
-       assert(object->copy == VM_OBJECT_NULL);
-       assert(object->copy_strategy == MEMORY_OBJECT_COPY_NONE);
+       retval = KERN_SUCCESS;
 
 
-       if(object->purgable == VM_PURGABLE_VOLATILE) {
-               unsigned int delta;
-               assert(object->resident_page_count >=
-                      object->wired_page_count);
-               delta = (object->resident_page_count -
-                        object->wired_page_count);
-               if (delta != 0) {
-                       assert(vm_page_purgeable_count >=
-                              delta);
-                       OSAddAtomic(-delta,
-                                   (SInt32 *)&vm_page_purgeable_count);
-               }
-               if (object->wired_page_count != 0) {
-                       assert(vm_page_purgeable_wired_count >=
-                              object->wired_page_count);
-                       OSAddAtomic(-object->wired_page_count,
-                                   (SInt32 *)&vm_page_purgeable_wired_count);
-               }
+done:
+       /*
+        * Cleanup.
+        */
+       if (tmp_object != VM_OBJECT_NULL) {
+               vm_object_unlock(tmp_object);
+               /*
+                * Re-initialize the temporary object to avoid
+                * deallocating a real pager.
+                */
+               _vm_object_allocate(transpose_size, tmp_object);
+               vm_object_deallocate(tmp_object);
+               tmp_object = VM_OBJECT_NULL;
        }
        }
-       object->purgable = VM_PURGABLE_EMPTY;
-       
-       vm_object_reap_pages(object, REAP_PURGEABLE);
+
+       if (object1_locked) {
+               vm_object_unlock(object1);
+               object1_locked = FALSE;
+       }
+       if (object2_locked) {
+               vm_object_unlock(object2);
+               object2_locked = FALSE;
+       }
+
+       vm_object_transpose_count++;
+
+       return retval;
 }
 }
-                               
+
 
 /*
 
 /*
- * vm_object_purgeable_control() allows the caller to control and investigate the
- * state of a purgeable object.  A purgeable object is created via a call to
- * vm_allocate() with VM_FLAGS_PURGABLE specified.  A purgeable object will
- * never be coalesced with any other object -- even other purgeable objects --
- * and will thus always remain a distinct object.  A purgeable object has
- * special semantics when its reference count is exactly 1.  If its reference
- * count is greater than 1, then a purgeable object will behave like a normal
- * object and attempts to use this interface will result in an error return
- * of KERN_INVALID_ARGUMENT.
- *
- * A purgeable object may be put into a "volatile" state which will make the
- * object's pages elligable for being reclaimed without paging to backing
- * store if the system runs low on memory.  If the pages in a volatile
- * purgeable object are reclaimed, the purgeable object is said to have been
- * "emptied."  When a purgeable object is emptied the system will reclaim as
- * many pages from the object as it can in a convenient manner (pages already
- * en route to backing store or busy for other reasons are left as is).  When
- * a purgeable object is made volatile, its pages will generally be reclaimed
- * before other pages in the application's working set.  This semantic is
- * generally used by applications which can recreate the data in the object
- * faster than it can be paged in.  One such example might be media assets
- * which can be reread from a much faster RAID volume.
- *
- * A purgeable object may be designated as "non-volatile" which means it will
- * behave like all other objects in the system with pages being written to and
- * read from backing store as needed to satisfy system memory needs.  If the
- * object was emptied before the object was made non-volatile, that fact will
- * be returned as the old state of the purgeable object (see
- * VM_PURGABLE_SET_STATE below).  In this case, any pages of the object which
- * were reclaimed as part of emptying the object will be refaulted in as
- * zero-fill on demand.  It is up to the application to note that an object
- * was emptied and recreate the objects contents if necessary.  When a
- * purgeable object is made non-volatile, its pages will generally not be paged
- * out to backing store in the immediate future.  A purgeable object may also
- * be manually emptied.
- *
- * Finally, the current state (non-volatile, volatile, volatile & empty) of a
- * volatile purgeable object may be queried at any time.  This information may
- * be used as a control input to let the application know when the system is
- * experiencing memory pressure and is reclaiming memory.
+ *      vm_object_cluster_size
  *
  *
- * The specified address may be any address within the purgeable object.  If
- * the specified address does not represent any object in the target task's
- * virtual address space, then KERN_INVALID_ADDRESS will be returned.  If the
- * object containing the specified address is not a purgeable object, then
- * KERN_INVALID_ARGUMENT will be returned.  Otherwise, KERN_SUCCESS will be
- * returned.
+ *      Determine how big a cluster we should issue an I/O for...
  *
  *
- * The control parameter may be any one of VM_PURGABLE_SET_STATE or
- * VM_PURGABLE_GET_STATE.  For VM_PURGABLE_SET_STATE, the in/out parameter
- * state is used to set the new state of the purgeable object and return its
- * old state.  For VM_PURGABLE_GET_STATE, the current state of the purgeable
- * object is returned in the parameter state.
+ *     Inputs:   *start == offset of page needed
+ *               *length == maximum cluster pager can handle
+ *     Outputs:  *start == beginning offset of cluster
+ *               *length == length of cluster to try
  *
  *
- * The in/out parameter state may be one of VM_PURGABLE_NONVOLATILE,
- * VM_PURGABLE_VOLATILE or VM_PURGABLE_EMPTY.  These, respectively, represent
- * the non-volatile, volatile and volatile/empty states described above.
- * Setting the state of a purgeable object to VM_PURGABLE_EMPTY will
- * immediately reclaim as many pages in the object as can be conveniently
- * collected (some may have already been written to backing store or be
- * otherwise busy).
+ *     The original *start will be encompassed by the cluster
  *
  *
- * The process of making a purgeable object non-volatile and determining its
- * previous state is atomic.  Thus, if a purgeable object is made
- * VM_PURGABLE_NONVOLATILE and the old state is returned as
- * VM_PURGABLE_VOLATILE, then the purgeable object's previous contents are
- * completely intact and will remain so until the object is made volatile
- * again.  If the old state is returned as VM_PURGABLE_EMPTY then the object
- * was reclaimed while it was in a volatile state and its previous contents
- * have been lost.
  */
  */
+extern int speculative_reads_disabled;
+extern int ignore_is_ssd;
+
 /*
 /*
- * The object must be locked.
+ * Try to always keep these values an even multiple of PAGE_SIZE. We use these values
+ * to derive min_ph_bytes and max_ph_bytes (IMP: bytes not # of pages) and expect those values to
+ * always be page-aligned. The derivation could involve operations (e.g. division)
+ * that could give us non-page-size aligned values if we start out with values that
+ * are odd multiples of PAGE_SIZE.
  */
  */
-kern_return_t
-vm_object_purgable_control(
-       vm_object_t     object,
-       vm_purgable_t   control,
-       int             *state)
+       unsigned int preheat_max_bytes = MAX_UPL_TRANSFER_BYTES;
+unsigned int preheat_min_bytes = (1024 * 32);
+
+
+__private_extern__ void
+vm_object_cluster_size(vm_object_t object, vm_object_offset_t *start,
+                      vm_size_t *length, vm_object_fault_info_t fault_info, uint32_t *io_streaming)
 {
 {
-       int             old_state;
-       int             new_state;
+       vm_size_t               pre_heat_size;
+       vm_size_t               tail_size;
+       vm_size_t               head_size;
+       vm_size_t               max_length;
+       vm_size_t               cluster_size;
+       vm_object_offset_t      object_size;
+       vm_object_offset_t      orig_start;
+       vm_object_offset_t      target_start;
+       vm_object_offset_t      offset;
+       vm_behavior_t           behavior;
+       boolean_t               look_behind = TRUE;
+       boolean_t               look_ahead  = TRUE;
+       boolean_t               isSSD = FALSE;
+       uint32_t                throttle_limit;
+       int                     sequential_run;
+       int                     sequential_behavior = VM_BEHAVIOR_SEQUENTIAL;
+       vm_size_t               max_ph_size;
+       vm_size_t               min_ph_size;
 
 
-       if (object == VM_OBJECT_NULL) {
-               /*
-                * Object must already be present or it can't be purgeable.
-                */
-               return KERN_INVALID_ARGUMENT;
-       }
+       assert( !(*length & PAGE_MASK));
+       assert( !(*start & PAGE_MASK_64));
 
        /*
 
        /*
-        * Get current state of the purgeable object.
+        * remember maxiumum length of run requested
         */
         */
-       old_state = object->purgable;
-       if (old_state == VM_PURGABLE_DENY)
-               return KERN_INVALID_ARGUMENT;
-    
-       /* purgeable cant have delayed copies - now or in the future */
-       assert(object->copy == VM_OBJECT_NULL); 
-       assert(object->copy_strategy == MEMORY_OBJECT_COPY_NONE);
-
+       max_length = *length;
        /*
        /*
-        * Execute the desired operation.
+        * we'll always return a cluster size of at least
+        * 1 page, since the original fault must always
+        * be processed
         */
         */
-       if (control == VM_PURGABLE_GET_STATE) {
-               *state = old_state;
-               return KERN_SUCCESS;
-       }
-
-       if ((*state) & VM_PURGABLE_DEBUG_EMPTY) {
-               object->volatile_empty = TRUE;
-       }
-       if ((*state) & VM_PURGABLE_DEBUG_FAULT) {
-               object->volatile_fault = TRUE;
-       }
+       *length = PAGE_SIZE;
+       *io_streaming = 0;
 
 
-       new_state = *state & VM_PURGABLE_STATE_MASK;
-       if (new_state == VM_PURGABLE_VOLATILE &&
-           object->volatile_empty) {
-               new_state = VM_PURGABLE_EMPTY;
+       if (speculative_reads_disabled || fault_info == NULL) {
+               /*
+                * no cluster... just fault the page in
+                */
+               return;
        }
        }
+       orig_start = *start;
+       target_start = orig_start;
+       cluster_size = round_page(fault_info->cluster_size);
+       behavior = fault_info->behavior;
 
 
-       switch (new_state) {
-       case VM_PURGABLE_DENY:
-       case VM_PURGABLE_NONVOLATILE:
-               object->purgable = new_state;
-
-               if (old_state == VM_PURGABLE_VOLATILE) {
-                       unsigned int delta;
-
-                       assert(object->resident_page_count >=
-                              object->wired_page_count);
-                       delta = (object->resident_page_count -
-                                object->wired_page_count);
-
-                       assert(vm_page_purgeable_count >= delta);
-
-                       if (delta != 0) {
-                               OSAddAtomic(-delta,
-                                           (SInt32 *)&vm_page_purgeable_count);
-                       }
-                       if (object->wired_page_count != 0) {
-                               assert(vm_page_purgeable_wired_count >=
-                                      object->wired_page_count);
-                               OSAddAtomic(-object->wired_page_count,
-                                           (SInt32 *)&vm_page_purgeable_wired_count);
-                       }
-
-                       vm_page_lock_queues();
-
-                       assert(object->objq.next != NULL && object->objq.prev != NULL); /* object should be on a queue */
-                       purgeable_q_t queue = vm_purgeable_object_remove(object);
-                       assert(queue);
-
-                       vm_purgeable_token_delete_first(queue);
-                       assert(queue->debug_count_objects>=0);
-
-                       vm_page_unlock_queues();
-               }
-               break;
+       vm_object_lock(object);
 
 
-       case VM_PURGABLE_VOLATILE:
-               if (object->volatile_fault) {
-                       vm_page_t       p;
-                       int             refmod;
+       if (object->pager == MEMORY_OBJECT_NULL)
+               goto out;       /* pager is gone for this object, nothing more to do */
 
 
-                       queue_iterate(&object->memq, p, vm_page_t, listq) {
-                               if (p->busy ||
-                                   VM_PAGE_WIRED(p) ||
-                                   p->fictitious) {
-                                       continue;
-                               }
-                               refmod = pmap_disconnect(p->phys_page);
-                               if ((refmod & VM_MEM_MODIFIED) &&
-                                   !p->dirty) {
-                                       p->dirty = TRUE;
-                               }
-                       }
+       if (!ignore_is_ssd)
+               vnode_pager_get_isSSD(object->pager, &isSSD);
+
+       min_ph_size = round_page(preheat_min_bytes);
+       max_ph_size = round_page(preheat_max_bytes);
+
+       if (isSSD) {
+               min_ph_size /= 2;
+               max_ph_size /= 8;
+
+               if (min_ph_size & PAGE_MASK_64) {
+                       min_ph_size = trunc_page(min_ph_size);
                }
                }
-                                              
-               if (old_state == VM_PURGABLE_EMPTY &&
-                   object->resident_page_count == 0)
-                       break;
 
 
-               purgeable_q_t queue;
-        
-               /* find the correct queue */
-               if ((*state&VM_PURGABLE_ORDERING_MASK) == VM_PURGABLE_ORDERING_OBSOLETE)
-                       queue = &purgeable_queues[PURGEABLE_Q_TYPE_OBSOLETE];
-               else {
-                       if ((*state&VM_PURGABLE_BEHAVIOR_MASK) == VM_PURGABLE_BEHAVIOR_FIFO)
-                               queue = &purgeable_queues[PURGEABLE_Q_TYPE_FIFO];
-                       else
-                               queue = &purgeable_queues[PURGEABLE_Q_TYPE_LIFO];
+               if (max_ph_size & PAGE_MASK_64) {
+                       max_ph_size = trunc_page(max_ph_size);
                }
                }
-        
-               if (old_state == VM_PURGABLE_NONVOLATILE ||
-                   old_state == VM_PURGABLE_EMPTY) {
-                       unsigned int delta;
+       }
 
 
-                       /* try to add token... this can fail */
-                       vm_page_lock_queues();
+       if (min_ph_size < PAGE_SIZE)
+               min_ph_size = PAGE_SIZE;
 
 
-                       kern_return_t result = vm_purgeable_token_add(queue);
-                       if (result != KERN_SUCCESS) {
-                               vm_page_unlock_queues();
-                               return result;
-                       }
-                       vm_page_unlock_queues();
+       if (max_ph_size < PAGE_SIZE)
+               max_ph_size = PAGE_SIZE;
+       else if (max_ph_size > MAX_UPL_TRANSFER_BYTES)
+               max_ph_size = MAX_UPL_TRANSFER_BYTES;
 
 
-                       assert(object->resident_page_count >=
-                              object->wired_page_count);
-                       delta = (object->resident_page_count -
-                                object->wired_page_count);
+       if (max_length > max_ph_size) 
+               max_length = max_ph_size;
 
 
-                       if (delta != 0) {
-                               OSAddAtomic(delta,
-                                           &vm_page_purgeable_count);
-                       }
-                       if (object->wired_page_count != 0) {
-                               OSAddAtomic(object->wired_page_count,
-                                           &vm_page_purgeable_wired_count);
-                       }
+       if (max_length <= PAGE_SIZE)
+               goto out;
 
 
-                       object->purgable = new_state;
+       if (object->internal)
+               object_size = object->vo_size;
+       else
+               vnode_pager_get_object_size(object->pager, &object_size);
 
 
-                       /* object should not be on a queue */
-                       assert(object->objq.next == NULL && object->objq.prev == NULL);
-               }
-               else if (old_state == VM_PURGABLE_VOLATILE) {
-                       /*
-                        * if reassigning priorities / purgeable groups, we don't change the
-                        * token queue. So moving priorities will not make pages stay around longer.
-                        * Reasoning is that the algorithm gives most priority to the most important
-                        * object. If a new token is added, the most important object' priority is boosted.
-                        * This biases the system already for purgeable queues that move a lot.
-                        * It doesn't seem more biasing is neccessary in this case, where no new object is added.
-                        */
-                       assert(object->objq.next != NULL && object->objq.prev != NULL); /* object should be on a queue */
-            
-                       purgeable_q_t old_queue=vm_purgeable_object_remove(object);
-                       assert(old_queue);
-            
-                       if (old_queue != queue) {
-                               kern_return_t result;
+       object_size = round_page_64(object_size);
 
 
-                               /* Changing queue. Have to move token. */
-                               vm_page_lock_queues();
-                               vm_purgeable_token_delete_first(old_queue);
-                               result = vm_purgeable_token_add(queue);
-                               vm_page_unlock_queues();
+       if (orig_start >= object_size) {
+               /*
+                * fault occurred beyond the EOF...
+                * we need to punt w/o changing the
+                * starting offset
+                */
+               goto out;
+       }
+       if (object->pages_used > object->pages_created) {
+               /*
+                * must have wrapped our 32 bit counters
+                * so reset
+                */
+               object->pages_used = object->pages_created = 0;
+       }
+       if ((sequential_run = object->sequential)) {
+                 if (sequential_run < 0) {
+                         sequential_behavior = VM_BEHAVIOR_RSEQNTL;
+                         sequential_run = 0 - sequential_run;
+                 } else {
+                         sequential_behavior = VM_BEHAVIOR_SEQUENTIAL;
+                 }
 
 
-                               assert(result==KERN_SUCCESS);   /* this should never fail since we just freed a token */
-                       }
-               };
-               vm_purgeable_object_add(object, queue, (*state&VM_VOLATILE_GROUP_MASK)>>VM_VOLATILE_GROUP_SHIFT );
+       }
+       switch (behavior) {
 
 
-               assert(queue->debug_count_objects>=0);
-        
-               break;
+       default:
+               behavior = VM_BEHAVIOR_DEFAULT;
 
 
+       case VM_BEHAVIOR_DEFAULT:
+               if (object->internal && fault_info->user_tag == VM_MEMORY_STACK)
+                       goto out;
 
 
-       case VM_PURGABLE_EMPTY:
-               if (object->volatile_fault) {
-                       vm_page_t       p;
-                       int             refmod;
+               if (sequential_run >= (3 * PAGE_SIZE)) {
+                       pre_heat_size = sequential_run + PAGE_SIZE;
 
 
-                       queue_iterate(&object->memq, p, vm_page_t, listq) {
-                               if (p->busy ||
-                                   VM_PAGE_WIRED(p) ||
-                                   p->fictitious) {
-                                       continue;
-                               }
-                               refmod = pmap_disconnect(p->phys_page);
-                               if ((refmod & VM_MEM_MODIFIED) &&
-                                   !p->dirty) {
-                                       p->dirty = TRUE;
-                               }
-                       }
-               }
+                       if (sequential_behavior == VM_BEHAVIOR_SEQUENTIAL)
+                               look_behind = FALSE;
+                       else
+                               look_ahead = FALSE;
 
 
-               if (old_state != new_state) {
-                       assert(old_state == VM_PURGABLE_NONVOLATILE ||
-                              old_state == VM_PURGABLE_VOLATILE);
-                       if (old_state == VM_PURGABLE_VOLATILE) {
-                               purgeable_q_t old_queue;
+                       *io_streaming = 1;
+               } else {
 
 
-                               /* object should be on a queue */
-                               assert(object->objq.next != NULL &&
-                                      object->objq.prev != NULL);
-                               old_queue = vm_purgeable_object_remove(object);
-                               assert(old_queue);
-                               vm_page_lock_queues();
-                               vm_purgeable_token_delete_first(old_queue);
-                               vm_page_unlock_queues();
+                       if (object->pages_created < (20 * (min_ph_size >> PAGE_SHIFT))) {
+                               /*
+                                * prime the pump
+                                */
+                               pre_heat_size = min_ph_size;
+                       } else {
+                               /*
+                                * Linear growth in PH size: The maximum size is max_length...
+                                * this cacluation will result in a size that is neither a 
+                                * power of 2 nor a multiple of PAGE_SIZE... so round
+                                * it up to the nearest PAGE_SIZE boundary
+                                */
+                               pre_heat_size = (max_length * (uint64_t)object->pages_used) / object->pages_created;
+
+                               if (pre_heat_size < min_ph_size)
+                                       pre_heat_size = min_ph_size;
+                               else
+                                       pre_heat_size = round_page(pre_heat_size);
                        }
                        }
-                       (void) vm_object_purge(object);
                }
                break;
 
                }
                break;
 
+       case VM_BEHAVIOR_RANDOM:
+               if ((pre_heat_size = cluster_size) <= PAGE_SIZE)
+                       goto out;
+               break;
+
+       case VM_BEHAVIOR_SEQUENTIAL:
+               if ((pre_heat_size = cluster_size) == 0)
+                       pre_heat_size = sequential_run + PAGE_SIZE;
+               look_behind = FALSE;
+               *io_streaming = 1;
+
+               break;
+
+       case VM_BEHAVIOR_RSEQNTL:
+               if ((pre_heat_size = cluster_size) == 0)
+                       pre_heat_size = sequential_run + PAGE_SIZE;
+               look_ahead = FALSE;
+               *io_streaming = 1;
+
+               break;
+
        }
        }
-       *state = old_state;
+       throttle_limit = (uint32_t) max_length;
+       assert(throttle_limit == max_length);
 
 
-       return KERN_SUCCESS;
-}
+       if (vnode_pager_get_throttle_io_limit(object->pager, &throttle_limit) == KERN_SUCCESS) {
+               if (max_length > throttle_limit)
+                       max_length = throttle_limit;
+       }
+       if (pre_heat_size > max_length)
+               pre_heat_size = max_length;
 
 
-#if    TASK_SWAPPER
-/*
- * vm_object_res_deallocate
- *
- * (recursively) decrement residence counts on vm objects and their shadows.
- * Called from vm_object_deallocate and when swapping out an object.
- *
- * The object is locked, and remains locked throughout the function,
- * even as we iterate down the shadow chain.  Locks on intermediate objects
- * will be dropped, but not the original object.
- *
- * NOTE: this function used to use recursion, rather than iteration.
- */
+       if (behavior == VM_BEHAVIOR_DEFAULT && (pre_heat_size > min_ph_size)) {
 
 
-__private_extern__ void
-vm_object_res_deallocate(
-       vm_object_t     object)
-{
-       vm_object_t orig_object = object;
-       /*
-        * Object is locked so it can be called directly
-        * from vm_object_deallocate.  Original object is never
-        * unlocked.
-        */
-       assert(object->res_count > 0);
-       while  (--object->res_count == 0) {
-               assert(object->ref_count >= object->res_count);
-               vm_object_deactivate_all_pages(object);
-               /* iterate on shadow, if present */
-               if (object->shadow != VM_OBJECT_NULL) {
-                       vm_object_t tmp_object = object->shadow;
-                       vm_object_lock(tmp_object);
-                       if (object != orig_object)
-                               vm_object_unlock(object);
-                       object = tmp_object;
-                       assert(object->res_count > 0);
-               } else
-                       break;
+               unsigned int consider_free = vm_page_free_count + vm_page_cleaned_count;
+               
+               if (consider_free < vm_page_throttle_limit) {
+                       pre_heat_size = trunc_page(pre_heat_size / 16);
+               } else if (consider_free < vm_page_free_target) {
+                       pre_heat_size = trunc_page(pre_heat_size / 4);
+               }
+               
+               if (pre_heat_size < min_ph_size)
+                       pre_heat_size = min_ph_size;
        }
        }
-       if (object != orig_object)
-               vm_object_unlock(object);
-}
+       if (look_ahead == TRUE) {
+               if (look_behind == TRUE) { 
+                       /*
+                        * if we get here its due to a random access... 
+                        * so we want to center the original fault address
+                        * within the cluster we will issue... make sure
+                        * to calculate 'head_size' as a multiple of PAGE_SIZE...
+                        * 'pre_heat_size' is a multiple of PAGE_SIZE but not
+                        * necessarily an even number of pages so we need to truncate
+                        * the result to a PAGE_SIZE boundary
+                        */
+                       head_size = trunc_page(pre_heat_size / 2);
 
 
-/*
- * vm_object_res_reference
- *
- * Internal function to increment residence count on a vm object
- * and its shadows.  It is called only from vm_object_reference, and
- * when swapping in a vm object, via vm_map_swap.
- *
- * The object is locked, and remains locked throughout the function,
- * even as we iterate down the shadow chain.  Locks on intermediate objects
- * will be dropped, but not the original object.
- *
- * NOTE: this function used to use recursion, rather than iteration.
- */
+                       if (target_start > head_size)
+                               target_start -= head_size;
+                       else
+                               target_start = 0;
+
+                       /*
+                        * 'target_start' at this point represents the beginning offset
+                        * of the cluster we are considering... 'orig_start' will be in
+                        * the center of this cluster if we didn't have to clip the start
+                        * due to running into the start of the file
+                        */
+               }
+               if ((target_start + pre_heat_size) > object_size)
+                       pre_heat_size = (vm_size_t)(round_page_64(object_size - target_start));
+               /*
+                * at this point caclulate the number of pages beyond the original fault
+                * address that we want to consider... this is guaranteed not to extend beyond
+                * the current EOF...
+                */
+               assert((vm_size_t)(orig_start - target_start) == (orig_start - target_start));
+               tail_size = pre_heat_size - (vm_size_t)(orig_start - target_start) - PAGE_SIZE;
+       } else {
+               if (pre_heat_size > target_start) {
+                       /*
+                        * since pre_heat_size is always smaller then 2^32,
+                        * if it is larger then target_start (a 64 bit value)
+                        * it is safe to clip target_start to 32 bits
+                        */
+                       pre_heat_size = (vm_size_t) target_start;
+               }
+               tail_size = 0;
+       }
+       assert( !(target_start & PAGE_MASK_64));
+       assert( !(pre_heat_size & PAGE_MASK_64));
+
+       if (pre_heat_size <= PAGE_SIZE)
+               goto out;
 
 
-__private_extern__ void
-vm_object_res_reference(
-       vm_object_t     object)
-{
-       vm_object_t orig_object = object;
-       /* 
-        * Object is locked, so this can be called directly
-        * from vm_object_reference.  This lock is never released.
-        */
-       while  ((++object->res_count == 1)  && 
-               (object->shadow != VM_OBJECT_NULL)) {
-               vm_object_t tmp_object = object->shadow;
+       if (look_behind == TRUE) {
+               /*
+                * take a look at the pages before the original
+                * faulting offset... recalculate this in case
+                * we had to clip 'pre_heat_size' above to keep 
+                * from running past the EOF.
+                */
+               head_size = pre_heat_size - tail_size - PAGE_SIZE;
 
 
-               assert(object->ref_count >= object->res_count);
-               vm_object_lock(tmp_object);
-               if (object != orig_object)
-                       vm_object_unlock(object);
-               object = tmp_object;
+               for (offset = orig_start - PAGE_SIZE_64; head_size; offset -= PAGE_SIZE_64, head_size -= PAGE_SIZE) {
+                       /*
+                        * don't poke below the lowest offset 
+                        */
+                       if (offset < fault_info->lo_offset)
+                               break;
+                       /*
+                        * for external objects or internal objects w/o a pager,
+                        * VM_COMPRESSOR_PAGER_STATE_GET will return VM_EXTERNAL_STATE_UNKNOWN
+                        */
+                       if (VM_COMPRESSOR_PAGER_STATE_GET(object, offset) == VM_EXTERNAL_STATE_ABSENT) {
+                               break;
+                       }
+                       if (vm_page_lookup(object, offset) != VM_PAGE_NULL) {
+                               /*
+                                * don't bridge resident pages
+                                */
+                               break;
+                       }
+                       *start = offset;
+                       *length += PAGE_SIZE;
+               }
        }
        }
-       if (object != orig_object)
-               vm_object_unlock(object);
-       assert(orig_object->ref_count >= orig_object->res_count);
-}
-#endif /* TASK_SWAPPER */
+       if (look_ahead == TRUE) {
+               for (offset = orig_start + PAGE_SIZE_64; tail_size; offset += PAGE_SIZE_64, tail_size -= PAGE_SIZE) {
+                       /*
+                        * don't poke above the highest offset 
+                        */
+                       if (offset >= fault_info->hi_offset)
+                               break;
+                       assert(offset < object_size);
 
 
-/*
- *     vm_object_reference:
- *
- *     Gets another reference to the given object.
- */
-#ifdef vm_object_reference
-#undef vm_object_reference
-#endif
-__private_extern__ void
-vm_object_reference(
-       register vm_object_t    object)
-{
-       if (object == VM_OBJECT_NULL)
-               return;
+                       /*
+                        * for external objects or internal objects w/o a pager,
+                        * VM_COMPRESSOR_PAGER_STATE_GET will return VM_EXTERNAL_STATE_UNKNOWN
+                        */
+                       if (VM_COMPRESSOR_PAGER_STATE_GET(object, offset) == VM_EXTERNAL_STATE_ABSENT) {
+                               break;
+                       }
+                       if (vm_page_lookup(object, offset) != VM_PAGE_NULL) {
+                               /*
+                                * don't bridge resident pages
+                                */
+                               break;
+                       }
+                       *length += PAGE_SIZE;
+               }
+       }
+out:
+       if (*length > max_length)
+               *length = max_length;
 
 
-       vm_object_lock(object);
-       assert(object->ref_count > 0);
-       vm_object_reference_locked(object);
        vm_object_unlock(object);
        vm_object_unlock(object);
+       
+       DTRACE_VM1(clustersize, vm_size_t, *length);
 }
 
 }
 
-#ifdef MACH_BSD
-/*
- * Scale the vm_object_cache
- * This is required to make sure that the vm_object_cache is big
- * enough to effectively cache the mapped file.
- * This is really important with UBC as all the regular file vnodes
- * have memory object associated with them. Havving this cache too
- * small results in rapid reclaim of vnodes and hurts performance a LOT!
- *
- * This is also needed as number of vnodes can be dynamically scaled.
- */
-kern_return_t
-adjust_vm_object_cache(
-       __unused vm_size_t oval,
-       __unused vm_size_t nval)
-{
-#if VM_OBJECT_CACHE
-       vm_object_cached_max = nval;
-       vm_object_cache_trim(FALSE);
-#endif
-       return (KERN_SUCCESS);
-}
-#endif /* MACH_BSD */
-
 
 /*
 
 /*
- * vm_object_transpose
- *
- * This routine takes two VM objects of the same size and exchanges
- * their backing store.
- * The objects should be "quiesced" via a UPL operation with UPL_SET_IO_WIRE
- * and UPL_BLOCK_ACCESS if they are referenced anywhere.
- *
- * The VM objects must not be locked by caller.
+ * Allow manipulation of individual page state.  This is actually part of
+ * the UPL regimen but takes place on the VM object rather than on a UPL
  */
  */
-unsigned int vm_object_transpose_count = 0;
+
 kern_return_t
 kern_return_t
-vm_object_transpose(
-       vm_object_t             object1,
-       vm_object_t             object2,
-       vm_object_size_t        transpose_size)
+vm_object_page_op(
+       vm_object_t             object,
+       vm_object_offset_t      offset,
+       int                     ops,
+       ppnum_t                 *phys_entry,
+       int                     *flags)
 {
 {
-       vm_object_t             tmp_object;
-       kern_return_t           retval;
-       boolean_t               object1_locked, object2_locked;
-       vm_page_t               page;
-       vm_object_offset_t      page_offset;
-       lck_mtx_t               *hash_lck;
-       vm_object_hash_entry_t  hash_entry;
-
-       tmp_object = VM_OBJECT_NULL;
-       object1_locked = FALSE; object2_locked = FALSE;
+       vm_page_t               dst_page;
 
 
-       if (object1 == object2 ||
-           object1 == VM_OBJECT_NULL ||
-           object2 == VM_OBJECT_NULL) {
-               /*
-                * If the 2 VM objects are the same, there's
-                * no point in exchanging their backing store.
-                */
-               retval = KERN_INVALID_VALUE;
-               goto done;
-       }
+       vm_object_lock(object);
 
 
-       /*
-        * Since we need to lock both objects at the same time,
-        * make sure we always lock them in the same order to
-        * avoid deadlocks.
-        */
-       if (object1 >  object2) {
-               tmp_object = object1;
-               object1 = object2;
-               object2 = tmp_object;
+       if(ops & UPL_POP_PHYSICAL) {
+               if(object->phys_contiguous) {
+                       if (phys_entry) {
+                               *phys_entry = (ppnum_t)
+                                       (object->vo_shadow_offset >> PAGE_SHIFT);
+                       }
+                       vm_object_unlock(object);
+                       return KERN_SUCCESS;
+               } else {
+                       vm_object_unlock(object);
+                       return KERN_INVALID_OBJECT;
+               }
        }
        }
-
-       /*
-        * Allocate a temporary VM object to hold object1's contents
-        * while we copy object2 to object1.
-        */
-       tmp_object = vm_object_allocate(transpose_size);
-       vm_object_lock(tmp_object);
-       tmp_object->can_persist = FALSE;
-
-
-       /*
-        * Grab control of the 1st VM object.
-        */
-       vm_object_lock(object1);
-       object1_locked = TRUE;
-       if (!object1->alive || object1->terminating ||
-           object1->copy || object1->shadow || object1->shadowed ||
-           object1->purgable != VM_PURGABLE_DENY) {
-               /*
-                * We don't deal with copy or shadow objects (yet).
-                */
-               retval = KERN_INVALID_VALUE;
-               goto done;
+       if(object->phys_contiguous) {
+               vm_object_unlock(object);
+               return KERN_INVALID_OBJECT;
        }
        }
-       /*
-        * We're about to mess with the object's backing store and 
-        * taking a "paging_in_progress" reference wouldn't be enough
-        * to prevent any paging activity on this object, so the caller should
-        * have "quiesced" the objects beforehand, via a UPL operation with
-        * UPL_SET_IO_WIRE (to make sure all the pages are there and wired)
-        * and UPL_BLOCK_ACCESS (to mark the pages "busy").
-        * 
-        * Wait for any paging operation to complete (but only paging, not 
-        * other kind of activities not linked to the pager).  After we're
-        * statisfied that there's no more paging in progress, we keep the
-        * object locked, to guarantee that no one tries to access its pager.
-        */
-       vm_object_paging_only_wait(object1, THREAD_UNINT);
 
 
-       /*
-        * Same as above for the 2nd object...
-        */
-       vm_object_lock(object2);
-       object2_locked = TRUE;
-       if (! object2->alive || object2->terminating ||
-           object2->copy || object2->shadow || object2->shadowed ||
-           object2->purgable != VM_PURGABLE_DENY) {
-               retval = KERN_INVALID_VALUE;
-               goto done;
-       }
-       vm_object_paging_only_wait(object2, THREAD_UNINT);
+       while(TRUE) {
+               if((dst_page = vm_page_lookup(object,offset)) == VM_PAGE_NULL) {
+                       vm_object_unlock(object);
+                       return KERN_FAILURE;
+               }
 
 
+               /* Sync up on getting the busy bit */
+               if((dst_page->busy || dst_page->cleaning) && 
+                          (((ops & UPL_POP_SET) && 
+                          (ops & UPL_POP_BUSY)) || (ops & UPL_POP_DUMP))) {
+                       /* someone else is playing with the page, we will */
+                       /* have to wait */
+                       PAGE_SLEEP(object, dst_page, THREAD_UNINT);
+                       continue;
+               }
 
 
-       if (object1->size != object2->size ||
-           object1->size != transpose_size) {
-               /*
-                * If the 2 objects don't have the same size, we can't
-                * exchange their backing stores or one would overflow.
-                * If their size doesn't match the caller's
-                * "transpose_size", we can't do it either because the
-                * transpose operation will affect the entire span of 
-                * the objects.
-                */
-               retval = KERN_INVALID_VALUE;
-               goto done;
-       }
+               if (ops & UPL_POP_DUMP) {
+                       if (dst_page->pmapped == TRUE)
+                               pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(dst_page));
 
 
+                       VM_PAGE_FREE(dst_page);
+                       break;
+               }
 
 
-       /*
-        * Transpose the lists of resident pages.
-        * This also updates the resident_page_count and the memq_hint.
-        */
-       if (object1->phys_contiguous || queue_empty(&object1->memq)) {
-               /*
-                * No pages in object1, just transfer pages
-                * from object2 to object1.  No need to go through
-                * an intermediate object.
-                */
-               while (!queue_empty(&object2->memq)) {
-                       page = (vm_page_t) queue_first(&object2->memq);
-                       vm_page_rename(page, object1, page->offset, FALSE);
+               if (flags) {
+                       *flags = 0;
+
+                       /* Get the condition of flags before requested ops */
+                       /* are undertaken */
+
+                       if(dst_page->dirty) *flags |= UPL_POP_DIRTY;
+                       if(dst_page->free_when_done) *flags |= UPL_POP_PAGEOUT;
+                       if(dst_page->precious) *flags |= UPL_POP_PRECIOUS;
+                       if(dst_page->absent) *flags |= UPL_POP_ABSENT;
+                       if(dst_page->busy) *flags |= UPL_POP_BUSY;
                }
                }
-               assert(queue_empty(&object2->memq));
-       } else if (object2->phys_contiguous || queue_empty(&object2->memq)) {
-               /*
-                * No pages in object2, just transfer pages
-                * from object1 to object2.  No need to go through
-                * an intermediate object.
-                */
-               while (!queue_empty(&object1->memq)) {
-                       page = (vm_page_t) queue_first(&object1->memq);
-                       vm_page_rename(page, object2, page->offset, FALSE);
+
+               /* The caller should have made a call either contingent with */
+               /* or prior to this call to set UPL_POP_BUSY */
+               if(ops & UPL_POP_SET) {
+                       /* The protection granted with this assert will */
+                       /* not be complete.  If the caller violates the */
+                       /* convention and attempts to change page state */
+                       /* without first setting busy we may not see it */
+                       /* because the page may already be busy.  However */
+                       /* if such violations occur we will assert sooner */
+                       /* or later. */
+                       assert(dst_page->busy || (ops & UPL_POP_BUSY));
+                       if (ops & UPL_POP_DIRTY) {
+                               SET_PAGE_DIRTY(dst_page, FALSE);
+                       }
+                       if (ops & UPL_POP_PAGEOUT) dst_page->free_when_done = TRUE;
+                       if (ops & UPL_POP_PRECIOUS) dst_page->precious = TRUE;
+                       if (ops & UPL_POP_ABSENT) dst_page->absent = TRUE;
+                       if (ops & UPL_POP_BUSY) dst_page->busy = TRUE;
                }
                }
-               assert(queue_empty(&object1->memq));
-       } else {
-               /* transfer object1's pages to tmp_object */
-               while (!queue_empty(&object1->memq)) {
-                       page = (vm_page_t) queue_first(&object1->memq);
-                       page_offset = page->offset;
-                       vm_page_remove(page, TRUE);
-                       page->offset = page_offset;
-                       queue_enter(&tmp_object->memq, page, vm_page_t, listq);
+
+               if(ops & UPL_POP_CLR) {
+                       assert(dst_page->busy);
+                       if (ops & UPL_POP_DIRTY) dst_page->dirty = FALSE;
+                       if (ops & UPL_POP_PAGEOUT) dst_page->free_when_done = FALSE;
+                       if (ops & UPL_POP_PRECIOUS) dst_page->precious = FALSE;
+                       if (ops & UPL_POP_ABSENT) dst_page->absent = FALSE;
+                       if (ops & UPL_POP_BUSY) {
+                               dst_page->busy = FALSE;
+                               PAGE_WAKEUP(dst_page);
+                       }
                }
                }
-               assert(queue_empty(&object1->memq));
-               /* transfer object2's pages to object1 */
-               while (!queue_empty(&object2->memq)) {
-                       page = (vm_page_t) queue_first(&object2->memq);
-                       vm_page_rename(page, object1, page->offset, FALSE);
+
+               if (dst_page->encrypted) {
+                       /*
+                        * ENCRYPTED SWAP:
+                        * We need to decrypt this encrypted page before the
+                        * caller can access its contents.
+                        * But if the caller really wants to access the page's
+                        * contents, they have to keep the page "busy".
+                        * Otherwise, the page could get recycled or re-encrypted
+                        * at any time.
+                        */
+                       if ((ops & UPL_POP_SET) && (ops & UPL_POP_BUSY) &&
+                           dst_page->busy) {
+                               /*
+                                * The page is stable enough to be accessed by
+                                * the caller, so make sure its contents are
+                                * not encrypted.
+                                */
+                               vm_page_decrypt(dst_page, 0);
+                       } else {
+                               /*
+                                * The page is not busy, so don't bother
+                                * decrypting it, since anything could
+                                * happen to it between now and when the
+                                * caller wants to access it.
+                                * We should not give the caller access
+                                * to this page.
+                                */
+                               assert(!phys_entry);
+                       }
                }
                }
-               assert(queue_empty(&object2->memq));
-               /* transfer tmp_object's pages to object1 */
-               while (!queue_empty(&tmp_object->memq)) {
-                       page = (vm_page_t) queue_first(&tmp_object->memq);
-                       queue_remove(&tmp_object->memq, page,
-                                    vm_page_t, listq);
-                       vm_page_insert(page, object2, page->offset);
+
+               if (phys_entry) {
+                       /*
+                        * The physical page number will remain valid
+                        * only if the page is kept busy.
+                        * ENCRYPTED SWAP: make sure we don't let the
+                        * caller access an encrypted page.
+                        */
+                       assert(dst_page->busy);
+                       assert(!dst_page->encrypted);
+                       *phys_entry = VM_PAGE_GET_PHYS_PAGE(dst_page);
                }
                }
-               assert(queue_empty(&tmp_object->memq));
+
+               break;
        }
 
        }
 
-#define __TRANSPOSE_FIELD(field)                               \
-MACRO_BEGIN                                                    \
-       tmp_object->field = object1->field;                     \
-       object1->field = object2->field;                        \
-       object2->field = tmp_object->field;                     \
-MACRO_END
+       vm_object_unlock(object);
+       return KERN_SUCCESS;
+                               
+}
 
 
-       /* "Lock" refers to the object not its contents */
-       /* "size" should be identical */
-       assert(object1->size == object2->size);
-       /* "memq_hint" was updated above when transposing pages */
-       /* "ref_count" refers to the object not its contents */
-#if TASK_SWAPPER
-       /* "res_count" refers to the object not its contents */
-#endif
-       /* "resident_page_count" was updated above when transposing pages */
-       /* "wired_page_count" was updated above when transposing pages */
-       /* "reusable_page_count" was updated above when transposing pages */
-       /* there should be no "copy" */
-       assert(!object1->copy);
-       assert(!object2->copy);
-       /* there should be no "shadow" */
-       assert(!object1->shadow);
-       assert(!object2->shadow);
-       __TRANSPOSE_FIELD(shadow_offset); /* used by phys_contiguous objects */
-       __TRANSPOSE_FIELD(pager);
-       __TRANSPOSE_FIELD(paging_offset);
-       __TRANSPOSE_FIELD(pager_control);
-       /* update the memory_objects' pointers back to the VM objects */
-       if (object1->pager_control != MEMORY_OBJECT_CONTROL_NULL) {
-               memory_object_control_collapse(object1->pager_control,
-                                              object1);
+/*
+ * vm_object_range_op offers performance enhancement over 
+ * vm_object_page_op for page_op functions which do not require page 
+ * level state to be returned from the call.  Page_op was created to provide 
+ * a low-cost alternative to page manipulation via UPLs when only a single 
+ * page was involved.  The range_op call establishes the ability in the _op 
+ * family of functions to work on multiple pages where the lack of page level
+ * state handling allows the caller to avoid the overhead of the upl structures.
+ */
+
+kern_return_t
+vm_object_range_op(
+       vm_object_t             object,
+       vm_object_offset_t      offset_beg,
+       vm_object_offset_t      offset_end,
+       int                     ops,
+       uint32_t                *range)
+{
+        vm_object_offset_t     offset;
+       vm_page_t               dst_page;
+
+       if (offset_end - offset_beg > (uint32_t) -1) {
+               /* range is too big and would overflow "*range" */
+               return KERN_INVALID_ARGUMENT;
+       } 
+       if (object->resident_page_count == 0) {
+               if (range) {
+                       if (ops & UPL_ROP_PRESENT) {
+                               *range = 0;
+                       } else {
+                               *range = (uint32_t) (offset_end - offset_beg);
+                               assert(*range == (offset_end - offset_beg));
+                       }
+               }
+               return KERN_SUCCESS;
        }
        }
-       if (object2->pager_control != MEMORY_OBJECT_CONTROL_NULL) {
-               memory_object_control_collapse(object2->pager_control,
-                                              object2);
+       vm_object_lock(object);
+
+       if (object->phys_contiguous) {
+               vm_object_unlock(object);
+               return KERN_INVALID_OBJECT;
        }
        }
-       __TRANSPOSE_FIELD(copy_strategy);
-       /* "paging_in_progress" refers to the object not its contents */
-       assert(!object1->paging_in_progress);
-       assert(!object2->paging_in_progress);
-       assert(object1->activity_in_progress);
-       assert(object2->activity_in_progress);
-       /* "all_wanted" refers to the object not its contents */
-       __TRANSPOSE_FIELD(pager_created);
-       __TRANSPOSE_FIELD(pager_initialized);
-       __TRANSPOSE_FIELD(pager_ready);
-       __TRANSPOSE_FIELD(pager_trusted);
-       __TRANSPOSE_FIELD(can_persist);
-       __TRANSPOSE_FIELD(internal);
-       __TRANSPOSE_FIELD(temporary);
-       __TRANSPOSE_FIELD(private);
-       __TRANSPOSE_FIELD(pageout);
-       /* "alive" should be set */
-       assert(object1->alive);
-       assert(object2->alive);
-       /* "purgeable" should be non-purgeable */
-       assert(object1->purgable == VM_PURGABLE_DENY);
-       assert(object2->purgable == VM_PURGABLE_DENY);
-       /* "shadowed" refers to the the object not its contents */
-       __TRANSPOSE_FIELD(silent_overwrite);
-       __TRANSPOSE_FIELD(advisory_pageout);
-       __TRANSPOSE_FIELD(true_share);
-       /* "terminating" should not be set */
-       assert(!object1->terminating);
-       assert(!object2->terminating);
-       __TRANSPOSE_FIELD(named);
-       /* "shadow_severed" refers to the object not its contents */
-       __TRANSPOSE_FIELD(phys_contiguous);
-       __TRANSPOSE_FIELD(nophyscache);
-       /* "cached_list.next" points to transposed object */
-       object1->cached_list.next = (queue_entry_t) object2;
-       object2->cached_list.next = (queue_entry_t) object1;
-       /* "cached_list.prev" should be NULL */
-       assert(object1->cached_list.prev == NULL);
-       assert(object2->cached_list.prev == NULL);
-       /* "msr_q" is linked to the object not its contents */
-       assert(queue_empty(&object1->msr_q));
-       assert(queue_empty(&object2->msr_q));
-       __TRANSPOSE_FIELD(last_alloc);
-       __TRANSPOSE_FIELD(sequential);
-       __TRANSPOSE_FIELD(pages_created);
-       __TRANSPOSE_FIELD(pages_used);
-#if MACH_PAGEMAP
-       __TRANSPOSE_FIELD(existence_map);
-#endif
-       __TRANSPOSE_FIELD(cow_hint);
-#if MACH_ASSERT
-       __TRANSPOSE_FIELD(paging_object);
-#endif
-       __TRANSPOSE_FIELD(wimg_bits);
-       __TRANSPOSE_FIELD(code_signed);
-       if (object1->hashed) {
-               hash_lck = vm_object_hash_lock_spin(object2->pager);
-               hash_entry = vm_object_hash_lookup(object2->pager, FALSE);
-               assert(hash_entry != VM_OBJECT_HASH_ENTRY_NULL);
-               hash_entry->object = object2;
-               vm_object_hash_unlock(hash_lck);
+       
+       offset = offset_beg & ~PAGE_MASK_64;
+
+       while (offset < offset_end) {
+               dst_page = vm_page_lookup(object, offset);
+               if (dst_page != VM_PAGE_NULL) {
+                       if (ops & UPL_ROP_DUMP) {
+                               if (dst_page->busy || dst_page->cleaning) {
+                                       /*
+                                        * someone else is playing with the 
+                                        * page, we will have to wait
+                                        */
+                                       PAGE_SLEEP(object, dst_page, THREAD_UNINT);
+                                       /*
+                                        * need to relook the page up since it's
+                                        * state may have changed while we slept
+                                        * it might even belong to a different object
+                                        * at this point
+                                        */
+                                       continue;
+                               }
+                               if (dst_page->laundry)
+                                       vm_pageout_steal_laundry(dst_page, FALSE);
+
+                               if (dst_page->pmapped == TRUE)
+                                       pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(dst_page));
+
+                               VM_PAGE_FREE(dst_page);
+
+                       } else if ((ops & UPL_ROP_ABSENT)
+                                          && (!dst_page->absent || dst_page->busy)) {
+                               break;
+                       }
+               } else if (ops & UPL_ROP_PRESENT)
+                       break;
+
+               offset += PAGE_SIZE;
        }
        }
-       if (object2->hashed) {
-               hash_lck = vm_object_hash_lock_spin(object1->pager);
-               hash_entry = vm_object_hash_lookup(object1->pager, FALSE);
-               assert(hash_entry != VM_OBJECT_HASH_ENTRY_NULL);
-               hash_entry->object = object1;
-               vm_object_hash_unlock(hash_lck);
+       vm_object_unlock(object);
+
+       if (range) {
+               if (offset > offset_end)
+                       offset = offset_end;
+               if(offset > offset_beg) {
+                       *range = (uint32_t) (offset - offset_beg);
+                       assert(*range == (offset - offset_beg));
+               } else {
+                       *range = 0;
+               }
        }
        }
-       __TRANSPOSE_FIELD(hashed);
-       object1->transposed = TRUE;
-       object2->transposed = TRUE;
-       __TRANSPOSE_FIELD(mapping_in_progress);
-       __TRANSPOSE_FIELD(volatile_empty);
-       __TRANSPOSE_FIELD(volatile_fault);
-       __TRANSPOSE_FIELD(all_reusable);
-       assert(object1->blocked_access);
-       assert(object2->blocked_access);
-       assert(object1->__object2_unused_bits == 0);
-       assert(object2->__object2_unused_bits == 0);
-#if UPL_DEBUG
-       /* "uplq" refers to the object not its contents (see upl_transpose()) */
-#endif
-       assert(object1->objq.next == NULL);
-       assert(object1->objq.prev == NULL);
-       assert(object2->objq.next == NULL);
-       assert(object2->objq.prev == NULL);
+       return KERN_SUCCESS;
+}
 
 
-#undef __TRANSPOSE_FIELD
+/*
+ * Used to point a pager directly to a range of memory (when the pager may be associated
+ *   with a non-device vnode).  Takes a virtual address, an offset, and a size.  We currently
+ *   expect that the virtual address will denote the start of a range that is physically contiguous.
+ */
+kern_return_t pager_map_to_phys_contiguous(
+       memory_object_control_t object,
+       memory_object_offset_t  offset,
+       addr64_t                base_vaddr,
+       vm_size_t               size)
+{
+       ppnum_t page_num;
+       boolean_t clobbered_private;
+       kern_return_t retval;
+       vm_object_t pager_object;
 
 
-       retval = KERN_SUCCESS;
+       page_num = pmap_find_phys(kernel_pmap, base_vaddr);
 
 
-done:
-       /*
-        * Cleanup.
-        */
-       if (tmp_object != VM_OBJECT_NULL) {
-               vm_object_unlock(tmp_object);
-               /*
-                * Re-initialize the temporary object to avoid
-                * deallocating a real pager.
-                */
-               _vm_object_allocate(transpose_size, tmp_object);
-               vm_object_deallocate(tmp_object);
-               tmp_object = VM_OBJECT_NULL;
+       if (!page_num) {
+               retval = KERN_FAILURE;
+               goto out;
        }
 
        }
 
-       if (object1_locked) {
-               vm_object_unlock(object1);
-               object1_locked = FALSE;
+       pager_object = memory_object_control_to_vm_object(object);
+
+       if (!pager_object) {
+               retval = KERN_FAILURE;
+               goto out;
        }
        }
-       if (object2_locked) {
-               vm_object_unlock(object2);
-               object2_locked = FALSE;
+
+       clobbered_private = pager_object->private;
+       if (pager_object->private != TRUE) {
+               vm_object_lock(pager_object);
+               pager_object->private = TRUE;
+               vm_object_unlock(pager_object);
        }
        }
+       retval = vm_object_populate_with_private(pager_object, offset, page_num, size);
 
 
-       vm_object_transpose_count++;
+       if (retval != KERN_SUCCESS) {
+               if (pager_object->private != clobbered_private) {
+                       vm_object_lock(pager_object);
+                       pager_object->private = clobbered_private;
+                       vm_object_unlock(pager_object);
+               }
+       }
 
 
+out:
        return retval;
 }
 
        return retval;
 }
 
+uint32_t scan_object_collision = 0;
 
 
-/*
- *      vm_object_cluster_size
- *
- *      Determine how big a cluster we should issue an I/O for...
- *
- *     Inputs:   *start == offset of page needed
- *               *length == maximum cluster pager can handle
- *     Outputs:  *start == beginning offset of cluster
- *               *length == length of cluster to try
- *
- *     The original *start will be encompassed by the cluster
- *
- */
-extern int speculative_reads_disabled;
-#if CONFIG_EMBEDDED
-unsigned int preheat_pages_max = MAX_UPL_TRANSFER;
-unsigned int preheat_pages_min = 8;
-unsigned int preheat_pages_mult = 4;
-#else
-unsigned int preheat_pages_max = MAX_UPL_TRANSFER;
-unsigned int preheat_pages_min = 8;
-unsigned int preheat_pages_mult = 4;
+void
+vm_object_lock(vm_object_t object)
+{
+        if (object == vm_pageout_scan_wants_object) {
+               scan_object_collision++;
+               mutex_pause(2);
+       }
+        lck_rw_lock_exclusive(&object->Lock);
+#if DEVELOPMENT || DEBUG
+       object->Lock_owner = current_thread();
 #endif
 #endif
+}
 
 
-uint32_t pre_heat_scaling[MAX_UPL_TRANSFER + 1];
-uint32_t pre_heat_cluster[MAX_UPL_TRANSFER + 1];
-
-
-__private_extern__ void
-vm_object_cluster_size(vm_object_t object, vm_object_offset_t *start,
-                      vm_size_t *length, vm_object_fault_info_t fault_info, uint32_t *io_streaming)
+boolean_t
+vm_object_lock_avoid(vm_object_t object)
 {
 {
-       vm_size_t               pre_heat_size;
-       vm_size_t               tail_size;
-       vm_size_t               head_size;
-       vm_size_t               max_length;
-       vm_size_t               cluster_size;
-       vm_object_offset_t      object_size;
-       vm_object_offset_t      orig_start;
-       vm_object_offset_t      target_start;
-       vm_object_offset_t      offset;
-       vm_behavior_t           behavior;
-       boolean_t               look_behind = TRUE;
-       boolean_t               look_ahead  = TRUE;
-       uint32_t                throttle_limit;
-       int                     sequential_run;
-       int                     sequential_behavior = VM_BEHAVIOR_SEQUENTIAL;
-       unsigned int            max_ph_size;
-       unsigned int            min_ph_size;
-       unsigned int            ph_mult;
+        if (object == vm_pageout_scan_wants_object) {
+               scan_object_collision++;
+               return TRUE;
+       }
+       return FALSE;
+}
 
 
-       assert( !(*length & PAGE_MASK));
-       assert( !(*start & PAGE_MASK_64));
+boolean_t
+_vm_object_lock_try(vm_object_t object)
+{
+       boolean_t       retval;
 
 
-       if ( (ph_mult = preheat_pages_mult) < 1 ) 
-               ph_mult = 1;
-       if ( (min_ph_size = preheat_pages_min) < 1 ) 
-               min_ph_size = 1;
-       if ( (max_ph_size = preheat_pages_max) > MAX_UPL_TRANSFER ) 
-               max_ph_size = MAX_UPL_TRANSFER;
-       
-       if ( (max_length = *length) > (max_ph_size * PAGE_SIZE) ) 
-               max_length = (max_ph_size * PAGE_SIZE);
+       retval = lck_rw_try_lock_exclusive(&object->Lock);
+#if DEVELOPMENT || DEBUG
+       if (retval == TRUE)
+               object->Lock_owner = current_thread();
+#endif
+       return (retval);
+}
 
 
+boolean_t
+vm_object_lock_try(vm_object_t object)
+{
        /*
        /*
-        * we'll always return a cluster size of at least
-        * 1 page, since the original fault must always
-        * be processed
+        * Called from hibernate path so check before blocking.
         */
         */
-       *length = PAGE_SIZE;
-       *io_streaming = 0;
+       if (vm_object_lock_avoid(object) && ml_get_interrupts_enabled() && get_preemption_level()==0) {
+               mutex_pause(2);
+       }
+       return _vm_object_lock_try(object);
+}
 
 
-       if (speculative_reads_disabled || fault_info == NULL || max_length == 0) {
-               /*
-                * no cluster... just fault the page in
-                */
-               return;
+void
+vm_object_lock_shared(vm_object_t object)
+{
+        if (vm_object_lock_avoid(object)) {
+               mutex_pause(2);
        }
        }
-       orig_start = *start;
-       target_start = orig_start;
-       cluster_size = round_page(fault_info->cluster_size);
-       behavior = fault_info->behavior;
+       lck_rw_lock_shared(&object->Lock);
+}
 
 
-       vm_object_lock(object);
+boolean_t
+vm_object_lock_try_shared(vm_object_t object)
+{
+        if (vm_object_lock_avoid(object)) {
+               mutex_pause(2);
+       }
+       return (lck_rw_try_lock_shared(&object->Lock));
+}
 
 
-       if (object->internal)
-               object_size = object->size;
-       else if (object->pager != MEMORY_OBJECT_NULL)
-               vnode_pager_get_object_size(object->pager, &object_size);
-       else
-               goto out;       /* pager is gone for this object, nothing more to do */
+boolean_t
+vm_object_lock_upgrade(vm_object_t object)
+{      boolean_t       retval;
 
 
-       object_size = round_page_64(object_size);
+       retval = lck_rw_lock_shared_to_exclusive(&object->Lock);
+#if DEVELOPMENT || DEBUG
+       if (retval == TRUE)
+               object->Lock_owner = current_thread();
+#endif
+       return (retval);
+}
 
 
-       if (orig_start >= object_size) {
-               /*
-                * fault occurred beyond the EOF...
-                * we need to punt w/o changing the
-                * starting offset
-                */
-               goto out;
+void
+vm_object_unlock(vm_object_t object)
+{
+#if DEVELOPMENT || DEBUG
+       if (object->Lock_owner) {
+               if (object->Lock_owner != current_thread())
+                       panic("vm_object_unlock: not owner - %p\n", object);
+               object->Lock_owner = 0;
        }
        }
-       if (object->pages_used > object->pages_created) {
-               /*
-                * must have wrapped our 32 bit counters
-                * so reset
-                */
-               object->pages_used = object->pages_created = 0;
+#endif
+       lck_rw_done(&object->Lock);
+}
+
+
+unsigned int vm_object_change_wimg_mode_count = 0;
+
+/*
+ * The object must be locked
+ */
+void
+vm_object_change_wimg_mode(vm_object_t object, unsigned int wimg_mode)
+{
+       vm_page_t p;
+
+       vm_object_lock_assert_exclusive(object);
+
+       vm_object_paging_wait(object, THREAD_UNINT);
+
+       vm_page_queue_iterate(&object->memq, p, vm_page_t, listq) {
+
+               if (!p->fictitious)
+                       pmap_set_cache_attributes(VM_PAGE_GET_PHYS_PAGE(p), wimg_mode);
        }
        }
-       if ((sequential_run = object->sequential)) {
-                 if (sequential_run < 0) {
-                         sequential_behavior = VM_BEHAVIOR_RSEQNTL;
-                         sequential_run = 0 - sequential_run;
-                 } else {
-                         sequential_behavior = VM_BEHAVIOR_SEQUENTIAL;
-                 }
+       if (wimg_mode == VM_WIMG_USE_DEFAULT)
+               object->set_cache_attr = FALSE;
+       else
+               object->set_cache_attr = TRUE;
+
+       object->wimg_bits = wimg_mode;
+
+       vm_object_change_wimg_mode_count++;
+}
+
+#if CONFIG_FREEZE
+
+/*
+ * This routine does the "relocation" of previously
+ * compressed pages belonging to this object that are
+ * residing in a number of compressed segments into
+ * a set of compressed segments dedicated to hold
+ * compressed pages belonging to this object.
+ */
+
+extern void *freezer_chead;
+extern char *freezer_compressor_scratch_buf;
+extern int c_freezer_compression_count;
+extern AbsoluteTime c_freezer_last_yield_ts;
+
+#define        MAX_FREE_BATCH  32
+#define FREEZER_DUTY_CYCLE_ON_MS       5
+#define FREEZER_DUTY_CYCLE_OFF_MS      5
+
+static int c_freezer_should_yield(void);
+
+
+static int
+c_freezer_should_yield()
+{
+       AbsoluteTime    cur_time;
+       uint64_t        nsecs;
+
+       assert(c_freezer_last_yield_ts);
+       clock_get_uptime(&cur_time);
+
+       SUB_ABSOLUTETIME(&cur_time, &c_freezer_last_yield_ts);
+       absolutetime_to_nanoseconds(cur_time, &nsecs);
+
+       if (nsecs > 1000 * 1000 * FREEZER_DUTY_CYCLE_ON_MS)
+               return (1);
+       return (0);
+}
+
+
+void
+vm_object_compressed_freezer_done()
+{
+       vm_compressor_finished_filling(&freezer_chead);
+}
+
+
+void
+vm_object_compressed_freezer_pageout(
+       vm_object_t object)
+{
+       vm_page_t                       p;
+       vm_page_t                       local_freeq = NULL;
+       int                             local_freed = 0;
+       kern_return_t                   retval = KERN_SUCCESS;
+       int                             obj_resident_page_count_snapshot = 0;
+
+       assert(object != VM_OBJECT_NULL);
+       assert(object->internal);
+
+       vm_object_lock(object);
+
+       if (!object->pager_initialized || object->pager == MEMORY_OBJECT_NULL)  {
+                       
+               if (!object->pager_initialized) {
 
 
-       }
-       switch(behavior) {
+                       vm_object_collapse(object, (vm_object_offset_t) 0, TRUE);
 
 
-       default:
-               behavior = VM_BEHAVIOR_DEFAULT;
+                       if (!object->pager_initialized)
+                               vm_object_compressor_pager_create(object);
+               }
 
 
-       case VM_BEHAVIOR_DEFAULT:
-               if (object->internal && fault_info->user_tag == VM_MEMORY_STACK)
-                       goto out;
+               if (!object->pager_initialized || object->pager == MEMORY_OBJECT_NULL)  {
+                       vm_object_unlock(object);
+                       return;
+               }
+       }
+                       
+       if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
+               vm_object_offset_t      curr_offset = 0;
 
 
-               if (sequential_run >= (3 * PAGE_SIZE)) {
-                       pre_heat_size = sequential_run + PAGE_SIZE;
+               /*
+                * Go through the object and make sure that any
+                * previously compressed pages are relocated into
+                * a compressed segment associated with our "freezer_chead".
+                */
+               while (curr_offset < object->vo_size) {
 
 
-                       if (sequential_behavior == VM_BEHAVIOR_SEQUENTIAL)
-                               look_behind = FALSE;
-                       else
-                               look_ahead = FALSE;
+                       curr_offset = vm_compressor_pager_next_compressed(object->pager, curr_offset);
+       
+                       if (curr_offset == (vm_object_offset_t) -1)
+                               break;
 
 
-                       *io_streaming = 1;
-               } else {
+                       retval = vm_compressor_pager_relocate(object->pager, curr_offset, &freezer_chead);
 
 
-                       if (object->pages_created < 32 * ph_mult) {
-                               /*
-                                * prime the pump
-                                */
-                               pre_heat_size = PAGE_SIZE * 8 * ph_mult;
-                               break;
-                       }
-                       /*
-                        * Linear growth in PH size: The maximum size is max_length...
-                        * this cacluation will result in a size that is neither a 
-                        * power of 2 nor a multiple of PAGE_SIZE... so round
-                        * it up to the nearest PAGE_SIZE boundary
-                        */
-                       pre_heat_size = (ph_mult * (max_length * object->pages_used) / object->pages_created);
-                       
-                       if (pre_heat_size < PAGE_SIZE * min_ph_size)
-                               pre_heat_size = PAGE_SIZE * min_ph_size;
-                       else
-                               pre_heat_size = round_page(pre_heat_size);
+                       if (retval != KERN_SUCCESS)
+                               break;
+
+                       curr_offset += PAGE_SIZE_64;
                }
                }
-               break;
+       }
 
 
-       case VM_BEHAVIOR_RANDOM:
-               if ((pre_heat_size = cluster_size) <= PAGE_SIZE)
-                       goto out;
-               break;
+       /*
+        * We can't hold the object lock while heading down into the compressed pager
+        * layer because we might need the kernel map lock down there to allocate new
+        * compressor data structures. And if this same object is mapped in the kernel
+        * and there's a fault on it, then that thread will want the object lock while
+        * holding the kernel map lock.
+        *
+        * Since we are going to drop/grab the object lock repeatedly, we must make sure
+        * we won't be stuck in an infinite loop if the same page(s) keep getting
+        * decompressed. So we grab a snapshot of the number of pages in the object and
+        * we won't process any more than that number of pages.
+        */
 
 
-       case VM_BEHAVIOR_SEQUENTIAL:
-               if ((pre_heat_size = cluster_size) == 0)
-                       pre_heat_size = sequential_run + PAGE_SIZE;
-               look_behind = FALSE;
-               *io_streaming = 1;
+       obj_resident_page_count_snapshot = object->resident_page_count;
 
 
-               break;
+       vm_object_activity_begin(object);
 
 
-       case VM_BEHAVIOR_RSEQNTL:
-               if ((pre_heat_size = cluster_size) == 0)
-                       pre_heat_size = sequential_run + PAGE_SIZE;
-               look_ahead = FALSE;
-               *io_streaming = 1;
+       while ((obj_resident_page_count_snapshot--) && !vm_page_queue_empty(&object->memq)) {
 
 
-               break;
+               p = (vm_page_t)vm_page_queue_first(&object->memq);
 
 
-       }
-       throttle_limit = (uint32_t) max_length;
-       assert(throttle_limit == max_length);
+               KERNEL_DEBUG(0xe0430004 | DBG_FUNC_START, object, local_freed, 0, 0, 0);
 
 
-       if (vnode_pager_check_hard_throttle(object->pager, &throttle_limit, *io_streaming) == KERN_SUCCESS) {
-               if (max_length > throttle_limit)
-                       max_length = throttle_limit;
-       }
-       if (pre_heat_size > max_length)
-               pre_heat_size = max_length;
+               vm_page_lockspin_queues();
 
 
-       if (behavior == VM_BEHAVIOR_DEFAULT) {
-               if (vm_page_free_count < vm_page_throttle_limit)
-                       pre_heat_size = trunc_page(pre_heat_size / 8);
-               else if (vm_page_free_count < vm_page_free_target)
-                       pre_heat_size = trunc_page(pre_heat_size / 2);
+               if (p->cleaning || p->fictitious || p->busy || p->absent || p->unusual || p->error || VM_PAGE_WIRED(p)) {
 
 
-               if (pre_heat_size <= PAGE_SIZE)
-                       goto out;
-       }
-       if (look_ahead == TRUE) {
-               if (look_behind == TRUE) { 
-                       /*
-                        * if we get here its due to a random access... 
-                        * so we want to center the original fault address
-                        * within the cluster we will issue... make sure
-                        * to calculate 'head_size' as a multiple of PAGE_SIZE...
-                        * 'pre_heat_size' is a multiple of PAGE_SIZE but not
-                        * necessarily an even number of pages so we need to truncate
-                        * the result to a PAGE_SIZE boundary
-                        */
-                       head_size = trunc_page(pre_heat_size / 2);
+                       vm_page_unlock_queues();
 
 
-                       if (target_start > head_size)
-                               target_start -= head_size;
-                       else
-                               target_start = 0;
+                       KERNEL_DEBUG(0xe0430004 | DBG_FUNC_END, object, local_freed, 1, 0, 0);
+
+                       vm_page_queue_remove(&object->memq, p, vm_page_t, listq);
+                       vm_page_queue_enter(&object->memq, p, vm_page_t, listq);
+
+                       continue;
+               }
 
 
+               if (p->pmapped == TRUE) {
+                       int refmod_state, pmap_flags;
+
+                       if (p->dirty || p->precious) {
+                               pmap_flags = PMAP_OPTIONS_COMPRESSOR;
+                       } else {
+                               pmap_flags = PMAP_OPTIONS_COMPRESSOR_IFF_MODIFIED;
+                       }
+
+                       refmod_state = pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(p), pmap_flags, NULL);
+                       if (refmod_state & VM_MEM_MODIFIED) {
+                               SET_PAGE_DIRTY(p, FALSE);
+                       }
+               }
+               
+               if (p->dirty == FALSE && p->precious == FALSE) {
                        /*
                        /*
-                        * 'target_start' at this point represents the beginning offset
-                        * of the cluster we are considering... 'orig_start' will be in
-                        * the center of this cluster if we didn't have to clip the start
-                        * due to running into the start of the file
+                        * Clean and non-precious page.
                         */
                         */
+                       vm_page_unlock_queues();
+                       VM_PAGE_FREE(p);
+
+                       KERNEL_DEBUG(0xe0430004 | DBG_FUNC_END, object, local_freed, 2, 0, 0);
+                       continue;
                }
                }
-               if ((target_start + pre_heat_size) > object_size)
-                       pre_heat_size = (vm_size_t)(round_page_64(object_size - target_start));
+
+               if (p->laundry)
+                       vm_pageout_steal_laundry(p, TRUE);
+
+               vm_page_queues_remove(p, TRUE);
+
+               vm_page_unlock_queues();
+
+
                /*
                /*
-                * at this point caclulate the number of pages beyond the original fault
-                * address that we want to consider... this is guaranteed not to extend beyond
-                * the current EOF...
+                * In case the compressor fails to compress this page, we need it at
+                * the back of the object memq so that we don't keep trying to process it.
+                * Make the move here while we have the object lock held.
                 */
                 */
-               assert((vm_size_t)(orig_start - target_start) == (orig_start - target_start));
-               tail_size = pre_heat_size - (vm_size_t)(orig_start - target_start) - PAGE_SIZE;
-       } else {
-               if (pre_heat_size > target_start)
-                       pre_heat_size = (vm_size_t) target_start; /* XXX: 32-bit vs 64-bit ? Joe ? */
-               tail_size = 0;
-       }
-       assert( !(target_start & PAGE_MASK_64));
-       assert( !(pre_heat_size & PAGE_MASK));
-
-       pre_heat_scaling[pre_heat_size / PAGE_SIZE]++;
 
 
-       if (pre_heat_size <= PAGE_SIZE)
-               goto out;
+               vm_page_queue_remove(&object->memq, p, vm_page_t, listq);
+               vm_page_queue_enter(&object->memq, p, vm_page_t, listq);
 
 
-       if (look_behind == TRUE) {
-               /*
-                * take a look at the pages before the original
-                * faulting offset... recalculate this in case
-                * we had to clip 'pre_heat_size' above to keep 
-                * from running past the EOF.
+               /*
+                * Grab an activity_in_progress here for vm_pageout_compress_page() to consume.
+                *
+                * Mark the page busy so no one messes with it while we have the object lock dropped.
                 */
                 */
-               head_size = pre_heat_size - tail_size - PAGE_SIZE;
 
 
-               for (offset = orig_start - PAGE_SIZE_64; head_size; offset -= PAGE_SIZE_64, head_size -= PAGE_SIZE) {
-                       /*
-                        * don't poke below the lowest offset 
-                        */
-                       if (offset < fault_info->lo_offset)
-                               break;
-                       /*
-                        * for external objects and internal objects w/o an existence map
-                        * vm_externl_state_get will return VM_EXTERNAL_STATE_UNKNOWN
+               p->busy = TRUE;
+
+               vm_object_activity_begin(object);
+
+               vm_object_unlock(object);
+
+               /*
+                * arg3 == FALSE  tells vm_pageout_compress_page that we don't hold the object lock and the pager may not be initialized.
+                */
+               if (vm_pageout_compress_page(&freezer_chead, freezer_compressor_scratch_buf, p, FALSE) == KERN_SUCCESS) {
+                       /*
+                        * page has already been un-tabled from the object via 'vm_page_remove'
                         */
                         */
-#if MACH_PAGEMAP
-                       if (vm_external_state_get(object->existence_map, offset) == VM_EXTERNAL_STATE_ABSENT) {
-                               /*
-                                * we know for a fact that the pager can't provide the page
-                                * so don't include it or any pages beyond it in this cluster
-                                */
-                               break;
-                       }
-#endif
-                       if (vm_page_lookup(object, offset) != VM_PAGE_NULL) {
-                               /*
-                                * don't bridge resident pages
-                                */
-                               break;
+                       p->snext = local_freeq;
+                       local_freeq = p;
+                       local_freed++;
+
+                       if (local_freed >= MAX_FREE_BATCH) {
+               
+                               vm_page_free_list(local_freeq, TRUE);
+                               
+                               local_freeq = NULL;
+                               local_freed = 0;
                        }
                        }
-                       *start = offset;
-                       *length += PAGE_SIZE;
+                       c_freezer_compression_count++;
                }
                }
-       }
-       if (look_ahead == TRUE) {
-               for (offset = orig_start + PAGE_SIZE_64; tail_size; offset += PAGE_SIZE_64, tail_size -= PAGE_SIZE) {
-                       /*
-                        * don't poke above the highest offset 
-                        */
-                       if (offset >= fault_info->hi_offset)
-                               break;
-                       assert(offset < object_size);
+               KERNEL_DEBUG(0xe0430004 | DBG_FUNC_END, object, local_freed, 0, 0, 0);
 
 
-                       /*
-                        * for external objects and internal objects w/o an existence map
-                        * vm_externl_state_get will return VM_EXTERNAL_STATE_UNKNOWN
-                        */
-#if MACH_PAGEMAP
-                       if (vm_external_state_get(object->existence_map, offset) == VM_EXTERNAL_STATE_ABSENT) {
-                               /*
-                                * we know for a fact that the pager can't provide the page
-                                * so don't include it or any pages beyond it in this cluster
-                                */
-                               break;
-                       }
-#endif
-                       if (vm_page_lookup(object, offset) != VM_PAGE_NULL) {
-                               /*
-                                * don't bridge resident pages
-                                */
-                               break;
-                       }
-                       *length += PAGE_SIZE;
+               if (local_freed == 0 && c_freezer_should_yield()) {
+
+                       thread_yield_internal(FREEZER_DUTY_CYCLE_OFF_MS);
+                       clock_get_uptime(&c_freezer_last_yield_ts);
                }
                }
+
+               vm_object_lock(object);
        }
        }
-out:
-       if (*length > max_length)
-               *length = max_length;
 
 
-       pre_heat_cluster[*length / PAGE_SIZE]++;
+       if (local_freeq) {
+               vm_page_free_list(local_freeq, TRUE);
+                               
+               local_freeq = NULL;
+               local_freed = 0;
+       }
+       
+       vm_object_activity_end(object);
 
        vm_object_unlock(object);
 
        vm_object_unlock(object);
+
+       if (c_freezer_should_yield()) {
+
+               thread_yield_internal(FREEZER_DUTY_CYCLE_OFF_MS);
+               clock_get_uptime(&c_freezer_last_yield_ts);
+       }
 }
 
 }
 
+#endif /* CONFIG_FREEZE */
 
 
-/*
- * Allow manipulation of individual page state.  This is actually part of
- * the UPL regimen but takes place on the VM object rather than on a UPL
- */
 
 
-kern_return_t
-vm_object_page_op(
-       vm_object_t             object,
-       vm_object_offset_t      offset,
-       int                     ops,
-       ppnum_t                 *phys_entry,
-       int                     *flags)
+void
+vm_object_pageout(
+       vm_object_t object)
 {
 {
-       vm_page_t               dst_page;
+       vm_page_t                       p, next;
+       struct  vm_pageout_queue        *iq;
+       boolean_t                       need_unlock = TRUE;
+
+       if (!VM_CONFIG_COMPRESSOR_IS_PRESENT)
+               return;
 
 
+       iq = &vm_pageout_queue_internal;
+       
+       assert(object != VM_OBJECT_NULL );
+       
        vm_object_lock(object);
 
        vm_object_lock(object);
 
-       if(ops & UPL_POP_PHYSICAL) {
-               if(object->phys_contiguous) {
-                       if (phys_entry) {
-                               *phys_entry = (ppnum_t)
-                                       (object->shadow_offset >> PAGE_SHIFT);
-                       }
-                       vm_object_unlock(object);
-                       return KERN_SUCCESS;
-               } else {
-                       vm_object_unlock(object);
-                       return KERN_INVALID_OBJECT;
-               }
-       }
-       if(object->phys_contiguous) {
+       if (!object->internal ||
+           object->terminating ||
+           !object->alive) {
                vm_object_unlock(object);
                vm_object_unlock(object);
-               return KERN_INVALID_OBJECT;
+               return;
        }
 
        }
 
-       while(TRUE) {
-               if((dst_page = vm_page_lookup(object,offset)) == VM_PAGE_NULL) {
-                       vm_object_unlock(object);
-                       return KERN_FAILURE;
-               }
+       if (!object->pager_initialized || object->pager == MEMORY_OBJECT_NULL)  {
+                       
+               if (!object->pager_initialized) {
 
 
-               /* Sync up on getting the busy bit */
-               if((dst_page->busy || dst_page->cleaning) && 
-                          (((ops & UPL_POP_SET) && 
-                          (ops & UPL_POP_BUSY)) || (ops & UPL_POP_DUMP))) {
-                       /* someone else is playing with the page, we will */
-                       /* have to wait */
-                       PAGE_SLEEP(object, dst_page, THREAD_UNINT);
-                       continue;
+                       vm_object_collapse(object, (vm_object_offset_t) 0, TRUE);
+
+                       if (!object->pager_initialized)
+                               vm_object_compressor_pager_create(object);
                }
 
                }
 
-               if (ops & UPL_POP_DUMP) {
-                       if (dst_page->pmapped == TRUE)
-                               pmap_disconnect(dst_page->phys_page);
+               if (!object->pager_initialized || object->pager == MEMORY_OBJECT_NULL)  {
+                       vm_object_unlock(object);
+                       return;
+               }
+       }
+                       
+ReScan:        
+       next = (vm_page_t)vm_page_queue_first(&object->memq);
 
 
-                       VM_PAGE_FREE(dst_page);
-                       break;
+       while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)next)) {
+               p = next;
+               next = (vm_page_t)vm_page_queue_next(&next->listq);
+               
+               assert(p->vm_page_q_state != VM_PAGE_ON_FREE_Q);
+               
+               if ((p->vm_page_q_state == VM_PAGE_ON_THROTTLED_Q) ||
+                   p->encrypted_cleaning ||
+                   p->cleaning ||
+                   p->laundry ||
+                   p->busy ||
+                   p->absent ||
+                   p->error ||
+                   p->fictitious ||
+                   VM_PAGE_WIRED(p)) {
+                       /*
+                        * Page is already being cleaned or can't be cleaned.
+                        */
+                       continue;
                }
 
                }
 
-               if (flags) {
-                       *flags = 0;
+               /* Throw to the pageout queue */
 
 
-                       /* Get the condition of flags before requested ops */
-                       /* are undertaken */
+               vm_page_lockspin_queues();
+               need_unlock = TRUE;
 
 
-                       if(dst_page->dirty) *flags |= UPL_POP_DIRTY;
-                       if(dst_page->pageout) *flags |= UPL_POP_PAGEOUT;
-                       if(dst_page->precious) *flags |= UPL_POP_PRECIOUS;
-                       if(dst_page->absent) *flags |= UPL_POP_ABSENT;
-                       if(dst_page->busy) *flags |= UPL_POP_BUSY;
+               if (vm_compressor_low_on_space()) {
+                       vm_page_unlock_queues();
+                       break;          
                }
 
                }
 
-               /* The caller should have made a call either contingent with */
-               /* or prior to this call to set UPL_POP_BUSY */
-               if(ops & UPL_POP_SET) {
-                       /* The protection granted with this assert will */
-                       /* not be complete.  If the caller violates the */
-                       /* convention and attempts to change page state */
-                       /* without first setting busy we may not see it */
-                       /* because the page may already be busy.  However */
-                       /* if such violations occur we will assert sooner */
-                       /* or later. */
-                       assert(dst_page->busy || (ops & UPL_POP_BUSY));
-                       if (ops & UPL_POP_DIRTY) dst_page->dirty = TRUE;
-                       if (ops & UPL_POP_PAGEOUT) dst_page->pageout = TRUE;
-                       if (ops & UPL_POP_PRECIOUS) dst_page->precious = TRUE;
-                       if (ops & UPL_POP_ABSENT) dst_page->absent = TRUE;
-                       if (ops & UPL_POP_BUSY) dst_page->busy = TRUE;
-               }
+               if (VM_PAGE_Q_THROTTLED(iq)) {
+                                       
+                       iq->pgo_draining = TRUE;
+                                       
+                       assert_wait((event_t) (&iq->pgo_laundry + 1),
+                                   THREAD_INTERRUPTIBLE);
+                       vm_page_unlock_queues();
+                       vm_object_unlock(object);
+                                       
+                       thread_block(THREAD_CONTINUE_NULL);
 
 
-               if(ops & UPL_POP_CLR) {
-                       assert(dst_page->busy);
-                       if (ops & UPL_POP_DIRTY) dst_page->dirty = FALSE;
-                       if (ops & UPL_POP_PAGEOUT) dst_page->pageout = FALSE;
-                       if (ops & UPL_POP_PRECIOUS) dst_page->precious = FALSE;
-                       if (ops & UPL_POP_ABSENT) dst_page->absent = FALSE;
-                       if (ops & UPL_POP_BUSY) {
-                               dst_page->busy = FALSE;
-                               PAGE_WAKEUP(dst_page);
-                       }
+                       vm_object_lock(object);
+                       goto ReScan;
                }
 
                }
 
-               if (dst_page->encrypted) {
+               assert(!p->fictitious);
+               assert(!p->busy);
+               assert(!p->absent);
+               assert(!p->unusual);
+               assert(!p->error);
+               assert(!VM_PAGE_WIRED(p));
+               assert(!p->cleaning);
+
+               if (p->pmapped == TRUE) {
+                       int refmod_state;
+                       int pmap_options;
+
                        /*
                        /*
-                        * ENCRYPTED SWAP:
-                        * We need to decrypt this encrypted page before the
-                        * caller can access its contents.
-                        * But if the caller really wants to access the page's
-                        * contents, they have to keep the page "busy".
-                        * Otherwise, the page could get recycled or re-encrypted
-                        * at any time.
+                        * Tell pmap the page should be accounted
+                        * for as "compressed" if it's been modified.
                         */
                         */
-                       if ((ops & UPL_POP_SET) && (ops & UPL_POP_BUSY) &&
-                           dst_page->busy) {
-                               /*
-                                * The page is stable enough to be accessed by
-                                * the caller, so make sure its contents are
-                                * not encrypted.
-                                */
-                               vm_page_decrypt(dst_page, 0);
-                       } else {
+                       pmap_options =
+                               PMAP_OPTIONS_COMPRESSOR_IFF_MODIFIED;
+                       if (p->dirty || p->precious) {
                                /*
                                /*
-                                * The page is not busy, so don't bother
-                                * decrypting it, since anything could
-                                * happen to it between now and when the
-                                * caller wants to access it.
-                                * We should not give the caller access
-                                * to this page.
+                                * We already know it's been modified,
+                                * so tell pmap to account for it
+                                * as "compressed".
                                 */
                                 */
-                               assert(!phys_entry);
+                               pmap_options = PMAP_OPTIONS_COMPRESSOR;
+                       }
+                       refmod_state = pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(p),
+                                                              pmap_options,
+                                                              NULL);
+                       if (refmod_state & VM_MEM_MODIFIED) {
+                               SET_PAGE_DIRTY(p, FALSE);
                        }
                }
 
                        }
                }
 
-               if (phys_entry) {
-                       /*
-                        * The physical page number will remain valid
-                        * only if the page is kept busy.
-                        * ENCRYPTED SWAP: make sure we don't let the
-                        * caller access an encrypted page.
-                        */
-                       assert(dst_page->busy);
-                       assert(!dst_page->encrypted);
-                       *phys_entry = dst_page->phys_page;
+               if (!p->dirty && !p->precious) {
+                       vm_page_unlock_queues();
+                       VM_PAGE_FREE(p);
+                       continue;
                }
 
                }
 
-               break;
+               vm_page_queues_remove(p, TRUE);
+
+               if (vm_pageout_cluster(p, FALSE, TRUE))
+                       need_unlock = FALSE;
+
+               if (need_unlock == TRUE)
+                       vm_page_unlock_queues();
        }
 
        vm_object_unlock(object);
        }
 
        vm_object_unlock(object);
-       return KERN_SUCCESS;
-                               
 }
 
 }
 
-/*
- * vm_object_range_op offers performance enhancement over 
- * vm_object_page_op for page_op functions which do not require page 
- * level state to be returned from the call.  Page_op was created to provide 
- * a low-cost alternative to page manipulation via UPLs when only a single 
- * page was involved.  The range_op call establishes the ability in the _op 
- * family of functions to work on multiple pages where the lack of page level
- * state handling allows the caller to avoid the overhead of the upl structures.
- */
 
 
-kern_return_t
-vm_object_range_op(
-       vm_object_t             object,
-       vm_object_offset_t      offset_beg,
-       vm_object_offset_t      offset_end,
-       int                     ops,
-       uint32_t                *range)
+#if CONFIG_IOSCHED
+void
+vm_page_request_reprioritize(vm_object_t o, uint64_t blkno, uint32_t len, int prio)
 {
 {
-        vm_object_offset_t     offset;
-       vm_page_t               dst_page;
+       io_reprioritize_req_t   req;
+       struct vnode            *devvp = NULL;  
 
 
-       if (offset_end - offset_beg > (uint32_t) -1) {
-               /* range is too big and would overflow "*range" */
-               return KERN_INVALID_ARGUMENT;
-       } 
-       if (object->resident_page_count == 0) {
-               if (range) {
-                       if (ops & UPL_ROP_PRESENT) {
-                               *range = 0;
-                       } else {
-                               *range = (uint32_t) (offset_end - offset_beg);
-                               assert(*range == (offset_end - offset_beg));
-                       }
-               }
-               return KERN_SUCCESS;
-       }
-       vm_object_lock(object);
+       if(vnode_pager_get_object_devvp(o->pager, (uintptr_t *)&devvp) != KERN_SUCCESS)
+               return;
+       
+       /*
+        * Create the request for I/O reprioritization.
+        * We use the noblock variant of zalloc because we're holding the object
+        * lock here and we could cause a deadlock in low memory conditions.
+        */
+       req = (io_reprioritize_req_t)zalloc_noblock(io_reprioritize_req_zone);
+       if (req == NULL)
+               return;
+       req->blkno = blkno;
+       req->len = len;
+       req->priority = prio;
+       req->devvp = devvp;
 
 
-       if (object->phys_contiguous) {
-               vm_object_unlock(object);
-               return KERN_INVALID_OBJECT;
+       /* Insert request into the reprioritization list */
+       IO_REPRIORITIZE_LIST_LOCK();
+       queue_enter(&io_reprioritize_list, req, io_reprioritize_req_t, io_reprioritize_list);
+       IO_REPRIORITIZE_LIST_UNLOCK();
+
+       /* Wakeup reprioritize thread */
+       IO_REPRIO_THREAD_WAKEUP();      
+
+       return;         
+}      
+
+void
+vm_decmp_upl_reprioritize(upl_t upl, int prio)
+{
+       int offset;
+       vm_object_t object;
+       io_reprioritize_req_t   req;
+       struct vnode            *devvp = NULL;
+       uint64_t                blkno;
+       uint32_t                len;
+       upl_t                   io_upl;
+       uint64_t                *io_upl_reprio_info;
+       int                     io_upl_size;
+
+       if ((upl->flags & UPL_TRACKED_BY_OBJECT) == 0 || (upl->flags & UPL_EXPEDITE_SUPPORTED) == 0)
+               return;
+
+       /* 
+        * We dont want to perform any allocations with the upl lock held since that might 
+        * result in a deadlock. If the system is low on memory, the pageout thread would 
+        * try to pageout stuff and might wait on this lock. If we are waiting for the memory to
+        * be freed up by the pageout thread, it would be a deadlock.
+        */
+
+
+       /* First step is just to get the size of the upl to find out how big the reprio info is */
+       if(!upl_try_lock(upl))
+               return;
+
+       if (upl->decmp_io_upl == NULL) {
+               /* The real I/O upl was destroyed by the time we came in here. Nothing to do. */
+               upl_unlock(upl);
+               return;
        }
        }
+
+       io_upl = upl->decmp_io_upl;
+       assert((io_upl->flags & UPL_DECMP_REAL_IO) != 0);
+       io_upl_size = io_upl->size;
+       upl_unlock(upl);
        
        
-       offset = offset_beg & ~PAGE_MASK_64;
+       /* Now perform the allocation */
+       io_upl_reprio_info = (uint64_t *)kalloc(sizeof(uint64_t) * (io_upl_size / PAGE_SIZE));
+       if (io_upl_reprio_info == NULL)
+               return;
 
 
-       while (offset < offset_end) {
-               dst_page = vm_page_lookup(object, offset);
-               if (dst_page != VM_PAGE_NULL) {
-                       if (ops & UPL_ROP_DUMP) {
-                               if (dst_page->busy || dst_page->cleaning) {
-                                       /*
-                                        * someone else is playing with the 
-                                        * page, we will have to wait
-                                        */
-                                       PAGE_SLEEP(object, dst_page, THREAD_UNINT);
-                                       /*
-                                        * need to relook the page up since it's
-                                        * state may have changed while we slept
-                                        * it might even belong to a different object
-                                        * at this point
-                                        */
-                                       continue;
-                               }
-                               if (dst_page->pmapped == TRUE)
-                                       pmap_disconnect(dst_page->phys_page);
+       /* Now again take the lock, recheck the state and grab out the required info */
+       if(!upl_try_lock(upl))
+               goto out;
 
 
-                               VM_PAGE_FREE(dst_page);
+       if (upl->decmp_io_upl == NULL || upl->decmp_io_upl != io_upl) {
+               /* The real I/O upl was destroyed by the time we came in here. Nothing to do. */
+               upl_unlock(upl);
+               goto out;
+       }
+       memcpy(io_upl_reprio_info, io_upl->upl_reprio_info, sizeof(uint64_t) * (io_upl_size / PAGE_SIZE));
 
 
-                       } else if ((ops & UPL_ROP_ABSENT) && !dst_page->absent)
-                               break;
-               } else if (ops & UPL_ROP_PRESENT)
-                       break;
+       /* Get the VM object for this UPL */
+       if (io_upl->flags & UPL_SHADOWED) {
+               object = io_upl->map_object->shadow;
+       } else {
+               object = io_upl->map_object;
+       }
 
 
-               offset += PAGE_SIZE;
+       /* Get the dev vnode ptr for this object */
+       if(!object || !object->pager || 
+          vnode_pager_get_object_devvp(object->pager, (uintptr_t *)&devvp) != KERN_SUCCESS) {
+               upl_unlock(upl);
+               goto out;
        }
        }
-       vm_object_unlock(object);
 
 
-       if (range) {
-               if (offset > offset_end)
-                       offset = offset_end;
-               if(offset > offset_beg) {
-                       *range = (uint32_t) (offset - offset_beg);
-                       assert(*range == (offset - offset_beg));
+       upl_unlock(upl);
+
+       /* Now we have all the information needed to do the expedite */
+
+       offset = 0;
+       while (offset < io_upl_size) {
+               blkno   = io_upl_reprio_info[(offset / PAGE_SIZE)] & UPL_REPRIO_INFO_MASK;
+               len     = (io_upl_reprio_info[(offset / PAGE_SIZE)] >> UPL_REPRIO_INFO_SHIFT) & UPL_REPRIO_INFO_MASK;   
+
+               /*
+                * This implementation may cause some spurious expedites due to the 
+                * fact that we dont cleanup the blkno & len from the upl_reprio_info 
+                * even after the I/O is complete. 
+                */
+               
+               if (blkno != 0 && len != 0) {
+                       /* Create the request for I/O reprioritization */
+                               req = (io_reprioritize_req_t)zalloc(io_reprioritize_req_zone);
+                       assert(req != NULL);
+                       req->blkno = blkno;
+                       req->len = len;
+                       req->priority = prio;
+                       req->devvp = devvp;
+
+                       /* Insert request into the reprioritization list */
+                       IO_REPRIORITIZE_LIST_LOCK();
+                       queue_enter(&io_reprioritize_list, req, io_reprioritize_req_t, io_reprioritize_list);
+                       IO_REPRIORITIZE_LIST_UNLOCK();          
+                       
+                       offset += len;
                } else {
                } else {
-                       *range = 0;
+                       offset += PAGE_SIZE;
                }
        }
                }
        }
-       return KERN_SUCCESS;
-}
 
 
+       /* Wakeup reprioritize thread */
+        IO_REPRIO_THREAD_WAKEUP();
 
 
-uint32_t scan_object_collision = 0;
+out:
+       kfree(io_upl_reprio_info, sizeof(uint64_t) * (io_upl_size / PAGE_SIZE));
+       return;
+}
 
 void
 
 void
-vm_object_lock(vm_object_t object)
+vm_page_handle_prio_inversion(vm_object_t o, vm_page_t m)
 {
 {
-        if (object == vm_pageout_scan_wants_object) {
-               scan_object_collision++;
-               mutex_pause(2);
-       }
-        lck_rw_lock_exclusive(&object->Lock);
-}
+       upl_t upl;
+        upl_page_info_t *pl;
+        unsigned int i, num_pages;
+        int cur_tier;
 
 
-boolean_t
-vm_object_lock_avoid(vm_object_t object)
-{
-        if (object == vm_pageout_scan_wants_object) {
-               scan_object_collision++;
-               return TRUE;
+       cur_tier = proc_get_effective_thread_policy(current_thread(), TASK_POLICY_IO);
+
+       /* 
+       Scan through all UPLs associated with the object to find the 
+       UPL containing the contended page.
+       */ 
+       queue_iterate(&o->uplq, upl, upl_t, uplq) {
+               if (((upl->flags & UPL_EXPEDITE_SUPPORTED) == 0) || upl->upl_priority <= cur_tier)
+                       continue;
+               pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
+                num_pages = (upl->size / PAGE_SIZE);
+                
+               /*
+               For each page in the UPL page list, see if it matches the contended
+               page and was issued as a low prio I/O. 
+               */
+               for(i=0; i < num_pages; i++) {
+                       if(UPL_PAGE_PRESENT(pl,i) && VM_PAGE_GET_PHYS_PAGE(m) == pl[i].phys_addr) {
+                               if ((upl->flags & UPL_DECMP_REQ) && upl->decmp_io_upl) {
+                                       KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, VM_PAGE_EXPEDITE)) | DBG_FUNC_NONE, VM_KERNEL_UNSLIDE_OR_PERM(upl->upl_creator), VM_KERNEL_UNSLIDE_OR_PERM(m),
+                                               VM_KERNEL_UNSLIDE_OR_PERM(upl), upl->upl_priority, 0);
+                                       vm_decmp_upl_reprioritize(upl, cur_tier);
+                                       break;
+                               }
+                               KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, VM_PAGE_EXPEDITE)) | DBG_FUNC_NONE, VM_KERNEL_UNSLIDE_OR_PERM(upl->upl_creator), VM_KERNEL_UNSLIDE_OR_PERM(m),
+                                       upl->upl_reprio_info[i], upl->upl_priority, 0);
+                               if (UPL_REPRIO_INFO_BLKNO(upl, i) != 0 && UPL_REPRIO_INFO_LEN(upl, i) != 0) 
+                                       vm_page_request_reprioritize(o, UPL_REPRIO_INFO_BLKNO(upl, i), UPL_REPRIO_INFO_LEN(upl, i), cur_tier);
+                                break;
+                         }
+                }
+                /* Check if we found any hits */
+                 if (i != num_pages)
+                       break;
        }
        }
-       return FALSE;
-}
+       
+       return;
+}      
 
 
-boolean_t
-_vm_object_lock_try(vm_object_t object)
+wait_result_t
+vm_page_sleep(vm_object_t o, vm_page_t m, int interruptible)
 {
 {
-       return (lck_rw_try_lock_exclusive(&object->Lock));
-}
+       wait_result_t ret;
 
 
-boolean_t
-vm_object_lock_try(vm_object_t object)
-{
-       if (vm_object_lock_avoid(object)) {
-               mutex_pause(2);
-       }
-       return _vm_object_lock_try(object);
-}
-void
-vm_object_lock_shared(vm_object_t object)
-{
-        if (vm_object_lock_avoid(object)) {
-               mutex_pause(2);
-       }
-       lck_rw_lock_shared(&object->Lock);
+       KERNEL_DEBUG((MACHDBG_CODE(DBG_MACH_VM, VM_PAGE_SLEEP)) | DBG_FUNC_START, o, m, 0, 0, 0);
+       
+       if (o->io_tracking && ((m->busy == TRUE) || (m->cleaning == TRUE) || VM_PAGE_WIRED(m))) {
+               /* 
+               Indicates page is busy due to an I/O. Issue a reprioritize request if necessary.
+               */
+               vm_page_handle_prio_inversion(o,m);
+       }
+       m->wanted = TRUE;
+       ret = thread_sleep_vm_object(o, m, interruptible);
+       KERNEL_DEBUG((MACHDBG_CODE(DBG_MACH_VM, VM_PAGE_SLEEP)) | DBG_FUNC_END, o, m, 0, 0, 0);
+       return ret;
 }
 
 }
 
-boolean_t
-vm_object_lock_try_shared(vm_object_t object)
+static void
+io_reprioritize_thread(void *param __unused, wait_result_t wr __unused)
 {
 {
-        if (vm_object_lock_avoid(object)) {
-               mutex_pause(2);
-       }
-       return (lck_rw_try_lock_shared(&object->Lock));
+       io_reprioritize_req_t   req = NULL;
+       
+       while(1) {
+
+               IO_REPRIORITIZE_LIST_LOCK();
+               if (queue_empty(&io_reprioritize_list)) {
+                       IO_REPRIORITIZE_LIST_UNLOCK();
+                       break;
+               }
+                       
+               queue_remove_first(&io_reprioritize_list, req, io_reprioritize_req_t, io_reprioritize_list);   
+               IO_REPRIORITIZE_LIST_UNLOCK();
+               
+               vnode_pager_issue_reprioritize_io(req->devvp, req->blkno, req->len, req->priority);
+               zfree(io_reprioritize_req_zone, req);   
+       }       
+       
+       IO_REPRIO_THREAD_CONTINUATION();
 }
 }
+#endif