/*
- * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2009 Apple Inc. All rights reserved.
*
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ *
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
*
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* Please see the License for the specific language governing rights and
* limitations under the License.
*
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* @OSF_COPYRIGHT@
*/
#include <debug.h>
+#include <libkern/OSAtomic.h>
+#include <libkern/OSDebug.h>
#include <mach/clock_types.h>
#include <mach/vm_prot.h>
#include <mach/vm_statistics.h>
+#include <mach/sdt.h>
#include <kern/counters.h>
#include <kern/sched_prim.h>
#include <kern/task.h>
#include <kern/thread.h>
+#include <kern/kalloc.h>
#include <kern/zalloc.h>
#include <kern/xpr.h>
+#include <kern/ledger.h>
#include <vm/pmap.h>
#include <vm/vm_init.h>
#include <vm/vm_map.h>
#include <vm/vm_kern.h> /* kernel_memory_allocate() */
#include <kern/misc_protos.h>
#include <zone_debug.h>
+#include <mach_debug/zone_info.h>
#include <vm/cpm.h>
-#include <ppc/mappings.h> /* (BRINGUP) */
-#include <pexpert/pexpert.h> /* (BRINGUP) */
+#include <pexpert/pexpert.h>
#include <vm/vm_protos.h>
+#include <vm/memory_object.h>
+#include <vm/vm_purgeable_internal.h>
+#include <vm/vm_compressor.h>
-/* Variables used to indicate the relative age of pages in the
- * inactive list
- */
+#if CONFIG_PHANTOM_CACHE
+#include <vm/vm_phantom_cache.h>
+#endif
+
+#include <IOKit/IOHibernatePrivate.h>
+
+#include <sys/kdebug.h>
+
+boolean_t hibernate_cleaning_in_progress = FALSE;
+boolean_t vm_page_free_verify = TRUE;
+
+uint32_t vm_lopage_free_count = 0;
+uint32_t vm_lopage_free_limit = 0;
+uint32_t vm_lopage_lowater = 0;
+boolean_t vm_lopage_refill = FALSE;
+boolean_t vm_lopage_needed = FALSE;
+
+lck_mtx_ext_t vm_page_queue_lock_ext;
+lck_mtx_ext_t vm_page_queue_free_lock_ext;
+lck_mtx_ext_t vm_purgeable_queue_lock_ext;
+
+int speculative_age_index = 0;
+int speculative_steal_index = 0;
+struct vm_speculative_age_q vm_page_queue_speculative[VM_PAGE_MAX_SPECULATIVE_AGE_Q + 1];
+
+
+__private_extern__ void vm_page_init_lck_grp(void);
+
+static void vm_page_free_prepare(vm_page_t page);
+static vm_page_t vm_page_grab_fictitious_common(ppnum_t phys_addr);
+
+static void vm_tag_init(void);
+
+uint64_t vm_min_kernel_and_kext_address = VM_MIN_KERNEL_AND_KEXT_ADDRESS;
-unsigned int vm_page_ticket_roll = 0;
-unsigned int vm_page_ticket = 0;
/*
* Associated with page of user-allocatable memory is a
* page structure.
vm_offset_t virtual_space_start;
vm_offset_t virtual_space_end;
-int vm_page_pages;
+uint32_t vm_page_pages;
/*
* The vm_page_lookup() routine, which provides for fast
* or VP, table.]
*/
typedef struct {
- vm_page_t pages;
+ vm_page_packed_t page_list;
#if MACH_PAGE_HASH_STATS
int cur_count; /* current count */
int hi_count; /* high water mark */
#endif /* MACH_PAGE_HASH_STATS */
} vm_page_bucket_t;
+
+#define BUCKETS_PER_LOCK 16
+
vm_page_bucket_t *vm_page_buckets; /* Array of buckets */
unsigned int vm_page_bucket_count = 0; /* How big is array? */
unsigned int vm_page_hash_mask; /* Mask for hash function */
unsigned int vm_page_hash_shift; /* Shift for hash function */
-uint32_t vm_page_bucket_hash; /* Basic bucket hash */
-decl_simple_lock_data(,vm_page_bucket_lock)
+uint32_t vm_page_bucket_hash; /* Basic bucket hash */
+unsigned int vm_page_bucket_lock_count = 0; /* How big is array of locks? */
-vm_page_t
-vm_page_lookup_nohint(vm_object_t object, vm_object_offset_t offset);
+lck_spin_t *vm_page_bucket_locks;
+lck_spin_t vm_objects_wired_lock;
+lck_spin_t vm_allocation_sites_lock;
+
+#if VM_PAGE_BUCKETS_CHECK
+boolean_t vm_page_buckets_check_ready = FALSE;
+#if VM_PAGE_FAKE_BUCKETS
+vm_page_bucket_t *vm_page_fake_buckets; /* decoy buckets */
+vm_map_offset_t vm_page_fake_buckets_start, vm_page_fake_buckets_end;
+#endif /* VM_PAGE_FAKE_BUCKETS */
+#endif /* VM_PAGE_BUCKETS_CHECK */
+
+extern int not_in_kdp;
#if MACH_PAGE_HASH_STATS
*/
vm_size_t page_size = PAGE_SIZE;
vm_size_t page_mask = PAGE_MASK;
-int page_shift = PAGE_SHIFT;
+int page_shift = PAGE_SHIFT;
/*
* Resident page structures are initialized from
*/
struct vm_page vm_page_template;
+vm_page_t vm_pages = VM_PAGE_NULL;
+unsigned int vm_pages_count = 0;
+ppnum_t vm_page_lowest = 0;
+
/*
* Resident pages that represent real memory
- * are allocated from a free list.
+ * are allocated from a set of free lists,
+ * one per color.
*/
-vm_page_t vm_page_queue_free;
-vm_page_t vm_page_queue_fictitious;
+unsigned int vm_colors;
+unsigned int vm_color_mask; /* mask is == (vm_colors-1) */
+unsigned int vm_cache_geometry_colors = 0; /* set by hw dependent code during startup */
+unsigned int vm_free_magazine_refill_limit = 0;
+queue_head_t vm_page_queue_free[MAX_COLORS];
unsigned int vm_page_free_wanted;
+unsigned int vm_page_free_wanted_privileged;
unsigned int vm_page_free_count;
unsigned int vm_page_fictitious_count;
-unsigned int vm_page_free_count_minimum; /* debugging */
-
/*
* Occasionally, the virtual memory system uses
* resident page structures that do not refer to
* most other kernel structures are.
*/
zone_t vm_page_zone;
-decl_mutex_data(,vm_page_alloc_lock)
+vm_locks_array_t vm_page_locks;
+decl_lck_mtx_data(,vm_page_alloc_lock)
+lck_mtx_ext_t vm_page_alloc_lock_ext;
+
unsigned int io_throttle_zero_fill;
+unsigned int vm_page_local_q_count = 0;
+unsigned int vm_page_local_q_soft_limit = 250;
+unsigned int vm_page_local_q_hard_limit = 500;
+struct vplq *vm_page_local_q = NULL;
+
+/* N.B. Guard and fictitious pages must not
+ * be assigned a zero phys_page value.
+ */
/*
* Fictitious pages don't have a physical address,
* but we must initialize phys_page to something.
* For debugging, this should be a strange value
* that the pmap module can recognize in assertions.
*/
-vm_offset_t vm_page_fictitious_addr = (vm_offset_t) -1;
+ppnum_t vm_page_fictitious_addr = (ppnum_t) -1;
+
+/*
+ * Guard pages are not accessible so they don't
+ * need a physical address, but we need to enter
+ * one in the pmap.
+ * Let's make it recognizable and make sure that
+ * we don't use a real physical page with that
+ * physical address.
+ */
+ppnum_t vm_page_guard_addr = (ppnum_t) -2;
/*
* Resident page structures are also chained on
* system (pageout daemon). These queues are
* defined here, but are shared by the pageout
* module. The inactive queue is broken into
- * inactive and zf for convenience as the
+ * file backed and anonymous for convenience as the
* pageout daemon often assignes a higher
- * affinity to zf pages
+ * importance to anonymous pages (less likely to pick)
*/
queue_head_t vm_page_queue_active;
queue_head_t vm_page_queue_inactive;
+queue_head_t vm_page_queue_anonymous; /* inactive memory queue for anonymous pages */
+queue_head_t vm_page_queue_throttled;
+
+queue_head_t vm_objects_wired;
+
unsigned int vm_page_active_count;
unsigned int vm_page_inactive_count;
+unsigned int vm_page_anonymous_count;
+unsigned int vm_page_throttled_count;
+unsigned int vm_page_speculative_count;
+
unsigned int vm_page_wire_count;
+unsigned int vm_page_stolen_count;
+unsigned int vm_page_wire_count_initial;
+unsigned int vm_page_pages_initial;
unsigned int vm_page_gobble_count = 0;
-unsigned int vm_page_wire_count_warning = 0;
-unsigned int vm_page_gobble_count_warning = 0;
+
+#define VM_PAGE_WIRE_COUNT_WARNING 0
+#define VM_PAGE_GOBBLE_COUNT_WARNING 0
unsigned int vm_page_purgeable_count = 0; /* # of pages purgeable now */
+unsigned int vm_page_purgeable_wired_count = 0; /* # of purgeable pages that are wired now */
uint64_t vm_page_purged_count = 0; /* total count of purged pages */
+unsigned int vm_page_xpmapped_external_count = 0;
+unsigned int vm_page_external_count = 0;
+unsigned int vm_page_internal_count = 0;
+unsigned int vm_page_pageable_external_count = 0;
+unsigned int vm_page_pageable_internal_count = 0;
+
+#if DEVELOPMENT || DEBUG
+unsigned int vm_page_speculative_recreated = 0;
+unsigned int vm_page_speculative_created = 0;
+unsigned int vm_page_speculative_used = 0;
+#endif
+
+queue_head_t vm_page_queue_cleaned;
+
+unsigned int vm_page_cleaned_count = 0;
+unsigned int vm_pageout_enqueued_cleaned = 0;
+
+uint64_t max_valid_dma_address = 0xffffffffffffffffULL;
+ppnum_t max_valid_low_ppnum = 0xffffffff;
+
+
/*
* Several page replacement parameters are also
* shared with this module, so that page allocation
*/
unsigned int vm_page_free_target = 0;
unsigned int vm_page_free_min = 0;
+unsigned int vm_page_throttle_limit = 0;
unsigned int vm_page_inactive_target = 0;
+unsigned int vm_page_anonymous_min = 0;
+unsigned int vm_page_inactive_min = 0;
unsigned int vm_page_free_reserved = 0;
-unsigned int vm_page_throttled_count = 0;
+unsigned int vm_page_throttle_count = 0;
+
/*
* The VM system has a couple of heuristics for deciding
boolean_t vm_page_deactivate_hint = TRUE;
+struct vm_page_stats_reusable vm_page_stats_reusable;
+
/*
* vm_set_page_size:
*
void
vm_set_page_size(void)
{
- page_mask = page_size - 1;
+ page_size = PAGE_SIZE;
+ page_mask = PAGE_MASK;
+ page_shift = PAGE_SHIFT;
if ((page_mask & page_size) != 0)
panic("vm_set_page_size: page size not a power of two");
break;
}
+#define COLOR_GROUPS_TO_STEAL 4
+
+
+/* Called once during statup, once the cache geometry is known.
+ */
+static void
+vm_page_set_colors( void )
+{
+ unsigned int n, override;
+
+ if ( PE_parse_boot_argn("colors", &override, sizeof (override)) ) /* colors specified as a boot-arg? */
+ n = override;
+ else if ( vm_cache_geometry_colors ) /* do we know what the cache geometry is? */
+ n = vm_cache_geometry_colors;
+ else n = DEFAULT_COLORS; /* use default if all else fails */
+
+ if ( n == 0 )
+ n = 1;
+ if ( n > MAX_COLORS )
+ n = MAX_COLORS;
+
+ /* the count must be a power of 2 */
+ if ( ( n & (n - 1)) != 0 )
+ panic("vm_page_set_colors");
+
+ vm_colors = n;
+ vm_color_mask = n - 1;
+
+ vm_free_magazine_refill_limit = vm_colors * COLOR_GROUPS_TO_STEAL;
+}
+
+
+lck_grp_t vm_page_lck_grp_free;
+lck_grp_t vm_page_lck_grp_queue;
+lck_grp_t vm_page_lck_grp_local;
+lck_grp_t vm_page_lck_grp_purge;
+lck_grp_t vm_page_lck_grp_alloc;
+lck_grp_t vm_page_lck_grp_bucket;
+lck_grp_attr_t vm_page_lck_grp_attr;
+lck_attr_t vm_page_lck_attr;
+
+
+__private_extern__ void
+vm_page_init_lck_grp(void)
+{
+ /*
+ * initialze the vm_page lock world
+ */
+ lck_grp_attr_setdefault(&vm_page_lck_grp_attr);
+ lck_grp_init(&vm_page_lck_grp_free, "vm_page_free", &vm_page_lck_grp_attr);
+ lck_grp_init(&vm_page_lck_grp_queue, "vm_page_queue", &vm_page_lck_grp_attr);
+ lck_grp_init(&vm_page_lck_grp_local, "vm_page_queue_local", &vm_page_lck_grp_attr);
+ lck_grp_init(&vm_page_lck_grp_purge, "vm_page_purge", &vm_page_lck_grp_attr);
+ lck_grp_init(&vm_page_lck_grp_alloc, "vm_page_alloc", &vm_page_lck_grp_attr);
+ lck_grp_init(&vm_page_lck_grp_bucket, "vm_page_bucket", &vm_page_lck_grp_attr);
+ lck_attr_setdefault(&vm_page_lck_attr);
+ lck_mtx_init_ext(&vm_page_alloc_lock, &vm_page_alloc_lock_ext, &vm_page_lck_grp_alloc, &vm_page_lck_attr);
+
+ vm_compressor_init_locks();
+}
+
+void
+vm_page_init_local_q()
+{
+ unsigned int num_cpus;
+ unsigned int i;
+ struct vplq *t_local_q;
+
+ num_cpus = ml_get_max_cpus();
+
+ /*
+ * no point in this for a uni-processor system
+ */
+ if (num_cpus >= 2) {
+ t_local_q = (struct vplq *)kalloc(num_cpus * sizeof(struct vplq));
+
+ for (i = 0; i < num_cpus; i++) {
+ struct vpl *lq;
+
+ lq = &t_local_q[i].vpl_un.vpl;
+ VPL_LOCK_INIT(lq, &vm_page_lck_grp_local, &vm_page_lck_attr);
+ queue_init(&lq->vpl_queue);
+ lq->vpl_count = 0;
+ lq->vpl_internal_count = 0;
+ lq->vpl_external_count = 0;
+ }
+ vm_page_local_q_count = num_cpus;
+
+ vm_page_local_q = (struct vplq *)t_local_q;
+ }
+}
+
+
/*
* vm_page_bootstrap:
*
*/
m = &vm_page_template;
- m->object = VM_OBJECT_NULL; /* reset later */
- m->offset = (vm_object_offset_t) -1; /* reset later */
- m->wire_count = 0;
+ bzero(m, sizeof (*m));
m->pageq.next = NULL;
m->pageq.prev = NULL;
m->listq.next = NULL;
m->listq.prev = NULL;
+ m->next_m = VM_PAGE_PACK_PTR(VM_PAGE_NULL);
+
+ m->object = VM_OBJECT_NULL; /* reset later */
+ m->offset = (vm_object_offset_t) -1; /* reset later */
+ m->wire_count = 0;
+ m->local = FALSE;
m->inactive = FALSE;
m->active = FALSE;
+ m->pageout_queue = FALSE;
+ m->speculative = FALSE;
m->laundry = FALSE;
m->free = FALSE;
- m->no_isync = TRUE;
m->reference = FALSE;
- m->pageout = FALSE;
- m->dump_cleaning = FALSE;
- m->list_req_pending = FALSE;
+ m->gobbled = FALSE;
+ m->private = FALSE;
+ m->throttled = FALSE;
+ m->__unused_pageq_bits = 0;
+
+ m->phys_page = 0; /* reset later */
m->busy = TRUE;
m->wanted = FALSE;
m->tabled = FALSE;
+ m->hashed = FALSE;
m->fictitious = FALSE;
- m->private = FALSE;
+ m->pmapped = FALSE;
+ m->wpmapped = FALSE;
+ m->pageout = FALSE;
m->absent = FALSE;
m->error = FALSE;
m->dirty = FALSE;
m->cleaning = FALSE;
m->precious = FALSE;
m->clustered = FALSE;
- m->lock_supplied = FALSE;
- m->unusual = FALSE;
+ m->overwriting = FALSE;
m->restart = FALSE;
- m->zero_fill = FALSE;
+ m->unusual = FALSE;
m->encrypted = FALSE;
-
- m->phys_page = 0; /* reset later */
-
- m->page_lock = VM_PROT_NONE;
- m->unlock_request = VM_PROT_NONE;
- m->page_error = KERN_SUCCESS;
+ m->encrypted_cleaning = FALSE;
+ m->cs_validated = FALSE;
+ m->cs_tainted = FALSE;
+ m->cs_nx = FALSE;
+ m->no_cache = FALSE;
+ m->reusable = FALSE;
+ m->slid = FALSE;
+ m->xpmapped = FALSE;
+ m->compressor = FALSE;
+ m->written_by_kernel = FALSE;
+ m->__unused_object_bits = 0;
/*
* Initialize the page queues.
*/
-
- mutex_init(&vm_page_queue_free_lock, 0);
- mutex_init(&vm_page_queue_lock, 0);
-
- vm_page_queue_free = VM_PAGE_NULL;
- vm_page_queue_fictitious = VM_PAGE_NULL;
+ vm_page_init_lck_grp();
+
+ lck_mtx_init_ext(&vm_page_queue_free_lock, &vm_page_queue_free_lock_ext, &vm_page_lck_grp_free, &vm_page_lck_attr);
+ lck_mtx_init_ext(&vm_page_queue_lock, &vm_page_queue_lock_ext, &vm_page_lck_grp_queue, &vm_page_lck_attr);
+ lck_mtx_init_ext(&vm_purgeable_queue_lock, &vm_purgeable_queue_lock_ext, &vm_page_lck_grp_purge, &vm_page_lck_attr);
+
+ for (i = 0; i < PURGEABLE_Q_TYPE_MAX; i++) {
+ int group;
+
+ purgeable_queues[i].token_q_head = 0;
+ purgeable_queues[i].token_q_tail = 0;
+ for (group = 0; group < NUM_VOLATILE_GROUPS; group++)
+ queue_init(&purgeable_queues[i].objq[group]);
+
+ purgeable_queues[i].type = i;
+ purgeable_queues[i].new_pages = 0;
+#if MACH_ASSERT
+ purgeable_queues[i].debug_count_tokens = 0;
+ purgeable_queues[i].debug_count_objects = 0;
+#endif
+ };
+ purgeable_nonvolatile_count = 0;
+ queue_init(&purgeable_nonvolatile_queue);
+
+ for (i = 0; i < MAX_COLORS; i++ )
+ queue_init(&vm_page_queue_free[i]);
+
+ queue_init(&vm_lopage_queue_free);
queue_init(&vm_page_queue_active);
queue_init(&vm_page_queue_inactive);
- queue_init(&vm_page_queue_zf);
+ queue_init(&vm_page_queue_cleaned);
+ queue_init(&vm_page_queue_throttled);
+ queue_init(&vm_page_queue_anonymous);
+ queue_init(&vm_objects_wired);
+ for ( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ ) {
+ queue_init(&vm_page_queue_speculative[i].age_q);
+
+ vm_page_queue_speculative[i].age_ts.tv_sec = 0;
+ vm_page_queue_speculative[i].age_ts.tv_nsec = 0;
+ }
vm_page_free_wanted = 0;
+ vm_page_free_wanted_privileged = 0;
+
+ vm_page_set_colors();
+
/*
* Steal memory for the map and zone subsystems.
*/
-
- vm_map_steal_memory();
+ kernel_debug_string_simple("zone_steal_memory");
zone_steal_memory();
+ kernel_debug_string_simple("vm_map_steal_memory");
+ vm_map_steal_memory();
/*
* Allocate (and initialize) the virtual-to-physical
* than the number of physical pages in the system.
*/
- simple_lock_init(&vm_page_bucket_lock, 0);
-
if (vm_page_bucket_count == 0) {
unsigned int npages = pmap_free_pages();
while (vm_page_bucket_count < npages)
vm_page_bucket_count <<= 1;
}
+ vm_page_bucket_lock_count = (vm_page_bucket_count + BUCKETS_PER_LOCK - 1) / BUCKETS_PER_LOCK;
vm_page_hash_mask = vm_page_bucket_count - 1;
if (vm_page_hash_mask & vm_page_bucket_count)
printf("vm_page_bootstrap: WARNING -- strange page hash\n");
+#if VM_PAGE_BUCKETS_CHECK
+#if VM_PAGE_FAKE_BUCKETS
+ /*
+ * Allocate a decoy set of page buckets, to detect
+ * any stomping there.
+ */
+ vm_page_fake_buckets = (vm_page_bucket_t *)
+ pmap_steal_memory(vm_page_bucket_count *
+ sizeof(vm_page_bucket_t));
+ vm_page_fake_buckets_start = (vm_map_offset_t) vm_page_fake_buckets;
+ vm_page_fake_buckets_end =
+ vm_map_round_page((vm_page_fake_buckets_start +
+ (vm_page_bucket_count *
+ sizeof (vm_page_bucket_t))),
+ PAGE_MASK);
+ char *cp;
+ for (cp = (char *)vm_page_fake_buckets_start;
+ cp < (char *)vm_page_fake_buckets_end;
+ cp++) {
+ *cp = 0x5a;
+ }
+#endif /* VM_PAGE_FAKE_BUCKETS */
+#endif /* VM_PAGE_BUCKETS_CHECK */
+
+ kernel_debug_string_simple("vm_page_buckets");
vm_page_buckets = (vm_page_bucket_t *)
pmap_steal_memory(vm_page_bucket_count *
sizeof(vm_page_bucket_t));
+ kernel_debug_string_simple("vm_page_bucket_locks");
+ vm_page_bucket_locks = (lck_spin_t *)
+ pmap_steal_memory(vm_page_bucket_lock_count *
+ sizeof(lck_spin_t));
+
for (i = 0; i < vm_page_bucket_count; i++) {
register vm_page_bucket_t *bucket = &vm_page_buckets[i];
- bucket->pages = VM_PAGE_NULL;
+ bucket->page_list = VM_PAGE_PACK_PTR(VM_PAGE_NULL);
#if MACH_PAGE_HASH_STATS
bucket->cur_count = 0;
bucket->hi_count = 0;
#endif /* MACH_PAGE_HASH_STATS */
}
+ for (i = 0; i < vm_page_bucket_lock_count; i++)
+ lck_spin_init(&vm_page_bucket_locks[i], &vm_page_lck_grp_bucket, &vm_page_lck_attr);
+
+ lck_spin_init(&vm_objects_wired_lock, &vm_page_lck_grp_bucket, &vm_page_lck_attr);
+ lck_spin_init(&vm_allocation_sites_lock, &vm_page_lck_grp_bucket, &vm_page_lck_attr);
+ vm_tag_init();
+
+#if VM_PAGE_BUCKETS_CHECK
+ vm_page_buckets_check_ready = TRUE;
+#endif /* VM_PAGE_BUCKETS_CHECK */
+
/*
* Machine-dependent code allocates the resident page table.
* It uses vm_page_init to initialize the page frames.
* to get the alignment right.
*/
+ kernel_debug_string_simple("pmap_startup");
pmap_startup(&virtual_space_start, &virtual_space_end);
virtual_space_start = round_page(virtual_space_start);
virtual_space_end = trunc_page(virtual_space_end);
* wired, they nonetheless can't be moved. At this moment,
* all VM managed pages are "free", courtesy of pmap_startup.
*/
- vm_page_wire_count = atop_64(max_mem) - vm_page_free_count; /* initial value */
+ assert((unsigned int) atop_64(max_mem) == atop_64(max_mem));
+ vm_page_wire_count = ((unsigned int) atop_64(max_mem)) - vm_page_free_count - vm_lopage_free_count; /* initial value */
+ vm_page_wire_count_initial = vm_page_wire_count;
+ vm_page_pages_initial = vm_page_pages;
- printf("vm_page_bootstrap: %d free pages\n", vm_page_free_count);
- vm_page_free_count_minimum = vm_page_free_count;
+ printf("vm_page_bootstrap: %d free pages and %d wired pages\n",
+ vm_page_free_count, vm_page_wire_count);
+ kernel_debug_string_simple("vm_page_bootstrap complete");
simple_lock_init(&vm_paging_lock, 0);
}
addr = virtual_space_start;
virtual_space_start += size;
- kprintf("pmap_steal_memory: %08X - %08X; size=%08X\n", addr, virtual_space_start, size); /* (TEST/DEBUG) */
+ //kprintf("pmap_steal_memory: %08lX - %08lX; size=%08lX\n", (long)addr, (long)virtual_space_start, (long)size); /* (TEST/DEBUG) */
/*
* Allocate and map physical pages to back new virtual pages.
for (vaddr = round_page(addr);
vaddr < addr + size;
vaddr += PAGE_SIZE) {
- if (!pmap_next_page(&phys_page))
+
+ if (!pmap_next_page_hi(&phys_page))
panic("pmap_steal_memory");
/*
* XXX Logically, these mappings should be wired,
* but some pmap modules barf if they are.
*/
+#if defined(__LP64__)
+ pmap_pre_expand(kernel_pmap, vaddr);
+#endif
pmap_enter(kernel_pmap, vaddr, phys_page,
- VM_PROT_READ|VM_PROT_WRITE,
+ VM_PROT_READ|VM_PROT_WRITE, VM_PROT_NONE,
VM_WIMG_USE_DEFAULT, FALSE);
/*
* Account for newly stolen memory
*/
vm_page_wire_count++;
-
+ vm_page_stolen_count++;
}
return (void *) addr;
}
+void vm_page_release_startup(vm_page_t mem);
void
pmap_startup(
vm_offset_t *startp,
vm_offset_t *endp)
{
unsigned int i, npages, pages_initialized, fill, fillval;
- vm_page_t pages;
ppnum_t phys_page;
addr64_t tmpaddr;
+
+#if defined(__LP64__)
+ /*
+ * struct vm_page must be of size 64 due to VM_PAGE_PACK_PTR use
+ */
+ assert(sizeof(struct vm_page) == 64);
+
+ /*
+ * make sure we are aligned on a 64 byte boundary
+ * for VM_PAGE_PACK_PTR (it clips off the low-order
+ * 6 bits of the pointer)
+ */
+ if (virtual_space_start != virtual_space_end)
+ virtual_space_start = round_page(virtual_space_start);
+#endif
+
/*
* We calculate how many page frames we will have
* and then allocate the page structures in one chunk.
*/
tmpaddr = (addr64_t)pmap_free_pages() * (addr64_t)PAGE_SIZE; /* Get the amount of memory left */
- tmpaddr = tmpaddr + (addr64_t)(round_page_32(virtual_space_start) - virtual_space_start); /* Account for any slop */
- npages = (unsigned int)(tmpaddr / (addr64_t)(PAGE_SIZE + sizeof(*pages))); /* Figure size of all vm_page_ts, including enough to hold the vm_page_ts */
+ tmpaddr = tmpaddr + (addr64_t)(round_page(virtual_space_start) - virtual_space_start); /* Account for any slop */
+ npages = (unsigned int)(tmpaddr / (addr64_t)(PAGE_SIZE + sizeof(*vm_pages))); /* Figure size of all vm_page_ts, including enough to hold the vm_page_ts */
- pages = (vm_page_t) pmap_steal_memory(npages * sizeof *pages);
+ vm_pages = (vm_page_t) pmap_steal_memory(npages * sizeof *vm_pages);
/*
* Initialize the page frames.
*/
-
+ kernel_debug_string_simple("Initialize the page frames");
for (i = 0, pages_initialized = 0; i < npages; i++) {
if (!pmap_next_page(&phys_page))
break;
+ if (pages_initialized == 0 || phys_page < vm_page_lowest)
+ vm_page_lowest = phys_page;
- vm_page_init(&pages[i], phys_page);
+ vm_page_init(&vm_pages[i], phys_page, FALSE);
vm_page_pages++;
pages_initialized++;
}
+ vm_pages_count = pages_initialized;
+
+#if defined(__LP64__)
+
+ if (VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[0])) != &vm_pages[0])
+ panic("VM_PAGE_PACK_PTR failed on &vm_pages[0] - %p", (void *)&vm_pages[0]);
+
+ if (VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[vm_pages_count-1])) != &vm_pages[vm_pages_count-1])
+ panic("VM_PAGE_PACK_PTR failed on &vm_pages[vm_pages_count-1] - %p", (void *)&vm_pages[vm_pages_count-1]);
+#endif
+ kernel_debug_string_simple("page fill/release");
+ /*
+ * Check if we want to initialize pages to a known value
+ */
+ fill = 0; /* Assume no fill */
+ if (PE_parse_boot_argn("fill", &fillval, sizeof (fillval))) fill = 1; /* Set fill */
+#if DEBUG
+ /* This slows down booting the DEBUG kernel, particularly on
+ * large memory systems, but is worthwhile in deterministically
+ * trapping uninitialized memory usage.
+ */
+ if (fill == 0) {
+ fill = 1;
+ fillval = 0xDEB8F177;
+ }
+#endif
+ if (fill)
+ kprintf("Filling vm_pages with pattern: 0x%x\n", fillval);
+ // -debug code remove
+ if (2 == vm_himemory_mode) {
+ // free low -> high so high is preferred
+ for (i = 1; i <= pages_initialized; i++) {
+ if(fill) fillPage(vm_pages[i - 1].phys_page, fillval); /* Fill the page with a know value if requested at boot */
+ vm_page_release_startup(&vm_pages[i - 1]);
+ }
+ }
+ else
+ // debug code remove-
/*
* Release pages in reverse order so that physical pages
* the devices (which must address physical memory) happy if
* they require several consecutive pages.
*/
-
-/*
- * Check if we want to initialize pages to a known value
- */
-
- fill = 0; /* Assume no fill */
- if (PE_parse_boot_arg("fill", &fillval)) fill = 1; /* Set fill */
-
for (i = pages_initialized; i > 0; i--) {
- if(fill) fillPage(pages[i - 1].phys_page, fillval); /* Fill the page with a know value if requested at boot */
- vm_page_release(&pages[i - 1]);
+ if(fill) fillPage(vm_pages[i - 1].phys_page, fillval); /* Fill the page with a know value if requested at boot */
+ vm_page_release_startup(&vm_pages[i - 1]);
}
+ VM_CHECK_MEMORYSTATUS;
+
#if 0
{
vm_page_t xx, xxo, xxl;
- int j, k, l;
+ int i, j, k, l;
j = 0; /* (BRINGUP) */
xxl = 0;
- for(xx = vm_page_queue_free; xx; xxl = xx, xx = xx->pageq.next) { /* (BRINGUP) */
- j++; /* (BRINGUP) */
- if(j > vm_page_free_count) { /* (BRINGUP) */
- panic("pmap_startup: too many pages, xx = %08X, xxl = %08X\n", xx, xxl);
- }
-
- l = vm_page_free_count - j; /* (BRINGUP) */
- k = 0; /* (BRINGUP) */
-
- if(((j - 1) & 0xFFFF) == 0) kprintf("checking number %d of %d\n", j, vm_page_free_count);
-
- for(xxo = xx->pageq.next; xxo; xxo = xxo->pageq.next) { /* (BRINGUP) */
- k++;
- if(k > l) panic("pmap_startup: too many in secondary check %d %d\n", k, l);
- if((xx->phys_page & 0xFFFFFFFF) == (xxo->phys_page & 0xFFFFFFFF)) { /* (BRINGUP) */
- panic("pmap_startup: duplicate physaddr, xx = %08X, xxo = %08X\n", xx, xxo);
+ for( i = 0; i < vm_colors; i++ ) {
+ queue_iterate(&vm_page_queue_free[i],
+ xx,
+ vm_page_t,
+ pageq) { /* BRINGUP */
+ j++; /* (BRINGUP) */
+ if(j > vm_page_free_count) { /* (BRINGUP) */
+ panic("pmap_startup: too many pages, xx = %08X, xxl = %08X\n", xx, xxl);
+ }
+
+ l = vm_page_free_count - j; /* (BRINGUP) */
+ k = 0; /* (BRINGUP) */
+
+ if(((j - 1) & 0xFFFF) == 0) kprintf("checking number %d of %d\n", j, vm_page_free_count);
+
+ for(xxo = xx->pageq.next; xxo != &vm_page_queue_free[i]; xxo = xxo->pageq.next) { /* (BRINGUP) */
+ k++;
+ if(k > l) panic("pmap_startup: too many in secondary check %d %d\n", k, l);
+ if((xx->phys_page & 0xFFFFFFFF) == (xxo->phys_page & 0xFFFFFFFF)) { /* (BRINGUP) */
+ panic("pmap_startup: duplicate physaddr, xx = %08X, xxo = %08X\n", xx, xxo);
+ }
}
+
+ xxl = xx;
}
}
* because pmap_steal_memory has been using it.
*/
- virtual_space_start = round_page_32(virtual_space_start);
+ virtual_space_start = round_page(virtual_space_start);
*startp = virtual_space_start;
*endp = virtual_space_end;
void
vm_page_module_init(void)
{
+ uint64_t vm_page_zone_pages, vm_page_zone_data_size;
vm_page_zone = zinit((vm_size_t) sizeof(struct vm_page),
0, PAGE_SIZE, "vm pages");
zone_debug_disable(vm_page_zone);
#endif /* ZONE_DEBUG */
+ zone_change(vm_page_zone, Z_CALLERACCT, FALSE);
zone_change(vm_page_zone, Z_EXPAND, FALSE);
zone_change(vm_page_zone, Z_EXHAUST, TRUE);
zone_change(vm_page_zone, Z_FOREIGN, TRUE);
-
- /*
- * Adjust zone statistics to account for the real pages allocated
- * in vm_page_create(). [Q: is this really what we want?]
- */
- vm_page_zone->count += vm_page_pages;
- vm_page_zone->cur_size += vm_page_pages * vm_page_zone->elem_size;
-
- mutex_init(&vm_page_alloc_lock, 0);
+ zone_change(vm_page_zone, Z_GZALLOC_EXEMPT, TRUE);
+ /*
+ * Adjust zone statistics to account for the real pages allocated
+ * in vm_page_create(). [Q: is this really what we want?]
+ */
+ vm_page_zone->count += vm_page_pages;
+ vm_page_zone->sum_count += vm_page_pages;
+ vm_page_zone_data_size = vm_page_pages * vm_page_zone->elem_size;
+ vm_page_zone->cur_size += vm_page_zone_data_size;
+ vm_page_zone_pages = ((round_page(vm_page_zone_data_size)) / PAGE_SIZE);
+ OSAddAtomic64(vm_page_zone_pages, &(vm_page_zone->page_count));
+ /* since zone accounts for these, take them out of stolen */
+ VM_PAGE_MOVE_STOLEN(vm_page_zone_pages);
}
/*
for (phys_page = start;
phys_page < end;
phys_page++) {
- while ((m = (vm_page_t) vm_page_grab_fictitious())
+ while ((m = (vm_page_t) vm_page_grab_fictitious_common(phys_page))
== VM_PAGE_NULL)
vm_page_more_fictitious();
- vm_page_init(m, phys_page);
+ m->fictitious = FALSE;
+ pmap_clear_noencrypt(phys_page);
+
vm_page_pages++;
vm_page_release(m);
}
* NOTE: The bucket count must be a power of 2
*/
#define vm_page_hash(object, offset) (\
- ( (natural_t)((uint32_t)object * vm_page_bucket_hash) + ((uint32_t)atop_64(offset) ^ vm_page_bucket_hash))\
+ ( (natural_t)((uintptr_t)object * vm_page_bucket_hash) + ((uint32_t)atop_64(offset) ^ vm_page_bucket_hash))\
& vm_page_hash_mask)
+
/*
* vm_page_insert: [ internal use only ]
*
*
* The object must be locked.
*/
-
void
vm_page_insert(
- register vm_page_t mem,
- register vm_object_t object,
- register vm_object_offset_t offset)
+ vm_page_t mem,
+ vm_object_t object,
+ vm_object_offset_t offset)
+{
+ vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, FALSE, FALSE, NULL);
+}
+
+void
+vm_page_insert_wired(
+ vm_page_t mem,
+ vm_object_t object,
+ vm_object_offset_t offset,
+ vm_tag_t tag)
+{
+ vm_page_insert_internal(mem, object, offset, tag, FALSE, TRUE, FALSE, FALSE, NULL);
+}
+
+void
+vm_page_insert_internal(
+ vm_page_t mem,
+ vm_object_t object,
+ vm_object_offset_t offset,
+ vm_tag_t tag,
+ boolean_t queues_lock_held,
+ boolean_t insert_in_hash,
+ boolean_t batch_pmap_op,
+ boolean_t batch_accounting,
+ uint64_t *delayed_ledger_update)
{
- register vm_page_bucket_t *bucket;
+ vm_page_bucket_t *bucket;
+ lck_spin_t *bucket_lock;
+ int hash_id;
+ task_t owner;
XPR(XPR_VM_PAGE,
"vm_page_insert, object 0x%X offset 0x%X page 0x%X\n",
- (integer_t)object, (integer_t)offset, (integer_t)mem, 0,0);
-
+ object, offset, mem, 0,0);
+#if 0
+ /*
+ * we may not hold the page queue lock
+ * so this check isn't safe to make
+ */
VM_PAGE_CHECK(mem);
-#if DEBUG
- _mutex_assert(&object->Lock, MA_OWNED);
+#endif
- if (mem->tabled || mem->object != VM_OBJECT_NULL)
- panic("vm_page_insert: page %p for (obj=%p,off=0x%llx) "
- "already in (obj=%p,off=0x%llx)",
- mem, object, offset, mem->object, mem->offset);
+ assert(page_aligned(offset));
+
+ assert(!VM_PAGE_WIRED(mem) || mem->private || mem->fictitious || (tag != VM_KERN_MEMORY_NONE));
+
+ /* the vm_submap_object is only a placeholder for submaps */
+ assert(object != vm_submap_object);
+
+ vm_object_lock_assert_exclusive(object);
+#if DEBUG
+ lck_mtx_assert(&vm_page_queue_lock,
+ queues_lock_held ? LCK_MTX_ASSERT_OWNED
+ : LCK_MTX_ASSERT_NOTOWNED);
+#endif /* DEBUG */
+
+ if (insert_in_hash == TRUE) {
+#if DEBUG || VM_PAGE_CHECK_BUCKETS
+ if (mem->tabled || mem->object != VM_OBJECT_NULL)
+ panic("vm_page_insert: page %p for (obj=%p,off=0x%llx) "
+ "already in (obj=%p,off=0x%llx)",
+ mem, object, offset, mem->object, mem->offset);
#endif
- assert(!object->internal || offset < object->size);
+ assert(!object->internal || offset < object->vo_size);
- /* only insert "pageout" pages into "pageout" objects,
- * and normal pages into normal objects */
- assert(object->pageout == mem->pageout);
+ /* only insert "pageout" pages into "pageout" objects,
+ * and normal pages into normal objects */
+#if 00
+ /*
+ * For some reason, this assertion gets tripped
+ * but it's mostly harmless, so let's disable it
+ * for now.
+ */
+ assert(object->pageout == mem->pageout);
+#endif /* 00 */
- assert(vm_page_lookup(object, offset) == VM_PAGE_NULL);
+ assert(vm_page_lookup(object, offset) == VM_PAGE_NULL);
+
+ /*
+ * Record the object/offset pair in this page
+ */
- /*
- * Record the object/offset pair in this page
- */
+ mem->object = object;
+ mem->offset = offset;
- mem->object = object;
- mem->offset = offset;
+ /*
+ * Insert it into the object_object/offset hash table
+ */
+ hash_id = vm_page_hash(object, offset);
+ bucket = &vm_page_buckets[hash_id];
+ bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
+
+ lck_spin_lock(bucket_lock);
- /*
- * Insert it into the object_object/offset hash table
- */
+ mem->next_m = bucket->page_list;
+ bucket->page_list = VM_PAGE_PACK_PTR(mem);
+ assert(mem == VM_PAGE_UNPACK_PTR(bucket->page_list));
- bucket = &vm_page_buckets[vm_page_hash(object, offset)];
- simple_lock(&vm_page_bucket_lock);
- mem->next = bucket->pages;
- bucket->pages = mem;
#if MACH_PAGE_HASH_STATS
- if (++bucket->cur_count > bucket->hi_count)
- bucket->hi_count = bucket->cur_count;
+ if (++bucket->cur_count > bucket->hi_count)
+ bucket->hi_count = bucket->cur_count;
#endif /* MACH_PAGE_HASH_STATS */
- simple_unlock(&vm_page_bucket_lock);
+ mem->hashed = TRUE;
+ lck_spin_unlock(bucket_lock);
+ }
+ {
+ unsigned int cache_attr;
+
+ cache_attr = object->wimg_bits & VM_WIMG_MASK;
+
+ if (cache_attr != VM_WIMG_USE_DEFAULT) {
+ PMAP_SET_CACHE_ATTR(mem, object, cache_attr, batch_pmap_op);
+ }
+ }
/*
* Now link into the object's list of backed pages.
*/
-
- VM_PAGE_INSERT(mem, object);
+ queue_enter(&object->memq, mem, vm_page_t, listq);
+ object->memq_hint = mem;
mem->tabled = TRUE;
/*
*/
object->resident_page_count++;
+ if (VM_PAGE_WIRED(mem)) {
+ if (!mem->private && !mem->fictitious)
+ {
+ if (!object->wired_page_count)
+ {
+ assert(VM_KERN_MEMORY_NONE != tag);
+ object->wire_tag = tag;
+ VM_OBJECT_WIRED(object);
+ }
+ }
+ object->wired_page_count++;
+ }
+ assert(object->resident_page_count >= object->wired_page_count);
- if (object->purgable == VM_OBJECT_PURGABLE_VOLATILE ||
- object->purgable == VM_OBJECT_PURGABLE_EMPTY) {
- vm_page_lock_queues();
- vm_page_purgeable_count++;
- vm_page_unlock_queues();
+ if (batch_accounting == FALSE) {
+ if (object->internal) {
+ OSAddAtomic(1, &vm_page_internal_count);
+ } else {
+ OSAddAtomic(1, &vm_page_external_count);
+ }
+ }
+
+ /*
+ * It wouldn't make sense to insert a "reusable" page in
+ * an object (the page would have been marked "reusable" only
+ * at the time of a madvise(MADV_FREE_REUSABLE) if it was already
+ * in the object at that time).
+ * But a page could be inserted in a "all_reusable" object, if
+ * something faults it in (a vm_read() from another task or a
+ * "use-after-free" issue in user space, for example). It can
+ * also happen if we're relocating a page from that object to
+ * a different physical page during a physically-contiguous
+ * allocation.
+ */
+ assert(!mem->reusable);
+ if (mem->object->all_reusable) {
+ OSAddAtomic(+1, &vm_page_stats_reusable.reusable_count);
+ }
+
+ if (object->purgable == VM_PURGABLE_DENY) {
+ owner = TASK_NULL;
+ } else {
+ owner = object->vo_purgeable_owner;
+ }
+ if (owner &&
+ (object->purgable == VM_PURGABLE_NONVOLATILE ||
+ VM_PAGE_WIRED(mem))) {
+
+ if (delayed_ledger_update)
+ *delayed_ledger_update += PAGE_SIZE;
+ else {
+ /* more non-volatile bytes */
+ ledger_credit(owner->ledger,
+ task_ledgers.purgeable_nonvolatile,
+ PAGE_SIZE);
+ /* more footprint */
+ ledger_credit(owner->ledger,
+ task_ledgers.phys_footprint,
+ PAGE_SIZE);
+ }
+
+ } else if (owner &&
+ (object->purgable == VM_PURGABLE_VOLATILE ||
+ object->purgable == VM_PURGABLE_EMPTY)) {
+ assert(! VM_PAGE_WIRED(mem));
+ /* more volatile bytes */
+ ledger_credit(owner->ledger,
+ task_ledgers.purgeable_volatile,
+ PAGE_SIZE);
+ }
+
+ if (object->purgable == VM_PURGABLE_VOLATILE) {
+ if (VM_PAGE_WIRED(mem)) {
+ OSAddAtomic(+1, &vm_page_purgeable_wired_count);
+ } else {
+ OSAddAtomic(+1, &vm_page_purgeable_count);
+ }
+ } else if (object->purgable == VM_PURGABLE_EMPTY &&
+ mem->throttled) {
+ /*
+ * This page belongs to a purged VM object but hasn't
+ * been purged (because it was "busy").
+ * It's in the "throttled" queue and hence not
+ * visible to vm_pageout_scan(). Move it to a pageable
+ * queue, so that it can eventually be reclaimed, instead
+ * of lingering in the "empty" object.
+ */
+ if (queues_lock_held == FALSE)
+ vm_page_lockspin_queues();
+ vm_page_deactivate(mem);
+ if (queues_lock_held == FALSE)
+ vm_page_unlock_queues();
}
+
+#if VM_OBJECT_TRACKING_OP_MODIFIED
+ if (vm_object_tracking_inited &&
+ object->internal &&
+ object->resident_page_count == 0 &&
+ object->pager == NULL &&
+ object->shadow != NULL &&
+ object->shadow->copy == object) {
+ 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_MODIFIED,
+ bt,
+ numsaved);
+ }
+#endif /* VM_OBJECT_TRACKING_OP_MODIFIED */
}
/*
* Exactly like vm_page_insert, except that we first
* remove any existing page at the given offset in object.
*
- * The object and page queues must be locked.
+ * The object must be locked.
*/
-
void
vm_page_replace(
register vm_page_t mem,
register vm_object_t object,
register vm_object_offset_t offset)
{
- register vm_page_bucket_t *bucket;
+ vm_page_bucket_t *bucket;
+ vm_page_t found_m = VM_PAGE_NULL;
+ lck_spin_t *bucket_lock;
+ int hash_id;
+#if 0
+ /*
+ * we don't hold the page queue lock
+ * so this check isn't safe to make
+ */
VM_PAGE_CHECK(mem);
-#if DEBUG
- _mutex_assert(&object->Lock, MA_OWNED);
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
-
+#endif
+ vm_object_lock_assert_exclusive(object);
+#if DEBUG || VM_PAGE_CHECK_BUCKETS
if (mem->tabled || mem->object != VM_OBJECT_NULL)
panic("vm_page_replace: page %p for (obj=%p,off=0x%llx) "
"already in (obj=%p,off=0x%llx)",
mem, object, offset, mem->object, mem->offset);
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
#endif
/*
* Record the object/offset pair in this page
* replacing any page that might have been there.
*/
- bucket = &vm_page_buckets[vm_page_hash(object, offset)];
- simple_lock(&vm_page_bucket_lock);
- if (bucket->pages) {
- vm_page_t *mp = &bucket->pages;
- register vm_page_t m = *mp;
+ hash_id = vm_page_hash(object, offset);
+ bucket = &vm_page_buckets[hash_id];
+ bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
+
+ lck_spin_lock(bucket_lock);
+
+ if (bucket->page_list) {
+ vm_page_packed_t *mp = &bucket->page_list;
+ vm_page_t m = VM_PAGE_UNPACK_PTR(*mp);
+
do {
if (m->object == object && m->offset == offset) {
/*
- * Remove page from bucket and from object,
- * and return it to the free list.
- */
- *mp = m->next;
- VM_PAGE_REMOVE(m);
- m->tabled = FALSE;
- m->object = VM_OBJECT_NULL;
- m->offset = (vm_object_offset_t) -1;
- object->resident_page_count--;
-
- if (object->purgable == VM_OBJECT_PURGABLE_VOLATILE ||
- object->purgable == VM_OBJECT_PURGABLE_EMPTY) {
- assert(vm_page_purgeable_count > 0);
- vm_page_purgeable_count--;
- }
-
- /*
- * Return page to the free list.
- * Note the page is not tabled now, so this
- * won't self-deadlock on the bucket lock.
+ * Remove old page from hash list
*/
+ *mp = m->next_m;
+ m->hashed = FALSE;
- vm_page_free(m);
+ found_m = m;
break;
}
- mp = &m->next;
- } while ((m = *mp));
- mem->next = bucket->pages;
+ mp = &m->next_m;
+ } while ((m = VM_PAGE_UNPACK_PTR(*mp)));
+
+ mem->next_m = bucket->page_list;
} else {
- mem->next = VM_PAGE_NULL;
+ mem->next_m = VM_PAGE_PACK_PTR(VM_PAGE_NULL);
}
- bucket->pages = mem;
- simple_unlock(&vm_page_bucket_lock);
-
/*
- * Now link into the object's list of backed pages.
+ * insert new page at head of hash list
*/
+ bucket->page_list = VM_PAGE_PACK_PTR(mem);
+ mem->hashed = TRUE;
- VM_PAGE_INSERT(mem, object);
- mem->tabled = TRUE;
-
- /*
- * And show that the object has one more resident
- * page.
- */
+ lck_spin_unlock(bucket_lock);
- object->resident_page_count++;
-
- if (object->purgable == VM_OBJECT_PURGABLE_VOLATILE ||
- object->purgable == VM_OBJECT_PURGABLE_EMPTY) {
- vm_page_purgeable_count++;
+ if (found_m) {
+ /*
+ * there was already a page at the specified
+ * offset for this object... remove it from
+ * the object and free it back to the free list
+ */
+ vm_page_free_unlocked(found_m, FALSE);
}
+ vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, FALSE, FALSE, FALSE, NULL);
}
/*
* Removes the given mem entry from the object/offset-page
* table and the object page list.
*
- * The object and page queues must be locked.
+ * The object must be locked.
*/
void
vm_page_remove(
- register vm_page_t mem)
+ vm_page_t mem,
+ boolean_t remove_from_hash)
{
- register vm_page_bucket_t *bucket;
- register vm_page_t this;
+ vm_page_bucket_t *bucket;
+ vm_page_t this;
+ lck_spin_t *bucket_lock;
+ int hash_id;
+ task_t owner;
XPR(XPR_VM_PAGE,
"vm_page_remove, object 0x%X offset 0x%X page 0x%X\n",
- (integer_t)mem->object, (integer_t)mem->offset,
- (integer_t)mem, 0,0);
-#if DEBUG
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
- _mutex_assert(&mem->object->Lock, MA_OWNED);
-#endif
+ mem->object, mem->offset,
+ mem, 0,0);
+
+ vm_object_lock_assert_exclusive(mem->object);
assert(mem->tabled);
assert(!mem->cleaning);
- VM_PAGE_CHECK(mem);
-
-
+ assert(!mem->laundry);
+#if 0
/*
- * Remove from the object_object/offset hash table
+ * we don't hold the page queue lock
+ * so this check isn't safe to make
*/
+ VM_PAGE_CHECK(mem);
+#endif
+ if (remove_from_hash == TRUE) {
+ /*
+ * Remove from the object_object/offset hash table
+ */
+ hash_id = vm_page_hash(mem->object, mem->offset);
+ bucket = &vm_page_buckets[hash_id];
+ bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
- bucket = &vm_page_buckets[vm_page_hash(mem->object, mem->offset)];
- simple_lock(&vm_page_bucket_lock);
- if ((this = bucket->pages) == mem) {
- /* optimize for common case */
+ lck_spin_lock(bucket_lock);
- bucket->pages = mem->next;
- } else {
- register vm_page_t *prev;
+ if ((this = VM_PAGE_UNPACK_PTR(bucket->page_list)) == mem) {
+ /* optimize for common case */
- for (prev = &this->next;
- (this = *prev) != mem;
- prev = &this->next)
- continue;
- *prev = this->next;
- }
+ bucket->page_list = mem->next_m;
+ } else {
+ vm_page_packed_t *prev;
+
+ for (prev = &this->next_m;
+ (this = VM_PAGE_UNPACK_PTR(*prev)) != mem;
+ prev = &this->next_m)
+ continue;
+ *prev = this->next_m;
+ }
#if MACH_PAGE_HASH_STATS
- bucket->cur_count--;
+ bucket->cur_count--;
#endif /* MACH_PAGE_HASH_STATS */
- simple_unlock(&vm_page_bucket_lock);
-
+ mem->hashed = FALSE;
+ lck_spin_unlock(bucket_lock);
+ }
/*
* Now remove from the object's list of backed pages.
*/
- VM_PAGE_REMOVE(mem);
+ vm_page_remove_internal(mem);
/*
* And show that the object has one fewer resident
* page.
*/
+ assert(mem->object->resident_page_count > 0);
mem->object->resident_page_count--;
- if (mem->object->purgable == VM_OBJECT_PURGABLE_VOLATILE ||
- mem->object->purgable == VM_OBJECT_PURGABLE_EMPTY) {
- assert(vm_page_purgeable_count > 0);
- vm_page_purgeable_count--;
+ if (mem->object->internal) {
+#if DEBUG
+ assert(vm_page_internal_count);
+#endif /* DEBUG */
+
+ OSAddAtomic(-1, &vm_page_internal_count);
+ } else {
+ assert(vm_page_external_count);
+ OSAddAtomic(-1, &vm_page_external_count);
+
+ if (mem->xpmapped) {
+ assert(vm_page_xpmapped_external_count);
+ OSAddAtomic(-1, &vm_page_xpmapped_external_count);
+ }
+ }
+ if (!mem->object->internal && (mem->object->objq.next || mem->object->objq.prev)) {
+ if (mem->object->resident_page_count == 0)
+ vm_object_cache_remove(mem->object);
+ }
+
+ if (VM_PAGE_WIRED(mem)) {
+ assert(mem->object->wired_page_count > 0);
+ mem->object->wired_page_count--;
+ if (!mem->object->wired_page_count) {
+ VM_OBJECT_UNWIRED(mem->object);
+ }
+ }
+ assert(mem->object->resident_page_count >=
+ mem->object->wired_page_count);
+ if (mem->reusable) {
+ assert(mem->object->reusable_page_count > 0);
+ mem->object->reusable_page_count--;
+ assert(mem->object->reusable_page_count <=
+ mem->object->resident_page_count);
+ mem->reusable = FALSE;
+ OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
+ vm_page_stats_reusable.reused_remove++;
+ } else if (mem->object->all_reusable) {
+ OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
+ vm_page_stats_reusable.reused_remove++;
+ }
+
+ if (mem->object->purgable == VM_PURGABLE_DENY) {
+ owner = TASK_NULL;
+ } else {
+ owner = mem->object->vo_purgeable_owner;
+ }
+ if (owner &&
+ (mem->object->purgable == VM_PURGABLE_NONVOLATILE ||
+ VM_PAGE_WIRED(mem))) {
+ /* less non-volatile bytes */
+ ledger_debit(owner->ledger,
+ task_ledgers.purgeable_nonvolatile,
+ PAGE_SIZE);
+ /* less footprint */
+ ledger_debit(owner->ledger,
+ task_ledgers.phys_footprint,
+ PAGE_SIZE);
+ } else if (owner &&
+ (mem->object->purgable == VM_PURGABLE_VOLATILE ||
+ mem->object->purgable == VM_PURGABLE_EMPTY)) {
+ assert(! VM_PAGE_WIRED(mem));
+ /* less volatile bytes */
+ ledger_debit(owner->ledger,
+ task_ledgers.purgeable_volatile,
+ PAGE_SIZE);
}
+ if (mem->object->purgable == VM_PURGABLE_VOLATILE) {
+ if (VM_PAGE_WIRED(mem)) {
+ assert(vm_page_purgeable_wired_count > 0);
+ OSAddAtomic(-1, &vm_page_purgeable_wired_count);
+ } else {
+ assert(vm_page_purgeable_count > 0);
+ OSAddAtomic(-1, &vm_page_purgeable_count);
+ }
+ }
+ if (mem->object->set_cache_attr == TRUE)
+ pmap_set_cache_attributes(mem->phys_page, 0);
mem->tabled = FALSE;
mem->object = VM_OBJECT_NULL;
mem->offset = (vm_object_offset_t) -1;
}
+
/*
* vm_page_lookup:
*
* The object must be locked. No side effects.
*/
-unsigned long vm_page_lookup_hint = 0;
-unsigned long vm_page_lookup_hint_next = 0;
-unsigned long vm_page_lookup_hint_prev = 0;
-unsigned long vm_page_lookup_hint_miss = 0;
+#define VM_PAGE_HASH_LOOKUP_THRESHOLD 10
+
+#if DEBUG_VM_PAGE_LOOKUP
+
+struct {
+ uint64_t vpl_total;
+ uint64_t vpl_empty_obj;
+ uint64_t vpl_bucket_NULL;
+ uint64_t vpl_hit_hint;
+ uint64_t vpl_hit_hint_next;
+ uint64_t vpl_hit_hint_prev;
+ uint64_t vpl_fast;
+ uint64_t vpl_slow;
+ uint64_t vpl_hit;
+ uint64_t vpl_miss;
+
+ uint64_t vpl_fast_elapsed;
+ uint64_t vpl_slow_elapsed;
+} vm_page_lookup_stats __attribute__((aligned(8)));
+
+#endif
+
+#define KDP_VM_PAGE_WALK_MAX 1000
+
+vm_page_t
+kdp_vm_page_lookup(
+ vm_object_t object,
+ vm_object_offset_t offset)
+{
+ vm_page_t cur_page;
+ int num_traversed = 0;
+
+ if (not_in_kdp) {
+ panic("panic: kdp_vm_page_lookup done outside of kernel debugger");
+ }
+
+ queue_iterate(&object->memq, cur_page, vm_page_t, listq) {
+ if (cur_page->offset == offset) {
+ return cur_page;
+ }
+ num_traversed++;
+
+ if (num_traversed >= KDP_VM_PAGE_WALK_MAX) {
+ return VM_PAGE_NULL;
+ }
+ }
+
+ return VM_PAGE_NULL;
+}
vm_page_t
vm_page_lookup(
- register vm_object_t object,
- register vm_object_offset_t offset)
+ vm_object_t object,
+ vm_object_offset_t offset)
{
- register vm_page_t mem;
- register vm_page_bucket_t *bucket;
- queue_entry_t qe;
-#if 0
- _mutex_assert(&object->Lock, MA_OWNED);
+ vm_page_t mem;
+ vm_page_bucket_t *bucket;
+ queue_entry_t qe;
+ lck_spin_t *bucket_lock = NULL;
+ int hash_id;
+#if DEBUG_VM_PAGE_LOOKUP
+ uint64_t start, elapsed;
+
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_total);
+#endif
+ vm_object_lock_assert_held(object);
+
+ if (object->resident_page_count == 0) {
+#if DEBUG_VM_PAGE_LOOKUP
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_empty_obj);
#endif
+ return (VM_PAGE_NULL);
+ }
mem = object->memq_hint;
+
if (mem != VM_PAGE_NULL) {
assert(mem->object == object);
+
if (mem->offset == offset) {
- vm_page_lookup_hint++;
- return mem;
+#if DEBUG_VM_PAGE_LOOKUP
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint);
+#endif
+ return (mem);
}
qe = queue_next(&mem->listq);
+
if (! queue_end(&object->memq, qe)) {
vm_page_t next_page;
next_page = (vm_page_t) qe;
assert(next_page->object == object);
+
if (next_page->offset == offset) {
- vm_page_lookup_hint_next++;
object->memq_hint = next_page; /* new hint */
- return next_page;
+#if DEBUG_VM_PAGE_LOOKUP
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_next);
+#endif
+ return (next_page);
}
}
qe = queue_prev(&mem->listq);
+
if (! queue_end(&object->memq, qe)) {
vm_page_t prev_page;
prev_page = (vm_page_t) qe;
assert(prev_page->object == object);
+
if (prev_page->offset == offset) {
- vm_page_lookup_hint_prev++;
object->memq_hint = prev_page; /* new hint */
- return prev_page;
+#if DEBUG_VM_PAGE_LOOKUP
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_prev);
+#endif
+ return (prev_page);
}
}
}
+ /*
+ * Search the hash table for this object/offset pair
+ */
+ hash_id = vm_page_hash(object, offset);
+ bucket = &vm_page_buckets[hash_id];
/*
- * Search the hash table for this object/offset pair
+ * since we hold the object lock, we are guaranteed that no
+ * new pages can be inserted into this object... this in turn
+ * guarantess that the page we're looking for can't exist
+ * if the bucket it hashes to is currently NULL even when looked
+ * at outside the scope of the hash bucket lock... this is a
+ * really cheap optimiztion to avoid taking the lock
*/
+ if (!bucket->page_list) {
+#if DEBUG_VM_PAGE_LOOKUP
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_bucket_NULL);
+#endif
+ return (VM_PAGE_NULL);
+ }
- bucket = &vm_page_buckets[vm_page_hash(object, offset)];
+#if DEBUG_VM_PAGE_LOOKUP
+ start = mach_absolute_time();
+#endif
+ if (object->resident_page_count <= VM_PAGE_HASH_LOOKUP_THRESHOLD) {
+ /*
+ * on average, it's roughly 3 times faster to run a short memq list
+ * than to take the spin lock and go through the hash list
+ */
+ mem = (vm_page_t)queue_first(&object->memq);
- simple_lock(&vm_page_bucket_lock);
- for (mem = bucket->pages; mem != VM_PAGE_NULL; mem = mem->next) {
- VM_PAGE_CHECK(mem);
- if ((mem->object == object) && (mem->offset == offset))
- break;
- }
- simple_unlock(&vm_page_bucket_lock);
+ while (!queue_end(&object->memq, (queue_entry_t)mem)) {
- if (mem != VM_PAGE_NULL) {
- if (object->memq_hint != VM_PAGE_NULL) {
- vm_page_lookup_hint_miss++;
- }
- assert(mem->object == object);
- object->memq_hint = mem;
- }
+ if (mem->offset == offset)
+ break;
- return(mem);
-}
+ mem = (vm_page_t)queue_next(&mem->listq);
+ }
+ if (queue_end(&object->memq, (queue_entry_t)mem))
+ mem = NULL;
+ } else {
+ bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
-vm_page_t
-vm_page_lookup_nohint(
- vm_object_t object,
- vm_object_offset_t offset)
-{
- register vm_page_t mem;
- register vm_page_bucket_t *bucket;
+ lck_spin_lock(bucket_lock);
+ for (mem = VM_PAGE_UNPACK_PTR(bucket->page_list); mem != VM_PAGE_NULL; mem = VM_PAGE_UNPACK_PTR(mem->next_m)) {
#if 0
- _mutex_assert(&object->Lock, MA_OWNED);
+ /*
+ * we don't hold the page queue lock
+ * so this check isn't safe to make
+ */
+ VM_PAGE_CHECK(mem);
#endif
- /*
- * Search the hash table for this object/offset pair
- */
+ if ((mem->object == object) && (mem->offset == offset))
+ break;
+ }
+ lck_spin_unlock(bucket_lock);
+ }
- bucket = &vm_page_buckets[vm_page_hash(object, offset)];
+#if DEBUG_VM_PAGE_LOOKUP
+ elapsed = mach_absolute_time() - start;
- simple_lock(&vm_page_bucket_lock);
- for (mem = bucket->pages; mem != VM_PAGE_NULL; mem = mem->next) {
- VM_PAGE_CHECK(mem);
- if ((mem->object == object) && (mem->offset == offset))
- break;
+ if (bucket_lock) {
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_slow);
+ OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_slow_elapsed);
+ } else {
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_fast);
+ OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_fast_elapsed);
}
- simple_unlock(&vm_page_bucket_lock);
+ if (mem != VM_PAGE_NULL)
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit);
+ else
+ OSAddAtomic64(1, &vm_page_lookup_stats.vpl_miss);
+#endif
+ if (mem != VM_PAGE_NULL) {
+ assert(mem->object == object);
- return(mem);
+ object->memq_hint = mem;
+ }
+ return (mem);
}
+
/*
* vm_page_rename:
*
vm_page_rename(
register vm_page_t mem,
register vm_object_t new_object,
- vm_object_offset_t new_offset)
+ vm_object_offset_t new_offset,
+ boolean_t encrypted_ok)
{
+ boolean_t internal_to_external, external_to_internal;
+ vm_tag_t tag;
+
assert(mem->object != new_object);
+
+ assert(mem->object);
+
/*
* ENCRYPTED SWAP:
* The encryption key is based on the page's memory object
* (aka "pager") and paging offset. Moving the page to
* another VM object changes its "pager" and "paging_offset"
- * so it has to be decrypted first.
+ * so it has to be decrypted first, or we would lose the key.
+ *
+ * One exception is VM object collapsing, where we transfer pages
+ * from one backing object to its parent object. This operation also
+ * transfers the paging information, so the <pager,paging_offset> info
+ * should remain consistent. The caller (vm_object_do_collapse())
+ * sets "encrypted_ok" in this case.
*/
- if (mem->encrypted) {
+ if (!encrypted_ok && mem->encrypted) {
panic("vm_page_rename: page %p is encrypted\n", mem);
}
+
+ XPR(XPR_VM_PAGE,
+ "vm_page_rename, new object 0x%X, offset 0x%X page 0x%X\n",
+ new_object, new_offset,
+ mem, 0,0);
+
/*
* Changes to mem->object require the page lock because
* the pageout daemon uses that lock to get the object.
*/
+ vm_page_lockspin_queues();
- XPR(XPR_VM_PAGE,
- "vm_page_rename, new object 0x%X, offset 0x%X page 0x%X\n",
- (integer_t)new_object, (integer_t)new_offset,
- (integer_t)mem, 0,0);
+ internal_to_external = FALSE;
+ external_to_internal = FALSE;
+
+ if (mem->local) {
+ /*
+ * it's much easier to get the vm_page_pageable_xxx accounting correct
+ * if we first move the page to the active queue... it's going to end
+ * up there anyway, and we don't do vm_page_rename's frequently enough
+ * for this to matter.
+ */
+ vm_page_queues_remove(mem);
+ vm_page_activate(mem);
+ }
+ if (mem->active || mem->inactive || mem->speculative) {
+ if (mem->object->internal && !new_object->internal) {
+ internal_to_external = TRUE;
+ }
+ if (!mem->object->internal && new_object->internal) {
+ external_to_internal = TRUE;
+ }
+ }
+
+ tag = mem->object->wire_tag;
+ vm_page_remove(mem, TRUE);
+ vm_page_insert_internal(mem, new_object, new_offset, tag, TRUE, TRUE, FALSE, FALSE, NULL);
+
+ if (internal_to_external) {
+ vm_page_pageable_internal_count--;
+ vm_page_pageable_external_count++;
+ } else if (external_to_internal) {
+ vm_page_pageable_external_count--;
+ vm_page_pageable_internal_count++;
+ }
- vm_page_lock_queues();
- vm_page_remove(mem);
- vm_page_insert(mem, new_object, new_offset);
vm_page_unlock_queues();
}
void
vm_page_init(
vm_page_t mem,
- ppnum_t phys_page)
+ ppnum_t phys_page,
+ boolean_t lopage)
{
assert(phys_page);
+
+#if DEBUG
+ if ((phys_page != vm_page_fictitious_addr) && (phys_page != vm_page_guard_addr)) {
+ if (!(pmap_valid_page(phys_page))) {
+ panic("vm_page_init: non-DRAM phys_page 0x%x\n", phys_page);
+ }
+ }
+#endif
*mem = vm_page_template;
mem->phys_page = phys_page;
+#if 0
+ /*
+ * we're leaving this turned off for now... currently pages
+ * come off the free list and are either immediately dirtied/referenced
+ * due to zero-fill or COW faults, or are used to read or write files...
+ * in the file I/O case, the UPL mechanism takes care of clearing
+ * the state of the HW ref/mod bits in a somewhat fragile way.
+ * Since we may change the way this works in the future (to toughen it up),
+ * I'm leaving this as a reminder of where these bits could get cleared
+ */
+
+ /*
+ * make sure both the h/w referenced and modified bits are
+ * clear at this point... we are especially dependent on
+ * not finding a 'stale' h/w modified in a number of spots
+ * once this page goes back into use
+ */
+ pmap_clear_refmod(phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED);
+#endif
+ mem->lopage = lopage;
}
/*
* Returns VM_PAGE_NULL if there are no free pages.
*/
int c_vm_page_grab_fictitious = 0;
+int c_vm_page_grab_fictitious_failed = 0;
int c_vm_page_release_fictitious = 0;
int c_vm_page_more_fictitious = 0;
vm_page_t
-vm_page_grab_fictitious(void)
+vm_page_grab_fictitious_common(
+ ppnum_t phys_addr)
{
- register vm_page_t m;
+ vm_page_t m;
+
+ if ((m = (vm_page_t)zget(vm_page_zone))) {
- m = (vm_page_t)zget(vm_page_zone);
- if (m) {
- vm_page_init(m, vm_page_fictitious_addr);
+ vm_page_init(m, phys_addr, FALSE);
m->fictitious = TRUE;
- }
- c_vm_page_grab_fictitious++;
+ c_vm_page_grab_fictitious++;
+ } else
+ c_vm_page_grab_fictitious_failed++;
+
return m;
}
+vm_page_t
+vm_page_grab_fictitious(void)
+{
+ return vm_page_grab_fictitious_common(vm_page_fictitious_addr);
+}
+
+vm_page_t
+vm_page_grab_guard(void)
+{
+ return vm_page_grab_fictitious_common(vm_page_guard_addr);
+}
+
+
/*
* vm_page_release_fictitious:
*
- * Release a fictitious page to the free list.
+ * Release a fictitious page to the zone pool
*/
-
void
vm_page_release_fictitious(
- register vm_page_t m)
+ vm_page_t m)
{
assert(!m->free);
- assert(m->busy);
assert(m->fictitious);
- assert(m->phys_page == vm_page_fictitious_addr);
+ assert(m->phys_page == vm_page_fictitious_addr ||
+ m->phys_page == vm_page_guard_addr);
c_vm_page_release_fictitious++;
-#if DEBUG
- if (m->free)
- panic("vm_page_release_fictitious");
-#endif
- m->free = TRUE;
+
zfree(vm_page_zone, m);
}
/*
* vm_page_more_fictitious:
*
- * Add more fictitious pages to the free list.
+ * Add more fictitious pages to the zone.
* Allowed to block. This routine is way intimate
* with the zones code, for several reasons:
* 1. we need to carve some page structures out of physical
* permanent allocation of a resource.
* 3. To smooth allocation humps, we allocate single pages
* with kernel_memory_allocate(), and cram them into the
- * zone. This also allows us to initialize the vm_page_t's
- * on the way into the zone, so that zget() always returns
- * an initialized structure. The zone free element pointer
- * and the free page pointer are both the first item in the
- * vm_page_t.
- * 4. By having the pages in the zone pre-initialized, we need
- * not keep 2 levels of lists. The garbage collector simply
- * scans our list, and reduces physical memory usage as it
- * sees fit.
+ * zone.
*/
void vm_page_more_fictitious(void)
{
- register vm_page_t m;
- vm_offset_t addr;
- kern_return_t retval;
- int i;
+ vm_offset_t addr;
+ kern_return_t retval;
c_vm_page_more_fictitious++;
* If winner is not vm-privileged, then the page allocation will fail,
* and it will temporarily block here in the vm_page_wait().
*/
- mutex_lock(&vm_page_alloc_lock);
+ lck_mtx_lock(&vm_page_alloc_lock);
/*
* If another thread allocated space, just bail out now.
*/
* of fictitious pages required in this manner is 2. 5 is
* simply a somewhat larger number.
*/
- mutex_unlock(&vm_page_alloc_lock);
+ lck_mtx_unlock(&vm_page_alloc_lock);
return;
}
retval = kernel_memory_allocate(zone_map,
&addr, PAGE_SIZE, VM_PROT_ALL,
- KMA_KOBJECT|KMA_NOPAGEWAIT);
+ KMA_KOBJECT|KMA_NOPAGEWAIT, VM_KERN_MEMORY_ZONE);
if (retval != KERN_SUCCESS) {
/*
- * No page was available. Tell the pageout daemon, drop the
+ * No page was available. Drop the
* lock to give another thread a chance at it, and
* wait for the pageout daemon to make progress.
*/
- mutex_unlock(&vm_page_alloc_lock);
+ lck_mtx_unlock(&vm_page_alloc_lock);
vm_page_wait(THREAD_UNINT);
return;
}
- /*
- * Initialize as many vm_page_t's as will fit on this page. This
- * depends on the zone code disturbing ONLY the first item of
- * each zone element.
- */
- m = (vm_page_t)addr;
- for (i = PAGE_SIZE/sizeof(struct vm_page); i > 0; i--) {
- vm_page_init(m, vm_page_fictitious_addr);
- m->fictitious = TRUE;
- m++;
- }
- zcram(vm_page_zone, (void *) addr, PAGE_SIZE);
- mutex_unlock(&vm_page_alloc_lock);
-}
-
-/*
- * vm_page_convert:
- *
- * Attempt to convert a fictitious page into a real page.
- */
-
-boolean_t
-vm_page_convert(
- register vm_page_t m)
-{
- register vm_page_t real_m;
-
- assert(m->busy);
- assert(m->fictitious);
- assert(!m->dirty);
-
- real_m = vm_page_grab();
- if (real_m == VM_PAGE_NULL)
- return FALSE;
-
- m->phys_page = real_m->phys_page;
- m->fictitious = FALSE;
- m->no_isync = TRUE;
-
- vm_page_lock_queues();
- if (m->active)
- vm_page_active_count++;
- else if (m->inactive)
- vm_page_inactive_count++;
- vm_page_unlock_queues();
- real_m->phys_page = vm_page_fictitious_addr;
- real_m->fictitious = TRUE;
+ zcram(vm_page_zone, addr, PAGE_SIZE);
- vm_page_release_fictitious(real_m);
- return TRUE;
+ lck_mtx_unlock(&vm_page_alloc_lock);
}
+
/*
* vm_pool_low():
*
vm_pool_low(void)
{
/* No locking, at worst we will fib. */
- return( vm_page_free_count < vm_page_free_reserved );
+ return( vm_page_free_count <= vm_page_free_reserved );
+}
+
+
+
+/*
+ * this is an interface to support bring-up of drivers
+ * on platforms with physical memory > 4G...
+ */
+int vm_himemory_mode = 2;
+
+
+/*
+ * this interface exists to support hardware controllers
+ * incapable of generating DMAs with more than 32 bits
+ * of address on platforms with physical memory > 4G...
+ */
+unsigned int vm_lopages_allocated_q = 0;
+unsigned int vm_lopages_allocated_cpm_success = 0;
+unsigned int vm_lopages_allocated_cpm_failed = 0;
+queue_head_t vm_lopage_queue_free;
+
+vm_page_t
+vm_page_grablo(void)
+{
+ vm_page_t mem;
+
+ if (vm_lopage_needed == FALSE)
+ return (vm_page_grab());
+
+ lck_mtx_lock_spin(&vm_page_queue_free_lock);
+
+ if ( !queue_empty(&vm_lopage_queue_free)) {
+ queue_remove_first(&vm_lopage_queue_free,
+ mem,
+ vm_page_t,
+ pageq);
+ assert(vm_lopage_free_count);
+
+ vm_lopage_free_count--;
+ vm_lopages_allocated_q++;
+
+ if (vm_lopage_free_count < vm_lopage_lowater)
+ vm_lopage_refill = TRUE;
+
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+ } else {
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+
+ if (cpm_allocate(PAGE_SIZE, &mem, atop(0xffffffff), 0, FALSE, KMA_LOMEM) != KERN_SUCCESS) {
+
+ lck_mtx_lock_spin(&vm_page_queue_free_lock);
+ vm_lopages_allocated_cpm_failed++;
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+
+ return (VM_PAGE_NULL);
+ }
+ mem->busy = TRUE;
+
+ vm_page_lockspin_queues();
+
+ mem->gobbled = FALSE;
+ vm_page_gobble_count--;
+ vm_page_wire_count--;
+
+ vm_lopages_allocated_cpm_success++;
+ vm_page_unlock_queues();
+ }
+ assert(mem->busy);
+ assert(!mem->free);
+ assert(!mem->pmapped);
+ assert(!mem->wpmapped);
+ assert(!pmap_is_noencrypt(mem->phys_page));
+
+ mem->pageq.next = NULL;
+ mem->pageq.prev = NULL;
+
+ return (mem);
}
+
/*
* vm_page_grab:
*
- * Remove a page from the free list.
- * Returns VM_PAGE_NULL if the free list is too small.
+ * first try to grab a page from the per-cpu free list...
+ * this must be done while pre-emption is disabled... if
+ * a page is available, we're done...
+ * if no page is available, grab the vm_page_queue_free_lock
+ * and see if current number of free pages would allow us
+ * to grab at least 1... if not, return VM_PAGE_NULL as before...
+ * if there are pages available, disable preemption and
+ * recheck the state of the per-cpu free list... we could
+ * have been preempted and moved to a different cpu, or
+ * some other thread could have re-filled it... if still
+ * empty, figure out how many pages we can steal from the
+ * global free queue and move to the per-cpu queue...
+ * return 1 of these pages when done... only wakeup the
+ * pageout_scan thread if we moved pages from the global
+ * list... no need for the wakeup if we've satisfied the
+ * request from the per-cpu queue.
*/
-unsigned long vm_page_grab_count = 0; /* measure demand */
vm_page_t
-vm_page_grab(void)
+vm_page_grab( void )
{
- register vm_page_t mem;
+ vm_page_t mem;
+
+
+ disable_preemption();
+
+ if ((mem = PROCESSOR_DATA(current_processor(), free_pages))) {
+return_page_from_cpu_list:
+ PROCESSOR_DATA(current_processor(), page_grab_count) += 1;
+ PROCESSOR_DATA(current_processor(), free_pages) = mem->pageq.next;
+
+ enable_preemption();
+ mem->pageq.next = NULL;
+
+ assert(mem->listq.next == NULL && mem->listq.prev == NULL);
+ assert(mem->tabled == FALSE);
+ assert(mem->object == VM_OBJECT_NULL);
+ assert(!mem->laundry);
+ assert(!mem->free);
+ assert(pmap_verify_free(mem->phys_page));
+ assert(mem->busy);
+ assert(!mem->encrypted);
+ assert(!mem->pmapped);
+ assert(!mem->wpmapped);
+ assert(!mem->active);
+ assert(!mem->inactive);
+ assert(!mem->throttled);
+ assert(!mem->speculative);
+ assert(!pmap_is_noencrypt(mem->phys_page));
+
+ return mem;
+ }
+ enable_preemption();
- mutex_lock(&vm_page_queue_free_lock);
- vm_page_grab_count++;
/*
* Optionally produce warnings if the wire or gobble
* counts exceed some threshold.
*/
- if (vm_page_wire_count_warning > 0
- && vm_page_wire_count >= vm_page_wire_count_warning) {
+#if VM_PAGE_WIRE_COUNT_WARNING
+ if (vm_page_wire_count >= VM_PAGE_WIRE_COUNT_WARNING) {
printf("mk: vm_page_grab(): high wired page count of %d\n",
vm_page_wire_count);
- assert(vm_page_wire_count < vm_page_wire_count_warning);
}
- if (vm_page_gobble_count_warning > 0
- && vm_page_gobble_count >= vm_page_gobble_count_warning) {
+#endif
+#if VM_PAGE_GOBBLE_COUNT_WARNING
+ if (vm_page_gobble_count >= VM_PAGE_GOBBLE_COUNT_WARNING) {
printf("mk: vm_page_grab(): high gobbled page count of %d\n",
vm_page_gobble_count);
- assert(vm_page_gobble_count < vm_page_gobble_count_warning);
}
+#endif
+ lck_mtx_lock_spin(&vm_page_queue_free_lock);
/*
* Only let privileged threads (involved in pageout)
* dip into the reserved pool.
*/
-
if ((vm_page_free_count < vm_page_free_reserved) &&
!(current_thread()->options & TH_OPT_VMPRIV)) {
- mutex_unlock(&vm_page_queue_free_lock);
+ lck_mtx_unlock(&vm_page_queue_free_lock);
mem = VM_PAGE_NULL;
- goto wakeup_pageout;
}
+ else {
+ vm_page_t head;
+ vm_page_t tail;
+ unsigned int pages_to_steal;
+ unsigned int color;
- while (vm_page_queue_free == VM_PAGE_NULL) {
- mutex_unlock(&vm_page_queue_free_lock);
- VM_PAGE_WAIT();
- mutex_lock(&vm_page_queue_free_lock);
- }
+ while ( vm_page_free_count == 0 ) {
- if (--vm_page_free_count < vm_page_free_count_minimum)
- vm_page_free_count_minimum = vm_page_free_count;
- mem = vm_page_queue_free;
- vm_page_queue_free = (vm_page_t) mem->pageq.next;
- mem->pageq.next = NULL;
- mem->pageq.prev = NULL;
- assert(mem->listq.next == NULL && mem->listq.prev == NULL);
- assert(mem->tabled == FALSE);
- assert(mem->object == VM_OBJECT_NULL);
- assert(!mem->laundry);
- mem->free = FALSE;
- mem->no_isync = TRUE;
- mutex_unlock(&vm_page_queue_free_lock);
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+ /*
+ * must be a privileged thread to be
+ * in this state since a non-privileged
+ * thread would have bailed if we were
+ * under the vm_page_free_reserved mark
+ */
+ VM_PAGE_WAIT();
+ lck_mtx_lock_spin(&vm_page_queue_free_lock);
+ }
- assert(pmap_verify_free(mem->phys_page));
+ disable_preemption();
- /*
- * Decide if we should poke the pageout daemon.
- * We do this if the free count is less than the low
+ if ((mem = PROCESSOR_DATA(current_processor(), free_pages))) {
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+
+ /*
+ * we got preempted and moved to another processor
+ * or we got preempted and someone else ran and filled the cache
+ */
+ goto return_page_from_cpu_list;
+ }
+ if (vm_page_free_count <= vm_page_free_reserved)
+ pages_to_steal = 1;
+ else {
+ if (vm_free_magazine_refill_limit <= (vm_page_free_count - vm_page_free_reserved))
+ pages_to_steal = vm_free_magazine_refill_limit;
+ else
+ pages_to_steal = (vm_page_free_count - vm_page_free_reserved);
+ }
+ color = PROCESSOR_DATA(current_processor(), start_color);
+ head = tail = NULL;
+
+ vm_page_free_count -= pages_to_steal;
+
+ while (pages_to_steal--) {
+
+ while (queue_empty(&vm_page_queue_free[color]))
+ color = (color + 1) & vm_color_mask;
+
+ queue_remove_first(&vm_page_queue_free[color],
+ mem,
+ vm_page_t,
+ pageq);
+ mem->pageq.next = NULL;
+ mem->pageq.prev = NULL;
+
+ assert(!mem->active);
+ assert(!mem->inactive);
+ assert(!mem->throttled);
+ assert(!mem->speculative);
+
+ color = (color + 1) & vm_color_mask;
+
+ if (head == NULL)
+ head = mem;
+ else
+ tail->pageq.next = (queue_t)mem;
+ tail = mem;
+
+ assert(mem->listq.next == NULL && mem->listq.prev == NULL);
+ assert(mem->tabled == FALSE);
+ assert(mem->object == VM_OBJECT_NULL);
+ assert(!mem->laundry);
+ assert(mem->free);
+ mem->free = FALSE;
+
+ assert(pmap_verify_free(mem->phys_page));
+ assert(mem->busy);
+ assert(!mem->free);
+ assert(!mem->encrypted);
+ assert(!mem->pmapped);
+ assert(!mem->wpmapped);
+ assert(!pmap_is_noencrypt(mem->phys_page));
+ }
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+
+ PROCESSOR_DATA(current_processor(), free_pages) = head->pageq.next;
+ PROCESSOR_DATA(current_processor(), start_color) = color;
+
+ /*
+ * satisfy this request
+ */
+ PROCESSOR_DATA(current_processor(), page_grab_count) += 1;
+ mem = head;
+ mem->pageq.next = NULL;
+
+ enable_preemption();
+ }
+ /*
+ * Decide if we should poke the pageout daemon.
+ * We do this if the free count is less than the low
* water mark, or if the free count is less than the high
* water mark (but above the low water mark) and the inactive
* count is less than its target.
* We don't have the counts locked ... if they change a little,
* it doesn't really matter.
*/
-
-wakeup_pageout:
if ((vm_page_free_count < vm_page_free_min) ||
- ((vm_page_free_count < vm_page_free_target) &&
- (vm_page_inactive_count < vm_page_inactive_target)))
- thread_wakeup((event_t) &vm_page_free_wanted);
+ ((vm_page_free_count < vm_page_free_target) &&
+ ((vm_page_inactive_count + vm_page_speculative_count) < vm_page_inactive_min)))
+ thread_wakeup((event_t) &vm_page_free_wanted);
+ VM_CHECK_MEMORYSTATUS;
+
// dbgLog(mem->phys_page, vm_page_free_count, vm_page_wire_count, 4); /* (TEST/DEBUG) */
return mem;
vm_page_release(
register vm_page_t mem)
{
+ unsigned int color;
+ int need_wakeup = 0;
+ int need_priv_wakeup = 0;
-#if 0
- unsigned int pindex;
- phys_entry *physent;
- physent = mapping_phys_lookup(mem->phys_page, &pindex); /* (BRINGUP) */
- if(physent->ppLink & ppN) { /* (BRINGUP) */
- panic("vm_page_release: already released - %08X %08X\n", mem, mem->phys_page);
- }
- physent->ppLink = physent->ppLink | ppN; /* (BRINGUP) */
-#endif
assert(!mem->private && !mem->fictitious);
-
+ if (vm_page_free_verify) {
+ assert(pmap_verify_free(mem->phys_page));
+ }
// dbgLog(mem->phys_page, vm_page_free_count, vm_page_wire_count, 5); /* (TEST/DEBUG) */
- mutex_lock(&vm_page_queue_free_lock);
+ pmap_clear_noencrypt(mem->phys_page);
+
+ lck_mtx_lock_spin(&vm_page_queue_free_lock);
#if DEBUG
if (mem->free)
panic("vm_page_release");
#endif
- mem->free = TRUE;
+
+ assert(mem->busy);
assert(!mem->laundry);
assert(mem->object == VM_OBJECT_NULL);
assert(mem->pageq.next == NULL &&
mem->pageq.prev == NULL);
- mem->pageq.next = (queue_entry_t) vm_page_queue_free;
- vm_page_queue_free = mem;
- vm_page_free_count++;
+ assert(mem->listq.next == NULL &&
+ mem->listq.prev == NULL);
+
+ if ((mem->lopage == TRUE || vm_lopage_refill == TRUE) &&
+ vm_lopage_free_count < vm_lopage_free_limit &&
+ mem->phys_page < max_valid_low_ppnum) {
+ /*
+ * this exists to support hardware controllers
+ * incapable of generating DMAs with more than 32 bits
+ * of address on platforms with physical memory > 4G...
+ */
+ queue_enter_first(&vm_lopage_queue_free,
+ mem,
+ vm_page_t,
+ pageq);
+ vm_lopage_free_count++;
+
+ if (vm_lopage_free_count >= vm_lopage_free_limit)
+ vm_lopage_refill = FALSE;
+
+ mem->lopage = TRUE;
+ } else {
+ mem->lopage = FALSE;
+ mem->free = TRUE;
+
+ color = mem->phys_page & vm_color_mask;
+ queue_enter_first(&vm_page_queue_free[color],
+ mem,
+ vm_page_t,
+ pageq);
+ vm_page_free_count++;
+ /*
+ * Check if we should wake up someone waiting for page.
+ * But don't bother waking them unless they can allocate.
+ *
+ * We wakeup only one thread, to prevent starvation.
+ * Because the scheduling system handles wait queues FIFO,
+ * if we wakeup all waiting threads, one greedy thread
+ * can starve multiple niceguy threads. When the threads
+ * all wakeup, the greedy threads runs first, grabs the page,
+ * and waits for another page. It will be the first to run
+ * when the next page is freed.
+ *
+ * However, there is a slight danger here.
+ * The thread we wake might not use the free page.
+ * Then the other threads could wait indefinitely
+ * while the page goes unused. To forestall this,
+ * the pageout daemon will keep making free pages
+ * as long as vm_page_free_wanted is non-zero.
+ */
- /*
- * Check if we should wake up someone waiting for page.
- * But don't bother waking them unless they can allocate.
- *
- * We wakeup only one thread, to prevent starvation.
- * Because the scheduling system handles wait queues FIFO,
- * if we wakeup all waiting threads, one greedy thread
- * can starve multiple niceguy threads. When the threads
- * all wakeup, the greedy threads runs first, grabs the page,
- * and waits for another page. It will be the first to run
- * when the next page is freed.
- *
- * However, there is a slight danger here.
- * The thread we wake might not use the free page.
- * Then the other threads could wait indefinitely
- * while the page goes unused. To forestall this,
- * the pageout daemon will keep making free pages
- * as long as vm_page_free_wanted is non-zero.
- */
+ assert(vm_page_free_count > 0);
+ if (vm_page_free_wanted_privileged > 0) {
+ vm_page_free_wanted_privileged--;
+ need_priv_wakeup = 1;
+ } else if (vm_page_free_wanted > 0 &&
+ vm_page_free_count > vm_page_free_reserved) {
+ vm_page_free_wanted--;
+ need_wakeup = 1;
+ }
+ }
+ lck_mtx_unlock(&vm_page_queue_free_lock);
- if ((vm_page_free_wanted > 0) &&
- (vm_page_free_count >= vm_page_free_reserved)) {
- vm_page_free_wanted--;
+ if (need_priv_wakeup)
+ thread_wakeup_one((event_t) &vm_page_free_wanted_privileged);
+ else if (need_wakeup)
thread_wakeup_one((event_t) &vm_page_free_count);
- }
- mutex_unlock(&vm_page_queue_free_lock);
+ VM_CHECK_MEMORYSTATUS;
+}
+
+/*
+ * This version of vm_page_release() is used only at startup
+ * when we are single-threaded and pages are being released
+ * for the first time. Hence, no locking or unnecessary checks are made.
+ * Note: VM_CHECK_MEMORYSTATUS invoked by the caller.
+ */
+void
+vm_page_release_startup(
+ register vm_page_t mem)
+{
+ queue_t queue_free;
+
+ if (vm_lopage_free_count < vm_lopage_free_limit &&
+ mem->phys_page < max_valid_low_ppnum) {
+ mem->lopage = TRUE;
+ vm_lopage_free_count++;
+ queue_free = &vm_lopage_queue_free;
+ } else {
+ mem->lopage = FALSE;
+ mem->free = TRUE;
+ vm_page_free_count++;
+ queue_free = &vm_page_queue_free[mem->phys_page & vm_color_mask];
+ }
+ queue_enter_first(queue_free, mem, vm_page_t, pageq);
}
/*
*/
kern_return_t wait_result;
int need_wakeup = 0;
+ int is_privileged = current_thread()->options & TH_OPT_VMPRIV;
+
+ lck_mtx_lock_spin(&vm_page_queue_free_lock);
- mutex_lock(&vm_page_queue_free_lock);
+ if (is_privileged && vm_page_free_count) {
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+ return TRUE;
+ }
if (vm_page_free_count < vm_page_free_target) {
- if (vm_page_free_wanted++ == 0)
- need_wakeup = 1;
- wait_result = assert_wait((event_t)&vm_page_free_count, interruptible);
- mutex_unlock(&vm_page_queue_free_lock);
+
+ if (is_privileged) {
+ if (vm_page_free_wanted_privileged++ == 0)
+ need_wakeup = 1;
+ wait_result = assert_wait((event_t)&vm_page_free_wanted_privileged, interruptible);
+ } else {
+ if (vm_page_free_wanted++ == 0)
+ need_wakeup = 1;
+ wait_result = assert_wait((event_t)&vm_page_free_count, interruptible);
+ }
+ lck_mtx_unlock(&vm_page_queue_free_lock);
counter(c_vm_page_wait_block++);
if (need_wakeup)
thread_wakeup((event_t)&vm_page_free_wanted);
- if (wait_result == THREAD_WAITING)
+ if (wait_result == THREAD_WAITING) {
+ VM_DEBUG_EVENT(vm_page_wait_block, VM_PAGE_WAIT_BLOCK, DBG_FUNC_START,
+ vm_page_free_wanted_privileged, vm_page_free_wanted, 0, 0);
wait_result = thread_block(THREAD_CONTINUE_NULL);
+ VM_DEBUG_EVENT(vm_page_wait_block, VM_PAGE_WAIT_BLOCK, DBG_FUNC_END, 0, 0, 0, 0);
+ }
return(wait_result == THREAD_AWAKENED);
} else {
- mutex_unlock(&vm_page_queue_free_lock);
+ lck_mtx_unlock(&vm_page_queue_free_lock);
return TRUE;
}
}
{
register vm_page_t mem;
-#if DEBUG
- _mutex_assert(&object->Lock, MA_OWNED);
-#endif
+ vm_object_lock_assert_exclusive(object);
mem = vm_page_grab();
if (mem == VM_PAGE_NULL)
return VM_PAGE_NULL;
return(mem);
}
+/*
+ * vm_page_alloc_guard:
+ *
+ * Allocate a fictitious page which will be used
+ * as a guard page. The page will be inserted into
+ * the object and returned to the caller.
+ */
+
+vm_page_t
+vm_page_alloc_guard(
+ vm_object_t object,
+ vm_object_offset_t offset)
+{
+ register vm_page_t mem;
+
+ vm_object_lock_assert_exclusive(object);
+ mem = vm_page_grab_guard();
+ if (mem == VM_PAGE_NULL)
+ return VM_PAGE_NULL;
+
+ vm_page_insert(mem, object, offset);
+
+ return(mem);
+}
+
+
counter(unsigned int c_laundry_pages_freed = 0;)
-int vm_pagein_cluster_unused = 0;
-boolean_t vm_page_free_verify = TRUE;
/*
- * vm_page_free:
+ * vm_page_free_prepare:
*
- * Returns the given page to the free list,
- * disassociating it with any VM object.
+ * Removes page from any queue it may be on
+ * and disassociates it from its VM object.
*
* Object and page queues must be locked prior to entry.
*/
-void
-vm_page_free(
- register vm_page_t mem)
+static void
+vm_page_free_prepare(
+ vm_page_t mem)
{
- vm_object_t object = mem->object;
+ vm_page_free_prepare_queues(mem);
+ vm_page_free_prepare_object(mem, TRUE);
+}
+
+void
+vm_page_free_prepare_queues(
+ vm_page_t mem)
+{
+ VM_PAGE_CHECK(mem);
assert(!mem->free);
assert(!mem->cleaning);
- assert(!mem->pageout);
- if (vm_page_free_verify && !mem->fictitious && !mem->private) {
- assert(pmap_verify_free(mem->phys_page));
- }
-
-#if DEBUG
- if (mem->object)
- _mutex_assert(&mem->object->Lock, MA_OWNED);
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
+#if MACH_ASSERT || DEBUG
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
if (mem->free)
- panic("vm_page_free: freeing page on free list\n");
-#endif
- if (mem->tabled)
- vm_page_remove(mem); /* clears tabled, object, offset */
- VM_PAGE_QUEUES_REMOVE(mem); /* clears active or inactive */
-
- if (mem->clustered) {
- mem->clustered = FALSE;
- vm_pagein_cluster_unused++;
+ panic("vm_page_free: freeing page on free list\n");
+#endif /* MACH_ASSERT || DEBUG */
+ if (mem->object) {
+ vm_object_lock_assert_exclusive(mem->object);
+ }
+ if (mem->laundry) {
+ /*
+ * We may have to free a page while it's being laundered
+ * if we lost its pager (due to a forced unmount, for example).
+ * We need to call vm_pageout_steal_laundry() before removing
+ * the page from its VM object, so that we can remove it
+ * from its pageout queue and adjust the laundry accounting
+ */
+ vm_pageout_steal_laundry(mem, TRUE);
+ counter(++c_laundry_pages_freed);
}
+
+ vm_page_queues_remove(mem); /* clears local/active/inactive/throttled/speculative */
+
+ if (VM_PAGE_WIRED(mem)) {
+ if (mem->object) {
+ assert(mem->object->wired_page_count > 0);
+ mem->object->wired_page_count--;
+ if (!mem->object->wired_page_count) {
+ VM_OBJECT_UNWIRED(mem->object);
+ }
- if (mem->wire_count) {
+ assert(mem->object->resident_page_count >=
+ mem->object->wired_page_count);
+
+ if (mem->object->purgable == VM_PURGABLE_VOLATILE) {
+ OSAddAtomic(+1, &vm_page_purgeable_count);
+ assert(vm_page_purgeable_wired_count > 0);
+ OSAddAtomic(-1, &vm_page_purgeable_wired_count);
+ }
+ if ((mem->object->purgable == VM_PURGABLE_VOLATILE ||
+ mem->object->purgable == VM_PURGABLE_EMPTY) &&
+ mem->object->vo_purgeable_owner != TASK_NULL) {
+ task_t owner;
+
+ owner = mem->object->vo_purgeable_owner;
+ /*
+ * While wired, this page was accounted
+ * as "non-volatile" but it should now
+ * be accounted as "volatile".
+ */
+ /* one less "non-volatile"... */
+ ledger_debit(owner->ledger,
+ task_ledgers.purgeable_nonvolatile,
+ PAGE_SIZE);
+ /* ... and "phys_footprint" */
+ ledger_debit(owner->ledger,
+ task_ledgers.phys_footprint,
+ PAGE_SIZE);
+ /* one more "volatile" */
+ ledger_credit(owner->ledger,
+ task_ledgers.purgeable_volatile,
+ PAGE_SIZE);
+ }
+ }
if (!mem->private && !mem->fictitious)
vm_page_wire_count--;
mem->wire_count = 0;
vm_page_wire_count--;
vm_page_gobble_count--;
}
- mem->gobbled = FALSE;
-
- if (mem->laundry) {
- vm_pageout_throttle_up(mem);
- counter(++c_laundry_pages_freed);
- }
-
- PAGE_WAKEUP(mem); /* clears wanted */
+}
- if (mem->absent)
- vm_object_absent_release(object);
- /* Some of these may be unnecessary */
- mem->page_lock = 0;
- mem->unlock_request = 0;
- mem->busy = TRUE;
- mem->absent = FALSE;
- mem->error = FALSE;
- mem->dirty = FALSE;
- mem->precious = FALSE;
- mem->reference = FALSE;
- mem->encrypted = FALSE;
+void
+vm_page_free_prepare_object(
+ vm_page_t mem,
+ boolean_t remove_from_hash)
+{
+ if (mem->tabled)
+ vm_page_remove(mem, remove_from_hash); /* clears tabled, object, offset */
- mem->page_error = KERN_SUCCESS;
+ PAGE_WAKEUP(mem); /* clears wanted */
if (mem->private) {
mem->private = FALSE;
mem->fictitious = TRUE;
mem->phys_page = vm_page_fictitious_addr;
}
+ if ( !mem->fictitious) {
+ vm_page_init(mem, mem->phys_page, mem->lopage);
+ }
+}
+
+
+/*
+ * vm_page_free:
+ *
+ * Returns the given page to the free list,
+ * disassociating it with any VM object.
+ *
+ * Object and page queues must be locked prior to entry.
+ */
+void
+vm_page_free(
+ vm_page_t mem)
+{
+ vm_page_free_prepare(mem);
+
if (mem->fictitious) {
vm_page_release_fictitious(mem);
} else {
- /* depends on the queues lock */
- if(mem->zero_fill) {
- vm_zf_count-=1;
- mem->zero_fill = FALSE;
- }
- vm_page_init(mem, mem->phys_page);
vm_page_release(mem);
}
}
void
-vm_page_free_list(
- register vm_page_t mem)
+vm_page_free_unlocked(
+ vm_page_t mem,
+ boolean_t remove_from_hash)
{
- register vm_page_t nxt;
- register vm_page_t first = NULL;
- register vm_page_t last = VM_PAGE_NULL;
- register int pg_count = 0;
+ vm_page_lockspin_queues();
+ vm_page_free_prepare_queues(mem);
+ vm_page_unlock_queues();
-#if DEBUG
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
-#endif
- while (mem) {
-#if DEBUG
- if (mem->tabled || mem->object)
- panic("vm_page_free_list: freeing tabled page\n");
- if (mem->inactive || mem->active || mem->free)
- panic("vm_page_free_list: freeing page on list\n");
-#endif
- assert(mem->pageq.prev == NULL);
- nxt = (vm_page_t)(mem->pageq.next);
+ vm_page_free_prepare_object(mem, remove_from_hash);
+
+ if (mem->fictitious) {
+ vm_page_release_fictitious(mem);
+ } else {
+ vm_page_release(mem);
+ }
+}
- if (mem->clustered)
- vm_pagein_cluster_unused++;
- if (mem->laundry) {
- vm_pageout_throttle_up(mem);
- counter(++c_laundry_pages_freed);
- }
- mem->busy = TRUE;
+/*
+ * Free a list of pages. The list can be up to several hundred pages,
+ * as blocked up by vm_pageout_scan().
+ * The big win is not having to take the free list lock once
+ * per page.
+ */
+void
+vm_page_free_list(
+ vm_page_t freeq,
+ boolean_t prepare_object)
+{
+ vm_page_t mem;
+ vm_page_t nxt;
+ vm_page_t local_freeq;
+ int pg_count;
- PAGE_WAKEUP(mem); /* clears wanted */
+ while (freeq) {
- if (mem->private)
- mem->fictitious = TRUE;
+ pg_count = 0;
+ local_freeq = VM_PAGE_NULL;
+ mem = freeq;
- if (!mem->fictitious) {
- /* depends on the queues lock */
- if (mem->zero_fill)
- vm_zf_count -= 1;
- assert(!mem->laundry);
- vm_page_init(mem, mem->phys_page);
+ /*
+ * break up the processing into smaller chunks so
+ * that we can 'pipeline' the pages onto the
+ * free list w/o introducing too much
+ * contention on the global free queue lock
+ */
+ while (mem && pg_count < 64) {
- mem->free = TRUE;
+ assert(!mem->inactive);
+ assert(!mem->active);
+ assert(!mem->throttled);
+ assert(!mem->free);
+ assert(!mem->speculative);
+ assert(!VM_PAGE_WIRED(mem));
+ assert(mem->pageq.prev == NULL);
- if (first == NULL)
- last = mem;
- mem->pageq.next = (queue_t) first;
- first = mem;
+ nxt = (vm_page_t)(mem->pageq.next);
+
+ if (vm_page_free_verify && !mem->fictitious && !mem->private) {
+ assert(pmap_verify_free(mem->phys_page));
+ }
+ if (prepare_object == TRUE)
+ vm_page_free_prepare_object(mem, TRUE);
- pg_count++;
- } else {
- mem->phys_page = vm_page_fictitious_addr;
- vm_page_release_fictitious(mem);
+ if (!mem->fictitious) {
+ assert(mem->busy);
+
+ if ((mem->lopage == TRUE || vm_lopage_refill == TRUE) &&
+ vm_lopage_free_count < vm_lopage_free_limit &&
+ mem->phys_page < max_valid_low_ppnum) {
+ mem->pageq.next = NULL;
+ vm_page_release(mem);
+ } else {
+ /*
+ * IMPORTANT: we can't set the page "free" here
+ * because that would make the page eligible for
+ * a physically-contiguous allocation (see
+ * vm_page_find_contiguous()) right away (we don't
+ * hold the vm_page_queue_free lock). That would
+ * cause trouble because the page is not actually
+ * in the free queue yet...
+ */
+ mem->pageq.next = (queue_entry_t)local_freeq;
+ local_freeq = mem;
+ pg_count++;
+
+ pmap_clear_noencrypt(mem->phys_page);
+ }
+ } else {
+ assert(mem->phys_page == vm_page_fictitious_addr ||
+ mem->phys_page == vm_page_guard_addr);
+ vm_page_release_fictitious(mem);
+ }
+ mem = nxt;
}
- mem = nxt;
- }
- if (first) {
-
- mutex_lock(&vm_page_queue_free_lock);
+ freeq = mem;
+
+ if ( (mem = local_freeq) ) {
+ unsigned int avail_free_count;
+ unsigned int need_wakeup = 0;
+ unsigned int need_priv_wakeup = 0;
+
+ lck_mtx_lock_spin(&vm_page_queue_free_lock);
+
+ while (mem) {
+ int color;
+
+ nxt = (vm_page_t)(mem->pageq.next);
+
+ assert(!mem->free);
+ assert(mem->busy);
+ mem->free = TRUE;
+
+ color = mem->phys_page & vm_color_mask;
+ queue_enter_first(&vm_page_queue_free[color],
+ mem,
+ vm_page_t,
+ pageq);
+ mem = nxt;
+ }
+ vm_page_free_count += pg_count;
+ avail_free_count = vm_page_free_count;
- last->pageq.next = (queue_entry_t) vm_page_queue_free;
- vm_page_queue_free = first;
+ if (vm_page_free_wanted_privileged > 0 && avail_free_count > 0) {
- vm_page_free_count += pg_count;
+ if (avail_free_count < vm_page_free_wanted_privileged) {
+ need_priv_wakeup = avail_free_count;
+ vm_page_free_wanted_privileged -= avail_free_count;
+ avail_free_count = 0;
+ } else {
+ need_priv_wakeup = vm_page_free_wanted_privileged;
+ vm_page_free_wanted_privileged = 0;
+ avail_free_count -= vm_page_free_wanted_privileged;
+ }
+ }
+ if (vm_page_free_wanted > 0 && avail_free_count > vm_page_free_reserved) {
+ unsigned int available_pages;
- if ((vm_page_free_wanted > 0) &&
- (vm_page_free_count >= vm_page_free_reserved)) {
- unsigned int available_pages;
+ available_pages = avail_free_count - vm_page_free_reserved;
- if (vm_page_free_count >= vm_page_free_reserved) {
- available_pages = (vm_page_free_count
- - vm_page_free_reserved);
- } else {
- available_pages = 0;
+ if (available_pages >= vm_page_free_wanted) {
+ need_wakeup = vm_page_free_wanted;
+ vm_page_free_wanted = 0;
+ } else {
+ need_wakeup = available_pages;
+ vm_page_free_wanted -= available_pages;
+ }
}
+ lck_mtx_unlock(&vm_page_queue_free_lock);
- if (available_pages >= vm_page_free_wanted) {
- vm_page_free_wanted = 0;
+ if (need_priv_wakeup != 0) {
+ /*
+ * There shouldn't be that many VM-privileged threads,
+ * so let's wake them all up, even if we don't quite
+ * have enough pages to satisfy them all.
+ */
+ thread_wakeup((event_t)&vm_page_free_wanted_privileged);
+ }
+ if (need_wakeup != 0 && vm_page_free_wanted == 0) {
+ /*
+ * We don't expect to have any more waiters
+ * after this, so let's wake them all up at
+ * once.
+ */
thread_wakeup((event_t) &vm_page_free_count);
- } else {
- while (available_pages--) {
- vm_page_free_wanted--;
- thread_wakeup_one((event_t) &vm_page_free_count);
- }
+ } else for (; need_wakeup != 0; need_wakeup--) {
+ /*
+ * Wake up one waiter per page we just released.
+ */
+ thread_wakeup_one((event_t) &vm_page_free_count);
}
+
+ VM_CHECK_MEMORYSTATUS;
}
- mutex_unlock(&vm_page_queue_free_lock);
}
}
*
* The page's object and the page queues must be locked.
*/
+
+
void
vm_page_wire(
- register vm_page_t mem)
+ register vm_page_t mem,
+ vm_tag_t tag,
+ boolean_t check_memorystatus)
{
// dbgLog(current_thread(), mem->offset, mem->object, 1); /* (TEST/DEBUG) */
VM_PAGE_CHECK(mem);
+ if (mem->object) {
+ vm_object_lock_assert_exclusive(mem->object);
+ } else {
+ /*
+ * In theory, the page should be in an object before it
+ * gets wired, since we need to hold the object lock
+ * to update some fields in the page structure.
+ * However, some code (i386 pmap, for example) might want
+ * to wire a page before it gets inserted into an object.
+ * That's somewhat OK, as long as nobody else can get to
+ * that page and update it at the same time.
+ */
+ }
#if DEBUG
- if (mem->object)
- _mutex_assert(&mem->object->Lock, MA_OWNED);
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
#endif
- if (mem->wire_count == 0) {
- VM_PAGE_QUEUES_REMOVE(mem);
+ if ( !VM_PAGE_WIRED(mem)) {
+
+ if (mem->pageout_queue) {
+ mem->pageout = FALSE;
+ vm_pageout_throttle_up(mem);
+ }
+ vm_page_queues_remove(mem);
+
+ if (mem->object) {
+
+ if (!mem->private && !mem->fictitious)
+ {
+ if (!mem->object->wired_page_count)
+ {
+ assert(VM_KERN_MEMORY_NONE != tag);
+ mem->object->wire_tag = tag;
+ VM_OBJECT_WIRED(mem->object);
+ }
+ }
+ mem->object->wired_page_count++;
+
+ assert(mem->object->resident_page_count >=
+ mem->object->wired_page_count);
+ if (mem->object->purgable == VM_PURGABLE_VOLATILE) {
+ assert(vm_page_purgeable_count > 0);
+ OSAddAtomic(-1, &vm_page_purgeable_count);
+ OSAddAtomic(1, &vm_page_purgeable_wired_count);
+ }
+ if ((mem->object->purgable == VM_PURGABLE_VOLATILE ||
+ mem->object->purgable == VM_PURGABLE_EMPTY) &&
+ mem->object->vo_purgeable_owner != TASK_NULL) {
+ task_t owner;
+
+ owner = mem->object->vo_purgeable_owner;
+ /* less volatile bytes */
+ ledger_debit(owner->ledger,
+ task_ledgers.purgeable_volatile,
+ PAGE_SIZE);
+ /* more not-quite-volatile bytes */
+ ledger_credit(owner->ledger,
+ task_ledgers.purgeable_nonvolatile,
+ PAGE_SIZE);
+ /* more footprint */
+ ledger_credit(owner->ledger,
+ task_ledgers.phys_footprint,
+ PAGE_SIZE);
+ }
+ if (mem->object->all_reusable) {
+ /*
+ * Wired pages are not counted as "re-usable"
+ * in "all_reusable" VM objects, so nothing
+ * to do here.
+ */
+ } else if (mem->reusable) {
+ /*
+ * This page is not "re-usable" when it's
+ * wired, so adjust its state and the
+ * accounting.
+ */
+ vm_object_reuse_pages(mem->object,
+ mem->offset,
+ mem->offset+PAGE_SIZE_64,
+ FALSE);
+ }
+ }
+ assert(!mem->reusable);
+
if (!mem->private && !mem->fictitious && !mem->gobbled)
vm_page_wire_count++;
if (mem->gobbled)
vm_page_gobble_count--;
mem->gobbled = FALSE;
- if(mem->zero_fill) {
- /* depends on the queues lock */
- vm_zf_count-=1;
- mem->zero_fill = FALSE;
+
+ if (check_memorystatus == TRUE) {
+ VM_CHECK_MEMORYSTATUS;
}
/*
* ENCRYPTED SWAP:
}
assert(!mem->gobbled);
mem->wire_count++;
-}
-
-/*
- * vm_page_gobble:
- *
- * Mark this page as consumed by the vm/ipc/xmm subsystems.
- *
- * Called only for freshly vm_page_grab()ed pages - w/ nothing locked.
- */
-void
-vm_page_gobble(
- register vm_page_t mem)
-{
- vm_page_lock_queues();
- VM_PAGE_CHECK(mem);
-
- assert(!mem->gobbled);
- assert(mem->wire_count == 0);
-
- if (!mem->gobbled && mem->wire_count == 0) {
- if (!mem->private && !mem->fictitious)
- vm_page_wire_count++;
- }
- vm_page_gobble_count++;
- mem->gobbled = TRUE;
- vm_page_unlock_queues();
+ VM_PAGE_CHECK(mem);
}
/*
*/
void
vm_page_unwire(
- register vm_page_t mem)
+ vm_page_t mem,
+ boolean_t queueit)
{
// dbgLog(current_thread(), mem->offset, mem->object, 0); /* (TEST/DEBUG) */
VM_PAGE_CHECK(mem);
- assert(mem->wire_count > 0);
+ assert(VM_PAGE_WIRED(mem));
+ assert(!mem->gobbled);
+ assert(mem->object != VM_OBJECT_NULL);
#if DEBUG
- if (mem->object)
- _mutex_assert(&mem->object->Lock, MA_OWNED);
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
+ vm_object_lock_assert_exclusive(mem->object);
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
#endif
if (--mem->wire_count == 0) {
- assert(!mem->private && !mem->fictitious);
- vm_page_wire_count--;
- assert(!mem->laundry);
+ if (!mem->private && !mem->fictitious) {
+ vm_page_wire_count--;
+ }
+ assert(mem->object->wired_page_count > 0);
+ mem->object->wired_page_count--;
+ if (!mem->object->wired_page_count) {
+ VM_OBJECT_UNWIRED(mem->object);
+ }
+ assert(mem->object->resident_page_count >=
+ mem->object->wired_page_count);
+ if (mem->object->purgable == VM_PURGABLE_VOLATILE) {
+ OSAddAtomic(+1, &vm_page_purgeable_count);
+ assert(vm_page_purgeable_wired_count > 0);
+ OSAddAtomic(-1, &vm_page_purgeable_wired_count);
+ }
+ if ((mem->object->purgable == VM_PURGABLE_VOLATILE ||
+ mem->object->purgable == VM_PURGABLE_EMPTY) &&
+ mem->object->vo_purgeable_owner != TASK_NULL) {
+ task_t owner;
+
+ owner = mem->object->vo_purgeable_owner;
+ /* more volatile bytes */
+ ledger_credit(owner->ledger,
+ task_ledgers.purgeable_volatile,
+ PAGE_SIZE);
+ /* less not-quite-volatile bytes */
+ ledger_debit(owner->ledger,
+ task_ledgers.purgeable_nonvolatile,
+ PAGE_SIZE);
+ /* less footprint */
+ ledger_debit(owner->ledger,
+ task_ledgers.phys_footprint,
+ PAGE_SIZE);
+ }
assert(mem->object != kernel_object);
assert(mem->pageq.next == NULL && mem->pageq.prev == NULL);
- queue_enter(&vm_page_queue_active, mem, vm_page_t, pageq);
- vm_page_active_count++;
- mem->active = TRUE;
- mem->reference = TRUE;
+
+ if (queueit == TRUE) {
+ if (mem->object->purgable == VM_PURGABLE_EMPTY) {
+ vm_page_deactivate(mem);
+ } else {
+ vm_page_activate(mem);
+ }
+ }
+
+ VM_CHECK_MEMORYSTATUS;
+
}
+ VM_PAGE_CHECK(mem);
}
/*
*/
void
vm_page_deactivate(
- register vm_page_t m)
+ vm_page_t m)
{
+ vm_page_deactivate_internal(m, TRUE);
+}
+
+
+void
+vm_page_deactivate_internal(
+ vm_page_t m,
+ boolean_t clear_hw_reference)
+{
+
VM_PAGE_CHECK(m);
assert(m->object != kernel_object);
+ assert(m->phys_page != vm_page_guard_addr);
// dbgLog(m->phys_page, vm_page_free_count, vm_page_wire_count, 6); /* (TEST/DEBUG) */
#if DEBUG
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
#endif
/*
* This page is no longer very interesting. If it was
* inactive queue. Note wired pages should not have
* their reference bit cleared.
*/
+ assert ( !(m->absent && !m->unusual));
+
if (m->gobbled) { /* can this happen? */
- assert(m->wire_count == 0);
+ assert( !VM_PAGE_WIRED(m));
+
if (!m->private && !m->fictitious)
vm_page_wire_count--;
vm_page_gobble_count--;
m->gobbled = FALSE;
}
- if (m->private || (m->wire_count != 0))
+ /*
+ * if this page is currently on the pageout queue, we can't do the
+ * vm_page_queues_remove (which doesn't handle the pageout queue case)
+ * and we can't remove it manually since we would need the object lock
+ * (which is not required here) to decrement the activity_in_progress
+ * reference which is held on the object while the page is in the pageout queue...
+ * just let the normal laundry processing proceed
+ */
+ if (m->laundry || m->pageout_queue || m->private || m->fictitious || m->compressor || (VM_PAGE_WIRED(m)))
return;
- if (m->active || (m->inactive && m->reference)) {
- if (!m->fictitious && !m->absent)
- pmap_clear_reference(m->phys_page);
- m->reference = FALSE;
- VM_PAGE_QUEUES_REMOVE(m);
- }
- if (m->wire_count == 0 && !m->inactive) {
- m->page_ticket = vm_page_ticket;
- vm_page_ticket_roll++;
-
- if(vm_page_ticket_roll == VM_PAGE_TICKETS_IN_ROLL) {
- vm_page_ticket_roll = 0;
- if(vm_page_ticket == VM_PAGE_TICKET_ROLL_IDS)
- vm_page_ticket= 0;
- else
- vm_page_ticket++;
- }
-
- assert(!m->laundry);
- assert(m->pageq.next == NULL && m->pageq.prev == NULL);
- if(m->zero_fill) {
- queue_enter(&vm_page_queue_zf, m, vm_page_t, pageq);
+
+ if (!m->absent && clear_hw_reference == TRUE)
+ pmap_clear_reference(m->phys_page);
+
+ m->reference = FALSE;
+ m->no_cache = FALSE;
+
+ if (!m->inactive) {
+ vm_page_queues_remove(m);
+
+ if (!VM_DYNAMIC_PAGING_ENABLED(memory_manager_default) &&
+ m->dirty && m->object->internal &&
+ (m->object->purgable == VM_PURGABLE_DENY ||
+ m->object->purgable == VM_PURGABLE_NONVOLATILE ||
+ m->object->purgable == VM_PURGABLE_VOLATILE)) {
+ vm_page_check_pageable_safe(m);
+ queue_enter(&vm_page_queue_throttled, m, vm_page_t, pageq);
+ m->throttled = TRUE;
+ vm_page_throttled_count++;
} else {
- queue_enter(&vm_page_queue_inactive,
- m, vm_page_t, pageq);
+ if (m->object->named && m->object->ref_count == 1) {
+ vm_page_speculate(m, FALSE);
+#if DEVELOPMENT || DEBUG
+ vm_page_speculative_recreated++;
+#endif
+ } else {
+ vm_page_enqueue_inactive(m, FALSE);
+ }
}
+ }
+}
+
+/*
+ * vm_page_enqueue_cleaned
+ *
+ * Put the page on the cleaned queue, mark it cleaned, etc.
+ * Being on the cleaned queue (and having m->clean_queue set)
+ * does ** NOT ** guarantee that the page is clean!
+ *
+ * Call with the queues lock held.
+ */
+
+void vm_page_enqueue_cleaned(vm_page_t m)
+{
+ assert(m->phys_page != vm_page_guard_addr);
+#if DEBUG
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
+#endif
+ assert( !(m->absent && !m->unusual));
+
+ if (m->gobbled) {
+ assert( !VM_PAGE_WIRED(m));
+ if (!m->private && !m->fictitious)
+ vm_page_wire_count--;
+ vm_page_gobble_count--;
+ m->gobbled = FALSE;
+ }
+ /*
+ * if this page is currently on the pageout queue, we can't do the
+ * vm_page_queues_remove (which doesn't handle the pageout queue case)
+ * and we can't remove it manually since we would need the object lock
+ * (which is not required here) to decrement the activity_in_progress
+ * reference which is held on the object while the page is in the pageout queue...
+ * just let the normal laundry processing proceed
+ */
+ if (m->laundry || m->clean_queue || m->pageout_queue || m->private || m->fictitious)
+ return;
+
+ vm_page_queues_remove(m);
- m->inactive = TRUE;
- if (!m->fictitious)
- vm_page_inactive_count++;
+ vm_page_check_pageable_safe(m);
+ queue_enter(&vm_page_queue_cleaned, m, vm_page_t, pageq);
+ m->clean_queue = TRUE;
+ vm_page_cleaned_count++;
+
+ m->inactive = TRUE;
+ vm_page_inactive_count++;
+ if (m->object->internal) {
+ vm_page_pageable_internal_count++;
+ } else {
+ vm_page_pageable_external_count++;
}
+
+ vm_pageout_enqueued_cleaned++;
}
/*
register vm_page_t m)
{
VM_PAGE_CHECK(m);
+#ifdef FIXME_4778297
assert(m->object != kernel_object);
+#endif
+ assert(m->phys_page != vm_page_guard_addr);
#if DEBUG
- _mutex_assert(&vm_page_queue_lock, MA_OWNED);
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
#endif
+ assert( !(m->absent && !m->unusual));
+
if (m->gobbled) {
- assert(m->wire_count == 0);
+ assert( !VM_PAGE_WIRED(m));
if (!m->private && !m->fictitious)
vm_page_wire_count--;
vm_page_gobble_count--;
m->gobbled = FALSE;
}
- if (m->private)
+ /*
+ * if this page is currently on the pageout queue, we can't do the
+ * vm_page_queues_remove (which doesn't handle the pageout queue case)
+ * and we can't remove it manually since we would need the object lock
+ * (which is not required here) to decrement the activity_in_progress
+ * reference which is held on the object while the page is in the pageout queue...
+ * just let the normal laundry processing proceed
+ */
+ if (m->laundry || m->pageout_queue || m->private || m->fictitious || m->compressor)
+ return;
+
+#if DEBUG
+ if (m->active)
+ panic("vm_page_activate: already active");
+#endif
+
+ if (m->speculative) {
+ DTRACE_VM2(pgrec, int, 1, (uint64_t *), NULL);
+ DTRACE_VM2(pgfrec, int, 1, (uint64_t *), NULL);
+ }
+
+ vm_page_queues_remove(m);
+
+ if ( !VM_PAGE_WIRED(m)) {
+ vm_page_check_pageable_safe(m);
+ if (!VM_DYNAMIC_PAGING_ENABLED(memory_manager_default) &&
+ m->dirty && m->object->internal &&
+ (m->object->purgable == VM_PURGABLE_DENY ||
+ m->object->purgable == VM_PURGABLE_NONVOLATILE ||
+ m->object->purgable == VM_PURGABLE_VOLATILE)) {
+ queue_enter(&vm_page_queue_throttled, m, vm_page_t, pageq);
+ m->throttled = TRUE;
+ vm_page_throttled_count++;
+ } else {
+ queue_enter(&vm_page_queue_active, m, vm_page_t, pageq);
+ m->active = TRUE;
+ vm_page_active_count++;
+ if (m->object->internal) {
+ vm_page_pageable_internal_count++;
+ } else {
+ vm_page_pageable_external_count++;
+ }
+ }
+ m->reference = TRUE;
+ m->no_cache = FALSE;
+ }
+ VM_PAGE_CHECK(m);
+}
+
+
+/*
+ * vm_page_speculate:
+ *
+ * Put the specified page on the speculative list (if appropriate).
+ *
+ * The page queues must be locked.
+ */
+void
+vm_page_speculate(
+ vm_page_t m,
+ boolean_t new)
+{
+ struct vm_speculative_age_q *aq;
+
+ VM_PAGE_CHECK(m);
+ vm_page_check_pageable_safe(m);
+
+ assert(m->phys_page != vm_page_guard_addr);
+#if DEBUG
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
+#endif
+ assert( !(m->absent && !m->unusual));
+
+ /*
+ * if this page is currently on the pageout queue, we can't do the
+ * vm_page_queues_remove (which doesn't handle the pageout queue case)
+ * and we can't remove it manually since we would need the object lock
+ * (which is not required here) to decrement the activity_in_progress
+ * reference which is held on the object while the page is in the pageout queue...
+ * just let the normal laundry processing proceed
+ */
+ if (m->laundry || m->pageout_queue || m->private || m->fictitious || m->compressor)
return;
- if (m->inactive) {
- assert(!m->laundry);
- if (m->zero_fill) {
- queue_remove(&vm_page_queue_zf, m, vm_page_t, pageq);
+ vm_page_queues_remove(m);
+
+ if ( !VM_PAGE_WIRED(m)) {
+ mach_timespec_t ts;
+ clock_sec_t sec;
+ clock_nsec_t nsec;
+
+ clock_get_system_nanotime(&sec, &nsec);
+ ts.tv_sec = (unsigned int) sec;
+ ts.tv_nsec = nsec;
+
+ if (vm_page_speculative_count == 0) {
+
+ speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
+ speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
+
+ aq = &vm_page_queue_speculative[speculative_age_index];
+
+ /*
+ * set the timer to begin a new group
+ */
+ aq->age_ts.tv_sec = vm_page_speculative_q_age_ms / 1000;
+ aq->age_ts.tv_nsec = (vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
+
+ ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
+ } else {
+ aq = &vm_page_queue_speculative[speculative_age_index];
+
+ if (CMP_MACH_TIMESPEC(&ts, &aq->age_ts) >= 0) {
+
+ speculative_age_index++;
+
+ if (speculative_age_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q)
+ speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
+ if (speculative_age_index == speculative_steal_index) {
+ speculative_steal_index = speculative_age_index + 1;
+
+ if (speculative_steal_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q)
+ speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
+ }
+ aq = &vm_page_queue_speculative[speculative_age_index];
+
+ if (!queue_empty(&aq->age_q))
+ vm_page_speculate_ageit(aq);
+
+ aq->age_ts.tv_sec = vm_page_speculative_q_age_ms / 1000;
+ aq->age_ts.tv_nsec = (vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
+
+ ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
+ }
+ }
+ enqueue_tail(&aq->age_q, &m->pageq);
+ m->speculative = TRUE;
+ vm_page_speculative_count++;
+ if (m->object->internal) {
+ vm_page_pageable_internal_count++;
} else {
- queue_remove(&vm_page_queue_inactive,
- m, vm_page_t, pageq);
+ vm_page_pageable_external_count++;
+ }
+
+ if (new == TRUE) {
+ vm_object_lock_assert_exclusive(m->object);
+
+ m->object->pages_created++;
+#if DEVELOPMENT || DEBUG
+ vm_page_speculative_created++;
+#endif
}
- m->pageq.next = NULL;
- m->pageq.prev = NULL;
- if (!m->fictitious)
- vm_page_inactive_count--;
- m->inactive = FALSE;
}
- if (m->wire_count == 0) {
+ VM_PAGE_CHECK(m);
+}
+
+
+/*
+ * move pages from the specified aging bin to
+ * the speculative bin that pageout_scan claims from
+ *
+ * The page queues must be locked.
+ */
+void
+vm_page_speculate_ageit(struct vm_speculative_age_q *aq)
+{
+ struct vm_speculative_age_q *sq;
+ vm_page_t t;
+
+ sq = &vm_page_queue_speculative[VM_PAGE_SPECULATIVE_AGED_Q];
+
+ if (queue_empty(&sq->age_q)) {
+ sq->age_q.next = aq->age_q.next;
+ sq->age_q.prev = aq->age_q.prev;
+
+ t = (vm_page_t)sq->age_q.next;
+ t->pageq.prev = &sq->age_q;
+
+ t = (vm_page_t)sq->age_q.prev;
+ t->pageq.next = &sq->age_q;
+ } else {
+ t = (vm_page_t)sq->age_q.prev;
+ t->pageq.next = aq->age_q.next;
+
+ t = (vm_page_t)aq->age_q.next;
+ t->pageq.prev = sq->age_q.prev;
+
+ t = (vm_page_t)aq->age_q.prev;
+ t->pageq.next = &sq->age_q;
+
+ sq->age_q.prev = aq->age_q.prev;
+ }
+ queue_init(&aq->age_q);
+}
+
+
+void
+vm_page_lru(
+ vm_page_t m)
+{
+ VM_PAGE_CHECK(m);
+ assert(m->object != kernel_object);
+ assert(m->phys_page != vm_page_guard_addr);
+
#if DEBUG
- if (m->active)
- panic("vm_page_activate: already active");
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
#endif
- assert(!m->laundry);
- assert(m->pageq.next == NULL && m->pageq.prev == NULL);
- queue_enter(&vm_page_queue_active, m, vm_page_t, pageq);
- m->active = TRUE;
- m->reference = TRUE;
- if (!m->fictitious)
- vm_page_active_count++;
+ /*
+ * if this page is currently on the pageout queue, we can't do the
+ * vm_page_queues_remove (which doesn't handle the pageout queue case)
+ * and we can't remove it manually since we would need the object lock
+ * (which is not required here) to decrement the activity_in_progress
+ * reference which is held on the object while the page is in the pageout queue...
+ * just let the normal laundry processing proceed
+ */
+ if (m->laundry || m->pageout_queue || m->private || m->compressor || (VM_PAGE_WIRED(m)))
+ return;
+
+ m->no_cache = FALSE;
+
+ vm_page_queues_remove(m);
+
+ vm_page_enqueue_inactive(m, FALSE);
+}
+
+
+void
+vm_page_reactivate_all_throttled(void)
+{
+ vm_page_t first_throttled, last_throttled;
+ vm_page_t first_active;
+ vm_page_t m;
+ int extra_active_count;
+ int extra_internal_count, extra_external_count;
+
+ if (!VM_DYNAMIC_PAGING_ENABLED(memory_manager_default))
+ return;
+
+ extra_active_count = 0;
+ extra_internal_count = 0;
+ extra_external_count = 0;
+ vm_page_lock_queues();
+ if (! queue_empty(&vm_page_queue_throttled)) {
+ /*
+ * Switch "throttled" pages to "active".
+ */
+ queue_iterate(&vm_page_queue_throttled, m, vm_page_t, pageq) {
+ VM_PAGE_CHECK(m);
+ assert(m->throttled);
+ assert(!m->active);
+ assert(!m->inactive);
+ assert(!m->speculative);
+ assert(!VM_PAGE_WIRED(m));
+
+ extra_active_count++;
+ if (m->object->internal) {
+ extra_internal_count++;
+ } else {
+ extra_external_count++;
+ }
+
+ m->throttled = FALSE;
+ m->active = TRUE;
+ VM_PAGE_CHECK(m);
+ }
+
+ /*
+ * Transfer the entire throttled queue to a regular LRU page queues.
+ * We insert it at the head of the active queue, so that these pages
+ * get re-evaluated by the LRU algorithm first, since they've been
+ * completely out of it until now.
+ */
+ first_throttled = (vm_page_t) queue_first(&vm_page_queue_throttled);
+ last_throttled = (vm_page_t) queue_last(&vm_page_queue_throttled);
+ first_active = (vm_page_t) queue_first(&vm_page_queue_active);
+ if (queue_empty(&vm_page_queue_active)) {
+ queue_last(&vm_page_queue_active) = (queue_entry_t) last_throttled;
+ } else {
+ queue_prev(&first_active->pageq) = (queue_entry_t) last_throttled;
+ }
+ queue_first(&vm_page_queue_active) = (queue_entry_t) first_throttled;
+ queue_prev(&first_throttled->pageq) = (queue_entry_t) &vm_page_queue_active;
+ queue_next(&last_throttled->pageq) = (queue_entry_t) first_active;
+
+#if DEBUG
+ printf("reactivated %d throttled pages\n", vm_page_throttled_count);
+#endif
+ queue_init(&vm_page_queue_throttled);
+ /*
+ * Adjust the global page counts.
+ */
+ vm_page_active_count += extra_active_count;
+ vm_page_pageable_internal_count += extra_internal_count;
+ vm_page_pageable_external_count += extra_external_count;
+ vm_page_throttled_count = 0;
+ }
+ assert(vm_page_throttled_count == 0);
+ assert(queue_empty(&vm_page_queue_throttled));
+ vm_page_unlock_queues();
+}
+
+
+/*
+ * move pages from the indicated local queue to the global active queue
+ * its ok to fail if we're below the hard limit and force == FALSE
+ * the nolocks == TRUE case is to allow this function to be run on
+ * the hibernate path
+ */
+
+void
+vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks)
+{
+ struct vpl *lq;
+ vm_page_t first_local, last_local;
+ vm_page_t first_active;
+ vm_page_t m;
+ uint32_t count = 0;
+
+ if (vm_page_local_q == NULL)
+ return;
+
+ lq = &vm_page_local_q[lid].vpl_un.vpl;
+
+ if (nolocks == FALSE) {
+ if (lq->vpl_count < vm_page_local_q_hard_limit && force == FALSE) {
+ if ( !vm_page_trylockspin_queues())
+ return;
+ } else
+ vm_page_lockspin_queues();
+
+ VPL_LOCK(&lq->vpl_lock);
+ }
+ if (lq->vpl_count) {
+ /*
+ * Switch "local" pages to "active".
+ */
+ assert(!queue_empty(&lq->vpl_queue));
+
+ queue_iterate(&lq->vpl_queue, m, vm_page_t, pageq) {
+ VM_PAGE_CHECK(m);
+ vm_page_check_pageable_safe(m);
+ assert(m->local);
+ assert(!m->active);
+ assert(!m->inactive);
+ assert(!m->speculative);
+ assert(!VM_PAGE_WIRED(m));
+ assert(!m->throttled);
+ assert(!m->fictitious);
+
+ if (m->local_id != lid)
+ panic("vm_page_reactivate_local: found vm_page_t(%p) with wrong cpuid", m);
+
+ m->local_id = 0;
+ m->local = FALSE;
+ m->active = TRUE;
+ VM_PAGE_CHECK(m);
+
+ count++;
+ }
+ if (count != lq->vpl_count)
+ panic("vm_page_reactivate_local: count = %d, vm_page_local_count = %d\n", count, lq->vpl_count);
+
+ /*
+ * Transfer the entire local queue to a regular LRU page queues.
+ */
+ first_local = (vm_page_t) queue_first(&lq->vpl_queue);
+ last_local = (vm_page_t) queue_last(&lq->vpl_queue);
+ first_active = (vm_page_t) queue_first(&vm_page_queue_active);
+
+ if (queue_empty(&vm_page_queue_active)) {
+ queue_last(&vm_page_queue_active) = (queue_entry_t) last_local;
+ } else {
+ queue_prev(&first_active->pageq) = (queue_entry_t) last_local;
+ }
+ queue_first(&vm_page_queue_active) = (queue_entry_t) first_local;
+ queue_prev(&first_local->pageq) = (queue_entry_t) &vm_page_queue_active;
+ queue_next(&last_local->pageq) = (queue_entry_t) first_active;
+
+ queue_init(&lq->vpl_queue);
+ /*
+ * Adjust the global page counts.
+ */
+ vm_page_active_count += lq->vpl_count;
+ vm_page_pageable_internal_count += lq->vpl_internal_count;
+ vm_page_pageable_external_count += lq->vpl_external_count;
+ lq->vpl_count = 0;
+ lq->vpl_internal_count = 0;
+ lq->vpl_external_count = 0;
+ }
+ assert(queue_empty(&lq->vpl_queue));
+
+ if (nolocks == FALSE) {
+ VPL_UNLOCK(&lq->vpl_lock);
+ vm_page_unlock_queues();
}
}
*
* Zero-fill a part of the page.
*/
+#define PMAP_ZERO_PART_PAGE_IMPLEMENTED
void
vm_page_part_zero_fill(
vm_page_t m,
vm_offset_t m_pa,
vm_size_t len)
{
- vm_page_t tmp;
+#if 0
+ /*
+ * we don't hold the page queue lock
+ * so this check isn't safe to make
+ */
VM_PAGE_CHECK(m);
+#endif
+
#ifdef PMAP_ZERO_PART_PAGE_IMPLEMENTED
pmap_zero_part_page(m->phys_page, m_pa, len);
#else
+ vm_page_t tmp;
while (1) {
tmp = vm_page_grab();
if (tmp == VM_PAGE_NULL) {
m_pa + len, PAGE_SIZE - (m_pa + len));
}
vm_page_copy(tmp,m);
- vm_page_lock_queues();
- vm_page_free(tmp);
- vm_page_unlock_queues();
+ VM_PAGE_FREE(tmp);
#endif
}
{
XPR(XPR_VM_PAGE,
"vm_page_zero_fill, object 0x%X offset 0x%X page 0x%X\n",
- (integer_t)m->object, (integer_t)m->offset, (integer_t)m, 0,0);
-
+ m->object, m->offset, m, 0,0);
+#if 0
+ /*
+ * we don't hold the page queue lock
+ * so this check isn't safe to make
+ */
VM_PAGE_CHECK(m);
+#endif
// dbgTrace(0xAEAEAEAE, m->phys_page, 0); /* (BRINGUP) */
pmap_zero_page(m->phys_page);
vm_offset_t dst_pa,
vm_size_t len)
{
+#if 0
+ /*
+ * we don't hold the page queue lock
+ * so this check isn't safe to make
+ */
VM_PAGE_CHECK(src_m);
VM_PAGE_CHECK(dst_m);
-
+#endif
pmap_copy_part_page(src_m->phys_page, src_pa,
dst_m->phys_page, dst_pa, len);
}
* make sure the page is decrypted first, if necessary.
*/
+int vm_page_copy_cs_validations = 0;
+int vm_page_copy_cs_tainted = 0;
+
void
vm_page_copy(
vm_page_t src_m,
{
XPR(XPR_VM_PAGE,
"vm_page_copy, object 0x%X offset 0x%X to object 0x%X offset 0x%X\n",
- (integer_t)src_m->object, src_m->offset,
- (integer_t)dest_m->object, dest_m->offset,
+ src_m->object, src_m->offset,
+ dest_m->object, dest_m->offset,
0);
-
+#if 0
+ /*
+ * we don't hold the page queue lock
+ * so this check isn't safe to make
+ */
VM_PAGE_CHECK(src_m);
VM_PAGE_CHECK(dest_m);
+#endif
+ vm_object_lock_assert_held(src_m->object);
/*
* ENCRYPTED SWAP:
}
dest_m->encrypted = FALSE;
+ if (src_m->object != VM_OBJECT_NULL &&
+ src_m->object->code_signed) {
+ /*
+ * We're copying a page from a code-signed object.
+ * Whoever ends up mapping the copy page might care about
+ * the original page's integrity, so let's validate the
+ * source page now.
+ */
+ vm_page_copy_cs_validations++;
+ vm_page_validate_cs(src_m);
+ }
+
+ if (vm_page_is_slideable(src_m)) {
+ boolean_t was_busy = src_m->busy;
+ src_m->busy = TRUE;
+ (void) vm_page_slide(src_m, 0);
+ assert(src_m->busy);
+ if (!was_busy) {
+ PAGE_WAKEUP_DONE(src_m);
+ }
+ }
+
+ /*
+ * Propagate the cs_tainted bit to the copy page. Do not propagate
+ * the cs_validated bit.
+ */
+ dest_m->cs_tainted = src_m->cs_tainted;
+ if (dest_m->cs_tainted) {
+ vm_page_copy_cs_tainted++;
+ }
+ dest_m->slid = src_m->slid;
+ dest_m->error = src_m->error; /* sliding src_m might have failed... */
pmap_copy_page(src_m->phys_page, dest_m->phys_page);
}
-/*
- * Currently, this is a primitive allocator that grabs
- * free pages from the system, sorts them by physical
- * address, then searches for a region large enough to
- * satisfy the user's request.
- *
- * Additional levels of effort:
- * + steal clean active/inactive pages
- * + force pageouts of dirty pages
- * + maintain a map of available physical
- * memory
- */
+#if MACH_ASSERT
+static void
+_vm_page_print(
+ vm_page_t p)
+{
+ printf("vm_page %p: \n", p);
+ printf(" pageq: next=%p prev=%p\n", p->pageq.next, p->pageq.prev);
+ printf(" listq: next=%p prev=%p\n", p->listq.next, p->listq.prev);
+ printf(" next=%p\n", VM_PAGE_UNPACK_PTR(p->next_m));
+ printf(" object=%p offset=0x%llx\n", p->object, p->offset);
+ printf(" wire_count=%u\n", p->wire_count);
+
+ printf(" %slocal, %sinactive, %sactive, %spageout_queue, %sspeculative, %slaundry\n",
+ (p->local ? "" : "!"),
+ (p->inactive ? "" : "!"),
+ (p->active ? "" : "!"),
+ (p->pageout_queue ? "" : "!"),
+ (p->speculative ? "" : "!"),
+ (p->laundry ? "" : "!"));
+ printf(" %sfree, %sref, %sgobbled, %sprivate, %sthrottled\n",
+ (p->free ? "" : "!"),
+ (p->reference ? "" : "!"),
+ (p->gobbled ? "" : "!"),
+ (p->private ? "" : "!"),
+ (p->throttled ? "" : "!"));
+ printf(" %sbusy, %swanted, %stabled, %sfictitious, %spmapped, %swpmapped\n",
+ (p->busy ? "" : "!"),
+ (p->wanted ? "" : "!"),
+ (p->tabled ? "" : "!"),
+ (p->fictitious ? "" : "!"),
+ (p->pmapped ? "" : "!"),
+ (p->wpmapped ? "" : "!"));
+ printf(" %spageout, %sabsent, %serror, %sdirty, %scleaning, %sprecious, %sclustered\n",
+ (p->pageout ? "" : "!"),
+ (p->absent ? "" : "!"),
+ (p->error ? "" : "!"),
+ (p->dirty ? "" : "!"),
+ (p->cleaning ? "" : "!"),
+ (p->precious ? "" : "!"),
+ (p->clustered ? "" : "!"));
+ printf(" %soverwriting, %srestart, %sunusual, %sencrypted, %sencrypted_cleaning\n",
+ (p->overwriting ? "" : "!"),
+ (p->restart ? "" : "!"),
+ (p->unusual ? "" : "!"),
+ (p->encrypted ? "" : "!"),
+ (p->encrypted_cleaning ? "" : "!"));
+ printf(" %scs_validated, %scs_tainted, %scs_nx, %sno_cache\n",
+ (p->cs_validated ? "" : "!"),
+ (p->cs_tainted ? "" : "!"),
+ (p->cs_nx ? "" : "!"),
+ (p->no_cache ? "" : "!"));
+
+ printf("phys_page=0x%x\n", p->phys_page);
+}
-#if MACH_ASSERT
/*
* Check that the list of pages is ordered by
* ascending physical address and has no holes.
*/
-int vm_page_verify_contiguous(
- vm_page_t pages,
- unsigned int npages);
-
-int
+static int
vm_page_verify_contiguous(
vm_page_t pages,
unsigned int npages)
page_count = 1;
for (m = NEXT_PAGE(pages); m != VM_PAGE_NULL; m = NEXT_PAGE(m)) {
if (m->phys_page != prev_addr + 1) {
- printf("m 0x%x prev_addr 0x%x, current addr 0x%x\n",
- m, prev_addr, m->phys_page);
- printf("pages 0x%x page_count %d\n", pages, page_count);
+ printf("m %p prev_addr 0x%lx, current addr 0x%x\n",
+ m, (long)prev_addr, m->phys_page);
+ printf("pages %p page_count %d npages %d\n", pages, page_count, npages);
panic("vm_page_verify_contiguous: not contiguous!");
}
prev_addr = m->phys_page;
++page_count;
}
if (page_count != npages) {
- printf("pages 0x%x actual count 0x%x but requested 0x%x\n",
+ printf("pages %p actual count 0x%x but requested 0x%x\n",
pages, page_count, npages);
panic("vm_page_verify_contiguous: count error");
}
return 1;
}
-#endif /* MACH_ASSERT */
-cpm_counter(unsigned int vpfls_pages_handled = 0;)
-cpm_counter(unsigned int vpfls_head_insertions = 0;)
-cpm_counter(unsigned int vpfls_tail_insertions = 0;)
-cpm_counter(unsigned int vpfls_general_insertions = 0;)
-cpm_counter(unsigned int vpfc_failed = 0;)
-cpm_counter(unsigned int vpfc_satisfied = 0;)
-
/*
- * Find a region large enough to contain at least npages
- * of contiguous physical memory.
- *
- * Requirements:
- * - Called while holding vm_page_queue_free_lock.
- * - Doesn't respect vm_page_free_reserved; caller
- * must not ask for more pages than are legal to grab.
- *
- * Returns a pointer to a list of gobbled pages or VM_PAGE_NULL.
- *
- * Algorithm:
- * Loop over the free list, extracting one page at a time and
- * inserting those into a sorted sub-list. We stop as soon as
- * there's a contiguous range within the sorted list that can
- * satisfy the contiguous memory request. This contiguous sub-
- * list is chopped out of the sorted sub-list and the remainder
- * of the sorted sub-list is put back onto the beginning of the
- * free list.
+ * Check the free lists for proper length etc.
*/
-static vm_page_t
-vm_page_find_contiguous(
- unsigned int contig_pages)
+static boolean_t vm_page_verify_this_free_list_enabled = FALSE;
+static unsigned int
+vm_page_verify_free_list(
+ queue_head_t *vm_page_queue,
+ unsigned int color,
+ vm_page_t look_for_page,
+ boolean_t expect_page)
{
- vm_page_t sort_list;
- vm_page_t *contfirstprev, contlast;
- vm_page_t m, m1;
- ppnum_t prevcontaddr;
- ppnum_t nextcontaddr;
- unsigned int npages;
+ unsigned int npages;
+ vm_page_t m;
+ vm_page_t prev_m;
+ boolean_t found_page;
- m = NULL;
-#if DEBUG
- _mutex_assert(&vm_page_queue_free_lock, MA_OWNED);
-#endif
-#if MACH_ASSERT
- /*
- * Verify pages in the free list..
- */
+ if (! vm_page_verify_this_free_list_enabled)
+ return 0;
+
+ found_page = FALSE;
npages = 0;
- for (m = vm_page_queue_free; m != VM_PAGE_NULL; m = NEXT_PAGE(m))
+ prev_m = (vm_page_t) vm_page_queue;
+ queue_iterate(vm_page_queue,
+ m,
+ vm_page_t,
+ pageq) {
+
+ if (m == look_for_page) {
+ found_page = TRUE;
+ }
+ if ((vm_page_t) m->pageq.prev != prev_m)
+ panic("vm_page_verify_free_list(color=%u, npages=%u): page %p corrupted prev ptr %p instead of %p\n",
+ color, npages, m, m->pageq.prev, prev_m);
+ if ( ! m->busy )
+ panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not busy\n",
+ color, npages, m);
+ if (color != (unsigned int) -1) {
+ if ((m->phys_page & vm_color_mask) != color)
+ panic("vm_page_verify_free_list(color=%u, npages=%u): page %p wrong color %u instead of %u\n",
+ color, npages, m, m->phys_page & vm_color_mask, color);
+ if ( ! m->free )
+ panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not free\n",
+ color, npages, m);
+ }
++npages;
- if (npages != vm_page_free_count)
- panic("vm_sort_free_list: prelim: npages %u free_count %d",
- npages, vm_page_free_count);
+ prev_m = m;
+ }
+ if (look_for_page != VM_PAGE_NULL) {
+ unsigned int other_color;
+
+ if (expect_page && !found_page) {
+ printf("vm_page_verify_free_list(color=%u, npages=%u): page %p not found phys=%u\n",
+ color, npages, look_for_page, look_for_page->phys_page);
+ _vm_page_print(look_for_page);
+ for (other_color = 0;
+ other_color < vm_colors;
+ other_color++) {
+ if (other_color == color)
+ continue;
+ vm_page_verify_free_list(&vm_page_queue_free[other_color],
+ other_color, look_for_page, FALSE);
+ }
+ if (color == (unsigned int) -1) {
+ vm_page_verify_free_list(&vm_lopage_queue_free,
+ (unsigned int) -1, look_for_page, FALSE);
+ }
+ panic("vm_page_verify_free_list(color=%u)\n", color);
+ }
+ if (!expect_page && found_page) {
+ printf("vm_page_verify_free_list(color=%u, npages=%u): page %p found phys=%u\n",
+ color, npages, look_for_page, look_for_page->phys_page);
+ }
+ }
+ return npages;
+}
+
+static boolean_t vm_page_verify_all_free_lists_enabled = FALSE;
+static void
+vm_page_verify_free_lists( void )
+{
+ unsigned int color, npages, nlopages;
+ boolean_t toggle = TRUE;
+
+ if (! vm_page_verify_all_free_lists_enabled)
+ return;
+
+ npages = 0;
+
+ lck_mtx_lock(&vm_page_queue_free_lock);
+
+ if (vm_page_verify_this_free_list_enabled == TRUE) {
+ /*
+ * This variable has been set globally for extra checking of
+ * each free list Q. Since we didn't set it, we don't own it
+ * and we shouldn't toggle it.
+ */
+ toggle = FALSE;
+ }
+
+ if (toggle == TRUE) {
+ vm_page_verify_this_free_list_enabled = TRUE;
+ }
+
+ for( color = 0; color < vm_colors; color++ ) {
+ npages += vm_page_verify_free_list(&vm_page_queue_free[color],
+ color, VM_PAGE_NULL, FALSE);
+ }
+ nlopages = vm_page_verify_free_list(&vm_lopage_queue_free,
+ (unsigned int) -1,
+ VM_PAGE_NULL, FALSE);
+ if (npages != vm_page_free_count || nlopages != vm_lopage_free_count)
+ panic("vm_page_verify_free_lists: "
+ "npages %u free_count %d nlopages %u lo_free_count %u",
+ npages, vm_page_free_count, nlopages, vm_lopage_free_count);
+
+ if (toggle == TRUE) {
+ vm_page_verify_this_free_list_enabled = FALSE;
+ }
+
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+}
+
+void
+vm_page_queues_assert(
+ vm_page_t mem,
+ int val)
+{
+#if DEBUG
+ lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
+#endif
+ if (mem->free + mem->active + mem->inactive + mem->speculative +
+ mem->throttled + mem->pageout_queue > (val)) {
+ _vm_page_print(mem);
+ panic("vm_page_queues_assert(%p, %d)\n", mem, val);
+ }
+ if (VM_PAGE_WIRED(mem)) {
+ assert(!mem->active);
+ assert(!mem->inactive);
+ assert(!mem->speculative);
+ assert(!mem->throttled);
+ assert(!mem->pageout_queue);
+ }
+}
#endif /* MACH_ASSERT */
- if (contig_pages == 0 || vm_page_queue_free == VM_PAGE_NULL)
- return VM_PAGE_NULL;
-#define PPNUM_PREV(x) (((x) > 0) ? ((x) - 1) : 0)
-#define PPNUM_NEXT(x) (((x) < PPNUM_MAX) ? ((x) + 1) : PPNUM_MAX)
-#define SET_NEXT_PAGE(m,n) ((m)->pageq.next = (struct queue_entry *) (n))
- npages = 1;
- contfirstprev = &sort_list;
- contlast = sort_list = vm_page_queue_free;
- vm_page_queue_free = NEXT_PAGE(sort_list);
- SET_NEXT_PAGE(sort_list, VM_PAGE_NULL);
- prevcontaddr = PPNUM_PREV(sort_list->phys_page);
- nextcontaddr = PPNUM_NEXT(sort_list->phys_page);
- while (npages < contig_pages &&
- (m = vm_page_queue_free) != VM_PAGE_NULL)
- {
- cpm_counter(++vpfls_pages_handled);
- /* prepend to existing run? */
- if (m->phys_page == prevcontaddr)
- {
- vm_page_queue_free = NEXT_PAGE(m);
- cpm_counter(++vpfls_head_insertions);
- prevcontaddr = PPNUM_PREV(prevcontaddr);
- SET_NEXT_PAGE(m, *contfirstprev);
- *contfirstprev = m;
- npages++;
- continue; /* no tail expansion check needed */
- }
-
- /* append to tail of existing run? */
- else if (m->phys_page == nextcontaddr)
- {
- vm_page_queue_free = NEXT_PAGE(m);
- cpm_counter(++vpfls_tail_insertions);
- nextcontaddr = PPNUM_NEXT(nextcontaddr);
- SET_NEXT_PAGE(m, NEXT_PAGE(contlast));
- SET_NEXT_PAGE(contlast, m);
- contlast = m;
- npages++;
- }
+extern boolean_t (* volatile consider_buffer_cache_collect)(int);
+
+/*
+ * CONTIGUOUS PAGE ALLOCATION
+ *
+ * Find a region large enough to contain at least n pages
+ * of contiguous physical memory.
+ *
+ * This is done by traversing the vm_page_t array in a linear fashion
+ * we assume that the vm_page_t array has the avaiable physical pages in an
+ * ordered, ascending list... this is currently true of all our implementations
+ * and must remain so... there can be 'holes' in the array... we also can
+ * no longer tolerate the vm_page_t's in the list being 'freed' and reclaimed
+ * which use to happen via 'vm_page_convert'... that function was no longer
+ * being called and was removed...
+ *
+ * The basic flow consists of stabilizing some of the interesting state of
+ * a vm_page_t behind the vm_page_queue and vm_page_free locks... we start our
+ * sweep at the beginning of the array looking for pages that meet our criterea
+ * for a 'stealable' page... currently we are pretty conservative... if the page
+ * meets this criterea and is physically contiguous to the previous page in the 'run'
+ * we keep developing it. If we hit a page that doesn't fit, we reset our state
+ * and start to develop a new run... if at this point we've already considered
+ * at least MAX_CONSIDERED_BEFORE_YIELD pages, we'll drop the 2 locks we hold,
+ * and mutex_pause (which will yield the processor), to keep the latency low w/r
+ * to other threads trying to acquire free pages (or move pages from q to q),
+ * and then continue from the spot we left off... we only make 1 pass through the
+ * array. Once we have a 'run' that is long enough, we'll go into the loop which
+ * which steals the pages from the queues they're currently on... pages on the free
+ * queue can be stolen directly... pages that are on any of the other queues
+ * must be removed from the object they are tabled on... this requires taking the
+ * object lock... we do this as a 'try' to prevent deadlocks... if the 'try' fails
+ * or if the state of the page behind the vm_object lock is no longer viable, we'll
+ * dump the pages we've currently stolen back to the free list, and pick up our
+ * scan from the point where we aborted the 'current' run.
+ *
+ *
+ * Requirements:
+ * - neither vm_page_queue nor vm_free_list lock can be held on entry
+ *
+ * Returns a pointer to a list of gobbled/wired pages or VM_PAGE_NULL.
+ *
+ * Algorithm:
+ */
+
+#define MAX_CONSIDERED_BEFORE_YIELD 1000
+
+
+#define RESET_STATE_OF_RUN() \
+ MACRO_BEGIN \
+ prevcontaddr = -2; \
+ start_pnum = -1; \
+ free_considered = 0; \
+ substitute_needed = 0; \
+ npages = 0; \
+ MACRO_END
+
+/*
+ * Can we steal in-use (i.e. not free) pages when searching for
+ * physically-contiguous pages ?
+ */
+#define VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL 1
+
+static unsigned int vm_page_find_contiguous_last_idx = 0, vm_page_lomem_find_contiguous_last_idx = 0;
+#if DEBUG
+int vm_page_find_contig_debug = 0;
+#endif
+
+static vm_page_t
+vm_page_find_contiguous(
+ unsigned int contig_pages,
+ ppnum_t max_pnum,
+ ppnum_t pnum_mask,
+ boolean_t wire,
+ int flags)
+{
+ vm_page_t m = NULL;
+ ppnum_t prevcontaddr;
+ ppnum_t start_pnum;
+ unsigned int npages, considered, scanned;
+ unsigned int page_idx, start_idx, last_idx, orig_last_idx;
+ unsigned int idx_last_contig_page_found = 0;
+ int free_considered, free_available;
+ int substitute_needed;
+ boolean_t wrapped, zone_gc_called = FALSE;
+#if DEBUG
+ clock_sec_t tv_start_sec, tv_end_sec;
+ clock_usec_t tv_start_usec, tv_end_usec;
+#endif
+
+ int yielded = 0;
+ int dumped_run = 0;
+ int stolen_pages = 0;
+ int compressed_pages = 0;
+
+
+ if (contig_pages == 0)
+ return VM_PAGE_NULL;
+
+full_scan_again:
+
+#if MACH_ASSERT
+ vm_page_verify_free_lists();
+#endif
+#if DEBUG
+ clock_get_system_microtime(&tv_start_sec, &tv_start_usec);
+#endif
+ PAGE_REPLACEMENT_ALLOWED(TRUE);
+
+ vm_page_lock_queues();
+
+
+ lck_mtx_lock(&vm_page_queue_free_lock);
+
+ RESET_STATE_OF_RUN();
+
+ scanned = 0;
+ considered = 0;
+ free_available = vm_page_free_count - vm_page_free_reserved;
+
+ wrapped = FALSE;
+
+ if(flags & KMA_LOMEM)
+ idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx;
+ else
+ idx_last_contig_page_found = vm_page_find_contiguous_last_idx;
+
+ orig_last_idx = idx_last_contig_page_found;
+ last_idx = orig_last_idx;
+
+ for (page_idx = last_idx, start_idx = last_idx;
+ npages < contig_pages && page_idx < vm_pages_count;
+ page_idx++) {
+retry:
+ if (wrapped &&
+ npages == 0 &&
+ page_idx >= orig_last_idx) {
+ /*
+ * We're back where we started and we haven't
+ * found any suitable contiguous range. Let's
+ * give up.
+ */
+ break;
+ }
+ scanned++;
+ m = &vm_pages[page_idx];
+
+ assert(!m->fictitious);
+ assert(!m->private);
+
+ if (max_pnum && m->phys_page > max_pnum) {
+ /* no more low pages... */
+ break;
+ }
+ if (!npages & ((m->phys_page & pnum_mask) != 0)) {
+ /*
+ * not aligned
+ */
+ RESET_STATE_OF_RUN();
+
+ } else if (VM_PAGE_WIRED(m) || m->gobbled ||
+ m->encrypted_cleaning ||
+ m->pageout_queue || m->laundry || m->wanted ||
+ m->cleaning || m->overwriting || m->pageout) {
+ /*
+ * page is in a transient state
+ * or a state we don't want to deal
+ * with, so don't consider it which
+ * means starting a new run
+ */
+ RESET_STATE_OF_RUN();
+
+ } else if (!m->free && !m->active && !m->inactive && !m->speculative && !m->throttled && !m->compressor) {
+ /*
+ * page needs to be on one of our queues
+ * or it needs to belong to the compressor pool
+ * in order for it to be stable behind the
+ * locks we hold at this point...
+ * if not, don't consider it which
+ * means starting a new run
+ */
+ RESET_STATE_OF_RUN();
+
+ } else if (!m->free && (!m->tabled || m->busy)) {
+ /*
+ * pages on the free list are always 'busy'
+ * so we couldn't test for 'busy' in the check
+ * for the transient states... pages that are
+ * 'free' are never 'tabled', so we also couldn't
+ * test for 'tabled'. So we check here to make
+ * sure that a non-free page is not busy and is
+ * tabled on an object...
+ * if not, don't consider it which
+ * means starting a new run
+ */
+ RESET_STATE_OF_RUN();
+
+ } else {
+ if (m->phys_page != prevcontaddr + 1) {
+ if ((m->phys_page & pnum_mask) != 0) {
+ RESET_STATE_OF_RUN();
+ goto did_consider;
+ } else {
+ npages = 1;
+ start_idx = page_idx;
+ start_pnum = m->phys_page;
+ }
+ } else {
+ npages++;
+ }
+ prevcontaddr = m->phys_page;
+
+ VM_PAGE_CHECK(m);
+ if (m->free) {
+ free_considered++;
+ } else {
+ /*
+ * This page is not free.
+ * If we can't steal used pages,
+ * we have to give up this run
+ * and keep looking.
+ * Otherwise, we might need to
+ * move the contents of this page
+ * into a substitute page.
+ */
+#if VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
+ if (m->pmapped || m->dirty || m->precious) {
+ substitute_needed++;
+ }
+#else
+ RESET_STATE_OF_RUN();
+#endif
+ }
+
+ if ((free_considered + substitute_needed) > free_available) {
+ /*
+ * if we let this run continue
+ * we will end up dropping the vm_page_free_count
+ * below the reserve limit... we need to abort
+ * this run, but we can at least re-consider this
+ * page... thus the jump back to 'retry'
+ */
+ RESET_STATE_OF_RUN();
+
+ if (free_available && considered <= MAX_CONSIDERED_BEFORE_YIELD) {
+ considered++;
+ goto retry;
+ }
+ /*
+ * free_available == 0
+ * so can't consider any free pages... if
+ * we went to retry in this case, we'd
+ * get stuck looking at the same page
+ * w/o making any forward progress
+ * we also want to take this path if we've already
+ * reached our limit that controls the lock latency
+ */
+ }
+ }
+did_consider:
+ if (considered > MAX_CONSIDERED_BEFORE_YIELD && npages <= 1) {
+
+ PAGE_REPLACEMENT_ALLOWED(FALSE);
+
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+ vm_page_unlock_queues();
+
+ mutex_pause(0);
+
+ PAGE_REPLACEMENT_ALLOWED(TRUE);
+
+ vm_page_lock_queues();
+ lck_mtx_lock(&vm_page_queue_free_lock);
+
+ RESET_STATE_OF_RUN();
+ /*
+ * reset our free page limit since we
+ * dropped the lock protecting the vm_page_free_queue
+ */
+ free_available = vm_page_free_count - vm_page_free_reserved;
+ considered = 0;
+
+ yielded++;
+
+ goto retry;
+ }
+ considered++;
+ }
+ m = VM_PAGE_NULL;
+
+ if (npages != contig_pages) {
+ if (!wrapped) {
+ /*
+ * We didn't find a contiguous range but we didn't
+ * start from the very first page.
+ * Start again from the very first page.
+ */
+ RESET_STATE_OF_RUN();
+ if( flags & KMA_LOMEM)
+ idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx = 0;
+ else
+ idx_last_contig_page_found = vm_page_find_contiguous_last_idx = 0;
+ last_idx = 0;
+ page_idx = last_idx;
+ wrapped = TRUE;
+ goto retry;
+ }
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+ } else {
+ vm_page_t m1;
+ vm_page_t m2;
+ unsigned int cur_idx;
+ unsigned int tmp_start_idx;
+ vm_object_t locked_object = VM_OBJECT_NULL;
+ boolean_t abort_run = FALSE;
+
+ assert(page_idx - start_idx == contig_pages);
+
+ tmp_start_idx = start_idx;
+
+ /*
+ * first pass through to pull the free pages
+ * off of the free queue so that in case we
+ * need substitute pages, we won't grab any
+ * of the free pages in the run... we'll clear
+ * the 'free' bit in the 2nd pass, and even in
+ * an abort_run case, we'll collect all of the
+ * free pages in this run and return them to the free list
+ */
+ while (start_idx < page_idx) {
+
+ m1 = &vm_pages[start_idx++];
+
+#if !VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
+ assert(m1->free);
+#endif
+
+ if (m1->free) {
+ unsigned int color;
+
+ color = m1->phys_page & vm_color_mask;
+#if MACH_ASSERT
+ vm_page_verify_free_list(&vm_page_queue_free[color], color, m1, TRUE);
+#endif
+ queue_remove(&vm_page_queue_free[color],
+ m1,
+ vm_page_t,
+ pageq);
+ m1->pageq.next = NULL;
+ m1->pageq.prev = NULL;
+#if MACH_ASSERT
+ vm_page_verify_free_list(&vm_page_queue_free[color], color, VM_PAGE_NULL, FALSE);
+#endif
+ /*
+ * Clear the "free" bit so that this page
+ * does not get considered for another
+ * concurrent physically-contiguous allocation.
+ */
+ m1->free = FALSE;
+ assert(m1->busy);
+
+ vm_page_free_count--;
+ }
+ }
+ if( flags & KMA_LOMEM)
+ vm_page_lomem_find_contiguous_last_idx = page_idx;
+ else
+ vm_page_find_contiguous_last_idx = page_idx;
+
+ /*
+ * we can drop the free queue lock at this point since
+ * we've pulled any 'free' candidates off of the list
+ * we need it dropped so that we can do a vm_page_grab
+ * when substituing for pmapped/dirty pages
+ */
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+
+ start_idx = tmp_start_idx;
+ cur_idx = page_idx - 1;
+
+ while (start_idx++ < page_idx) {
+ /*
+ * must go through the list from back to front
+ * so that the page list is created in the
+ * correct order - low -> high phys addresses
+ */
+ m1 = &vm_pages[cur_idx--];
+
+ assert(!m1->free);
+
+ if (m1->object == VM_OBJECT_NULL) {
+ /*
+ * page has already been removed from
+ * the free list in the 1st pass
+ */
+ assert(m1->offset == (vm_object_offset_t) -1);
+ assert(m1->busy);
+ assert(!m1->wanted);
+ assert(!m1->laundry);
+ } else {
+ vm_object_t object;
+ int refmod;
+ boolean_t disconnected, reusable;
+
+ if (abort_run == TRUE)
+ continue;
+
+ object = m1->object;
+
+ if (object != locked_object) {
+ if (locked_object) {
+ vm_object_unlock(locked_object);
+ locked_object = VM_OBJECT_NULL;
+ }
+ if (vm_object_lock_try(object))
+ locked_object = object;
+ }
+ if (locked_object == VM_OBJECT_NULL ||
+ (VM_PAGE_WIRED(m1) || m1->gobbled ||
+ m1->encrypted_cleaning ||
+ m1->pageout_queue || m1->laundry || m1->wanted ||
+ m1->cleaning || m1->overwriting || m1->pageout || m1->busy)) {
+
+ if (locked_object) {
+ vm_object_unlock(locked_object);
+ locked_object = VM_OBJECT_NULL;
+ }
+ tmp_start_idx = cur_idx;
+ abort_run = TRUE;
+ continue;
+ }
+
+ disconnected = FALSE;
+ reusable = FALSE;
+
+ if ((m1->reusable ||
+ m1->object->all_reusable) &&
+ m1->inactive &&
+ !m1->dirty &&
+ !m1->reference) {
+ /* reusable page... */
+ refmod = pmap_disconnect(m1->phys_page);
+ disconnected = TRUE;
+ if (refmod == 0) {
+ /*
+ * ... not reused: can steal
+ * without relocating contents.
+ */
+ reusable = TRUE;
+ }
+ }
+
+ if ((m1->pmapped &&
+ ! reusable) ||
+ m1->dirty ||
+ m1->precious) {
+ vm_object_offset_t offset;
+
+ m2 = vm_page_grab();
+
+ if (m2 == VM_PAGE_NULL) {
+ if (locked_object) {
+ vm_object_unlock(locked_object);
+ locked_object = VM_OBJECT_NULL;
+ }
+ tmp_start_idx = cur_idx;
+ abort_run = TRUE;
+ continue;
+ }
+ if (! disconnected) {
+ if (m1->pmapped)
+ refmod = pmap_disconnect(m1->phys_page);
+ else
+ refmod = 0;
+ }
+
+ /* copy the page's contents */
+ pmap_copy_page(m1->phys_page, m2->phys_page);
+ /* copy the page's state */
+ assert(!VM_PAGE_WIRED(m1));
+ assert(!m1->free);
+ assert(!m1->pageout_queue);
+ assert(!m1->laundry);
+ m2->reference = m1->reference;
+ assert(!m1->gobbled);
+ assert(!m1->private);
+ m2->no_cache = m1->no_cache;
+ m2->xpmapped = 0;
+ assert(!m1->busy);
+ assert(!m1->wanted);
+ assert(!m1->fictitious);
+ m2->pmapped = m1->pmapped; /* should flush cache ? */
+ m2->wpmapped = m1->wpmapped;
+ assert(!m1->pageout);
+ m2->absent = m1->absent;
+ m2->error = m1->error;
+ m2->dirty = m1->dirty;
+ assert(!m1->cleaning);
+ m2->precious = m1->precious;
+ m2->clustered = m1->clustered;
+ assert(!m1->overwriting);
+ m2->restart = m1->restart;
+ m2->unusual = m1->unusual;
+ m2->encrypted = m1->encrypted;
+ assert(!m1->encrypted_cleaning);
+ m2->cs_validated = m1->cs_validated;
+ m2->cs_tainted = m1->cs_tainted;
+ m2->cs_nx = m1->cs_nx;
+
+ /*
+ * If m1 had really been reusable,
+ * we would have just stolen it, so
+ * let's not propagate it's "reusable"
+ * bit and assert that m2 is not
+ * marked as "reusable".
+ */
+ // m2->reusable = m1->reusable;
+ assert(!m2->reusable);
+
+ assert(!m1->lopage);
+ m2->slid = m1->slid;
+ m2->compressor = m1->compressor;
+
+ /*
+ * page may need to be flushed if
+ * it is marshalled into a UPL
+ * that is going to be used by a device
+ * that doesn't support coherency
+ */
+ m2->written_by_kernel = TRUE;
+
+ /*
+ * make sure we clear the ref/mod state
+ * from the pmap layer... else we risk
+ * inheriting state from the last time
+ * this page was used...
+ */
+ pmap_clear_refmod(m2->phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED);
+
+ if (refmod & VM_MEM_REFERENCED)
+ m2->reference = TRUE;
+ if (refmod & VM_MEM_MODIFIED) {
+ SET_PAGE_DIRTY(m2, TRUE);
+ }
+ offset = m1->offset;
+
+ /*
+ * completely cleans up the state
+ * of the page so that it is ready
+ * to be put onto the free list, or
+ * for this purpose it looks like it
+ * just came off of the free list
+ */
+ vm_page_free_prepare(m1);
+
+ /*
+ * now put the substitute page
+ * on the object
+ */
+ vm_page_insert_internal(m2, locked_object, offset, VM_KERN_MEMORY_NONE, TRUE, TRUE, FALSE, FALSE, NULL);
+
+ if (m2->compressor) {
+ m2->pmapped = TRUE;
+ m2->wpmapped = TRUE;
+
+ PMAP_ENTER(kernel_pmap, m2->offset, m2,
+ VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, 0, TRUE);
+
+ compressed_pages++;
+
+ } else {
+ if (m2->reference)
+ vm_page_activate(m2);
+ else
+ vm_page_deactivate(m2);
+ }
+ PAGE_WAKEUP_DONE(m2);
+
+ } else {
+ assert(!m1->compressor);
+
+ /*
+ * completely cleans up the state
+ * of the page so that it is ready
+ * to be put onto the free list, or
+ * for this purpose it looks like it
+ * just came off of the free list
+ */
+ vm_page_free_prepare(m1);
+ }
+
+ stolen_pages++;
+
+ }
+ m1->pageq.next = (queue_entry_t) m;
+ m1->pageq.prev = NULL;
+ m = m1;
+ }
+ if (locked_object) {
+ vm_object_unlock(locked_object);
+ locked_object = VM_OBJECT_NULL;
+ }
+
+ if (abort_run == TRUE) {
+ if (m != VM_PAGE_NULL) {
+ vm_page_free_list(m, FALSE);
+ }
+
+ dumped_run++;
+
+ /*
+ * want the index of the last
+ * page in this run that was
+ * successfully 'stolen', so back
+ * it up 1 for the auto-decrement on use
+ * and 1 more to bump back over this page
+ */
+ page_idx = tmp_start_idx + 2;
+ if (page_idx >= vm_pages_count) {
+ if (wrapped)
+ goto done_scanning;
+ page_idx = last_idx = 0;
+ wrapped = TRUE;
+ }
+ abort_run = FALSE;
+
+ /*
+ * We didn't find a contiguous range but we didn't
+ * start from the very first page.
+ * Start again from the very first page.
+ */
+ RESET_STATE_OF_RUN();
+
+ if( flags & KMA_LOMEM)
+ idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx = page_idx;
+ else
+ idx_last_contig_page_found = vm_page_find_contiguous_last_idx = page_idx;
+
+ last_idx = page_idx;
+
+ lck_mtx_lock(&vm_page_queue_free_lock);
+ /*
+ * reset our free page limit since we
+ * dropped the lock protecting the vm_page_free_queue
+ */
+ free_available = vm_page_free_count - vm_page_free_reserved;
+ goto retry;
+ }
+
+ for (m1 = m; m1 != VM_PAGE_NULL; m1 = NEXT_PAGE(m1)) {
+
+ if (wire == TRUE)
+ m1->wire_count++;
+ else
+ m1->gobbled = TRUE;
+ }
+ if (wire == FALSE)
+ vm_page_gobble_count += npages;
+
+ /*
+ * gobbled pages are also counted as wired pages
+ */
+ vm_page_wire_count += npages;
+
+ assert(vm_page_verify_contiguous(m, npages));
+ }
+done_scanning:
+ PAGE_REPLACEMENT_ALLOWED(FALSE);
+
+ vm_page_unlock_queues();
+
+#if DEBUG
+ clock_get_system_microtime(&tv_end_sec, &tv_end_usec);
+
+ tv_end_sec -= tv_start_sec;
+ if (tv_end_usec < tv_start_usec) {
+ tv_end_sec--;
+ tv_end_usec += 1000000;
+ }
+ tv_end_usec -= tv_start_usec;
+ if (tv_end_usec >= 1000000) {
+ tv_end_sec++;
+ tv_end_sec -= 1000000;
+ }
+ if (vm_page_find_contig_debug) {
+ printf("%s(num=%d,low=%d): found %d pages at 0x%llx in %ld.%06ds... started at %d... scanned %d pages... yielded %d times... dumped run %d times... stole %d pages... stole %d compressed pages\n",
+ __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
+ (long)tv_end_sec, tv_end_usec, orig_last_idx,
+ scanned, yielded, dumped_run, stolen_pages, compressed_pages);
+ }
+
+#endif
+#if MACH_ASSERT
+ vm_page_verify_free_lists();
+#endif
+ if (m == NULL && zone_gc_called == FALSE) {
+ printf("%s(num=%d,low=%d): found %d pages at 0x%llx...scanned %d pages... yielded %d times... dumped run %d times... stole %d pages... stole %d compressed pages... wired count is %d\n",
+ __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
+ scanned, yielded, dumped_run, stolen_pages, compressed_pages, vm_page_wire_count);
+
+ if (consider_buffer_cache_collect != NULL) {
+ (void)(*consider_buffer_cache_collect)(1);
+ }
+
+ consider_zone_gc(TRUE);
+
+ zone_gc_called = TRUE;
+
+ printf("vm_page_find_contiguous: zone_gc called... wired count is %d\n", vm_page_wire_count);
+ goto full_scan_again;
+ }
+
+ return m;
+}
+
+/*
+ * Allocate a list of contiguous, wired pages.
+ */
+kern_return_t
+cpm_allocate(
+ vm_size_t size,
+ vm_page_t *list,
+ ppnum_t max_pnum,
+ ppnum_t pnum_mask,
+ boolean_t wire,
+ int flags)
+{
+ vm_page_t pages;
+ unsigned int npages;
+
+ if (size % PAGE_SIZE != 0)
+ return KERN_INVALID_ARGUMENT;
+
+ npages = (unsigned int) (size / PAGE_SIZE);
+ if (npages != size / PAGE_SIZE) {
+ /* 32-bit overflow */
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ /*
+ * Obtain a pointer to a subset of the free
+ * list large enough to satisfy the request;
+ * the region will be physically contiguous.
+ */
+ pages = vm_page_find_contiguous(npages, max_pnum, pnum_mask, wire, flags);
+
+ if (pages == VM_PAGE_NULL)
+ return KERN_NO_SPACE;
+ /*
+ * determine need for wakeups
+ */
+ if ((vm_page_free_count < vm_page_free_min) ||
+ ((vm_page_free_count < vm_page_free_target) &&
+ ((vm_page_inactive_count + vm_page_speculative_count) < vm_page_inactive_min)))
+ thread_wakeup((event_t) &vm_page_free_wanted);
+
+ VM_CHECK_MEMORYSTATUS;
+
+ /*
+ * The CPM pages should now be available and
+ * ordered by ascending physical address.
+ */
+ assert(vm_page_verify_contiguous(pages, npages));
+
+ *list = pages;
+ return KERN_SUCCESS;
+}
+
+
+unsigned int vm_max_delayed_work_limit = DEFAULT_DELAYED_WORK_LIMIT;
+
+/*
+ * when working on a 'run' of 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... in fact there are certain
+ * collections of pages that don't require any work brokered by the
+ * vm_page_queue_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...
+ * then call vm_page_do_delayed_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
+ * if it's not already held so that vm_page_do_delayed_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.
+ * the operation names are modeled after the names of the routines that
+ * need to be called in order to make the changes very obvious in the
+ * original loop
+ */
+
+void
+vm_page_do_delayed_work(
+ vm_object_t object,
+ vm_tag_t tag,
+ struct vm_page_delayed_work *dwp,
+ int dw_count)
+{
+ int j;
+ vm_page_t m;
+ vm_page_t local_free_q = VM_PAGE_NULL;
+
+ /*
+ * 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.
+ */
+ if (!vm_page_trylockspin_queues()) {
+ vm_object_unlock(object);
+
+ vm_page_lockspin_queues();
+
+ 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++) {
+
+ m = dwp->dw_m;
+
+ if (dwp->dw_mask & DW_vm_pageout_throttle_up)
+ vm_pageout_throttle_up(m);
+#if CONFIG_PHANTOM_CACHE
+ if (dwp->dw_mask & DW_vm_phantom_cache_update)
+ vm_phantom_cache_update(m);
+#endif
+ if (dwp->dw_mask & DW_vm_page_wire)
+ vm_page_wire(m, tag, FALSE);
+ else if (dwp->dw_mask & DW_vm_page_unwire) {
+ boolean_t queueit;
+
+ queueit = (dwp->dw_mask & (DW_vm_page_free | DW_vm_page_deactivate_internal)) ? FALSE : TRUE;
+
+ vm_page_unwire(m, queueit);
+ }
+ if (dwp->dw_mask & DW_vm_page_free) {
+ vm_page_free_prepare_queues(m);
+
+ assert(m->pageq.next == NULL && m->pageq.prev == NULL);
+ /*
+ * Add this page to our list of reclaimed pages,
+ * to be freed later.
+ */
+ m->pageq.next = (queue_entry_t) local_free_q;
+ local_free_q = m;
+ } else {
+ if (dwp->dw_mask & DW_vm_page_deactivate_internal)
+ vm_page_deactivate_internal(m, FALSE);
+ else if (dwp->dw_mask & DW_vm_page_activate) {
+ if (m->active == FALSE) {
+ vm_page_activate(m);
+ }
+ }
+ else if (dwp->dw_mask & DW_vm_page_speculate)
+ vm_page_speculate(m, TRUE);
+ else if (dwp->dw_mask & DW_enqueue_cleaned) {
+ /*
+ * if we didn't hold the object lock and did this,
+ * we might disconnect the page, then someone might
+ * soft fault it back in, then we would put it on the
+ * cleaned queue, and so we would have a referenced (maybe even dirty)
+ * page on that queue, which we don't want
+ */
+ int refmod_state = pmap_disconnect(m->phys_page);
+
+ if ((refmod_state & VM_MEM_REFERENCED)) {
+ /*
+ * this page has been touched since it got cleaned; let's activate it
+ * if it hasn't already been
+ */
+ vm_pageout_enqueued_cleaned++;
+ vm_pageout_cleaned_reactivated++;
+ vm_pageout_cleaned_commit_reactivated++;
+
+ if (m->active == FALSE)
+ vm_page_activate(m);
+ } else {
+ m->reference = FALSE;
+ vm_page_enqueue_cleaned(m);
+ }
+ }
+ else if (dwp->dw_mask & DW_vm_page_lru)
+ vm_page_lru(m);
+ else if (dwp->dw_mask & DW_VM_PAGE_QUEUES_REMOVE) {
+ if ( !m->pageout_queue)
+ vm_page_queues_remove(m);
+ }
+ if (dwp->dw_mask & DW_set_reference)
+ m->reference = TRUE;
+ else if (dwp->dw_mask & DW_clear_reference)
+ m->reference = FALSE;
+
+ if (dwp->dw_mask & DW_move_page) {
+ if ( !m->pageout_queue) {
+ vm_page_queues_remove(m);
+
+ assert(m->object != kernel_object);
+
+ vm_page_enqueue_inactive(m, FALSE);
+ }
+ }
+ if (dwp->dw_mask & DW_clear_busy)
+ m->busy = FALSE;
+
+ if (dwp->dw_mask & DW_PAGE_WAKEUP)
+ PAGE_WAKEUP(m);
+ }
+ }
+ vm_page_unlock_queues();
+
+ if (local_free_q)
+ vm_page_free_list(local_free_q, TRUE);
+
+ VM_CHECK_MEMORYSTATUS;
+
+}
+
+kern_return_t
+vm_page_alloc_list(
+ int page_count,
+ int flags,
+ vm_page_t *list)
+{
+ vm_page_t lo_page_list = VM_PAGE_NULL;
+ vm_page_t mem;
+ int i;
+
+ if ( !(flags & KMA_LOMEM))
+ panic("vm_page_alloc_list: called w/o KMA_LOMEM");
+
+ for (i = 0; i < page_count; i++) {
+
+ mem = vm_page_grablo();
+
+ if (mem == VM_PAGE_NULL) {
+ if (lo_page_list)
+ vm_page_free_list(lo_page_list, FALSE);
+
+ *list = VM_PAGE_NULL;
+
+ return (KERN_RESOURCE_SHORTAGE);
+ }
+ mem->pageq.next = (queue_entry_t) lo_page_list;
+ lo_page_list = mem;
+ }
+ *list = lo_page_list;
+
+ return (KERN_SUCCESS);
+}
+
+void
+vm_page_set_offset(vm_page_t page, vm_object_offset_t offset)
+{
+ page->offset = offset;
+}
+
+vm_page_t
+vm_page_get_next(vm_page_t page)
+{
+ return ((vm_page_t) page->pageq.next);
+}
+
+vm_object_offset_t
+vm_page_get_offset(vm_page_t page)
+{
+ return (page->offset);
+}
+
+ppnum_t
+vm_page_get_phys_page(vm_page_t page)
+{
+ return (page->phys_page);
+}
+
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+#if HIBERNATION
+
+static vm_page_t hibernate_gobble_queue;
+
+static int hibernate_drain_pageout_queue(struct vm_pageout_queue *);
+static int hibernate_flush_dirty_pages(int);
+static int hibernate_flush_queue(queue_head_t *, int);
+
+void hibernate_flush_wait(void);
+void hibernate_mark_in_progress(void);
+void hibernate_clear_in_progress(void);
+
+void hibernate_free_range(int, int);
+void hibernate_hash_insert_page(vm_page_t);
+uint32_t hibernate_mark_as_unneeded(addr64_t, addr64_t, hibernate_page_list_t *, hibernate_page_list_t *);
+void hibernate_rebuild_vm_structs(void);
+uint32_t hibernate_teardown_vm_structs(hibernate_page_list_t *, hibernate_page_list_t *);
+ppnum_t hibernate_lookup_paddr(unsigned int);
+
+struct hibernate_statistics {
+ int hibernate_considered;
+ int hibernate_reentered_on_q;
+ int hibernate_found_dirty;
+ int hibernate_skipped_cleaning;
+ int hibernate_skipped_transient;
+ int hibernate_skipped_precious;
+ int hibernate_skipped_external;
+ int hibernate_queue_nolock;
+ int hibernate_queue_paused;
+ int hibernate_throttled;
+ int hibernate_throttle_timeout;
+ int hibernate_drained;
+ int hibernate_drain_timeout;
+ int cd_lock_failed;
+ int cd_found_precious;
+ int cd_found_wired;
+ int cd_found_busy;
+ int cd_found_unusual;
+ int cd_found_cleaning;
+ int cd_found_laundry;
+ int cd_found_dirty;
+ int cd_found_xpmapped;
+ int cd_skipped_xpmapped;
+ int cd_local_free;
+ int cd_total_free;
+ int cd_vm_page_wire_count;
+ int cd_vm_struct_pages_unneeded;
+ int cd_pages;
+ int cd_discarded;
+ int cd_count_wire;
+} hibernate_stats;
+
+
+/*
+ * clamp the number of 'xpmapped' pages we'll sweep into the hibernation image
+ * so that we don't overrun the estimated image size, which would
+ * result in a hibernation failure.
+ */
+#define HIBERNATE_XPMAPPED_LIMIT 40000
+
+
+static int
+hibernate_drain_pageout_queue(struct vm_pageout_queue *q)
+{
+ wait_result_t wait_result;
+
+ vm_page_lock_queues();
+
+ while ( !queue_empty(&q->pgo_pending) ) {
+
+ q->pgo_draining = TRUE;
+
+ assert_wait_timeout((event_t) (&q->pgo_laundry+1), THREAD_INTERRUPTIBLE, 5000, 1000*NSEC_PER_USEC);
+
+ vm_page_unlock_queues();
+
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
+
+ if (wait_result == THREAD_TIMED_OUT && !queue_empty(&q->pgo_pending)) {
+ hibernate_stats.hibernate_drain_timeout++;
+
+ if (q == &vm_pageout_queue_external)
+ return (0);
+
+ return (1);
+ }
+ vm_page_lock_queues();
+
+ hibernate_stats.hibernate_drained++;
+ }
+ vm_page_unlock_queues();
+
+ return (0);
+}
+
+
+boolean_t hibernate_skip_external = FALSE;
+
+static int
+hibernate_flush_queue(queue_head_t *q, int qcount)
+{
+ vm_page_t m;
+ vm_object_t l_object = NULL;
+ vm_object_t m_object = NULL;
+ int refmod_state = 0;
+ int try_failed_count = 0;
+ int retval = 0;
+ int current_run = 0;
+ struct vm_pageout_queue *iq;
+ struct vm_pageout_queue *eq;
+ struct vm_pageout_queue *tq;
+
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_START, q, qcount, 0, 0, 0);
+
+ iq = &vm_pageout_queue_internal;
+ eq = &vm_pageout_queue_external;
+
+ vm_page_lock_queues();
+
+ while (qcount && !queue_empty(q)) {
+
+ if (current_run++ == 1000) {
+ if (hibernate_should_abort()) {
+ retval = 1;
+ break;
+ }
+ current_run = 0;
+ }
+
+ m = (vm_page_t) queue_first(q);
+ m_object = m->object;
+
+ /*
+ * check to see if we currently are working
+ * with the same object... if so, we've
+ * already got the lock
+ */
+ if (m_object != l_object) {
+ /*
+ * the object associated with candidate page is
+ * different from the one we were just working
+ * with... dump the lock if we still own it
+ */
+ if (l_object != NULL) {
+ vm_object_unlock(l_object);
+ l_object = NULL;
+ }
+ /*
+ * Try to lock object; since we've alread got the
+ * page queues lock, we can only 'try' for this one.
+ * if the 'try' fails, we need to do a mutex_pause
+ * to allow the owner of the object lock a chance to
+ * run...
+ */
+ if ( !vm_object_lock_try_scan(m_object)) {
+
+ if (try_failed_count > 20) {
+ hibernate_stats.hibernate_queue_nolock++;
+
+ goto reenter_pg_on_q;
+ }
+
+ vm_page_unlock_queues();
+ mutex_pause(try_failed_count++);
+ vm_page_lock_queues();
+
+ hibernate_stats.hibernate_queue_paused++;
+ continue;
+ } else {
+ l_object = m_object;
+ }
+ }
+ if ( !m_object->alive || m->encrypted_cleaning || m->cleaning || m->laundry || m->busy || m->absent || m->error) {
+ /*
+ * page is not to be cleaned
+ * put it back on the head of its queue
+ */
+ if (m->cleaning)
+ hibernate_stats.hibernate_skipped_cleaning++;
+ else
+ hibernate_stats.hibernate_skipped_transient++;
+
+ goto reenter_pg_on_q;
+ }
+ if (m_object->copy == VM_OBJECT_NULL) {
+ if (m_object->purgable == VM_PURGABLE_VOLATILE || m_object->purgable == VM_PURGABLE_EMPTY) {
+ /*
+ * let the normal hibernate image path
+ * deal with these
+ */
+ goto reenter_pg_on_q;
+ }
+ }
+ if ( !m->dirty && m->pmapped) {
+ refmod_state = pmap_get_refmod(m->phys_page);
+
+ if ((refmod_state & VM_MEM_MODIFIED)) {
+ SET_PAGE_DIRTY(m, FALSE);
+ }
+ } else
+ refmod_state = 0;
+
+ if ( !m->dirty) {
+ /*
+ * page is not to be cleaned
+ * put it back on the head of its queue
+ */
+ if (m->precious)
+ hibernate_stats.hibernate_skipped_precious++;
+
+ goto reenter_pg_on_q;
+ }
+
+ if (hibernate_skip_external == TRUE && !m_object->internal) {
+
+ hibernate_stats.hibernate_skipped_external++;
+
+ goto reenter_pg_on_q;
+ }
+ tq = NULL;
+
+ if (m_object->internal) {
+ if (VM_PAGE_Q_THROTTLED(iq))
+ tq = iq;
+ } else if (VM_PAGE_Q_THROTTLED(eq))
+ tq = eq;
+
+ if (tq != NULL) {
+ wait_result_t wait_result;
+ int wait_count = 5;
+
+ if (l_object != NULL) {
+ vm_object_unlock(l_object);
+ l_object = NULL;
+ }
+
+ while (retval == 0) {
+
+ tq->pgo_throttled = TRUE;
+
+ assert_wait_timeout((event_t) &tq->pgo_laundry, THREAD_INTERRUPTIBLE, 1000, 1000*NSEC_PER_USEC);
+
+ vm_page_unlock_queues();
+
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
+
+ vm_page_lock_queues();
+
+ if (wait_result != THREAD_TIMED_OUT)
+ break;
+ if (!VM_PAGE_Q_THROTTLED(tq))
+ break;
+
+ if (hibernate_should_abort())
+ retval = 1;
+
+ if (--wait_count == 0) {
+
+ hibernate_stats.hibernate_throttle_timeout++;
+
+ if (tq == eq) {
+ hibernate_skip_external = TRUE;
+ break;
+ }
+ retval = 1;
+ }
+ }
+ if (retval)
+ break;
+
+ hibernate_stats.hibernate_throttled++;
+
+ continue;
+ }
+ /*
+ * we've already factored out pages in the laundry which
+ * means this page can't be on the pageout queue so it's
+ * safe to do the vm_page_queues_remove
+ */
+ assert(!m->pageout_queue);
+
+ vm_page_queues_remove(m);
+
+ if (COMPRESSED_PAGER_IS_ACTIVE && m_object->internal == TRUE)
+ pmap_disconnect_options(m->phys_page, PMAP_OPTIONS_COMPRESSOR, NULL);
+
+ (void)vm_pageout_cluster(m, FALSE, FALSE, FALSE);
+
+ hibernate_stats.hibernate_found_dirty++;
+
+ goto next_pg;
+
+reenter_pg_on_q:
+ queue_remove(q, m, vm_page_t, pageq);
+ queue_enter(q, m, vm_page_t, pageq);
+
+ hibernate_stats.hibernate_reentered_on_q++;
+next_pg:
+ hibernate_stats.hibernate_considered++;
+
+ qcount--;
+ try_failed_count = 0;
+ }
+ if (l_object != NULL) {
+ vm_object_unlock(l_object);
+ l_object = NULL;
+ }
+
+ vm_page_unlock_queues();
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_END, hibernate_stats.hibernate_found_dirty, retval, 0, 0, 0);
+
+ return (retval);
+}
+
+
+static int
+hibernate_flush_dirty_pages(int pass)
+{
+ struct vm_speculative_age_q *aq;
+ uint32_t i;
+
+ if (vm_page_local_q) {
+ for (i = 0; i < vm_page_local_q_count; i++)
+ vm_page_reactivate_local(i, TRUE, FALSE);
+ }
+
+ for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
+ int qcount;
+ vm_page_t m;
+
+ aq = &vm_page_queue_speculative[i];
+
+ if (queue_empty(&aq->age_q))
+ continue;
+ qcount = 0;
+
+ vm_page_lockspin_queues();
+
+ queue_iterate(&aq->age_q,
+ m,
+ vm_page_t,
+ pageq)
+ {
+ qcount++;
+ }
+ vm_page_unlock_queues();
+
+ if (qcount) {
+ if (hibernate_flush_queue(&aq->age_q, qcount))
+ return (1);
+ }
+ }
+ if (hibernate_flush_queue(&vm_page_queue_inactive, vm_page_inactive_count - vm_page_anonymous_count - vm_page_cleaned_count))
+ return (1);
+ if (hibernate_flush_queue(&vm_page_queue_anonymous, vm_page_anonymous_count))
+ return (1);
+ if (hibernate_flush_queue(&vm_page_queue_cleaned, vm_page_cleaned_count))
+ return (1);
+ if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal))
+ return (1);
+
+ if (COMPRESSED_PAGER_IS_ACTIVE && pass == 1)
+ vm_compressor_record_warmup_start();
+
+ if (hibernate_flush_queue(&vm_page_queue_active, vm_page_active_count)) {
+ if (COMPRESSED_PAGER_IS_ACTIVE && pass == 1)
+ vm_compressor_record_warmup_end();
+ return (1);
+ }
+ if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) {
+ if (COMPRESSED_PAGER_IS_ACTIVE && pass == 1)
+ vm_compressor_record_warmup_end();
+ return (1);
+ }
+ if (COMPRESSED_PAGER_IS_ACTIVE && pass == 1)
+ vm_compressor_record_warmup_end();
+
+ if (hibernate_skip_external == FALSE && hibernate_drain_pageout_queue(&vm_pageout_queue_external))
+ return (1);
+
+ return (0);
+}
+
+
+void
+hibernate_reset_stats()
+{
+ bzero(&hibernate_stats, sizeof(struct hibernate_statistics));
+}
+
+
+int
+hibernate_flush_memory()
+{
+ int retval;
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_START, vm_page_free_count, 0, 0, 0, 0);
+
+ hibernate_cleaning_in_progress = TRUE;
+ hibernate_skip_external = FALSE;
+
+ if ((retval = hibernate_flush_dirty_pages(1)) == 0) {
+
+ if (COMPRESSED_PAGER_IS_ACTIVE) {
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_START, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
+
+ vm_compressor_flush();
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_END, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
+ }
+ if (consider_buffer_cache_collect != NULL) {
+ unsigned int orig_wire_count;
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_START, 0, 0, 0, 0, 0);
+ orig_wire_count = vm_page_wire_count;
+
+ (void)(*consider_buffer_cache_collect)(1);
+ consider_zone_gc(TRUE);
+
+ HIBLOG("hibernate_flush_memory: buffer_cache_gc freed up %d wired pages\n", orig_wire_count - vm_page_wire_count);
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_END, orig_wire_count - vm_page_wire_count, 0, 0, 0, 0);
+ }
+ }
+ hibernate_cleaning_in_progress = FALSE;
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_END, vm_page_free_count, hibernate_stats.hibernate_found_dirty, retval, 0, 0);
+
+ if (retval && COMPRESSED_PAGER_IS_ACTIVE)
+ HIBLOG("hibernate_flush_memory() failed to finish - vm_page_compressor_count(%d)\n", VM_PAGE_COMPRESSOR_COUNT);
+
+
+ HIBPRINT("hibernate_flush_memory() considered(%d) reentered_on_q(%d) found_dirty(%d)\n",
+ hibernate_stats.hibernate_considered,
+ hibernate_stats.hibernate_reentered_on_q,
+ hibernate_stats.hibernate_found_dirty);
+ HIBPRINT(" skipped_cleaning(%d) skipped_transient(%d) skipped_precious(%d) skipped_external(%d) queue_nolock(%d)\n",
+ hibernate_stats.hibernate_skipped_cleaning,
+ hibernate_stats.hibernate_skipped_transient,
+ hibernate_stats.hibernate_skipped_precious,
+ hibernate_stats.hibernate_skipped_external,
+ hibernate_stats.hibernate_queue_nolock);
+ HIBPRINT(" queue_paused(%d) throttled(%d) throttle_timeout(%d) drained(%d) drain_timeout(%d)\n",
+ hibernate_stats.hibernate_queue_paused,
+ hibernate_stats.hibernate_throttled,
+ hibernate_stats.hibernate_throttle_timeout,
+ hibernate_stats.hibernate_drained,
+ hibernate_stats.hibernate_drain_timeout);
+
+ return (retval);
+}
+
+
+static void
+hibernate_page_list_zero(hibernate_page_list_t *list)
+{
+ uint32_t bank;
+ hibernate_bitmap_t * bitmap;
+
+ bitmap = &list->bank_bitmap[0];
+ for (bank = 0; bank < list->bank_count; bank++)
+ {
+ uint32_t last_bit;
+
+ bzero((void *) &bitmap->bitmap[0], bitmap->bitmapwords << 2);
+ // set out-of-bound bits at end of bitmap.
+ last_bit = ((bitmap->last_page - bitmap->first_page + 1) & 31);
+ if (last_bit)
+ bitmap->bitmap[bitmap->bitmapwords - 1] = (0xFFFFFFFF >> last_bit);
+
+ bitmap = (hibernate_bitmap_t *) &bitmap->bitmap[bitmap->bitmapwords];
+ }
+}
+
+void
+hibernate_free_gobble_pages(void)
+{
+ vm_page_t m, next;
+ uint32_t count = 0;
+
+ m = (vm_page_t) hibernate_gobble_queue;
+ while(m)
+ {
+ next = (vm_page_t) m->pageq.next;
+ vm_page_free(m);
+ count++;
+ m = next;
+ }
+ hibernate_gobble_queue = VM_PAGE_NULL;
+
+ if (count)
+ HIBLOG("Freed %d pages\n", count);
+}
+
+static boolean_t
+hibernate_consider_discard(vm_page_t m, boolean_t preflight)
+{
+ vm_object_t object = NULL;
+ int refmod_state;
+ boolean_t discard = FALSE;
+
+ do
+ {
+ if (m->private)
+ panic("hibernate_consider_discard: private");
+
+ if (!vm_object_lock_try(m->object)) {
+ if (!preflight) hibernate_stats.cd_lock_failed++;
+ break;
+ }
+ object = m->object;
+
+ if (VM_PAGE_WIRED(m)) {
+ if (!preflight) hibernate_stats.cd_found_wired++;
+ break;
+ }
+ if (m->precious) {
+ if (!preflight) hibernate_stats.cd_found_precious++;
+ break;
+ }
+ if (m->busy || !object->alive) {
+ /*
+ * Somebody is playing with this page.
+ */
+ if (!preflight) hibernate_stats.cd_found_busy++;
+ break;
+ }
+ if (m->absent || m->unusual || m->error) {
+ /*
+ * If it's unusual in anyway, ignore it
+ */
+ if (!preflight) hibernate_stats.cd_found_unusual++;
+ break;
+ }
+ if (m->cleaning) {
+ if (!preflight) hibernate_stats.cd_found_cleaning++;
+ break;
+ }
+ if (m->laundry) {
+ if (!preflight) hibernate_stats.cd_found_laundry++;
+ break;
+ }
+ if (!m->dirty)
+ {
+ refmod_state = pmap_get_refmod(m->phys_page);
+
+ if (refmod_state & VM_MEM_REFERENCED)
+ m->reference = TRUE;
+ if (refmod_state & VM_MEM_MODIFIED) {
+ SET_PAGE_DIRTY(m, FALSE);
+ }
+ }
+
+ /*
+ * If it's clean or purgeable we can discard the page on wakeup.
+ */
+ discard = (!m->dirty)
+ || (VM_PURGABLE_VOLATILE == object->purgable)
+ || (VM_PURGABLE_EMPTY == object->purgable);
+
+
+ if (discard == FALSE) {
+ if (!preflight)
+ hibernate_stats.cd_found_dirty++;
+ } else if (m->xpmapped && m->reference && !object->internal) {
+ if (hibernate_stats.cd_found_xpmapped < HIBERNATE_XPMAPPED_LIMIT) {
+ if (!preflight)
+ hibernate_stats.cd_found_xpmapped++;
+ discard = FALSE;
+ } else {
+ if (!preflight)
+ hibernate_stats.cd_skipped_xpmapped++;
+ }
+ }
+ }
+ while (FALSE);
+
+ if (object)
+ vm_object_unlock(object);
+
+ return (discard);
+}
+
+
+static void
+hibernate_discard_page(vm_page_t m)
+{
+ if (m->absent || m->unusual || m->error)
+ /*
+ * If it's unusual in anyway, ignore
+ */
+ return;
+
+#if MACH_ASSERT || DEBUG
+ vm_object_t object = m->object;
+ if (!vm_object_lock_try(m->object))
+ panic("hibernate_discard_page(%p) !vm_object_lock_try", m);
+#else
+ /* No need to lock page queue for token delete, hibernate_vm_unlock()
+ makes sure these locks are uncontended before sleep */
+#endif /* MACH_ASSERT || DEBUG */
+
+ if (m->pmapped == TRUE)
+ {
+ __unused int refmod_state = pmap_disconnect(m->phys_page);
+ }
+
+ if (m->laundry)
+ panic("hibernate_discard_page(%p) laundry", m);
+ if (m->private)
+ panic("hibernate_discard_page(%p) private", m);
+ if (m->fictitious)
+ panic("hibernate_discard_page(%p) fictitious", m);
+
+ if (VM_PURGABLE_VOLATILE == m->object->purgable)
+ {
+ /* object should be on a queue */
+ assert((m->object->objq.next != NULL) && (m->object->objq.prev != NULL));
+ purgeable_q_t old_queue = vm_purgeable_object_remove(m->object);
+ assert(old_queue);
+ if (m->object->purgeable_when_ripe) {
+ vm_purgeable_token_delete_first(old_queue);
+ }
+ m->object->purgable = VM_PURGABLE_EMPTY;
+
+ /*
+ * Purgeable ledgers: pages of VOLATILE and EMPTY objects are
+ * accounted in the "volatile" ledger, so no change here.
+ * We have to update vm_page_purgeable_count, though, since we're
+ * effectively purging this object.
+ */
+ unsigned int delta;
+ assert(m->object->resident_page_count >= m->object->wired_page_count);
+ delta = (m->object->resident_page_count - m->object->wired_page_count);
+ assert(vm_page_purgeable_count >= delta);
+ assert(delta > 0);
+ OSAddAtomic(-delta, (SInt32 *)&vm_page_purgeable_count);
+ }
+
+ vm_page_free(m);
+
+#if MACH_ASSERT || DEBUG
+ vm_object_unlock(object);
+#endif /* MACH_ASSERT || DEBUG */
+}
+
+/*
+ Grab locks for hibernate_page_list_setall()
+*/
+void
+hibernate_vm_lock_queues(void)
+{
+ vm_object_lock(compressor_object);
+ vm_page_lock_queues();
+ lck_mtx_lock(&vm_page_queue_free_lock);
+
+ if (vm_page_local_q) {
+ uint32_t i;
+ for (i = 0; i < vm_page_local_q_count; i++) {
+ struct vpl *lq;
+ lq = &vm_page_local_q[i].vpl_un.vpl;
+ VPL_LOCK(&lq->vpl_lock);
+ }
+ }
+}
+
+void
+hibernate_vm_unlock_queues(void)
+{
+ if (vm_page_local_q) {
+ uint32_t i;
+ for (i = 0; i < vm_page_local_q_count; i++) {
+ struct vpl *lq;
+ lq = &vm_page_local_q[i].vpl_un.vpl;
+ VPL_UNLOCK(&lq->vpl_lock);
+ }
+ }
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+ vm_page_unlock_queues();
+ vm_object_unlock(compressor_object);
+}
+
+/*
+ Bits zero in the bitmaps => page needs to be saved. All pages default to be saved,
+ pages known to VM to not need saving are subtracted.
+ Wired pages to be saved are present in page_list_wired, pageable in page_list.
+*/
+
+void
+hibernate_page_list_setall(hibernate_page_list_t * page_list,
+ hibernate_page_list_t * page_list_wired,
+ hibernate_page_list_t * page_list_pal,
+ boolean_t preflight,
+ boolean_t will_discard,
+ uint32_t * pagesOut)
+{
+ uint64_t start, end, nsec;
+ vm_page_t m;
+ vm_page_t next;
+ uint32_t pages = page_list->page_count;
+ uint32_t count_anonymous = 0, count_throttled = 0, count_compressor = 0;
+ uint32_t count_inactive = 0, count_active = 0, count_speculative = 0, count_cleaned = 0;
+ uint32_t count_wire = pages;
+ uint32_t count_discard_active = 0;
+ uint32_t count_discard_inactive = 0;
+ uint32_t count_discard_cleaned = 0;
+ uint32_t count_discard_purgeable = 0;
+ uint32_t count_discard_speculative = 0;
+ uint32_t count_discard_vm_struct_pages = 0;
+ uint32_t i;
+ uint32_t bank;
+ hibernate_bitmap_t * bitmap;
+ hibernate_bitmap_t * bitmap_wired;
+ boolean_t discard_all;
+ boolean_t discard;
+
+ HIBLOG("hibernate_page_list_setall(preflight %d) start\n", preflight);
+
+ if (preflight) {
+ page_list = NULL;
+ page_list_wired = NULL;
+ page_list_pal = NULL;
+ discard_all = FALSE;
+ } else {
+ discard_all = will_discard;
+ }
+
+#if MACH_ASSERT || DEBUG
+ if (!preflight)
+ {
+ vm_page_lock_queues();
+ if (vm_page_local_q) {
+ for (i = 0; i < vm_page_local_q_count; i++) {
+ struct vpl *lq;
+ lq = &vm_page_local_q[i].vpl_un.vpl;
+ VPL_LOCK(&lq->vpl_lock);
+ }
+ }
+ }
+#endif /* MACH_ASSERT || DEBUG */
+
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_START, count_wire, 0, 0, 0, 0);
+
+ clock_get_uptime(&start);
+
+ if (!preflight) {
+ hibernate_page_list_zero(page_list);
+ hibernate_page_list_zero(page_list_wired);
+ hibernate_page_list_zero(page_list_pal);
+
+ hibernate_stats.cd_vm_page_wire_count = vm_page_wire_count;
+ hibernate_stats.cd_pages = pages;
+ }
+
+ if (vm_page_local_q) {
+ for (i = 0; i < vm_page_local_q_count; i++)
+ vm_page_reactivate_local(i, TRUE, !preflight);
+ }
+
+ if (preflight) {
+ vm_object_lock(compressor_object);
+ vm_page_lock_queues();
+ lck_mtx_lock(&vm_page_queue_free_lock);
+ }
+
+ m = (vm_page_t) hibernate_gobble_queue;
+ while (m)
+ {
+ pages--;
+ count_wire--;
+ if (!preflight) {
+ hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+ }
+ m = (vm_page_t) m->pageq.next;
+ }
+
+ if (!preflight) for( i = 0; i < real_ncpus; i++ )
+ {
+ if (cpu_data_ptr[i] && cpu_data_ptr[i]->cpu_processor)
+ {
+ for (m = PROCESSOR_DATA(cpu_data_ptr[i]->cpu_processor, free_pages); m; m = (vm_page_t)m->pageq.next)
+ {
+ pages--;
+ count_wire--;
+ hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+
+ hibernate_stats.cd_local_free++;
+ hibernate_stats.cd_total_free++;
+ }
+ }
+ }
+
+ for( i = 0; i < vm_colors; i++ )
+ {
+ queue_iterate(&vm_page_queue_free[i],
+ m,
+ vm_page_t,
+ pageq)
+ {
+ pages--;
+ count_wire--;
+ if (!preflight) {
+ hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+
+ hibernate_stats.cd_total_free++;
+ }
+ }
+ }
+
+ queue_iterate(&vm_lopage_queue_free,
+ m,
+ vm_page_t,
+ pageq)
+ {
+ pages--;
+ count_wire--;
+ if (!preflight) {
+ hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+
+ hibernate_stats.cd_total_free++;
+ }
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_throttled);
+ while (m && !queue_end(&vm_page_queue_throttled, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ discard = FALSE;
+ if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
+ && hibernate_consider_discard(m, preflight))
+ {
+ if (!preflight) hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ count_discard_inactive++;
+ discard = discard_all;
+ }
+ else
+ count_throttled++;
+ count_wire--;
+ if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+
+ if (discard) hibernate_discard_page(m);
+ m = next;
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_anonymous);
+ while (m && !queue_end(&vm_page_queue_anonymous, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ discard = FALSE;
+ if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
+ && hibernate_consider_discard(m, preflight))
+ {
+ if (!preflight) hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_inactive++;
+ discard = discard_all;
+ }
+ else
+ count_anonymous++;
+ count_wire--;
+ if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+ if (discard) hibernate_discard_page(m);
+ m = next;
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_cleaned);
+ while (m && !queue_end(&vm_page_queue_cleaned, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ discard = FALSE;
+ if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
+ && hibernate_consider_discard(m, preflight))
+ {
+ if (!preflight) hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_cleaned++;
+ discard = discard_all;
+ }
+ else
+ count_cleaned++;
+ count_wire--;
+ if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+ if (discard) hibernate_discard_page(m);
+ m = next;
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_active);
+ while (m && !queue_end(&vm_page_queue_active, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ discard = FALSE;
+ if ((kIOHibernateModeDiscardCleanActive & gIOHibernateMode)
+ && hibernate_consider_discard(m, preflight))
+ {
+ if (!preflight) hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_active++;
+ discard = discard_all;
+ }
+ else
+ count_active++;
+ count_wire--;
+ if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+ if (discard) hibernate_discard_page(m);
+ m = next;
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_inactive);
+ while (m && !queue_end(&vm_page_queue_inactive, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ discard = FALSE;
+ if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
+ && hibernate_consider_discard(m, preflight))
+ {
+ if (!preflight) hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_inactive++;
+ discard = discard_all;
+ }
+ else
+ count_inactive++;
+ count_wire--;
+ if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+ if (discard) hibernate_discard_page(m);
+ m = next;
+ }
+
+ for( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ )
+ {
+ m = (vm_page_t) queue_first(&vm_page_queue_speculative[i].age_q);
+ while (m && !queue_end(&vm_page_queue_speculative[i].age_q, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ discard = FALSE;
+ if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
+ && hibernate_consider_discard(m, preflight))
+ {
+ if (!preflight) hibernate_page_bitset(page_list, TRUE, m->phys_page);
+ count_discard_speculative++;
+ discard = discard_all;
+ }
+ else
+ count_speculative++;
+ count_wire--;
+ if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+ if (discard) hibernate_discard_page(m);
+ m = next;
+ }
+ }
+
+ queue_iterate(&compressor_object->memq, m, vm_page_t, listq)
+ {
+ count_compressor++;
+ count_wire--;
+ if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, m->phys_page);
+ }
+
+ if (preflight == FALSE && discard_all == TRUE) {
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_START, 0, 0, 0, 0, 0);
+
+ HIBLOG("hibernate_teardown started\n");
+ count_discard_vm_struct_pages = hibernate_teardown_vm_structs(page_list, page_list_wired);
+ HIBLOG("hibernate_teardown completed - discarded %d\n", count_discard_vm_struct_pages);
+
+ pages -= count_discard_vm_struct_pages;
+ count_wire -= count_discard_vm_struct_pages;
+
+ hibernate_stats.cd_vm_struct_pages_unneeded = count_discard_vm_struct_pages;
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_END, 0, 0, 0, 0, 0);
+ }
+
+ if (!preflight) {
+ // pull wired from hibernate_bitmap
+ bitmap = &page_list->bank_bitmap[0];
+ bitmap_wired = &page_list_wired->bank_bitmap[0];
+ for (bank = 0; bank < page_list->bank_count; bank++)
+ {
+ for (i = 0; i < bitmap->bitmapwords; i++)
+ bitmap->bitmap[i] = bitmap->bitmap[i] | ~bitmap_wired->bitmap[i];
+ bitmap = (hibernate_bitmap_t *) &bitmap->bitmap [bitmap->bitmapwords];
+ bitmap_wired = (hibernate_bitmap_t *) &bitmap_wired->bitmap[bitmap_wired->bitmapwords];
+ }
+ }
+
+ // machine dependent adjustments
+ hibernate_page_list_setall_machine(page_list, page_list_wired, preflight, &pages);
+
+ if (!preflight) {
+ hibernate_stats.cd_count_wire = count_wire;
+ hibernate_stats.cd_discarded = count_discard_active + count_discard_inactive + count_discard_purgeable +
+ count_discard_speculative + count_discard_cleaned + count_discard_vm_struct_pages;
+ }
+
+ clock_get_uptime(&end);
+ absolutetime_to_nanoseconds(end - start, &nsec);
+ HIBLOG("hibernate_page_list_setall time: %qd ms\n", nsec / 1000000ULL);
+
+ HIBLOG("pages %d, wire %d, act %d, inact %d, cleaned %d spec %d, zf %d, throt %d, compr %d, xpmapped %d\n %s discard act %d inact %d purgeable %d spec %d cleaned %d\n",
+ pages, count_wire, count_active, count_inactive, count_cleaned, count_speculative, count_anonymous, count_throttled, count_compressor, hibernate_stats.cd_found_xpmapped,
+ discard_all ? "did" : "could",
+ count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned);
+
+ if (hibernate_stats.cd_skipped_xpmapped)
+ HIBLOG("WARNING: hibernate_page_list_setall skipped %d xpmapped pages\n", hibernate_stats.cd_skipped_xpmapped);
+
+ *pagesOut = pages - count_discard_active - count_discard_inactive - count_discard_purgeable - count_discard_speculative - count_discard_cleaned;
+
+ if (preflight && will_discard) *pagesOut -= count_compressor + count_throttled + count_anonymous + count_inactive + count_cleaned + count_speculative + count_active;
+
+#if MACH_ASSERT || DEBUG
+ if (!preflight)
+ {
+ if (vm_page_local_q) {
+ for (i = 0; i < vm_page_local_q_count; i++) {
+ struct vpl *lq;
+ lq = &vm_page_local_q[i].vpl_un.vpl;
+ VPL_UNLOCK(&lq->vpl_lock);
+ }
+ }
+ vm_page_unlock_queues();
+ }
+#endif /* MACH_ASSERT || DEBUG */
+
+ if (preflight) {
+ lck_mtx_unlock(&vm_page_queue_free_lock);
+ vm_page_unlock_queues();
+ vm_object_unlock(compressor_object);
+ }
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_END, count_wire, *pagesOut, 0, 0, 0);
+}
+
+void
+hibernate_page_list_discard(hibernate_page_list_t * page_list)
+{
+ uint64_t start, end, nsec;
+ vm_page_t m;
+ vm_page_t next;
+ uint32_t i;
+ uint32_t count_discard_active = 0;
+ uint32_t count_discard_inactive = 0;
+ uint32_t count_discard_purgeable = 0;
+ uint32_t count_discard_cleaned = 0;
+ uint32_t count_discard_speculative = 0;
+
+
+#if MACH_ASSERT || DEBUG
+ vm_page_lock_queues();
+ if (vm_page_local_q) {
+ for (i = 0; i < vm_page_local_q_count; i++) {
+ struct vpl *lq;
+ lq = &vm_page_local_q[i].vpl_un.vpl;
+ VPL_LOCK(&lq->vpl_lock);
+ }
+ }
+#endif /* MACH_ASSERT || DEBUG */
+
+ clock_get_uptime(&start);
+
+ m = (vm_page_t) queue_first(&vm_page_queue_anonymous);
+ while (m && !queue_end(&vm_page_queue_anonymous, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ if (hibernate_page_bittst(page_list, m->phys_page))
+ {
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_inactive++;
+ hibernate_discard_page(m);
+ }
+ m = next;
+ }
+
+ for( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ )
+ {
+ m = (vm_page_t) queue_first(&vm_page_queue_speculative[i].age_q);
+ while (m && !queue_end(&vm_page_queue_speculative[i].age_q, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ if (hibernate_page_bittst(page_list, m->phys_page))
+ {
+ count_discard_speculative++;
+ hibernate_discard_page(m);
+ }
+ m = next;
+ }
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_inactive);
+ while (m && !queue_end(&vm_page_queue_inactive, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ if (hibernate_page_bittst(page_list, m->phys_page))
+ {
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_inactive++;
+ hibernate_discard_page(m);
+ }
+ m = next;
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_active);
+ while (m && !queue_end(&vm_page_queue_active, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ if (hibernate_page_bittst(page_list, m->phys_page))
+ {
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_active++;
+ hibernate_discard_page(m);
+ }
+ m = next;
+ }
+
+ m = (vm_page_t) queue_first(&vm_page_queue_cleaned);
+ while (m && !queue_end(&vm_page_queue_cleaned, (queue_entry_t)m))
+ {
+ next = (vm_page_t) m->pageq.next;
+ if (hibernate_page_bittst(page_list, m->phys_page))
+ {
+ if (m->dirty)
+ count_discard_purgeable++;
+ else
+ count_discard_cleaned++;
+ hibernate_discard_page(m);
+ }
+ m = next;
+ }
+
+#if MACH_ASSERT || DEBUG
+ if (vm_page_local_q) {
+ for (i = 0; i < vm_page_local_q_count; i++) {
+ struct vpl *lq;
+ lq = &vm_page_local_q[i].vpl_un.vpl;
+ VPL_UNLOCK(&lq->vpl_lock);
+ }
+ }
+ vm_page_unlock_queues();
+#endif /* MACH_ASSERT || DEBUG */
+
+ clock_get_uptime(&end);
+ absolutetime_to_nanoseconds(end - start, &nsec);
+ HIBLOG("hibernate_page_list_discard time: %qd ms, discarded act %d inact %d purgeable %d spec %d cleaned %d\n",
+ nsec / 1000000ULL,
+ count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned);
+}
+
+boolean_t hibernate_paddr_map_inited = FALSE;
+boolean_t hibernate_rebuild_needed = FALSE;
+unsigned int hibernate_teardown_last_valid_compact_indx = -1;
+vm_page_t hibernate_rebuild_hash_list = NULL;
+
+unsigned int hibernate_teardown_found_tabled_pages = 0;
+unsigned int hibernate_teardown_found_created_pages = 0;
+unsigned int hibernate_teardown_found_free_pages = 0;
+unsigned int hibernate_teardown_vm_page_free_count;
+
+
+struct ppnum_mapping {
+ struct ppnum_mapping *ppnm_next;
+ ppnum_t ppnm_base_paddr;
+ unsigned int ppnm_sindx;
+ unsigned int ppnm_eindx;
+};
+
+struct ppnum_mapping *ppnm_head;
+struct ppnum_mapping *ppnm_last_found = NULL;
+
+
+void
+hibernate_create_paddr_map()
+{
+ unsigned int i;
+ ppnum_t next_ppnum_in_run = 0;
+ struct ppnum_mapping *ppnm = NULL;
+
+ if (hibernate_paddr_map_inited == FALSE) {
+
+ for (i = 0; i < vm_pages_count; i++) {
+
+ if (ppnm)
+ ppnm->ppnm_eindx = i;
+
+ if (ppnm == NULL || vm_pages[i].phys_page != next_ppnum_in_run) {
+
+ ppnm = kalloc(sizeof(struct ppnum_mapping));
+
+ ppnm->ppnm_next = ppnm_head;
+ ppnm_head = ppnm;
+
+ ppnm->ppnm_sindx = i;
+ ppnm->ppnm_base_paddr = vm_pages[i].phys_page;
+ }
+ next_ppnum_in_run = vm_pages[i].phys_page + 1;
+ }
+ ppnm->ppnm_eindx++;
+
+ hibernate_paddr_map_inited = TRUE;
+ }
+}
+
+ppnum_t
+hibernate_lookup_paddr(unsigned int indx)
+{
+ struct ppnum_mapping *ppnm = NULL;
+
+ ppnm = ppnm_last_found;
+
+ if (ppnm) {
+ if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx)
+ goto done;
+ }
+ for (ppnm = ppnm_head; ppnm; ppnm = ppnm->ppnm_next) {
+
+ if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) {
+ ppnm_last_found = ppnm;
+ break;
+ }
+ }
+ if (ppnm == NULL)
+ panic("hibernate_lookup_paddr of %d failed\n", indx);
+done:
+ return (ppnm->ppnm_base_paddr + (indx - ppnm->ppnm_sindx));
+}
+
+
+uint32_t
+hibernate_mark_as_unneeded(addr64_t saddr, addr64_t eaddr, hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
+{
+ addr64_t saddr_aligned;
+ addr64_t eaddr_aligned;
+ addr64_t addr;
+ ppnum_t paddr;
+ unsigned int mark_as_unneeded_pages = 0;
+
+ saddr_aligned = (saddr + PAGE_MASK_64) & ~PAGE_MASK_64;
+ eaddr_aligned = eaddr & ~PAGE_MASK_64;
+
+ for (addr = saddr_aligned; addr < eaddr_aligned; addr += PAGE_SIZE_64) {
+
+ paddr = pmap_find_phys(kernel_pmap, addr);
+
+ assert(paddr);
+
+ hibernate_page_bitset(page_list, TRUE, paddr);
+ hibernate_page_bitset(page_list_wired, TRUE, paddr);
+
+ mark_as_unneeded_pages++;
+ }
+ return (mark_as_unneeded_pages);
+}
+
+
+void
+hibernate_hash_insert_page(vm_page_t mem)
+{
+ vm_page_bucket_t *bucket;
+ int hash_id;
+
+ assert(mem->hashed);
+ assert(mem->object);
+ assert(mem->offset != (vm_object_offset_t) -1);
+
+ /*
+ * Insert it into the object_object/offset hash table
+ */
+ hash_id = vm_page_hash(mem->object, mem->offset);
+ bucket = &vm_page_buckets[hash_id];
+
+ mem->next_m = bucket->page_list;
+ bucket->page_list = VM_PAGE_PACK_PTR(mem);
+}
+
+
+void
+hibernate_free_range(int sindx, int eindx)
+{
+ vm_page_t mem;
+ unsigned int color;
+
+ while (sindx < eindx) {
+ mem = &vm_pages[sindx];
+
+ vm_page_init(mem, hibernate_lookup_paddr(sindx), FALSE);
+
+ mem->lopage = FALSE;
+ mem->free = TRUE;
+
+ color = mem->phys_page & vm_color_mask;
+ queue_enter_first(&vm_page_queue_free[color],
+ mem,
+ vm_page_t,
+ pageq);
+ vm_page_free_count++;
+
+ sindx++;
+ }
+}
+
+
+extern void hibernate_rebuild_pmap_structs(void);
- /* prepend to the very front of sorted list? */
- else if (m->phys_page < sort_list->phys_page)
- {
- vm_page_queue_free = NEXT_PAGE(m);
- cpm_counter(++vpfls_general_insertions);
- prevcontaddr = PPNUM_PREV(m->phys_page);
- nextcontaddr = PPNUM_NEXT(m->phys_page);
- SET_NEXT_PAGE(m, sort_list);
- contfirstprev = &sort_list;
- contlast = sort_list = m;
- npages = 1;
- }
+void
+hibernate_rebuild_vm_structs(void)
+{
+ int cindx, sindx, eindx;
+ vm_page_t mem, tmem, mem_next;
+ AbsoluteTime startTime, endTime;
+ uint64_t nsec;
- else /* get to proper place for insertion */
- {
- if (m->phys_page < nextcontaddr)
- {
- prevcontaddr = PPNUM_PREV(sort_list->phys_page);
- nextcontaddr = PPNUM_NEXT(sort_list->phys_page);
- contfirstprev = &sort_list;
- contlast = sort_list;
- npages = 1;
- }
- for (m1 = NEXT_PAGE(contlast);
- npages < contig_pages &&
- m1 != VM_PAGE_NULL && m1->phys_page < m->phys_page;
- m1 = NEXT_PAGE(m1))
- {
- if (m1->phys_page != nextcontaddr) {
- prevcontaddr = PPNUM_PREV(m1->phys_page);
- contfirstprev = NEXT_PAGE_PTR(contlast);
- npages = 1;
- } else {
- npages++;
- }
- nextcontaddr = PPNUM_NEXT(m1->phys_page);
- contlast = m1;
- }
+ if (hibernate_rebuild_needed == FALSE)
+ return;
- /*
- * We may actually already have enough.
- * This could happen if a previous prepend
- * joined up two runs to meet our needs.
- * If so, bail before we take the current
- * page off the free queue.
- */
- if (npages == contig_pages)
- break;
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_START, 0, 0, 0, 0, 0);
+ HIBLOG("hibernate_rebuild started\n");
- if (m->phys_page != nextcontaddr)
- {
- contfirstprev = NEXT_PAGE_PTR(contlast);
- prevcontaddr = PPNUM_PREV(m->phys_page);
- nextcontaddr = PPNUM_NEXT(m->phys_page);
- npages = 1;
- } else {
- nextcontaddr = PPNUM_NEXT(nextcontaddr);
- npages++;
- }
- vm_page_queue_free = NEXT_PAGE(m);
- cpm_counter(++vpfls_general_insertions);
- SET_NEXT_PAGE(m, NEXT_PAGE(contlast));
- SET_NEXT_PAGE(contlast, m);
- contlast = m;
- }
+ clock_get_uptime(&startTime);
+
+ hibernate_rebuild_pmap_structs();
+
+ bzero(&vm_page_buckets[0], vm_page_bucket_count * sizeof(vm_page_bucket_t));
+ eindx = vm_pages_count;
+
+ for (cindx = hibernate_teardown_last_valid_compact_indx; cindx >= 0; cindx--) {
- /* See how many pages are now contiguous after the insertion */
- for (m1 = NEXT_PAGE(m);
- npages < contig_pages &&
- m1 != VM_PAGE_NULL && m1->phys_page == nextcontaddr;
- m1 = NEXT_PAGE(m1))
- {
- nextcontaddr = PPNUM_NEXT(nextcontaddr);
- contlast = m1;
- npages++;
+ mem = &vm_pages[cindx];
+ /*
+ * hibernate_teardown_vm_structs leaves the location where
+ * this vm_page_t must be located in "next".
+ */
+ tmem = VM_PAGE_UNPACK_PTR(mem->next_m);
+ mem->next_m = VM_PAGE_PACK_PTR(NULL);
+
+ sindx = (int)(tmem - &vm_pages[0]);
+
+ if (mem != tmem) {
+ /*
+ * this vm_page_t was moved by hibernate_teardown_vm_structs,
+ * so move it back to its real location
+ */
+ *tmem = *mem;
+ mem = tmem;
}
- }
+ if (mem->hashed)
+ hibernate_hash_insert_page(mem);
+ /*
+ * the 'hole' between this vm_page_t and the previous
+ * vm_page_t we moved needs to be initialized as
+ * a range of free vm_page_t's
+ */
+ hibernate_free_range(sindx + 1, eindx);
- /* how did we do? */
- if (npages == contig_pages)
- {
- cpm_counter(++vpfc_satisfied);
+ eindx = sindx;
+ }
+ if (sindx)
+ hibernate_free_range(0, sindx);
- /* remove the contiguous range from the sorted list */
- m = *contfirstprev;
- *contfirstprev = NEXT_PAGE(contlast);
- SET_NEXT_PAGE(contlast, VM_PAGE_NULL);
- assert(vm_page_verify_contiguous(m, npages));
+ assert(vm_page_free_count == hibernate_teardown_vm_page_free_count);
- /* inline vm_page_gobble() for each returned page */
- for (m1 = m; m1 != VM_PAGE_NULL; m1 = NEXT_PAGE(m1)) {
- assert(m1->free);
- assert(!m1->wanted);
- assert(!m1->laundry);
- m1->free = FALSE;
- m1->no_isync = TRUE;
- m1->gobbled = TRUE;
- }
- vm_page_wire_count += npages;
- vm_page_gobble_count += npages;
- vm_page_free_count -= npages;
+ /*
+ * process the list of vm_page_t's that were entered in the hash,
+ * but were not located in the vm_pages arrary... these are
+ * vm_page_t's that were created on the fly (i.e. fictitious)
+ */
+ for (mem = hibernate_rebuild_hash_list; mem; mem = mem_next) {
+ mem_next = VM_PAGE_UNPACK_PTR(mem->next_m);
- /* stick free list at the tail of the sorted list */
- while ((m1 = *contfirstprev) != VM_PAGE_NULL)
- contfirstprev = (vm_page_t *)&m1->pageq.next;
- *contfirstprev = vm_page_queue_free;
+ mem->next_m = VM_PAGE_PACK_PTR(NULL);
+ hibernate_hash_insert_page(mem);
}
+ hibernate_rebuild_hash_list = NULL;
- vm_page_queue_free = sort_list;
- return m;
+ clock_get_uptime(&endTime);
+ SUB_ABSOLUTETIME(&endTime, &startTime);
+ absolutetime_to_nanoseconds(endTime, &nsec);
+
+ HIBLOG("hibernate_rebuild completed - took %qd msecs\n", nsec / 1000000ULL);
+
+ hibernate_rebuild_needed = FALSE;
+
+ KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_END, 0, 0, 0, 0, 0);
}
-/*
- * Allocate a list of contiguous, wired pages.
- */
-kern_return_t
-cpm_allocate(
- vm_size_t size,
- vm_page_t *list,
- boolean_t wire)
+
+extern void hibernate_teardown_pmap_structs(addr64_t *, addr64_t *);
+
+uint32_t
+hibernate_teardown_vm_structs(hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
{
- register vm_page_t m;
- vm_page_t pages;
- unsigned int npages;
- unsigned int vm_pages_available;
- boolean_t wakeup;
+ unsigned int i;
+ unsigned int compact_target_indx;
+ vm_page_t mem, mem_next;
+ vm_page_bucket_t *bucket;
+ unsigned int mark_as_unneeded_pages = 0;
+ unsigned int unneeded_vm_page_bucket_pages = 0;
+ unsigned int unneeded_vm_pages_pages = 0;
+ unsigned int unneeded_pmap_pages = 0;
+ addr64_t start_of_unneeded = 0;
+ addr64_t end_of_unneeded = 0;
- if (size % page_size != 0)
- return KERN_INVALID_ARGUMENT;
+
+ if (hibernate_should_abort())
+ return (0);
- vm_page_lock_queues();
- mutex_lock(&vm_page_queue_free_lock);
+ HIBLOG("hibernate_teardown: wired_pages %d, free_pages %d, active_pages %d, inactive_pages %d, speculative_pages %d, cleaned_pages %d, compressor_pages %d\n",
+ vm_page_wire_count, vm_page_free_count, vm_page_active_count, vm_page_inactive_count, vm_page_speculative_count,
+ vm_page_cleaned_count, compressor_object->resident_page_count);
- /*
- * Should also take active and inactive pages
- * into account... One day...
- */
- npages = size / page_size;
- vm_pages_available = vm_page_free_count - vm_page_free_reserved;
+ for (i = 0; i < vm_page_bucket_count; i++) {
- if (npages > vm_pages_available) {
- mutex_unlock(&vm_page_queue_free_lock);
- vm_page_unlock_queues();
- return KERN_RESOURCE_SHORTAGE;
+ bucket = &vm_page_buckets[i];
+
+ for (mem = VM_PAGE_UNPACK_PTR(bucket->page_list); mem != VM_PAGE_NULL; mem = mem_next) {
+ assert(mem->hashed);
+
+ mem_next = VM_PAGE_UNPACK_PTR(mem->next_m);
+
+ if (mem < &vm_pages[0] || mem >= &vm_pages[vm_pages_count]) {
+ mem->next_m = VM_PAGE_PACK_PTR(hibernate_rebuild_hash_list);
+ hibernate_rebuild_hash_list = mem;
+ }
+ }
}
+ unneeded_vm_page_bucket_pages = hibernate_mark_as_unneeded((addr64_t)&vm_page_buckets[0], (addr64_t)&vm_page_buckets[vm_page_bucket_count], page_list, page_list_wired);
+ mark_as_unneeded_pages += unneeded_vm_page_bucket_pages;
- /*
- * Obtain a pointer to a subset of the free
- * list large enough to satisfy the request;
- * the region will be physically contiguous.
- */
- pages = vm_page_find_contiguous(npages);
+ hibernate_teardown_vm_page_free_count = vm_page_free_count;
- /* adjust global freelist counts and determine need for wakeups */
- if (vm_page_free_count < vm_page_free_count_minimum)
- vm_page_free_count_minimum = vm_page_free_count;
+ compact_target_indx = 0;
- wakeup = ((vm_page_free_count < vm_page_free_min) ||
- ((vm_page_free_count < vm_page_free_target) &&
- (vm_page_inactive_count < vm_page_inactive_target)));
-
- mutex_unlock(&vm_page_queue_free_lock);
+ for (i = 0; i < vm_pages_count; i++) {
- if (pages == VM_PAGE_NULL) {
- vm_page_unlock_queues();
- return KERN_NO_SPACE;
- }
+ mem = &vm_pages[i];
- /*
- * Walk the returned list, wiring the pages.
- */
- if (wire == TRUE)
- for (m = pages; m != VM_PAGE_NULL; m = NEXT_PAGE(m)) {
+ if (mem->free) {
+ unsigned int color;
+
+ assert(mem->busy);
+ assert(!mem->lopage);
+
+ color = mem->phys_page & vm_color_mask;
+
+ queue_remove(&vm_page_queue_free[color],
+ mem,
+ vm_page_t,
+ pageq);
+ mem->pageq.next = NULL;
+ mem->pageq.prev = NULL;
+
+ vm_page_free_count--;
+
+ hibernate_teardown_found_free_pages++;
+
+ if ( !vm_pages[compact_target_indx].free)
+ compact_target_indx = i;
+ } else {
/*
- * Essentially inlined vm_page_wire.
+ * record this vm_page_t's original location
+ * we need this even if it doesn't get moved
+ * as an indicator to the rebuild function that
+ * we don't have to move it
*/
- assert(!m->active);
- assert(!m->inactive);
- assert(!m->private);
- assert(!m->fictitious);
- assert(m->wire_count == 0);
- assert(m->gobbled);
- m->gobbled = FALSE;
- m->wire_count++;
- --vm_page_gobble_count;
+ mem->next_m = VM_PAGE_PACK_PTR(mem);
+
+ if (vm_pages[compact_target_indx].free) {
+ /*
+ * we've got a hole to fill, so
+ * move this vm_page_t to it's new home
+ */
+ vm_pages[compact_target_indx] = *mem;
+ mem->free = TRUE;
+
+ hibernate_teardown_last_valid_compact_indx = compact_target_indx;
+ compact_target_indx++;
+ } else
+ hibernate_teardown_last_valid_compact_indx = i;
}
- vm_page_unlock_queues();
+ }
+ unneeded_vm_pages_pages = hibernate_mark_as_unneeded((addr64_t)&vm_pages[hibernate_teardown_last_valid_compact_indx+1],
+ (addr64_t)&vm_pages[vm_pages_count-1], page_list, page_list_wired);
+ mark_as_unneeded_pages += unneeded_vm_pages_pages;
- if (wakeup)
- thread_wakeup((event_t) &vm_page_free_wanted);
+ hibernate_teardown_pmap_structs(&start_of_unneeded, &end_of_unneeded);
- /*
- * The CPM pages should now be available and
- * ordered by ascending physical address.
- */
- assert(vm_page_verify_contiguous(pages, npages));
+ if (start_of_unneeded) {
+ unneeded_pmap_pages = hibernate_mark_as_unneeded(start_of_unneeded, end_of_unneeded, page_list, page_list_wired);
+ mark_as_unneeded_pages += unneeded_pmap_pages;
+ }
+ HIBLOG("hibernate_teardown: mark_as_unneeded_pages %d, %d, %d\n", unneeded_vm_page_bucket_pages, unneeded_vm_pages_pages, unneeded_pmap_pages);
- *list = pages;
- return KERN_SUCCESS;
+ hibernate_rebuild_needed = TRUE;
+
+ return (mark_as_unneeded_pages);
}
+#endif /* HIBERNATION */
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
#include <mach_vm_debug.h>
#if MACH_VM_DEBUG
unsigned int count)
{
unsigned int i;
+ lck_spin_t *bucket_lock;
if (vm_page_bucket_count < count)
count = vm_page_bucket_count;
unsigned int bucket_count = 0;
vm_page_t m;
- simple_lock(&vm_page_bucket_lock);
- for (m = bucket->pages; m != VM_PAGE_NULL; m = m->next)
+ bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
+ lck_spin_lock(bucket_lock);
+
+ for (m = VM_PAGE_UNPACK_PTR(bucket->page_list); m != VM_PAGE_NULL; m = VM_PAGE_UNPACK_PTR(m->next_m))
bucket_count++;
- simple_unlock(&vm_page_bucket_lock);
+
+ lck_spin_unlock(bucket_lock);
/* don't touch pageable memory while holding locks */
info[i].hib_count = bucket_count;
}
#endif /* MACH_VM_DEBUG */
-#include <mach_kdb.h>
-#if MACH_KDB
+#if VM_PAGE_BUCKETS_CHECK
+void
+vm_page_buckets_check(void)
+{
+ unsigned int i;
+ vm_page_t p;
+ unsigned int p_hash;
+ vm_page_bucket_t *bucket;
+ lck_spin_t *bucket_lock;
+
+ if (!vm_page_buckets_check_ready) {
+ return;
+ }
+
+#if HIBERNATION
+ if (hibernate_rebuild_needed ||
+ hibernate_rebuild_hash_list) {
+ panic("BUCKET_CHECK: hibernation in progress: "
+ "rebuild_needed=%d rebuild_hash_list=%p\n",
+ hibernate_rebuild_needed,
+ hibernate_rebuild_hash_list);
+ }
+#endif /* HIBERNATION */
+
+#if VM_PAGE_FAKE_BUCKETS
+ char *cp;
+ for (cp = (char *) vm_page_fake_buckets_start;
+ cp < (char *) vm_page_fake_buckets_end;
+ cp++) {
+ if (*cp != 0x5a) {
+ panic("BUCKET_CHECK: corruption at %p in fake buckets "
+ "[0x%llx:0x%llx]\n",
+ cp,
+ (uint64_t) vm_page_fake_buckets_start,
+ (uint64_t) vm_page_fake_buckets_end);
+ }
+ }
+#endif /* VM_PAGE_FAKE_BUCKETS */
+
+ for (i = 0; i < vm_page_bucket_count; i++) {
+ bucket = &vm_page_buckets[i];
+ if (!bucket->page_list) {
+ continue;
+ }
+
+ bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
+ lck_spin_lock(bucket_lock);
+ p = VM_PAGE_UNPACK_PTR(bucket->page_list);
+ while (p != VM_PAGE_NULL) {
+ if (!p->hashed) {
+ panic("BUCKET_CHECK: page %p (%p,0x%llx) "
+ "hash %d in bucket %d at %p "
+ "is not hashed\n",
+ p, p->object, p->offset,
+ p_hash, i, bucket);
+ }
+ p_hash = vm_page_hash(p->object, p->offset);
+ if (p_hash != i) {
+ panic("BUCKET_CHECK: corruption in bucket %d "
+ "at %p: page %p object %p offset 0x%llx "
+ "hash %d\n",
+ i, bucket, p, p->object, p->offset,
+ p_hash);
+ }
+ p = VM_PAGE_UNPACK_PTR(p->next_m);
+ }
+ lck_spin_unlock(bucket_lock);
+ }
-#include <ddb/db_output.h>
-#include <vm/vm_print.h>
-#define printf kdbprintf
+// printf("BUCKET_CHECK: checked buckets\n");
+}
+#endif /* VM_PAGE_BUCKETS_CHECK */
/*
- * Routine: vm_page_print [exported]
+ * 'vm_fault_enter' will place newly created pages (zero-fill and COW) onto the
+ * local queues if they exist... its the only spot in the system where we add pages
+ * to those queues... once on those queues, those pages can only move to one of the
+ * global page queues or the free queues... they NEVER move from local q to local q.
+ * the 'local' state is stable when vm_page_queues_remove is called since we're behind
+ * the global vm_page_queue_lock at this point... we still need to take the local lock
+ * in case this operation is being run on a different CPU then the local queue's identity,
+ * but we don't have to worry about the page moving to a global queue or becoming wired
+ * while we're grabbing the local lock since those operations would require the global
+ * vm_page_queue_lock to be held, and we already own it.
+ *
+ * this is why its safe to utilze the wire_count field in the vm_page_t as the local_id...
+ * 'wired' and local are ALWAYS mutually exclusive conditions.
*/
void
-vm_page_print(
- db_addr_t db_addr)
+vm_page_queues_remove(vm_page_t mem)
{
- vm_page_t p;
+ boolean_t was_pageable;
+
+ VM_PAGE_QUEUES_ASSERT(mem, 1);
+ assert(!mem->pageout_queue);
+ /*
+ * if (mem->pageout_queue)
+ * NOTE: vm_page_queues_remove does not deal with removing pages from the pageout queue...
+ * the caller is responsible for determing if the page is on that queue, and if so, must
+ * either first remove it (it needs both the page queues lock and the object lock to do
+ * this via vm_pageout_steal_laundry), or avoid the call to vm_page_queues_remove
+ */
+ if (mem->local) {
+ struct vpl *lq;
+ assert(mem->object != kernel_object);
+ assert(mem->object != compressor_object);
+ assert(!mem->inactive && !mem->speculative);
+ assert(!mem->active && !mem->throttled);
+ assert(!mem->clean_queue);
+ assert(!mem->fictitious);
+ lq = &vm_page_local_q[mem->local_id].vpl_un.vpl;
+ VPL_LOCK(&lq->vpl_lock);
+ queue_remove(&lq->vpl_queue,
+ mem, vm_page_t, pageq);
+ mem->local = FALSE;
+ mem->local_id = 0;
+ lq->vpl_count--;
+ if (mem->object->internal) {
+ lq->vpl_internal_count--;
+ } else {
+ lq->vpl_external_count--;
+ }
+ VPL_UNLOCK(&lq->vpl_lock);
+ was_pageable = FALSE;
+ }
- p = (vm_page_t) (long) db_addr;
+ else if (mem->active) {
+ assert(mem->object != kernel_object);
+ assert(mem->object != compressor_object);
+ assert(!mem->inactive && !mem->speculative);
+ assert(!mem->clean_queue);
+ assert(!mem->throttled);
+ assert(!mem->fictitious);
+ queue_remove(&vm_page_queue_active,
+ mem, vm_page_t, pageq);
+ mem->active = FALSE;
+ vm_page_active_count--;
+ was_pageable = TRUE;
+ }
- iprintf("page 0x%x\n", p);
+ else if (mem->inactive) {
+ assert(mem->object != kernel_object);
+ assert(mem->object != compressor_object);
+ assert(!mem->active && !mem->speculative);
+ assert(!mem->throttled);
+ assert(!mem->fictitious);
+ vm_page_inactive_count--;
+ if (mem->clean_queue) {
+ queue_remove(&vm_page_queue_cleaned,
+ mem, vm_page_t, pageq);
+ mem->clean_queue = FALSE;
+ vm_page_cleaned_count--;
+ } else {
+ if (mem->object->internal) {
+ queue_remove(&vm_page_queue_anonymous,
+ mem, vm_page_t, pageq);
+ vm_page_anonymous_count--;
+ } else {
+ queue_remove(&vm_page_queue_inactive,
+ mem, vm_page_t, pageq);
+ }
+ vm_purgeable_q_advance_all();
+ }
+ mem->inactive = FALSE;
+ was_pageable = TRUE;
+ }
- db_indent += 2;
+ else if (mem->throttled) {
+ assert(mem->object != compressor_object);
+ assert(!mem->active && !mem->inactive);
+ assert(!mem->speculative);
+ assert(!mem->fictitious);
+ queue_remove(&vm_page_queue_throttled,
+ mem, vm_page_t, pageq);
+ mem->throttled = FALSE;
+ vm_page_throttled_count--;
+ was_pageable = FALSE;
+ }
- iprintf("object=0x%x", p->object);
- printf(", offset=0x%x", p->offset);
- printf(", wire_count=%d", p->wire_count);
+ else if (mem->speculative) {
+ assert(mem->object != compressor_object);
+ assert(!mem->active && !mem->inactive);
+ assert(!mem->throttled);
+ assert(!mem->fictitious);
+ remque(&mem->pageq);
+ mem->speculative = FALSE;
+ vm_page_speculative_count--;
+ was_pageable = TRUE;
+ }
- iprintf("%sinactive, %sactive, %sgobbled, %slaundry, %sfree, %sref, %sencrypted\n",
- (p->inactive ? "" : "!"),
- (p->active ? "" : "!"),
- (p->gobbled ? "" : "!"),
- (p->laundry ? "" : "!"),
- (p->free ? "" : "!"),
- (p->reference ? "" : "!"),
- (p->encrypted ? "" : "!"));
- iprintf("%sbusy, %swanted, %stabled, %sfictitious, %sprivate, %sprecious\n",
- (p->busy ? "" : "!"),
- (p->wanted ? "" : "!"),
- (p->tabled ? "" : "!"),
- (p->fictitious ? "" : "!"),
- (p->private ? "" : "!"),
- (p->precious ? "" : "!"));
- iprintf("%sabsent, %serror, %sdirty, %scleaning, %spageout, %sclustered\n",
- (p->absent ? "" : "!"),
- (p->error ? "" : "!"),
- (p->dirty ? "" : "!"),
- (p->cleaning ? "" : "!"),
- (p->pageout ? "" : "!"),
- (p->clustered ? "" : "!"));
- iprintf("%slock_supplied, %soverwriting, %srestart, %sunusual\n",
- (p->lock_supplied ? "" : "!"),
- (p->overwriting ? "" : "!"),
- (p->restart ? "" : "!"),
- (p->unusual ? "" : "!"));
-
- iprintf("phys_page=0x%x", p->phys_page);
- printf(", page_error=0x%x", p->page_error);
- printf(", page_lock=0x%x", p->page_lock);
- printf(", unlock_request=%d\n", p->unlock_request);
-
- db_indent -= 2;
-}
-#endif /* MACH_KDB */
+ else if (mem->pageq.next || mem->pageq.prev) {
+ was_pageable = FALSE;
+ panic("vm_page_queues_remove: unmarked page on Q");
+ } else {
+ was_pageable = FALSE;
+ }
+
+ mem->pageq.next = NULL;
+ mem->pageq.prev = NULL;
+ VM_PAGE_QUEUES_ASSERT(mem, 0);
+ if (was_pageable) {
+ if (mem->object->internal) {
+ vm_page_pageable_internal_count--;
+ } else {
+ vm_page_pageable_external_count--;
+ }
+ }
+}
+
+void
+vm_page_remove_internal(vm_page_t page)
+{
+ vm_object_t __object = page->object;
+ if (page == __object->memq_hint) {
+ vm_page_t __new_hint;
+ queue_entry_t __qe;
+ __qe = queue_next(&page->listq);
+ if (queue_end(&__object->memq, __qe)) {
+ __qe = queue_prev(&page->listq);
+ if (queue_end(&__object->memq, __qe)) {
+ __qe = NULL;
+ }
+ }
+ __new_hint = (vm_page_t) __qe;
+ __object->memq_hint = __new_hint;
+ }
+ queue_remove(&__object->memq, page, vm_page_t, listq);
+}
+
+void
+vm_page_enqueue_inactive(vm_page_t mem, boolean_t first)
+{
+ VM_PAGE_QUEUES_ASSERT(mem, 0);
+ assert(!mem->fictitious);
+ assert(!mem->laundry);
+ assert(!mem->pageout_queue);
+ vm_page_check_pageable_safe(mem);
+ if (mem->object->internal) {
+ if (first == TRUE)
+ queue_enter_first(&vm_page_queue_anonymous, mem, vm_page_t, pageq);
+ else
+ queue_enter(&vm_page_queue_anonymous, mem, vm_page_t, pageq);
+ vm_page_anonymous_count++;
+ vm_page_pageable_internal_count++;
+ } else {
+ if (first == TRUE)
+ queue_enter_first(&vm_page_queue_inactive, mem, vm_page_t, pageq);
+ else
+ queue_enter(&vm_page_queue_inactive, mem, vm_page_t, pageq);
+ vm_page_pageable_external_count++;
+ }
+ mem->inactive = TRUE;
+ vm_page_inactive_count++;
+ token_new_pagecount++;
+}
+
+/*
+ * Pages from special kernel objects shouldn't
+ * be placed on pageable queues.
+ */
+void
+vm_page_check_pageable_safe(vm_page_t page)
+{
+ if (page->object == kernel_object) {
+ panic("vm_page_check_pageable_safe: trying to add page" \
+ "from kernel object (%p) to pageable queue", kernel_object);
+ }
+
+ if (page->object == compressor_object) {
+ panic("vm_page_check_pageable_safe: trying to add page" \
+ "from compressor object (%p) to pageable queue", compressor_object);
+ }
+
+ if (page->object == vm_submap_object) {
+ panic("vm_page_check_pageable_safe: trying to add page" \
+ "from submap object (%p) to pageable queue", vm_submap_object);
+ }
+}
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ * wired page diagnose
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+#include <libkern/OSKextLibPrivate.h>
+
+vm_allocation_site_t *
+vm_allocation_sites[VM_KERN_MEMORY_COUNT];
+
+vm_tag_t
+vm_tag_bt(void)
+{
+ uintptr_t* frameptr;
+ uintptr_t* frameptr_next;
+ uintptr_t retaddr;
+ uintptr_t kstackb, kstackt;
+ const vm_allocation_site_t * site;
+ thread_t cthread;
+
+ cthread = current_thread();
+ if (__improbable(cthread == NULL)) return VM_KERN_MEMORY_OSFMK;
+
+ kstackb = cthread->kernel_stack;
+ kstackt = kstackb + kernel_stack_size;
+
+ /* Load stack frame pointer (EBP on x86) into frameptr */
+ frameptr = __builtin_frame_address(0);
+ site = NULL;
+ while (frameptr != NULL)
+ {
+ /* Verify thread stack bounds */
+ if (((uintptr_t)(frameptr + 2) > kstackt) || ((uintptr_t)frameptr < kstackb)) break;
+
+ /* Next frame pointer is pointed to by the previous one */
+ frameptr_next = (uintptr_t*) *frameptr;
+
+ /* Pull return address from one spot above the frame pointer */
+ retaddr = *(frameptr + 1);
+
+ if ((retaddr < vm_kernel_stext) || (retaddr > vm_kernel_top))
+ {
+ site = OSKextGetAllocationSiteForCaller(retaddr);
+ break;
+ }
+
+ frameptr = frameptr_next;
+ }
+ return (site ? site->tag : VM_KERN_MEMORY_NONE);
+}
+
+static uint64_t free_tag_bits[256/64];
+
+void
+vm_tag_alloc_locked(vm_allocation_site_t * site)
+{
+ vm_tag_t tag;
+ uint64_t avail;
+ uint64_t idx;
+
+ if (site->tag) return;
+
+ idx = 0;
+ while (TRUE)
+ {
+ avail = free_tag_bits[idx];
+ if (avail)
+ {
+ tag = __builtin_clzll(avail);
+ avail &= ~(1ULL << (63 - tag));
+ free_tag_bits[idx] = avail;
+ tag += (idx << 6);
+ break;
+ }
+ idx++;
+ if (idx >= (sizeof(free_tag_bits) / sizeof(free_tag_bits[0])))
+ {
+ tag = VM_KERN_MEMORY_ANY;
+ break;
+ }
+ }
+ site->tag = tag;
+ if (VM_KERN_MEMORY_ANY != tag)
+ {
+ assert(!vm_allocation_sites[tag]);
+ vm_allocation_sites[tag] = site;
+ }
+}
+
+static void
+vm_tag_free_locked(vm_tag_t tag)
+{
+ uint64_t avail;
+ uint32_t idx;
+ uint64_t bit;
+
+ if (VM_KERN_MEMORY_ANY == tag) return;
+
+ idx = (tag >> 6);
+ avail = free_tag_bits[idx];
+ tag &= 63;
+ bit = (1ULL << (63 - tag));
+ assert(!(avail & bit));
+ free_tag_bits[idx] = (avail | bit);
+}
+
+static void
+vm_tag_init(void)
+{
+ vm_tag_t tag;
+ for (tag = VM_KERN_MEMORY_FIRST_DYNAMIC; tag < VM_KERN_MEMORY_ANY; tag++)
+ {
+ vm_tag_free_locked(tag);
+ }
+}
+
+vm_tag_t
+vm_tag_alloc(vm_allocation_site_t * site)
+{
+ vm_tag_t tag;
+
+ if (VM_TAG_BT & site->flags)
+ {
+ tag = vm_tag_bt();
+ if (VM_KERN_MEMORY_NONE != tag) return (tag);
+ }
+
+ if (!site->tag)
+ {
+ lck_spin_lock(&vm_allocation_sites_lock);
+ vm_tag_alloc_locked(site);
+ lck_spin_unlock(&vm_allocation_sites_lock);
+ }
+
+ return (site->tag);
+}
+
+static void
+vm_page_count_object(mach_memory_info_t * sites, unsigned int __unused num_sites, vm_object_t object)
+{
+ if (!object->wired_page_count) return;
+ if (object != kernel_object)
+ {
+ assert(object->wire_tag < num_sites);
+ sites[object->wire_tag].size += ptoa_64(object->wired_page_count);
+ }
+}
+
+typedef void (*vm_page_iterate_proc)(mach_memory_info_t * sites,
+ unsigned int num_sites, vm_object_t object);
+
+static void
+vm_page_iterate_purgeable_objects(mach_memory_info_t * sites, unsigned int num_sites,
+ vm_page_iterate_proc proc, purgeable_q_t queue,
+ int group)
+{
+ vm_object_t object;
+
+ for (object = (vm_object_t) queue_first(&queue->objq[group]);
+ !queue_end(&queue->objq[group], (queue_entry_t) object);
+ object = (vm_object_t) queue_next(&object->objq))
+ {
+ proc(sites, num_sites, object);
+ }
+}
+
+static void
+vm_page_iterate_objects(mach_memory_info_t * sites, unsigned int num_sites,
+ vm_page_iterate_proc proc)
+{
+ purgeable_q_t volatile_q;
+ queue_head_t * nonvolatile_q;
+ vm_object_t object;
+ int group;
+
+ lck_spin_lock(&vm_objects_wired_lock);
+ queue_iterate(&vm_objects_wired,
+ object,
+ vm_object_t,
+ objq)
+ {
+ proc(sites, num_sites, object);
+ }
+ lck_spin_unlock(&vm_objects_wired_lock);
+
+ lck_mtx_lock(&vm_purgeable_queue_lock);
+ nonvolatile_q = &purgeable_nonvolatile_queue;
+ for (object = (vm_object_t) queue_first(nonvolatile_q);
+ !queue_end(nonvolatile_q, (queue_entry_t) object);
+ object = (vm_object_t) queue_next(&object->objq))
+ {
+ proc(sites, num_sites, object);
+ }
+
+ volatile_q = &purgeable_queues[PURGEABLE_Q_TYPE_OBSOLETE];
+ vm_page_iterate_purgeable_objects(sites, num_sites, proc, volatile_q, 0);
+
+ volatile_q = &purgeable_queues[PURGEABLE_Q_TYPE_FIFO];
+ for (group = 0; group < NUM_VOLATILE_GROUPS; group++)
+ {
+ vm_page_iterate_purgeable_objects(sites, num_sites, proc, volatile_q, group);
+ }
+
+ volatile_q = &purgeable_queues[PURGEABLE_Q_TYPE_LIFO];
+ for (group = 0; group < NUM_VOLATILE_GROUPS; group++)
+ {
+ vm_page_iterate_purgeable_objects(sites, num_sites, proc, volatile_q, group);
+ }
+ lck_mtx_unlock(&vm_purgeable_queue_lock);
+}
+
+static uint64_t
+process_account(mach_memory_info_t * sites, unsigned int __unused num_sites)
+{
+ uint64_t found;
+ unsigned int idx;
+ vm_allocation_site_t * site;
+
+ assert(num_sites >= VM_KERN_MEMORY_COUNT);
+ found = 0;
+ for (idx = 0; idx < VM_KERN_MEMORY_COUNT; idx++)
+ {
+ found += sites[idx].size;
+ if (idx < VM_KERN_MEMORY_FIRST_DYNAMIC)
+ {
+ sites[idx].site = idx;
+ sites[idx].flags |= VM_KERN_SITE_TAG;
+ if (VM_KERN_MEMORY_ZONE == idx) sites[idx].flags |= VM_KERN_SITE_HIDE;
+ else sites[idx].flags |= VM_KERN_SITE_WIRED;
+ continue;
+ }
+ lck_spin_lock(&vm_allocation_sites_lock);
+ if ((site = vm_allocation_sites[idx]))
+ {
+ if (sites[idx].size)
+ {
+ sites[idx].flags |= VM_KERN_SITE_WIRED;
+ if (VM_TAG_KMOD == (VM_KERN_SITE_TYPE & site->flags))
+ {
+ sites[idx].site = OSKextGetKmodIDForSite(site);
+ sites[idx].flags |= VM_KERN_SITE_KMOD;
+ }
+ else
+ {
+ sites[idx].site = VM_KERNEL_UNSLIDE(site);
+ sites[idx].flags |= VM_KERN_SITE_KERNEL;
+ }
+ site = NULL;
+ }
+ else
+ {
+ vm_tag_free_locked(site->tag);
+ site->tag = VM_KERN_MEMORY_NONE;
+ vm_allocation_sites[idx] = NULL;
+ if (!(VM_TAG_UNLOAD & site->flags)) site = NULL;
+ }
+ }
+ lck_spin_unlock(&vm_allocation_sites_lock);
+ if (site) OSKextFreeSite(site);
+ }
+ return (found);
+}
+
+kern_return_t
+vm_page_diagnose(mach_memory_info_t * sites, unsigned int num_sites)
+{
+ enum { kMaxKernelDepth = 1 };
+ vm_map_t maps [kMaxKernelDepth];
+ vm_map_entry_t entries[kMaxKernelDepth];
+ vm_map_t map;
+ vm_map_entry_t entry;
+ vm_object_offset_t offset;
+ vm_page_t page;
+ int stackIdx, count;
+ uint64_t wired_size;
+ uint64_t wired_managed_size;
+ uint64_t wired_reserved_size;
+ mach_memory_info_t * counts;
+
+ bzero(sites, num_sites * sizeof(mach_memory_info_t));
+
+ vm_page_iterate_objects(sites, num_sites, &vm_page_count_object);
+
+ wired_size = ptoa_64(vm_page_wire_count + vm_lopage_free_count + vm_page_throttled_count);
+ wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count + vm_page_throttled_count);
+ wired_managed_size = ptoa_64(vm_page_wire_count - vm_page_wire_count_initial);
+
+ assert(num_sites >= (VM_KERN_MEMORY_COUNT + VM_KERN_COUNTER_COUNT));
+ counts = &sites[VM_KERN_MEMORY_COUNT];
+
+#define SET_COUNT(xcount, xsize, xflags) \
+ counts[xcount].site = (xcount); \
+ counts[xcount].size = (xsize); \
+ counts[xcount].flags = VM_KERN_SITE_COUNTER | xflags;
+
+ SET_COUNT(VM_KERN_COUNT_MANAGED, ptoa_64(vm_page_pages), 0);
+ SET_COUNT(VM_KERN_COUNT_WIRED, wired_size, 0);
+ SET_COUNT(VM_KERN_COUNT_WIRED_MANAGED, wired_managed_size, 0);
+ SET_COUNT(VM_KERN_COUNT_RESERVED, wired_reserved_size, VM_KERN_SITE_WIRED);
+ SET_COUNT(VM_KERN_COUNT_STOLEN, ptoa_64(vm_page_stolen_count), VM_KERN_SITE_WIRED);
+ SET_COUNT(VM_KERN_COUNT_LOPAGE, ptoa_64(vm_lopage_free_count), VM_KERN_SITE_WIRED);
+
+#define SET_MAP(xcount, xsize, xfree, xlargest) \
+ counts[xcount].site = (xcount); \
+ counts[xcount].size = (xsize); \
+ counts[xcount].free = (xfree); \
+ counts[xcount].largest = (xlargest); \
+ counts[xcount].flags = VM_KERN_SITE_COUNTER;
+
+ vm_map_size_t map_size, map_free, map_largest;
+
+ vm_map_sizes(kernel_map, &map_size, &map_free, &map_largest);
+ SET_MAP(VM_KERN_COUNT_MAP_KERNEL, map_size, map_free, map_largest);
+
+ vm_map_sizes(zone_map, &map_size, &map_free, &map_largest);
+ SET_MAP(VM_KERN_COUNT_MAP_ZONE, map_size, map_free, map_largest);
+
+ vm_map_sizes(kalloc_map, &map_size, &map_free, &map_largest);
+ SET_MAP(VM_KERN_COUNT_MAP_KALLOC, map_size, map_free, map_largest);
+
+ map = kernel_map;
+ stackIdx = 0;
+ while (map)
+ {
+ vm_map_lock(map);
+ for (entry = map->hdr.links.next; map; entry = entry->links.next)
+ {
+ if (entry->is_sub_map)
+ {
+ assert(stackIdx < kMaxKernelDepth);
+ maps[stackIdx] = map;
+ entries[stackIdx] = entry;
+ stackIdx++;
+ map = VME_SUBMAP(entry);
+ entry = NULL;
+ break;
+ }
+ if (VME_OBJECT(entry) == kernel_object)
+ {
+ count = 0;
+ vm_object_lock(VME_OBJECT(entry));
+ for (offset = entry->links.start; offset < entry->links.end; offset += page_size)
+ {
+ page = vm_page_lookup(VME_OBJECT(entry), offset);
+ if (page && VM_PAGE_WIRED(page)) count++;
+ }
+ vm_object_unlock(VME_OBJECT(entry));
+
+ if (count)
+ {
+ assert(VME_ALIAS(entry) < num_sites);
+ sites[VME_ALIAS(entry)].size += ptoa_64(count);
+ }
+ }
+ if (entry == vm_map_last_entry(map))
+ {
+ vm_map_unlock(map);
+ if (!stackIdx) map = NULL;
+ else
+ {
+ --stackIdx;
+ map = maps[stackIdx];
+ entry = entries[stackIdx];
+ }
+ }
+ }
+ }
+
+ process_account(sites, num_sites);
+
+ return (KERN_SUCCESS);
+}
projects / apple / xnu.git / blobdiff