#include <vm/vm_external.h>
#include <vm/vm_options.h>
+#include <vm/vm_page.h>
#if VM_OBJECT_TRACKING
#include <libkern/OSDebug.h>
extern void vm_object_tracking_init(void);
extern boolean_t vm_object_tracking_inited;
extern btlog_t *vm_object_tracking_btlog;
+#define VM_OBJECT_TRACKING_NUM_RECORDS 50000
#define VM_OBJECT_TRACKING_BTDEPTH 7
#define VM_OBJECT_TRACKING_OP_CREATED 1
#define VM_OBJECT_TRACKING_OP_MODIFIED 2
#endif /* VM_OBJECT_TRACKING */
struct vm_page;
-struct vm_shared_region_slide_info;
/*
* Types defined:
/* boolean_t */ stealth:1,
/* boolean_t */ io_sync:1,
/* boolean_t */ cs_bypass:1,
+ /* boolean_t */ pmap_cs_associated:1,
/* boolean_t */ mark_zf_absent:1,
/* boolean_t */ batch_pmap_op:1,
- __vm_object_fault_info_unused_bits:26;
+ __vm_object_fault_info_unused_bits:25;
int pmap_options;
};
#define vo_cache_pages_to_scan vo_un1.vou_cache_pages_to_scan
#define vo_shadow_offset vo_un2.vou_shadow_offset
#define vo_cache_ts vo_un2.vou_cache_ts
-#define vo_purgeable_owner vo_un2.vou_purgeable_owner
-#define vo_slide_info vo_un2.vou_slide_info
+#define vo_owner vo_un2.vou_owner
struct vm_object {
- queue_head_t memq; /* Resident memory */
+ /*
+ * on 64 bit systems we pack the pointers hung off the memq.
+ * those pointers have to be able to point back to the memq.
+ * the packed pointers are required to be on a 64 byte boundary
+ * which means 2 things for the vm_object... (1) the memq
+ * struct has to be the first element of the structure so that
+ * we can control it's alignment... (2) the vm_object must be
+ * aligned on a 64 byte boundary... for static vm_object's
+ * this is accomplished via the 'aligned' attribute... for
+ * vm_object's in the zone pool, this is accomplished by
+ * rounding the size of the vm_object element to the nearest
+ * 64 byte size before creating the zone.
+ */
+ vm_page_queue_head_t memq; /* Resident memory - must be first */
lck_rw_t Lock; /* Synchronization */
+#if DEVELOPMENT || DEBUG
+ thread_t Lock_owner;
+#endif
union {
vm_object_size_t vou_size; /* Object size (only valid if internal) */
int vou_cache_pages_to_scan; /* pages yet to be visited in an
struct vm_page *memq_hint;
int ref_count; /* Number of references */
-#if TASK_SWAPPER
- int res_count; /* Residency references (swap)*/
-#endif /* TASK_SWAPPER */
unsigned int resident_page_count;
/* number of resident pages */
- unsigned int wired_page_count; /* number of wired pages */
+ unsigned int wired_page_count; /* number of wired pages
+ use VM_OBJECT_WIRED_PAGE_UPDATE macros to update */
unsigned int reusable_page_count;
struct vm_object *copy; /* Object that should receive
clock_sec_t vou_cache_ts; /* age of an external object
* present in cache
*/
- task_t vou_purgeable_owner; /* If the purg'a'ble bits below are set
- * to volatile/emtpy, this is the task
- * that owns this purgeable object.
- */
- struct vm_shared_region_slide_info *vou_slide_info;
+ task_t vou_owner; /* If the object is purgeable
+ * or has a "ledger_tag", this
+ * is the task that owns it.
+ */
} vo_un2;
memory_object_t pager; /* Where to get data */
* therefore, managed by the
* default memory manger)
*/
- /* boolean_t */ temporary:1, /* Permanent objects may be
- * changed externally by the
- * memory manager, and changes
- * made in memory must be
- * reflected back to the memory
- * manager. Temporary objects
- * lack both of these
- * characteristics.
- */
/* boolean_t */ private:1, /* magic device_pager object,
* holds private pages only */
/* boolean_t */ pageout:1, /* pageout object. contains
/* boolean_t */ purgable:2, /* Purgable state. See
* VM_PURGABLE_*
*/
+ /* boolean_t */ purgeable_only_by_kernel:1,
/* boolean_t */ purgeable_when_ripe:1, /* Purgeable when a token
* becomes ripe.
*/
/* boolean_t */ shadowed:1, /* Shadow may exist */
- /* boolean_t */ advisory_pageout:1,
- /* Instead of sending page
- * via OOL, just notify
- * pager that the kernel
- * wants to discard it, page
- * remains in object */
/* boolean_t */ true_share:1,
/* This object is mapped
* in more than one place
* memory rules w.r.t pmap
* access bits.
*/
- /* boolean_t */ nophyscache:1;
+ /* boolean_t */ nophyscache:1,
/* When mapped at the
* pmap level, don't allow
* primary caching. (for
* I/O)
*/
-
-
+ /* boolean_t */ _object5_unused_bits:1;
queue_chain_t cached_list; /* Attachment point for the
* list of objects cached as a
* result of their can_persist
* value
*/
-
- queue_head_t msr_q; /* memory object synchronise
- request queue */
-
/*
* the following fields are not protected by any locks
* they are updated via atomic compare and swap
uint32_t pages_created;
uint32_t pages_used;
-#if MACH_PAGEMAP
- vm_external_map_t existence_map; /* bitmap of pages written to
- * backing storage */
-#endif /* MACH_PAGEMAP */
vm_offset_t cow_hint; /* last page present in */
/* shadow but not in object */
-#if MACH_ASSERT
- struct vm_object *paging_object; /* object which pages to be
- * swapped out are temporary
- * put in current object
- */
-#endif
/* hold object lock when altering */
unsigned int
wimg_bits:8, /* cache WIMG bits */
code_signed:1, /* pages are signed and should be
validated; the signatures are stored
with the pager */
- hashed:1, /* object/pager entered in hash */
transposed:1, /* object was transposed with another */
mapping_in_progress:1, /* pager being mapped/unmapped */
phantom_isssd:1,
all_reusable:1,
blocked_access:1,
set_cache_attr:1,
- object_slid:1,
+ object_is_shared_cache:1,
purgeable_queue_type:2,
purgeable_queue_group:3,
io_tracking:1,
- __object2_unused_bits:7; /* for expansion */
+ no_tag_update:1, /* */
+#if CONFIG_SECLUDED_MEMORY
+ eligible_for_secluded:1,
+ can_grab_secluded:1,
+#else /* CONFIG_SECLUDED_MEMORY */
+ __object3_unused_bits:2,
+#endif /* CONFIG_SECLUDED_MEMORY */
+#if VM_OBJECT_ACCESS_TRACKING
+ access_tracking:1,
+#else /* VM_OBJECT_ACCESS_TRACKING */
+ __unused_access_tracking:1,
+#endif /* VM_OBJECT_ACCESS_TRACKING */
+ vo_ledger_tag:2,
+ __object2_unused_bits:2; /* for expansion */
+
+#if VM_OBJECT_ACCESS_TRACKING
+ uint32_t access_tracking_reads;
+ uint32_t access_tracking_writes;
+#endif /* VM_OBJECT_ACCESS_TRACKING */
uint8_t scan_collisions;
vm_tag_t wire_tag;
#endif /* VM_PIP_DEBUG */
queue_chain_t objq; /* object queue - currently used for purgable queues */
+ queue_chain_t task_objq; /* objects owned by task - protected by task lock */
+
+#if !VM_TAG_ACTIVE_UPDATE
+ queue_chain_t wired_objq;
+#endif /* !VM_TAG_ACTIVE_UPDATE */
#if DEBUG
void *purgeable_owner_bt[16];
#endif /* DEBUG */
};
+/* values for object->vo_ledger_tag */
+#define VM_OBJECT_LEDGER_TAG_NONE 0
+#define VM_OBJECT_LEDGER_TAG_NETWORK 1
+#define VM_OBJECT_LEDGER_TAG_MEDIA 2
+#define VM_OBJECT_LEDGER_TAG_RESERVED 3
+
#define VM_OBJECT_PURGEABLE_FAULT_ERROR(object) \
((object)->volatile_fault && \
((object)->purgable == VM_PURGABLE_VOLATILE || \
(object)->purgable == VM_PURGABLE_EMPTY))
+#if VM_OBJECT_ACCESS_TRACKING
+extern uint64_t vm_object_access_tracking_reads;
+extern uint64_t vm_object_access_tracking_writes;
+extern void vm_object_access_tracking(vm_object_t object,
+ int *access_tracking,
+ uint32_t *access_tracking_reads,
+ uint32_t *acess_tracking_writes);
+#endif /* VM_OBJECT_ACCESS_TRACKING */
+
extern
vm_object_t kernel_object; /* the single kernel object */
# define VM_MSYNC_SYNCHRONIZING 1
# define VM_MSYNC_DONE 2
-struct msync_req {
- queue_chain_t msr_q; /* object request queue */
- queue_chain_t req_q; /* vm_msync request queue */
- unsigned int flag;
- vm_object_offset_t offset;
- vm_object_size_t length;
- vm_object_t object; /* back pointer */
- decl_lck_mtx_data(, msync_req_lock) /* Lock for this structure */
-};
-
-typedef struct msync_req *msync_req_t;
-#define MSYNC_REQ_NULL ((msync_req_t) 0)
-
extern lck_grp_t vm_map_lck_grp;
extern lck_attr_t vm_map_lck_attr;
-/*
- * Macros to allocate and free msync_reqs
- */
-#define msync_req_alloc(msr) \
- MACRO_BEGIN \
- (msr) = (msync_req_t)kalloc(sizeof(struct msync_req)); \
- lck_mtx_init(&(msr)->msync_req_lock, &vm_map_lck_grp, &vm_map_lck_attr); \
- msr->flag = VM_MSYNC_INITIALIZED; \
- MACRO_END
-
-#define msync_req_free(msr) \
- MACRO_BEGIN \
- lck_mtx_destroy(&(msr)->msync_req_lock, &vm_map_lck_grp); \
- kfree((msr), sizeof(struct msync_req)); \
- MACRO_END
-
-#define msr_lock(msr) lck_mtx_lock(&(msr)->msync_req_lock)
-#define msr_unlock(msr) lck_mtx_unlock(&(msr)->msync_req_lock)
+#ifndef VM_TAG_ACTIVE_UPDATE
+#error VM_TAG_ACTIVE_UPDATE
+#endif
-#define VM_OBJECT_WIRED(object) \
- MACRO_BEGIN \
- if ((object)->purgable == VM_PURGABLE_DENY) \
- { \
+#if VM_TAG_ACTIVE_UPDATE
+#define VM_OBJECT_WIRED_ENQUEUE(object) panic("VM_OBJECT_WIRED_ENQUEUE")
+#define VM_OBJECT_WIRED_DEQUEUE(object) panic("VM_OBJECT_WIRED_DEQUEUE")
+#else /* VM_TAG_ACTIVE_UPDATE */
+#define VM_OBJECT_WIRED_ENQUEUE(object) \
+ MACRO_BEGIN \
lck_spin_lock(&vm_objects_wired_lock); \
- assert(!(object)->objq.next); \
- queue_enter(&vm_objects_wired, (object), vm_object_t, objq); \
+ assert(!(object)->wired_objq.next); \
+ assert(!(object)->wired_objq.prev); \
+ queue_enter(&vm_objects_wired, (object), \
+ vm_object_t, wired_objq); \
lck_spin_unlock(&vm_objects_wired_lock); \
+ MACRO_END
+#define VM_OBJECT_WIRED_DEQUEUE(object) \
+ MACRO_BEGIN \
+ if ((object)->wired_objq.next) { \
+ lck_spin_lock(&vm_objects_wired_lock); \
+ queue_remove(&vm_objects_wired, (object), \
+ vm_object_t, wired_objq); \
+ lck_spin_unlock(&vm_objects_wired_lock); \
+ } \
+ MACRO_END
+#endif /* VM_TAG_ACTIVE_UPDATE */
+
+#define VM_OBJECT_WIRED(object, tag) \
+ MACRO_BEGIN \
+ assert(VM_KERN_MEMORY_NONE != (tag)); \
+ assert(VM_KERN_MEMORY_NONE == (object)->wire_tag); \
+ (object)->wire_tag = (tag); \
+ if (!VM_TAG_ACTIVE_UPDATE) { \
+ VM_OBJECT_WIRED_ENQUEUE((object)); \
} \
MACRO_END
-#define VM_OBJECT_UNWIRED(object) \
- MACRO_BEGIN \
- (object)->wire_tag = VM_KERN_MEMORY_NONE; \
- if (((object)->purgable == VM_PURGABLE_DENY) && (object)->objq.next) \
- { \
- lck_spin_lock(&vm_objects_wired_lock); \
- queue_remove(&vm_objects_wired, (object), vm_object_t, objq); \
- lck_spin_unlock(&vm_objects_wired_lock); \
- } \
+#define VM_OBJECT_UNWIRED(object) \
+ MACRO_BEGIN \
+ if (!VM_TAG_ACTIVE_UPDATE) { \
+ VM_OBJECT_WIRED_DEQUEUE((object)); \
+ } \
+ if (VM_KERN_MEMORY_NONE != (object)->wire_tag) { \
+ vm_tag_update_size((object)->wire_tag, -ptoa_64((object)->wired_page_count)); \
+ (object)->wire_tag = VM_KERN_MEMORY_NONE; \
+ } \
MACRO_END
+// These two macros start & end a C block
+#define VM_OBJECT_WIRED_PAGE_UPDATE_START(object) \
+ MACRO_BEGIN \
+ { \
+ int64_t __wireddelta = 0; vm_tag_t __waswired = (object)->wire_tag;
+
+#define VM_OBJECT_WIRED_PAGE_UPDATE_END(object, tag) \
+ if (__wireddelta) { \
+ boolean_t __overflow __assert_only = \
+ os_add_overflow((object)->wired_page_count, __wireddelta, \
+ &(object)->wired_page_count); \
+ assert(!__overflow); \
+ if (!(object)->pageout && !(object)->no_tag_update) { \
+ if (__wireddelta > 0) { \
+ assert (VM_KERN_MEMORY_NONE != (tag)); \
+ if (VM_KERN_MEMORY_NONE == __waswired) { \
+ VM_OBJECT_WIRED((object), (tag)); \
+ } \
+ vm_tag_update_size((object)->wire_tag, ptoa_64(__wireddelta)); \
+ } else if (VM_KERN_MEMORY_NONE != __waswired) { \
+ assert (VM_KERN_MEMORY_NONE != (object)->wire_tag); \
+ vm_tag_update_size((object)->wire_tag, ptoa_64(__wireddelta)); \
+ if (!(object)->wired_page_count) { \
+ VM_OBJECT_UNWIRED((object)); \
+ } \
+ } \
+ } \
+ } \
+ } \
+ MACRO_END
+
+#define VM_OBJECT_WIRED_PAGE_COUNT(object, delta) \
+ __wireddelta += delta; \
+
+#define VM_OBJECT_WIRED_PAGE_ADD(object, m) \
+ if (!(m)->vmp_private && !(m)->vmp_fictitious) __wireddelta++;
+
+#define VM_OBJECT_WIRED_PAGE_REMOVE(object, m) \
+ if (!(m)->vmp_private && !(m)->vmp_fictitious) __wireddelta--;
+
+
+
+#define OBJECT_LOCK_SHARED 0
+#define OBJECT_LOCK_EXCLUSIVE 1
+
+extern lck_grp_t vm_object_lck_grp;
+extern lck_grp_attr_t vm_object_lck_grp_attr;
+extern lck_attr_t vm_object_lck_attr;
+extern lck_attr_t kernel_object_lck_attr;
+extern lck_attr_t compressor_object_lck_attr;
+
+extern vm_object_t vm_pageout_scan_wants_object;
+
+extern void vm_object_lock(vm_object_t);
+extern boolean_t vm_object_lock_try(vm_object_t);
+extern boolean_t _vm_object_lock_try(vm_object_t);
+extern boolean_t vm_object_lock_avoid(vm_object_t);
+extern void vm_object_lock_shared(vm_object_t);
+extern boolean_t vm_object_lock_yield_shared(vm_object_t);
+extern boolean_t vm_object_lock_try_shared(vm_object_t);
+extern void vm_object_unlock(vm_object_t);
+extern boolean_t vm_object_lock_upgrade(vm_object_t);
+
+/*
+ * Object locking macros
+ */
+
+#define vm_object_lock_init(object) \
+ lck_rw_init(&(object)->Lock, &vm_object_lck_grp, \
+ (((object) == kernel_object || \
+ (object) == vm_submap_object) ? \
+ &kernel_object_lck_attr : \
+ (((object) == compressor_object) ? \
+ &compressor_object_lck_attr : \
+ &vm_object_lck_attr)))
+#define vm_object_lock_destroy(object) lck_rw_destroy(&(object)->Lock, &vm_object_lck_grp)
+
+#define vm_object_lock_try_scan(object) _vm_object_lock_try(object)
+
+/*
+ * CAUTION: the following vm_object_lock_assert_held*() macros merely
+ * check if anyone is holding the lock, but the holder may not necessarily
+ * be the caller...
+ */
+#if MACH_ASSERT || DEBUG
+#define vm_object_lock_assert_held(object) \
+ lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_HELD)
+#define vm_object_lock_assert_shared(object) \
+ lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_SHARED)
+#define vm_object_lock_assert_exclusive(object) \
+ lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_EXCLUSIVE)
+#define vm_object_lock_assert_notheld(object) \
+ lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_NOTHELD)
+#else /* MACH_ASSERT || DEBUG */
+#define vm_object_lock_assert_held(object)
+#define vm_object_lock_assert_shared(object)
+#define vm_object_lock_assert_exclusive(object)
+#define vm_object_lock_assert_notheld(object)
+#endif /* MACH_ASSERT || DEBUG */
+
/*
* Declare procedures that operate on VM objects.
vm_object_offset_t end_offset,
boolean_t allow_partial_reuse);
-__private_extern__ void vm_object_purge(
+__private_extern__ uint64_t vm_object_purge(
vm_object_t object,
int flags);
upl_t *upl,
upl_page_info_t *page_info,
unsigned int *count,
- upl_control_flags_t flags);
+ upl_control_flags_t flags,
+ vm_tag_t tag);
__private_extern__ kern_return_t vm_object_transpose(
vm_object_t object1,
-__private_extern__ vm_object_t vm_object_enter(
+__private_extern__ vm_object_t vm_object_memory_object_associate(
memory_object_t pager,
+ vm_object_t object,
vm_object_size_t size,
- boolean_t internal,
- boolean_t init,
boolean_t check_named);
#define REAP_DATA_FLUSH 3
#if CONFIG_FREEZE
-struct default_freezer_handle;
-
-__private_extern__ kern_return_t
-vm_object_pack(
- unsigned int *purgeable_count,
- unsigned int *wired_count,
- unsigned int *clean_count,
- unsigned int *dirty_count,
- unsigned int dirty_budget,
- boolean_t *shared,
- vm_object_t src_object,
- struct default_freezer_handle *df_handle);
-
-__private_extern__ void
-vm_object_pack_pages(
- unsigned int *wired_count,
- unsigned int *clean_count,
- unsigned int *dirty_count,
- unsigned int dirty_budget,
- vm_object_t src_object,
- struct default_freezer_handle *df_handle);
__private_extern__ void
vm_object_compressed_freezer_pageout(
vm_object_compressed_freezer_done(
void);
-__private_extern__ kern_return_t
-vm_object_pagein(
- vm_object_t object);
#endif /* CONFIG_FREEZE */
__private_extern__ void
#define VM_OBJECT_EVENT_UNBLOCKED 8
#define VM_OBJECT_EVENT_PAGING_ONLY_IN_PROGRESS 9
-#define vm_object_assert_wait(object, event, interruptible) \
- (((object)->all_wanted |= 1 << (event)), \
- assert_wait((event_t)((vm_offset_t)(object)+(event)),(interruptible)))
-
-#define vm_object_wait(object, event, interruptible) \
- (vm_object_assert_wait((object),(event),(interruptible)), \
- vm_object_unlock(object), \
- thread_block(THREAD_CONTINUE_NULL)) \
+#define VM_OBJECT_EVENT_MAX 10 /* 11 bits in "all_wanted", so 0->10 */
-#define thread_sleep_vm_object(object, event, interruptible) \
- lck_rw_sleep(&(object)->Lock, LCK_SLEEP_PROMOTED_PRI, (event_t)(event), (interruptible))
+static __inline__ wait_result_t
+vm_object_assert_wait(
+ vm_object_t object,
+ int event,
+ wait_interrupt_t interruptible)
+{
+ wait_result_t wr;
+
+ vm_object_lock_assert_exclusive(object);
+ assert(event >= 0 && event <= VM_OBJECT_EVENT_MAX);
+
+ object->all_wanted |= 1 << event;
+ wr = assert_wait((event_t)((vm_offset_t)object + event),
+ interruptible);
+ return wr;
+}
+
+static __inline__ wait_result_t
+vm_object_wait(
+ vm_object_t object,
+ int event,
+ wait_interrupt_t interruptible)
+{
+ wait_result_t wr;
+
+ vm_object_assert_wait(object, event, interruptible);
+ vm_object_unlock(object);
+ wr = thread_block(THREAD_CONTINUE_NULL);
+ return wr;
+}
+
+static __inline__ wait_result_t
+thread_sleep_vm_object(
+ vm_object_t object,
+ event_t event,
+ wait_interrupt_t interruptible)
+{
+ wait_result_t wr;
+
+#if DEVELOPMENT || DEBUG
+ if (object->Lock_owner != current_thread())
+ panic("thread_sleep_vm_object: now owner - %p\n", object);
+ object->Lock_owner = 0;
+#endif
+ wr = lck_rw_sleep(&object->Lock,
+ LCK_SLEEP_PROMOTED_PRI,
+ event,
+ interruptible);
+#if DEVELOPMENT || DEBUG
+ object->Lock_owner = current_thread();
+#endif
+ return wr;
+}
-#define vm_object_sleep(object, event, interruptible) \
- (((object)->all_wanted |= 1 << (event)), \
- thread_sleep_vm_object((object), \
- ((vm_offset_t)(object)+(event)), (interruptible)))
+static __inline__ wait_result_t
+vm_object_sleep(
+ vm_object_t object,
+ int event,
+ wait_interrupt_t interruptible)
+{
+ wait_result_t wr;
+
+ vm_object_lock_assert_exclusive(object);
+ assert(event >= 0 && event <= VM_OBJECT_EVENT_MAX);
+
+ object->all_wanted |= 1 << event;
+ wr = thread_sleep_vm_object(object,
+ (event_t)((vm_offset_t)object + event),
+ interruptible);
+ return wr;
+}
+
+static __inline__ void
+vm_object_wakeup(
+ vm_object_t object,
+ int event)
+{
+ vm_object_lock_assert_exclusive(object);
+ assert(event >= 0 && event <= VM_OBJECT_EVENT_MAX);
+
+ if (object->all_wanted & (1 << event))
+ thread_wakeup((event_t)((vm_offset_t)object + event));
+ object->all_wanted &= ~(1 << event);
+}
+
+static __inline__ void
+vm_object_set_wanted(
+ vm_object_t object,
+ int event)
+{
+ vm_object_lock_assert_exclusive(object);
+ assert(event >= 0 && event <= VM_OBJECT_EVENT_MAX);
-#define vm_object_wakeup(object, event) \
- MACRO_BEGIN \
- if ((object)->all_wanted & (1 << (event))) \
- thread_wakeup((event_t)((vm_offset_t)(object) + (event))); \
- (object)->all_wanted &= ~(1 << (event)); \
- MACRO_END
+ object->all_wanted |= (1 << event);
+}
-#define vm_object_set_wanted(object, event) \
- MACRO_BEGIN \
- ((object)->all_wanted |= (1 << (event))); \
- MACRO_END
+static __inline__ int
+vm_object_wanted(
+ vm_object_t object,
+ int event)
+{
+ vm_object_lock_assert_held(object);
+ assert(event >= 0 && event <= VM_OBJECT_EVENT_MAX);
-#define vm_object_wanted(object, event) \
- ((object)->all_wanted & (1 << (event)))
+ return object->all_wanted & (1 << event);
+}
/*
* Routines implemented as macros
-#define OBJECT_LOCK_SHARED 0
-#define OBJECT_LOCK_EXCLUSIVE 1
-
-extern lck_grp_t vm_object_lck_grp;
-extern lck_grp_attr_t vm_object_lck_grp_attr;
-extern lck_attr_t vm_object_lck_attr;
-extern lck_attr_t kernel_object_lck_attr;
-extern lck_attr_t compressor_object_lck_attr;
-
-extern vm_object_t vm_pageout_scan_wants_object;
-
-extern void vm_object_lock(vm_object_t);
-extern boolean_t vm_object_lock_try(vm_object_t);
-extern boolean_t _vm_object_lock_try(vm_object_t);
-extern boolean_t vm_object_lock_avoid(vm_object_t);
-extern void vm_object_lock_shared(vm_object_t);
-extern boolean_t vm_object_lock_try_shared(vm_object_t);
-
-/*
- * Object locking macros
- */
-
-#define vm_object_lock_init(object) \
- lck_rw_init(&(object)->Lock, &vm_object_lck_grp, \
- (((object) == kernel_object || \
- (object) == vm_submap_object) ? \
- &kernel_object_lck_attr : \
- (((object) == compressor_object) ? \
- &compressor_object_lck_attr : \
- &vm_object_lck_attr)))
-#define vm_object_lock_destroy(object) lck_rw_destroy(&(object)->Lock, &vm_object_lck_grp)
-
-#define vm_object_unlock(object) lck_rw_done(&(object)->Lock)
-#define vm_object_lock_upgrade(object) lck_rw_lock_shared_to_exclusive(&(object)->Lock)
-#define vm_object_lock_try_scan(object) _vm_object_lock_try(object)
-
-/*
- * CAUTION: the following vm_object_lock_assert_held*() macros merely
- * check if anyone is holding the lock, but the holder may not necessarily
- * be the caller...
- */
-#if MACH_ASSERT || DEBUG
-#define vm_object_lock_assert_held(object) \
- lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_HELD)
-#define vm_object_lock_assert_shared(object) \
- lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_SHARED)
-#define vm_object_lock_assert_exclusive(object) \
- lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_EXCLUSIVE)
-#define vm_object_lock_assert_notheld(object) \
- lck_rw_assert(&(object)->Lock, LCK_RW_ASSERT_NOTHELD)
-#else /* MACH_ASSERT || DEBUG */
-#define vm_object_lock_assert_held(object)
-#define vm_object_lock_assert_shared(object)
-#define vm_object_lock_assert_exclusive(object)
-#define vm_object_lock_assert_notheld(object)
-#endif /* MACH_ASSERT || DEBUG */
-
#define vm_object_round_page(x) (((vm_object_offset_t)(x) + PAGE_MASK) & ~((signed)PAGE_MASK))
#define vm_object_trunc_page(x) ((vm_object_offset_t)(x) & ~((signed)PAGE_MASK))
extern void vm_object_cache_remove(vm_object_t);
extern int vm_object_cache_evict(int, int);
+#define VM_OBJECT_OWNER_DISOWNED ((task_t) -1)
+#define VM_OBJECT_OWNER(object) \
+ ((((object)->purgable == VM_PURGABLE_DENY && \
+ (object)->vo_ledger_tag == 0) || \
+ (object)->vo_owner == TASK_NULL) \
+ ? TASK_NULL /* not owned */ \
+ : (((object)->vo_owner == VM_OBJECT_OWNER_DISOWNED) \
+ ? kernel_task /* disowned -> kernel */ \
+ : (object)->vo_owner)) /* explicit owner */ \
+
+extern void vm_object_ledger_tag_ledgers(
+ vm_object_t object,
+ int *ledger_idx_volatile,
+ int *ledger_idx_nonvolatile,
+ int *ledger_idx_volatile_compressed,
+ int *ledger_idx_nonvolatile_compressed,
+ boolean_t *do_footprint);
+extern kern_return_t vm_object_ownership_change(
+ vm_object_t object,
+ int ledger_tag,
+ task_t owner,
+ boolean_t task_objq_locked);
+
#endif /* _VM_VM_OBJECT_H_ */