X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..5c9f46613a83ebfc29a5b1f099448259e96a98f0:/osfmk/vm/vm_object.h diff --git a/osfmk/vm/vm_object.h b/osfmk/vm/vm_object.h index a093bf2f4..1ef57792d 100644 --- a/osfmk/vm/vm_object.h +++ b/osfmk/vm/vm_object.h @@ -66,6 +66,8 @@ #ifndef _VM_VM_OBJECT_H_ #define _VM_VM_OBJECT_H_ +#include +#include #include #include @@ -77,7 +79,6 @@ #include #include #include -#include #include #include #include @@ -85,11 +86,26 @@ #include #include -#if MACH_PAGEMAP #include -#endif /* MACH_PAGEMAP */ + +#include +#include + +#if VM_OBJECT_TRACKING +#include +#include +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 +#define VM_OBJECT_TRACKING_OP_TRUESHARE 3 +#endif /* VM_OBJECT_TRACKING */ struct vm_page; +struct vm_shared_region_slide_info; /* * Types defined: @@ -105,25 +121,59 @@ struct vm_object_fault_info { vm_behavior_t behavior; vm_map_offset_t lo_offset; vm_map_offset_t hi_offset; - boolean_t no_cache; + unsigned int + /* boolean_t */ no_cache:1, + /* boolean_t */ stealth:1, + /* boolean_t */ io_sync:1, + /* boolean_t */ cs_bypass:1, + /* boolean_t */ mark_zf_absent:1, + /* boolean_t */ batch_pmap_op:1, + __vm_object_fault_info_unused_bits:26; + int pmap_options; }; +#define vo_size vo_un1.vou_size +#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 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 */ - vm_object_size_t size; /* Object size (only valid - * if internal) - */ +#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 + * external object in cache + */ + } vo_un1; + 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 */ + const 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 * a copy of my changed pages, @@ -133,7 +183,18 @@ struct vm_object { * copy_call. */ struct vm_object *shadow; /* My shadow */ - vm_object_offset_t shadow_offset; /* Offset into shadow */ + + union { + vm_object_offset_t vou_shadow_offset; /* Offset into shadow */ + 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; + } vo_un2; memory_object_t pager; /* Where to get data */ vm_object_offset_t paging_offset; /* Offset into memory object */ @@ -142,7 +203,27 @@ struct vm_object { memory_object_copy_strategy_t copy_strategy; /* How to handle data copy */ - int paging_in_progress; +#if __LP64__ + /* + * Some user processes (mostly VirtualMachine software) take a large + * number of UPLs (via IOMemoryDescriptors) to wire pages in large + * VM objects and overflow the 16-bit "activity_in_progress" counter. + * Since we never enforced any limit there, let's give them 32 bits + * for backwards compatibility's sake. + */ + unsigned int paging_in_progress:16, + __object1_unused_bits:16; + unsigned int activity_in_progress; +#else /* __LP64__ */ + /* + * On 32-bit platforms, enlarging "activity_in_progress" would increase + * the size of "struct vm_object". Since we don't know of any actual + * overflow of these counters on these platforms, let's keep the + * counters as 16-bit integers. + */ + unsigned short paging_in_progress; + unsigned short activity_in_progress; +#endif /* __LP64__ */ /* The memory object ports are * being used (e.g., for pagein * or pageout) -- don't change @@ -150,6 +231,7 @@ struct vm_object { * don't collapse, destroy or * terminate) */ + unsigned int /* boolean_t array */ all_wanted:11, /* Bit array of "want to be * awakened" notations. See @@ -174,15 +256,6 @@ struct vm_object { * 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 @@ -193,17 +266,11 @@ struct vm_object { /* 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 */ silent_overwrite:1, - /* Allow full page overwrite - * without data_request if - * page is absent */ - /* 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 @@ -243,24 +310,19 @@ struct vm_object { * 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 @@ -270,29 +332,45 @@ struct vm_object { 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 */ - not_in_use:23; /* for expansion */ - -#ifdef UPL_DEBUG + transposed:1, /* object was transposed with another */ + mapping_in_progress:1, /* pager being mapped/unmapped */ + phantom_isssd:1, + volatile_empty:1, + volatile_fault:1, + all_reusable:1, + blocked_access:1, + set_cache_attr:1, + object_slid:1, + purgeable_queue_type:2, + purgeable_queue_group:3, + io_tracking:1, + 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 */ + __object2_unused_bits:5; /* for expansion */ + + uint8_t scan_collisions; + vm_tag_t wire_tag; + uint8_t __object4_unused_bits[2]; + +#if CONFIG_PHANTOM_CACHE + uint32_t phantom_object_id; +#endif +#if CONFIG_IOSCHED || UPL_DEBUG queue_head_t uplq; /* List of outstanding upls */ -#endif /* UPL_DEBUG */ +#endif #ifdef VM_PIP_DEBUG /* @@ -306,41 +384,27 @@ struct vm_object { } pip_holders[VM_PIP_DEBUG_MAX_REFS]; #endif /* VM_PIP_DEBUG */ - queue_chain_t objq; /* object queue - currently used for purgable queues */ -}; + queue_chain_t objq; /* object queue - currently used for purgable queues */ -#define VM_PAGE_REMOVE(page) \ - MACRO_BEGIN \ - vm_page_t __page = (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); \ - MACRO_END +#if DEBUG + void *purgeable_owner_bt[16]; + task_t vo_purgeable_volatilizer; /* who made it volatile? */ + void *purgeable_volatilizer_bt[16]; +#endif /* DEBUG */ +}; -#define VM_PAGE_INSERT(page, object) \ - MACRO_BEGIN \ - vm_page_t __page = (page); \ - vm_object_t __object = (object); \ - queue_enter(&__object->memq, __page, vm_page_t, listq); \ - __object->memq_hint = __page; \ - MACRO_END +#define VM_OBJECT_PURGEABLE_FAULT_ERROR(object) \ + ((object)->volatile_fault && \ + ((object)->purgable == VM_PURGABLE_VOLATILE || \ + (object)->purgable == VM_PURGABLE_EMPTY)) -__private_extern__ +extern vm_object_t kernel_object; /* the single kernel object */ -__private_extern__ +extern +vm_object_t compressor_object; /* the single compressor object */ + +extern unsigned int vm_object_absent_max; /* maximum number of absent pages at a time for each object */ @@ -348,40 +412,151 @@ unsigned int vm_object_absent_max; /* maximum number of absent pages # 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_mutex_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; + +#ifndef VM_TAG_ACTIVE_UPDATE +#error VM_TAG_ACTIVE_UPDATE +#endif + +#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 \ + && ((object)->purgable == VM_PURGABLE_DENY)) \ + { \ + lck_spin_lock(&vm_objects_wired_lock); \ + assert(!(object)->objq.next); \ + assert(!(object)->objq.prev); \ + queue_enter(&vm_objects_wired, (object), vm_object_t, objq); \ + lck_spin_unlock(&vm_objects_wired_lock); \ + } \ + MACRO_END + +#define VM_OBJECT_UNWIRED(object) \ + MACRO_BEGIN \ + if (!VM_TAG_ACTIVE_UPDATE \ + && ((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); \ + } \ + 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, \ + (unsigned int *)(uintptr_t)&(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->private && !m->fictitious) __wireddelta++; + +#define VM_OBJECT_WIRED_PAGE_REMOVE(object, m) \ + if (!m->private && !m->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); /* - * Macros to allocate and free msync_reqs + * Object locking macros */ -#define msync_req_alloc(msr) \ - MACRO_BEGIN \ - (msr) = (msync_req_t)kalloc(sizeof(struct msync_req)); \ - mutex_init(&(msr)->msync_req_lock, 0); \ - msr->flag = VM_MSYNC_INITIALIZED; \ - MACRO_END -#define msync_req_free(msr) \ - (kfree((msr), sizeof(struct msync_req))) +#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 */ -#define msr_lock(msr) mutex_lock(&(msr)->msync_req_lock) -#define msr_unlock(msr) mutex_unlock(&(msr)->msync_req_lock) /* * Declare procedures that operate on VM objects. */ -__private_extern__ void vm_object_bootstrap(void) __attribute__((section("__TEXT, initcode"))); +__private_extern__ void vm_object_bootstrap(void); __private_extern__ void vm_object_init(void); @@ -389,8 +564,7 @@ __private_extern__ void vm_object_init_lck_grp(void); __private_extern__ void vm_object_reaper_init(void); -__private_extern__ vm_object_t vm_object_allocate( - vm_object_size_t size); +__private_extern__ vm_object_t vm_object_allocate(vm_object_size_t size); __private_extern__ void _vm_object_allocate(vm_object_size_t size, vm_object_t object); @@ -429,8 +603,8 @@ __private_extern__ void vm_object_res_deallocate( vm_object_t RLObject = (object); \ vm_object_lock_assert_shared(object); \ assert((RLObject)->ref_count > 0); \ - OSAddAtomic(1, (SInt32 *)&(RLObject)->ref_count); \ - assert((RLObject)->ref_count > 1); \ + OSAddAtomic(1, &(RLObject)->ref_count); \ + assert((RLObject)->ref_count > 0); \ /* XXX we would need an atomic version of the following ... */ \ vm_object_res_reference(RLObject); \ MACRO_END @@ -445,8 +619,8 @@ __private_extern__ void vm_object_reference( MACRO_BEGIN \ vm_object_t RObject = (object); \ if (RObject) { \ - vm_object_lock(RObject); \ - vm_object_reference_locked(RObject); \ + vm_object_lock_shared(RObject); \ + vm_object_reference_shared(RObject); \ vm_object_unlock(RObject); \ } \ MACRO_END @@ -468,6 +642,15 @@ __private_extern__ void vm_object_pmap_protect( vm_map_offset_t pmap_start, vm_prot_t prot); +__private_extern__ void vm_object_pmap_protect_options( + vm_object_t object, + vm_object_offset_t offset, + vm_object_size_t size, + pmap_t pmap, + vm_map_offset_t pmap_start, + vm_prot_t prot, + int options); + __private_extern__ void vm_object_page_remove( vm_object_t object, vm_object_offset_t start, @@ -477,16 +660,33 @@ __private_extern__ void vm_object_deactivate_pages( vm_object_t object, vm_object_offset_t offset, vm_object_size_t size, - boolean_t kill_page); + boolean_t kill_page, + boolean_t reusable_page, + struct pmap *pmap, + vm_map_offset_t pmap_offset); + +__private_extern__ void vm_object_reuse_pages( + vm_object_t object, + vm_object_offset_t start_offset, + vm_object_offset_t end_offset, + boolean_t allow_partial_reuse); -__private_extern__ unsigned int vm_object_purge( - vm_object_t object); +__private_extern__ void vm_object_purge( + vm_object_t object, + int flags); __private_extern__ kern_return_t vm_object_purgable_control( vm_object_t object, vm_purgable_t control, int *state); +__private_extern__ kern_return_t vm_object_get_page_counts( + vm_object_t object, + vm_object_offset_t offset, + vm_object_size_t size, + unsigned int *resident_page_count, + unsigned int *dirty_page_count); + __private_extern__ boolean_t vm_object_coalesce( vm_object_t prev_object, vm_object_t next_object, @@ -524,7 +724,7 @@ __private_extern__ kern_return_t vm_object_copy_slowly( vm_object_t src_object, vm_object_offset_t src_offset, vm_object_size_t size, - int interruptible, + boolean_t interruptible, vm_object_t *_result_object); __private_extern__ vm_object_t vm_object_copy_delayed( @@ -542,6 +742,9 @@ __private_extern__ kern_return_t vm_object_destroy( __private_extern__ void vm_object_pager_create( vm_object_t object); +__private_extern__ void vm_object_compressor_pager_create( + vm_object_t object); + __private_extern__ void vm_object_page_map( vm_object_t object, vm_object_offset_t offset, @@ -557,7 +760,8 @@ __private_extern__ kern_return_t vm_object_upl_request( upl_t *upl, upl_page_info_t *page_info, unsigned int *count, - int flags); + upl_control_flags_t flags, + vm_tag_t tag); __private_extern__ kern_return_t vm_object_transpose( vm_object_t object1, @@ -592,11 +796,10 @@ __private_extern__ kern_return_t vm_object_lock_request( -__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); @@ -604,7 +807,8 @@ __private_extern__ void vm_object_cluster_size( vm_object_t object, vm_object_offset_t *start, vm_size_t *length, - vm_object_fault_info_t fault_info); + vm_object_fault_info_t fault_info, + uint32_t *io_streaming); __private_extern__ kern_return_t vm_object_populate_with_private( vm_object_t object, @@ -612,6 +816,10 @@ __private_extern__ kern_return_t vm_object_populate_with_private( ppnum_t phys_page, vm_size_t size); +__private_extern__ void vm_object_change_wimg_mode( + vm_object_t object, + unsigned int wimg_mode); + extern kern_return_t adjust_vm_object_cache( vm_size_t oval, vm_size_t nval); @@ -628,7 +836,45 @@ extern kern_return_t vm_object_range_op( vm_object_offset_t offset_beg, vm_object_offset_t offset_end, int ops, - int *range); + uint32_t *range); + + +__private_extern__ void vm_object_reap_pages( + vm_object_t object, + int reap_type); +#define REAP_REAP 0 +#define REAP_TERMINATE 1 +#define REAP_PURGEABLE 2 +#define REAP_DATA_FLUSH 3 + +#if CONFIG_FREEZE + +__private_extern__ void +vm_object_compressed_freezer_pageout( + vm_object_t object); + +__private_extern__ void +vm_object_compressed_freezer_done( + void); + +#endif /* CONFIG_FREEZE */ + +__private_extern__ void +vm_object_pageout( + vm_object_t object); + +#if CONFIG_IOSCHED +struct io_reprioritize_req { + uint64_t blkno; + uint32_t len; + int priority; + struct vnode *devvp; + queue_chain_t io_reprioritize_list; +}; +typedef struct io_reprioritize_req *io_reprioritize_req_t; + +extern void vm_io_reprioritize_init(void); +#endif /* * Event waiting handling @@ -637,42 +883,122 @@ extern kern_return_t vm_object_range_op( #define VM_OBJECT_EVENT_INITIALIZED 0 #define VM_OBJECT_EVENT_PAGER_READY 1 #define VM_OBJECT_EVENT_PAGING_IN_PROGRESS 2 +#define VM_OBJECT_EVENT_MAPPING_IN_PROGRESS 3 #define VM_OBJECT_EVENT_LOCK_IN_PROGRESS 4 #define VM_OBJECT_EVENT_UNCACHING 5 #define VM_OBJECT_EVENT_COPY_CALL 6 #define VM_OBJECT_EVENT_CACHING 7 +#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_EVENT_MAX 10 /* 11 bits in "all_wanted", so 0->10 */ -#define vm_object_wait(object, event, interruptible) \ - (vm_object_assert_wait((object),(event),(interruptible)), \ - vm_object_unlock(object), \ - thread_block(THREAD_CONTINUE_NULL)) \ +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 thread_sleep_vm_object(object, event, interruptible) \ - lck_rw_sleep(&(object)->Lock, LCK_SLEEP_DEFAULT, (event_t)(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_sleep(object, event, interruptible) \ - (((object)->all_wanted |= 1 << (event)), \ - thread_sleep_vm_object((object), \ - ((vm_offset_t)(object)+(event)), (interruptible))) + object->all_wanted |= (1 << event); +} -#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 - -#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 @@ -681,8 +1007,9 @@ extern kern_return_t vm_object_range_op( #include #define VM_PIP_DEBUG_BEGIN(object) \ MACRO_BEGIN \ - if ((object)->paging_in_progress < VM_PIP_DEBUG_MAX_REFS) { \ - int pip = (object)->paging_in_progress; \ + int pip = ((object)->paging_in_progress + \ + (object)->activity_in_progress); \ + if (pip < VM_PIP_DEBUG_MAX_REFS) { \ (void) OSBacktrace(&(object)->pip_holders[pip].pip_retaddr[0], \ VM_PIP_DEBUG_STACK_FRAMES); \ } \ @@ -691,28 +1018,60 @@ extern kern_return_t vm_object_range_op( #define VM_PIP_DEBUG_BEGIN(object) #endif /* VM_PIP_DEBUG */ -#define vm_object_paging_begin(object) \ +#define vm_object_activity_begin(object) \ + MACRO_BEGIN \ + vm_object_lock_assert_exclusive((object)); \ + VM_PIP_DEBUG_BEGIN((object)); \ + (object)->activity_in_progress++; \ + if ((object)->activity_in_progress == 0) { \ + panic("vm_object_activity_begin(%p): overflow\n", (object));\ + } \ + MACRO_END + +#define vm_object_activity_end(object) \ + MACRO_BEGIN \ + vm_object_lock_assert_exclusive((object)); \ + if ((object)->activity_in_progress == 0) { \ + panic("vm_object_activity_end(%p): underflow\n", (object));\ + } \ + (object)->activity_in_progress--; \ + if ((object)->paging_in_progress == 0 && \ + (object)->activity_in_progress == 0) \ + vm_object_wakeup((object), \ + VM_OBJECT_EVENT_PAGING_IN_PROGRESS); \ + MACRO_END + +#define vm_object_paging_begin(object) \ MACRO_BEGIN \ vm_object_lock_assert_exclusive((object)); \ - assert((object)->paging_in_progress >= 0); \ VM_PIP_DEBUG_BEGIN((object)); \ (object)->paging_in_progress++; \ + if ((object)->paging_in_progress == 0) { \ + panic("vm_object_paging_begin(%p): overflow\n", (object));\ + } \ MACRO_END -#define vm_object_paging_end(object) \ +#define vm_object_paging_end(object) \ MACRO_BEGIN \ vm_object_lock_assert_exclusive((object)); \ - assert((object)->paging_in_progress > 0); \ - if (--(object)->paging_in_progress == 0) { \ - vm_object_wakeup(object, \ - VM_OBJECT_EVENT_PAGING_IN_PROGRESS); \ + if ((object)->paging_in_progress == 0) { \ + panic("vm_object_paging_end(%p): underflow\n", (object));\ + } \ + (object)->paging_in_progress--; \ + if ((object)->paging_in_progress == 0) { \ + vm_object_wakeup((object), \ + VM_OBJECT_EVENT_PAGING_ONLY_IN_PROGRESS); \ + if ((object)->activity_in_progress == 0) \ + vm_object_wakeup((object), \ + VM_OBJECT_EVENT_PAGING_IN_PROGRESS); \ } \ MACRO_END #define vm_object_paging_wait(object, interruptible) \ MACRO_BEGIN \ vm_object_lock_assert_exclusive((object)); \ - while ((object)->paging_in_progress != 0) { \ + while ((object)->paging_in_progress != 0 || \ + (object)->activity_in_progress != 0) { \ wait_result_t _wr; \ \ _wr = vm_object_sleep((object), \ @@ -724,58 +1083,60 @@ extern kern_return_t vm_object_range_op( } \ MACRO_END +#define vm_object_paging_only_wait(object, interruptible) \ + MACRO_BEGIN \ + vm_object_lock_assert_exclusive((object)); \ + while ((object)->paging_in_progress != 0) { \ + wait_result_t _wr; \ + \ + _wr = vm_object_sleep((object), \ + VM_OBJECT_EVENT_PAGING_ONLY_IN_PROGRESS,\ + (interruptible)); \ + \ + /*XXX if ((interruptible) && (_wr != THREAD_AWAKENED))*/\ + /*XXX break; */ \ + } \ + MACRO_END -#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 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 void vm_object_lock_shared(vm_object_t); -extern boolean_t vm_object_lock_try_shared(vm_object_t); +#define vm_object_mapping_begin(object) \ + MACRO_BEGIN \ + vm_object_lock_assert_exclusive((object)); \ + assert(! (object)->mapping_in_progress); \ + (object)->mapping_in_progress = TRUE; \ + MACRO_END -/* - * Object locking macros - */ +#define vm_object_mapping_end(object) \ + MACRO_BEGIN \ + vm_object_lock_assert_exclusive((object)); \ + assert((object)->mapping_in_progress); \ + (object)->mapping_in_progress = FALSE; \ + vm_object_wakeup((object), \ + VM_OBJECT_EVENT_MAPPING_IN_PROGRESS); \ + MACRO_END -#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 : \ - &vm_object_lck_attr)) -#define vm_object_lock_destroy(object) lck_rw_destroy(&(object)->Lock, &vm_object_lck_grp) +#define vm_object_mapping_wait(object, interruptible) \ + MACRO_BEGIN \ + vm_object_lock_assert_exclusive((object)); \ + while ((object)->mapping_in_progress) { \ + wait_result_t _wr; \ + \ + _wr = vm_object_sleep((object), \ + VM_OBJECT_EVENT_MAPPING_IN_PROGRESS, \ + (interruptible)); \ + /*XXX if ((interruptible) && (_wr != THREAD_AWAKENED))*/\ + /*XXX break; */ \ + } \ + assert(!(object)->mapping_in_progress); \ + MACRO_END -#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) lck_rw_try_lock_exclusive(&(object)->Lock) -/* - * 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 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) -#else /* DEBUG */ -#define vm_object_lock_assert_held(object) -#define vm_object_lock_assert_shared(object) -#define vm_object_lock_assert_exclusive(object) -#endif /* 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_add(vm_object_t); +extern void vm_object_cache_remove(vm_object_t); +extern int vm_object_cache_evict(int, int); + #endif /* _VM_VM_OBJECT_H_ */