X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/0b4c1975fb5e4eccf1012a35081f7e7799b81046..0a7de7458d150b5d4dffc935ba399be265ef0a1a:/osfmk/vm/vm_resident.c diff --git a/osfmk/vm/vm_resident.c b/osfmk/vm/vm_resident.c index dae49ac1a..4cdb91692 100644 --- a/osfmk/vm/vm_resident.c +++ b/osfmk/vm/vm_resident.c @@ -2,7 +2,7 @@ * Copyright (c) 2000-2009 Apple Inc. All rights reserved. * * @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 @@ -11,10 +11,10 @@ * 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 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, @@ -22,34 +22,34 @@ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. - * + * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * @OSF_COPYRIGHT@ */ -/* +/* * Mach Operating System * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University * All Rights Reserved. - * + * * Permission to use, copy, modify and distribute this software and its * documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. - * + * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. - * + * * Carnegie Mellon requests users of this software to return to - * + * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 - * + * * any improvements or extensions that they make and grant Carnegie Mellon * the rights to redistribute these changes. */ @@ -64,6 +64,7 @@ #include #include +#include #include #include @@ -71,58 +72,90 @@ #include #include #include +#include #include #include #include #include #include +#include #include #include #include #include #include -#include /* kernel_memory_allocate() */ +#include /* kernel_memory_allocate() */ #include #include +#include #include -#include /* (BRINGUP) */ -#include /* (BRINGUP) */ +#include +#include #include #include #include +#include + +#if CONFIG_PHANTOM_CACHE +#include +#endif #include +#include -#if CONFIG_EMBEDDED -#include -#endif -#include +#if MACH_ASSERT + +#define ASSERT_PMAP_FREE(mem) pmap_assert_free(VM_PAGE_GET_PHYS_PAGE(mem)) + +#else /* MACH_ASSERT */ + +#define ASSERT_PMAP_FREE(mem) /* nothing */ -boolean_t vm_page_free_verify = TRUE; +#endif /* MACH_ASSERT */ -uint_t vm_lopage_free_count = 0; -uint_t vm_lopage_free_limit = 0; -uint_t vm_lopage_lowater = 0; -boolean_t vm_lopage_refill = FALSE; -boolean_t vm_lopage_needed = FALSE; +char vm_page_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; +char vm_page_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; +char vm_page_non_speculative_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; +char vm_page_active_or_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; -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; +#if CONFIG_SECLUDED_MEMORY +struct vm_page_secluded_data vm_page_secluded; +void secluded_suppression_init(void); +#endif /* CONFIG_SECLUDED_MEMORY */ -int speculative_age_index = 0; -int speculative_steal_index = 0; +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]; +boolean_t hibernation_vmqueues_inspection = FALSE; /* Tracks if the hibernation code is looking at the VM queues. + * Updated and checked behind the vm_page_queues_lock. */ -__private_extern__ void vm_page_init_lck_grp(void); +__private_extern__ void vm_page_init_lck_grp(void); -static void vm_page_free_prepare(vm_page_t page); +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; +uint32_t vm_packed_from_vm_pages_array_mask = VM_PACKED_FROM_VM_PAGES_ARRAY; +uint32_t vm_packed_pointer_shift = VM_PACKED_POINTER_SHIFT; /* * Associated with page of user-allocatable memory is a @@ -137,7 +170,7 @@ static void vm_page_free_prepare(vm_page_t page); 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 @@ -148,27 +181,54 @@ int vm_page_pages; * or VP, table.] */ typedef struct { - vm_page_t pages; -#if MACH_PAGE_HASH_STATS - int cur_count; /* current count */ - int hi_count; /* high water mark */ + 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 +#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 */ -unsigned int vm_page_bucket_lock_count = 0; /* How big is array of locks? */ +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 */ +unsigned int vm_page_bucket_lock_count = 0; /* How big is array of locks? */ -lck_spin_t *vm_page_bucket_locks; +#ifndef VM_TAG_ACTIVE_UPDATE +#error VM_TAG_ACTIVE_UPDATE +#endif +#ifndef VM_MAX_TAG_ZONES +#error VM_MAX_TAG_ZONES +#endif + +boolean_t vm_tag_active_update = VM_TAG_ACTIVE_UPDATE; +lck_spin_t *vm_page_bucket_locks; +lck_spin_t vm_objects_wired_lock; +lck_spin_t vm_allocation_sites_lock; + +vm_allocation_site_t vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC + 1]; +vm_allocation_site_t * vm_allocation_sites[VM_MAX_TAG_VALUE]; +#if VM_MAX_TAG_ZONES +vm_allocation_zone_total_t ** vm_allocation_zone_totals; +#endif /* VM_MAX_TAG_ZONES */ +vm_tag_t vm_allocation_tag_highest; -#if MACH_PAGE_HASH_STATS +#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 */ + + + +#if MACH_PAGE_HASH_STATS /* This routine is only for debug. It is intended to be called by * hand by a developer using a kernel debugger. This routine prints * out vm_page_hash table statistics to the kernel debug console. @@ -176,29 +236,30 @@ lck_spin_t *vm_page_bucket_locks; void hash_debug(void) { - int i; - int numbuckets = 0; - int highsum = 0; - int maxdepth = 0; + int i; + int numbuckets = 0; + int highsum = 0; + int maxdepth = 0; for (i = 0; i < vm_page_bucket_count; i++) { if (vm_page_buckets[i].hi_count) { numbuckets++; highsum += vm_page_buckets[i].hi_count; - if (vm_page_buckets[i].hi_count > maxdepth) + if (vm_page_buckets[i].hi_count > maxdepth) { maxdepth = vm_page_buckets[i].hi_count; + } } } printf("Total number of buckets: %d\n", vm_page_bucket_count); printf("Number used buckets: %d = %d%%\n", - numbuckets, 100*numbuckets/vm_page_bucket_count); + numbuckets, 100 * numbuckets / vm_page_bucket_count); printf("Number unused buckets: %d = %d%%\n", - vm_page_bucket_count - numbuckets, - 100*(vm_page_bucket_count-numbuckets)/vm_page_bucket_count); + vm_page_bucket_count - numbuckets, + 100 * (vm_page_bucket_count - numbuckets) / vm_page_bucket_count); printf("Sum of bucket max depth: %d\n", highsum); printf("Average bucket depth: %d.%2d\n", - highsum/vm_page_bucket_count, - highsum%vm_page_bucket_count); + highsum / vm_page_bucket_count, + highsum % vm_page_bucket_count); printf("Maximum bucket depth: %d\n", maxdepth); } #endif /* MACH_PAGE_HASH_STATS */ @@ -214,40 +275,46 @@ hash_debug(void) * module must use the PAGE_SIZE, PAGE_MASK and PAGE_SHIFT * constants. */ -vm_size_t page_size = PAGE_SIZE; -vm_size_t page_mask = PAGE_MASK; -int page_shift = PAGE_SHIFT; +#if defined(__arm__) || defined(__arm64__) +vm_size_t page_size; +vm_size_t page_mask; +int page_shift; +#else +vm_size_t page_size = PAGE_SIZE; +vm_size_t page_mask = PAGE_MASK; +int page_shift = PAGE_SHIFT; +#endif -/* - * Resident page structures are initialized from - * a template (see vm_page_alloc). - * - * When adding a new field to the virtual memory - * object structure, be sure to add initialization - * (see vm_page_bootstrap). - */ -struct vm_page vm_page_template; +vm_page_t vm_pages = VM_PAGE_NULL; +vm_page_t vm_page_array_beginning_addr; +vm_page_t vm_page_array_ending_addr; -vm_page_t vm_pages = VM_PAGE_NULL; -unsigned int vm_pages_count = 0; -ppnum_t vm_page_lowest = 0; +unsigned int vm_pages_count = 0; /* * Resident pages that represent real memory * are allocated from a set of free lists, * one per color. */ -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 */ -queue_head_t vm_page_queue_free[MAX_COLORS]; -vm_page_t vm_page_queue_fictitious; -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 */ +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; + + +struct vm_page_queue_free_head { + vm_page_queue_head_t qhead; +} __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); + +struct vm_page_queue_free_head vm_page_queue_free[MAX_COLORS]; + + +unsigned int vm_page_free_wanted; +unsigned int vm_page_free_wanted_privileged; +#if CONFIG_SECLUDED_MEMORY +unsigned int vm_page_free_wanted_secluded; +#endif /* CONFIG_SECLUDED_MEMORY */ +unsigned int vm_page_free_count; /* * Occasionally, the virtual memory system uses @@ -258,71 +325,116 @@ unsigned int vm_page_free_count_minimum; /* debugging */ * These page structures are allocated the way * most other kernel structures are. */ -zone_t vm_page_zone; +zone_t vm_page_array_zone; +zone_t vm_page_zone; vm_locks_array_t vm_page_locks; -decl_lck_mtx_data(,vm_page_alloc_lock) -unsigned int io_throttle_zero_fill; +decl_lck_mtx_data(, vm_page_alloc_lock) +lck_mtx_ext_t vm_page_alloc_lock_ext; -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; +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. */ -ppnum_t vm_page_fictitious_addr = (ppnum_t) -1; +const 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 + * 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; +const ppnum_t vm_page_guard_addr = (ppnum_t) -2; /* * Resident page structures are also chained on * queues that are used by the page replacement * 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 - * pageout daemon often assignes a higher - * affinity to zf pages + * module. The inactive queue is broken into + * file backed and anonymous for convenience as the + * pageout daemon often assignes a higher + * 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_zf; /* inactive memory queue for zero fill */ -queue_head_t vm_page_queue_throttled; - -unsigned int vm_page_active_count; -unsigned int vm_page_inactive_count; -unsigned int vm_page_throttled_count; -unsigned int vm_page_speculative_count; -unsigned int vm_page_wire_count; -unsigned int vm_page_wire_count_initial; -unsigned int vm_page_gobble_count = 0; -unsigned int vm_page_wire_count_warning = 0; -unsigned int 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 */ +vm_page_queue_head_t vm_page_queue_active __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); +vm_page_queue_head_t vm_page_queue_inactive __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); +#if CONFIG_SECLUDED_MEMORY +vm_page_queue_head_t vm_page_queue_secluded __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); +#endif /* CONFIG_SECLUDED_MEMORY */ +vm_page_queue_head_t vm_page_queue_anonymous __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); /* inactive memory queue for anonymous pages */ +vm_page_queue_head_t vm_page_queue_throttled __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); + +queue_head_t vm_objects_wired; + +void vm_update_darkwake_mode(boolean_t); + +#if CONFIG_BACKGROUND_QUEUE +vm_page_queue_head_t vm_page_queue_background __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); +uint32_t vm_page_background_target; +uint32_t vm_page_background_target_snapshot; +uint32_t vm_page_background_count; +uint64_t vm_page_background_promoted_count; + +uint32_t vm_page_background_internal_count; +uint32_t vm_page_background_external_count; + +uint32_t vm_page_background_mode; +uint32_t vm_page_background_exclude_external; +#endif + +unsigned int vm_page_active_count; +unsigned int vm_page_inactive_count; +#if CONFIG_SECLUDED_MEMORY +unsigned int vm_page_secluded_count; +unsigned int vm_page_secluded_count_free; +unsigned int vm_page_secluded_count_inuse; +#endif /* CONFIG_SECLUDED_MEMORY */ +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_wire_count_on_boot = 0; +unsigned int vm_page_stolen_count = 0; +unsigned int vm_page_wire_count_initial; +unsigned int vm_page_gobble_count = 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; +unsigned int vm_page_speculative_recreated = 0; +unsigned int vm_page_speculative_created = 0; +unsigned int vm_page_speculative_used = 0; #endif -uint64_t max_valid_dma_address = 0xffffffffffffffffULL; -ppnum_t max_valid_low_ppnum = 0xffffffff; +vm_page_queue_head_t vm_page_queue_cleaned __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); + +unsigned int vm_page_cleaned_count = 0; + +uint64_t max_valid_dma_address = 0xffffffffffffffffULL; +ppnum_t max_valid_low_ppnum = PPNUM_MAX; /* @@ -331,14 +443,16 @@ ppnum_t max_valid_low_ppnum = 0xffffffff; * (done here in vm_page_alloc) can trigger the * pageout daemon. */ -unsigned int vm_page_free_target = 0; -unsigned int vm_page_free_min = 0; -unsigned int vm_page_throttle_limit = 0; -uint32_t vm_page_creation_throttle = 0; -unsigned int vm_page_inactive_target = 0; -unsigned int vm_page_inactive_min = 0; -unsigned int vm_page_free_reserved = 0; -unsigned int vm_page_throttle_count = 0; +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; +#if CONFIG_SECLUDED_MEMORY +unsigned int vm_page_secluded_target = 0; +#endif /* CONFIG_SECLUDED_MEMORY */ +unsigned int vm_page_anonymous_min = 0; +unsigned int vm_page_free_reserved = 0; + /* * The VM system has a couple of heuristics for deciding @@ -351,7 +465,7 @@ unsigned int vm_page_throttle_count = 0; boolean_t vm_page_deactivate_hint = TRUE; struct vm_page_stats_reusable vm_page_stats_reusable; - + /* * vm_set_page_size: * @@ -364,52 +478,339 @@ struct vm_page_stats_reusable vm_page_stats_reusable; 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) + if ((page_mask & page_size) != 0) { panic("vm_set_page_size: page size not a power of two"); + } - for (page_shift = 0; ; page_shift++) - if ((1U << page_shift) == page_size) + for (page_shift = 0;; page_shift++) { + if ((1U << page_shift) == page_size) { break; + } + } +} + +#if defined (__x86_64__) + +#define MAX_CLUMP_SIZE 16 +#define DEFAULT_CLUMP_SIZE 4 + +unsigned int vm_clump_size, vm_clump_mask, vm_clump_shift, vm_clump_promote_threshold; + +#if DEVELOPMENT || DEBUG +unsigned long vm_clump_stats[MAX_CLUMP_SIZE + 1]; +unsigned long vm_clump_allocs, vm_clump_inserts, vm_clump_inrange, vm_clump_promotes; + +static inline void +vm_clump_update_stats(unsigned int c) +{ + assert(c <= vm_clump_size); + if (c > 0 && c <= vm_clump_size) { + vm_clump_stats[c] += c; + } + vm_clump_allocs += c; +} +#endif /* if DEVELOPMENT || DEBUG */ + +/* Called once to setup the VM clump knobs */ +static void +vm_page_setup_clump( void ) +{ + unsigned int override, n; + + vm_clump_size = DEFAULT_CLUMP_SIZE; + if (PE_parse_boot_argn("clump_size", &override, sizeof(override))) { + vm_clump_size = override; + } + + if (vm_clump_size > MAX_CLUMP_SIZE) { + panic("vm_page_setup_clump:: clump_size is too large!"); + } + if (vm_clump_size < 1) { + panic("vm_page_setup_clump:: clump_size must be >= 1"); + } + if ((vm_clump_size & (vm_clump_size - 1)) != 0) { + panic("vm_page_setup_clump:: clump_size must be a power of 2"); + } + + vm_clump_promote_threshold = vm_clump_size; + vm_clump_mask = vm_clump_size - 1; + for (vm_clump_shift = 0, n = vm_clump_size; n > 1; n >>= 1, vm_clump_shift++) { + ; + } + +#if DEVELOPMENT || DEBUG + bzero(vm_clump_stats, sizeof(vm_clump_stats)); + vm_clump_allocs = vm_clump_inserts = vm_clump_inrange = vm_clump_promotes = 0; +#endif /* if DEVELOPMENT || DEBUG */ } +#endif /* #if defined (__x86_64__) */ + +#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 */ + unsigned int n, override; - if ( n == 0 ) +#if defined (__x86_64__) + /* adjust #colors because we need to color outside the clump boundary */ + vm_cache_geometry_colors >>= vm_clump_shift; +#endif + 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 ) + } + 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"); - + if ((n & (n - 1)) != 0) { + n = DEFAULT_COLORS; /* use default if all else fails */ + } vm_colors = n; vm_color_mask = n - 1; + + vm_free_magazine_refill_limit = vm_colors * COLOR_GROUPS_TO_STEAL; + +#if defined (__x86_64__) + /* adjust for reduction in colors due to clumping and multiple cores */ + if (real_ncpus) { + vm_free_magazine_refill_limit *= (vm_clump_size * real_ncpus); + } +#endif +} + +/* + * During single threaded early boot we don't initialize all pages. + * This avoids some delay during boot. They'll be initialized and + * added to the free list as needed or after we are multithreaded by + * what becomes the pageout thread. + */ +static boolean_t fill = FALSE; +static unsigned int fillval; +uint_t vm_delayed_count = 0; /* when non-zero, indicates we may have more pages to init */ +ppnum_t delay_above_pnum = PPNUM_MAX; + +/* + * For x86 first 8 Gig initializes quickly and gives us lots of lowmem + mem above to start off with. + * If ARM ever uses delayed page initialization, this value may need to be quite different. + */ +#define DEFAULT_DELAY_ABOVE_PHYS_GB (8) + +/* + * When we have to dip into more delayed pages due to low memory, free up + * a large chunk to get things back to normal. This avoids contention on the + * delayed code allocating page by page. + */ +#define VM_DELAY_PAGE_CHUNK ((1024 * 1024 * 1024) / PAGE_SIZE) + +/* + * Get and initialize the next delayed page. + */ +static vm_page_t +vm_get_delayed_page(int grab_options) +{ + vm_page_t p; + ppnum_t pnum; + + /* + * Get a new page if we have one. + */ + lck_mtx_lock(&vm_page_queue_free_lock); + if (vm_delayed_count == 0) { + lck_mtx_unlock(&vm_page_queue_free_lock); + return NULL; + } + if (!pmap_next_page(&pnum)) { + vm_delayed_count = 0; + lck_mtx_unlock(&vm_page_queue_free_lock); + return NULL; + } + + assert(vm_delayed_count > 0); + --vm_delayed_count; + + p = &vm_pages[vm_pages_count]; + assert(p < vm_page_array_ending_addr); + vm_page_init(p, pnum, FALSE); + ++vm_pages_count; + ++vm_page_pages; + lck_mtx_unlock(&vm_page_queue_free_lock); + + /* + * These pages were initially counted as wired, undo that now. + */ + if (grab_options & VM_PAGE_GRAB_Q_LOCK_HELD) { + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + } else { + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); + vm_page_lockspin_queues(); + } + --vm_page_wire_count; + --vm_page_wire_count_initial; + if (vm_page_wire_count_on_boot != 0) { + --vm_page_wire_count_on_boot; + } + if (!(grab_options & VM_PAGE_GRAB_Q_LOCK_HELD)) { + vm_page_unlock_queues(); + } + + + if (fill) { + fillPage(pnum, fillval); + } + return p; } +static void vm_page_module_init_delayed(void); + +/* + * Free all remaining delayed pages to the free lists. + */ +void +vm_free_delayed_pages(void) +{ + vm_page_t p; + vm_page_t list = NULL; + uint_t cnt = 0; + vm_offset_t start_free_page; + vm_size_t free_size; + + while ((p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE)) != NULL) { + if (vm_himemory_mode) { + vm_page_release(p, FALSE); + } else { + p->vmp_snext = list; + list = p; + } + ++cnt; + } + + /* + * Free the pages in reverse order if not himemory mode. + * Hence the low memory pages will be first on free lists. (LIFO) + */ + while (list != NULL) { + p = list; + list = p->vmp_snext; + p->vmp_snext = NULL; + vm_page_release(p, FALSE); + } +#if DEVELOPMENT || DEBUG + kprintf("vm_free_delayed_pages: freed %d pages\n", cnt); +#endif + + /* + * Free up any unused full pages at the end of the vm_pages[] array + */ + start_free_page = round_page((vm_offset_t)&vm_pages[vm_pages_count]); + if (start_free_page < (vm_offset_t)vm_page_array_ending_addr) { + free_size = trunc_page((vm_offset_t)vm_page_array_ending_addr - start_free_page); + if (free_size > 0) { +#if DEVELOPMENT || DEBUG + kprintf("Freeing final unused %ld bytes from vm_pages[] at 0x%lx\n", + (long)free_size, (long)start_free_page); +#endif + pmap_pv_fixup(start_free_page, free_size); + ml_static_mfree(start_free_page, free_size); + vm_page_array_ending_addr = (void *)start_free_page; + + /* + * Note there's no locking here, as only this thread will ever change this value. + * The reader, vm_page_diagnose, doesn't grab any locks for the counts it looks at. + */ + --vm_page_stolen_count; + } + } + + + /* + * now we can create the VM page array zone + */ + vm_page_module_init_delayed(); +} + +/* + * Try and free up enough delayed pages to match a contig memory allocation. + */ +static void +vm_free_delayed_pages_contig( + uint_t npages, + ppnum_t max_pnum, + ppnum_t pnum_mask) +{ + vm_page_t p; + ppnum_t pnum; + uint_t cnt = 0; + + /* + * Treat 0 as the absolute max page number. + */ + if (max_pnum == 0) { + max_pnum = PPNUM_MAX; + } + + /* + * Free till we get a properly aligned start page + */ + for (;;) { + p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE); + if (p == NULL) { + return; + } + pnum = VM_PAGE_GET_PHYS_PAGE(p); + vm_page_release(p, FALSE); + if (pnum >= max_pnum) { + return; + } + if ((pnum & pnum_mask) == 0) { + break; + } + } + + /* + * Having a healthy pool of free pages will help performance. We don't + * want to fall back to the delayed code for every page allocation. + */ + if (vm_page_free_count < VM_DELAY_PAGE_CHUNK) { + npages += VM_DELAY_PAGE_CHUNK; + } + + /* + * Now free up the pages + */ + for (cnt = 1; cnt < npages; ++cnt) { + p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE); + if (p == NULL) { + return; + } + vm_page_release(p, FALSE); + } +} -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; + +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 @@ -426,14 +827,19 @@ vm_page_init_lck_grp(void) 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(); } +#define ROUNDUP_NEXTP2(X) (1U << (32 - __builtin_clz((X) - 1))) + void vm_page_init_local_q() { - unsigned int num_cpus; - unsigned int i; - struct vplq *t_local_q; + unsigned int num_cpus; + unsigned int i; + struct vplq *t_local_q; num_cpus = ml_get_max_cpus(); @@ -441,15 +847,25 @@ vm_page_init_local_q() * no point in this for a uni-processor system */ if (num_cpus >= 2) { - t_local_q = (struct vplq *)kalloc(num_cpus * sizeof(struct vplq)); +#if KASAN + /* KASAN breaks the expectation of a size-aligned object by adding a + * redzone, so explicitly align. */ + t_local_q = (struct vplq *)kalloc(num_cpus * sizeof(struct vplq) + VM_PACKED_POINTER_ALIGNMENT); + t_local_q = (void *)(((uintptr_t)t_local_q + (VM_PACKED_POINTER_ALIGNMENT - 1)) & ~(VM_PACKED_POINTER_ALIGNMENT - 1)); +#else + /* round the size up to the nearest power of two */ + t_local_q = (struct vplq *)kalloc(ROUNDUP_NEXTP2(num_cpus * sizeof(struct vplq))); +#endif for (i = 0; i < num_cpus; i++) { - struct vpl *lq; + 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); + vm_page_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; @@ -457,11 +873,19 @@ vm_page_init_local_q() } } +/* + * vm_init_before_launchd + * + * This should be called right before launchd is loaded. + */ +void +vm_init_before_launchd() +{ + vm_page_lockspin_queues(); + vm_page_wire_count_on_boot = vm_page_wire_count; + vm_page_unlock_queues(); +} -uint64_t initial_max_mem; -int initial_wire_count; -int initial_free_count; -int initial_lopage_count; /* * vm_page_bootstrap: @@ -476,91 +900,31 @@ int initial_lopage_count; void vm_page_bootstrap( - vm_offset_t *startp, - vm_offset_t *endp) + vm_offset_t *startp, + vm_offset_t *endp) { - register vm_page_t m; - unsigned int i; - unsigned int log1; - unsigned int log2; - unsigned int size; - - /* - * Initialize the vm_page template. - */ - - m = &vm_page_template; - bzero(m, sizeof (*m)); - - m->pageq.next = NULL; - m->pageq.prev = NULL; - m->listq.next = NULL; - m->listq.prev = NULL; - m->next = 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->reference = 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->fictitious = 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->overwriting = FALSE; - m->restart = FALSE; - m->unusual = FALSE; - m->encrypted = FALSE; - m->encrypted_cleaning = FALSE; - m->list_req_pending = FALSE; - m->dump_cleaning = FALSE; - m->cs_validated = FALSE; - m->cs_tainted = FALSE; - m->no_cache = FALSE; - m->zero_fill = FALSE; - m->reusable = FALSE; - m->__unused_object_bits = 0; - + unsigned int i; + unsigned int log1; + unsigned int log2; + unsigned int size; /* * Initialize the page queues. */ 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]); + 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; @@ -568,35 +932,122 @@ vm_page_bootstrap( purgeable_queues[i].debug_count_tokens = 0; purgeable_queues[i].debug_count_objects = 0; #endif - }; - - for (i = 0; i < MAX_COLORS; i++ ) - queue_init(&vm_page_queue_free[i]); - queue_init(&vm_lopage_queue_free); - vm_page_queue_fictitious = VM_PAGE_NULL; - queue_init(&vm_page_queue_active); - queue_init(&vm_page_queue_inactive); - queue_init(&vm_page_queue_throttled); - queue_init(&vm_page_queue_zf); - - for ( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ ) { - queue_init(&vm_page_queue_speculative[i].age_q); + } + ; + purgeable_nonvolatile_count = 0; + queue_init(&purgeable_nonvolatile_queue); + + for (i = 0; i < MAX_COLORS; i++) { + vm_page_queue_init(&vm_page_queue_free[i].qhead); + } + + vm_page_queue_init(&vm_lopage_queue_free); + vm_page_queue_init(&vm_page_queue_active); + vm_page_queue_init(&vm_page_queue_inactive); +#if CONFIG_SECLUDED_MEMORY + vm_page_queue_init(&vm_page_queue_secluded); +#endif /* CONFIG_SECLUDED_MEMORY */ + vm_page_queue_init(&vm_page_queue_cleaned); + vm_page_queue_init(&vm_page_queue_throttled); + vm_page_queue_init(&vm_page_queue_anonymous); + queue_init(&vm_objects_wired); + + for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) { + vm_page_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; } +#if CONFIG_BACKGROUND_QUEUE + vm_page_queue_init(&vm_page_queue_background); + + vm_page_background_count = 0; + vm_page_background_internal_count = 0; + vm_page_background_external_count = 0; + vm_page_background_promoted_count = 0; + + vm_page_background_target = (unsigned int)(atop_64(max_mem) / 25); + + if (vm_page_background_target > VM_PAGE_BACKGROUND_TARGET_MAX) { + vm_page_background_target = VM_PAGE_BACKGROUND_TARGET_MAX; + } + + vm_page_background_mode = VM_PAGE_BG_LEVEL_1; + vm_page_background_exclude_external = 0; + + PE_parse_boot_argn("vm_page_bg_mode", &vm_page_background_mode, sizeof(vm_page_background_mode)); + PE_parse_boot_argn("vm_page_bg_exclude_external", &vm_page_background_exclude_external, sizeof(vm_page_background_exclude_external)); + PE_parse_boot_argn("vm_page_bg_target", &vm_page_background_target, sizeof(vm_page_background_target)); + + if (vm_page_background_mode > VM_PAGE_BG_LEVEL_1) { + vm_page_background_mode = VM_PAGE_BG_LEVEL_1; + } +#endif vm_page_free_wanted = 0; vm_page_free_wanted_privileged = 0; - +#if CONFIG_SECLUDED_MEMORY + vm_page_free_wanted_secluded = 0; +#endif /* CONFIG_SECLUDED_MEMORY */ + +#if defined (__x86_64__) + /* this must be called before vm_page_set_colors() */ + vm_page_setup_clump(); +#endif + vm_page_set_colors(); + bzero(vm_page_inactive_states, sizeof(vm_page_inactive_states)); + vm_page_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; + vm_page_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; + vm_page_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; + + bzero(vm_page_pageable_states, sizeof(vm_page_pageable_states)); + vm_page_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; + vm_page_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; + vm_page_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; + vm_page_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1; + vm_page_pageable_states[VM_PAGE_ON_SPECULATIVE_Q] = 1; + vm_page_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1; +#if CONFIG_SECLUDED_MEMORY + vm_page_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1; +#endif /* CONFIG_SECLUDED_MEMORY */ + + bzero(vm_page_non_speculative_pageable_states, sizeof(vm_page_non_speculative_pageable_states)); + vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; + vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; + vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; + vm_page_non_speculative_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1; + vm_page_non_speculative_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1; +#if CONFIG_SECLUDED_MEMORY + vm_page_non_speculative_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1; +#endif /* CONFIG_SECLUDED_MEMORY */ + + bzero(vm_page_active_or_inactive_states, sizeof(vm_page_active_or_inactive_states)); + vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; + vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; + vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; + vm_page_active_or_inactive_states[VM_PAGE_ON_ACTIVE_Q] = 1; +#if CONFIG_SECLUDED_MEMORY + vm_page_active_or_inactive_states[VM_PAGE_ON_SECLUDED_Q] = 1; +#endif /* CONFIG_SECLUDED_MEMORY */ + + for (i = 0; i < VM_KERN_MEMORY_FIRST_DYNAMIC; i++) { + vm_allocation_sites_static[i].refcount = 2; + vm_allocation_sites_static[i].tag = i; + vm_allocation_sites[i] = &vm_allocation_sites_static[i]; + } + vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].refcount = 2; + vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].tag = VM_KERN_MEMORY_ANY; + vm_allocation_sites[VM_KERN_MEMORY_ANY] = &vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC]; /* * Steal memory for the map and zone subsystems. */ - +#if CONFIG_GZALLOC + gzalloc_configure(); +#endif + kernel_debug_string_early("vm_map_steal_memory"); vm_map_steal_memory(); - zone_steal_memory(); /* * Allocate (and initialize) the virtual-to-physical @@ -612,8 +1063,9 @@ vm_page_bootstrap( unsigned int npages = pmap_free_pages(); vm_page_bucket_count = 1; - while (vm_page_bucket_count < npages) + 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; @@ -627,40 +1079,78 @@ vm_page_bootstrap( * B/2 - O */ size = vm_page_bucket_count; - for (log1 = 0; size > 1; log1++) + for (log1 = 0; size > 1; log1++) { size /= 2; + } size = sizeof(struct vm_object); - for (log2 = 0; size > 1; log2++) + for (log2 = 0; size > 1; log2++) { size /= 2; - vm_page_hash_shift = log1/2 - log2 + 1; - - vm_page_bucket_hash = 1 << ((log1 + 1) >> 1); /* Get (ceiling of sqrt of table size) */ - vm_page_bucket_hash |= 1 << ((log1 + 1) >> 2); /* Get (ceiling of quadroot of table size) */ - vm_page_bucket_hash |= 1; /* Set bit and add 1 - always must be 1 to insure unique series */ + } + vm_page_hash_shift = log1 / 2 - log2 + 1; + + vm_page_bucket_hash = 1 << ((log1 + 1) >> 1); /* Get (ceiling of sqrt of table size) */ + vm_page_bucket_hash |= 1 << ((log1 + 1) >> 2); /* Get (ceiling of quadroot of table size) */ + vm_page_bucket_hash |= 1; /* Set bit and add 1 - always must be 1 to insure unique series */ - if (vm_page_hash_mask & vm_page_bucket_count) + 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_early("vm_page_buckets"); vm_page_buckets = (vm_page_bucket_t *) - pmap_steal_memory(vm_page_bucket_count * - sizeof(vm_page_bucket_t)); + pmap_steal_memory(vm_page_bucket_count * + sizeof(vm_page_bucket_t)); + kernel_debug_string_early("vm_page_bucket_locks"); vm_page_bucket_locks = (lck_spin_t *) - pmap_steal_memory(vm_page_bucket_lock_count * - sizeof(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]; + 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); + 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. @@ -670,6 +1160,7 @@ vm_page_bootstrap( * to get the alignment right. */ + kernel_debug_string_early("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); @@ -679,28 +1170,27 @@ vm_page_bootstrap( /* * Compute the initial "wire" count. - * Up until now, the pages which have been set aside are not under + * Up until now, the pages which have been set aside are not under * the VM system's control, so although they aren't explicitly * wired, they nonetheless can't be moved. At this moment, * all VM managed pages are "free", courtesy of pmap_startup. */ 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 = ((unsigned int) atop_64(max_mem)) - + vm_page_free_count - vm_lopage_free_count; +#if CONFIG_SECLUDED_MEMORY + vm_page_wire_count -= vm_page_secluded_count; +#endif vm_page_wire_count_initial = vm_page_wire_count; - vm_page_free_count_minimum = vm_page_free_count; - initial_max_mem = max_mem; - initial_wire_count = vm_page_wire_count; - initial_free_count = vm_page_free_count; - initial_lopage_count = vm_lopage_free_count; - - printf("vm_page_bootstrap: %d free pages and %d wired pages\n", - vm_page_free_count, vm_page_wire_count); + printf("vm_page_bootstrap: %d free pages, %d wired pages, (up to %d of which are delayed free)\n", + vm_page_free_count, vm_page_wire_count, vm_delayed_count); + kernel_debug_string_early("vm_page_bootstrap complete"); simple_lock_init(&vm_paging_lock, 0); } -#ifndef MACHINE_PAGES +#ifndef MACHINE_PAGES /* * We implement pmap_steal_memory and pmap_startup with the help * of two simpler functions, pmap_virtual_space and pmap_next_page. @@ -710,14 +1200,15 @@ void * pmap_steal_memory( vm_size_t size) { + kern_return_t kr; vm_offset_t addr, vaddr; - ppnum_t phys_page; + ppnum_t phys_page; /* * We round the size to a round multiple. */ - size = (size + sizeof (void *) - 1) &~ (sizeof (void *) - 1); + size = (size + sizeof(void *) - 1) & ~(sizeof(void *) - 1); /* * If this is the first call to pmap_steal_memory, @@ -743,185 +1234,358 @@ pmap_steal_memory( addr = virtual_space_start; virtual_space_start += size; - kprintf("pmap_steal_memory: %08lX - %08lX; size=%08lX\n", (long)addr, (long)virtual_space_start, (long)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_hi(&phys_page)) - panic("pmap_steal_memory"); + vaddr < addr + size; + vaddr += PAGE_SIZE) { + if (!pmap_next_page_hi(&phys_page)) { + panic("pmap_steal_memory() size: 0x%llx\n", (uint64_t)size); + } /* * XXX Logically, these mappings should be wired, * but some pmap modules barf if they are. */ #if defined(__LP64__) +#ifdef __arm64__ + /* ARM64_TODO: verify that we really don't need this */ +#else pmap_pre_expand(kernel_pmap, vaddr); #endif +#endif + + kr = pmap_enter(kernel_pmap, vaddr, phys_page, + VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, + VM_WIMG_USE_DEFAULT, FALSE); + + if (kr != KERN_SUCCESS) { + panic("pmap_steal_memory() pmap_enter failed, vaddr=%#lx, phys_page=%u", + (unsigned long)vaddr, phys_page); + } - pmap_enter(kernel_pmap, vaddr, phys_page, - VM_PROT_READ|VM_PROT_WRITE, - VM_WIMG_USE_DEFAULT, FALSE); /* * Account for newly stolen memory */ vm_page_wire_count++; - + vm_page_stolen_count++; } +#if KASAN + kasan_notify_address(round_page(addr), size); +#endif return (void *) addr; } +#if CONFIG_SECLUDED_MEMORY +/* boot-args to control secluded memory */ +unsigned int secluded_mem_mb = 0; /* # of MBs of RAM to seclude */ +int secluded_for_iokit = 1; /* IOKit can use secluded memory */ +int secluded_for_apps = 1; /* apps can use secluded memory */ +int secluded_for_filecache = 2; /* filecache can use seclude memory */ +#if 11 +int secluded_for_fbdp = 0; +#endif +uint64_t secluded_shutoff_trigger = 0; +#endif /* CONFIG_SECLUDED_MEMORY */ + + +#if defined(__arm__) || defined(__arm64__) +extern void patch_low_glo_vm_page_info(void *, void *, uint32_t); +unsigned int vm_first_phys_ppnum = 0; +#endif + +void vm_page_release_startup(vm_page_t mem); void pmap_startup( - vm_offset_t *startp, - vm_offset_t *endp) + vm_offset_t *startp, + vm_offset_t *endp) { - unsigned int i, npages, pages_initialized, fill, fillval; - ppnum_t phys_page; - addr64_t tmpaddr; - + unsigned int i, npages; + ppnum_t phys_page; + uint64_t mem_sz; + uint64_t start_ns; + uint64_t now_ns; + uint_t low_page_count = 0; + +#if defined(__LP64__) /* - * We calculate how many page frames we will have - * and then allocate the page structures in one chunk. + * 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 - tmpaddr = (addr64_t)pmap_free_pages() * (addr64_t)PAGE_SIZE; /* Get the amount of memory left */ - 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 */ + /* + * We calculate how many page frames we will have + * and then allocate the page structures in one chunk. + * + * Note that the calculation here doesn't take into account + * the memory needed to map what's being allocated, i.e. the page + * table entries. So the actual number of pages we get will be + * less than this. To do someday: include that in the computation. + */ + mem_sz = pmap_free_pages() * (uint64_t)PAGE_SIZE; + mem_sz += round_page(virtual_space_start) - virtual_space_start; /* Account for any slop */ + npages = (uint_t)(mem_sz / (PAGE_SIZE + sizeof(*vm_pages))); /* scaled to include the vm_page_ts */ vm_pages = (vm_page_t) pmap_steal_memory(npages * sizeof *vm_pages); /* - * Initialize the page frames. + * Check if we want to initialize pages to a known value */ - 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; + if (PE_parse_boot_argn("fill", &fillval, sizeof(fillval))) { + fill = TRUE; + } +#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) { + fill = TRUE; + fillval = 0xDEB8F177; + } +#endif + if (fill) { + kprintf("Filling vm_pages with pattern: 0x%x\n", fillval); + } - vm_page_init(&vm_pages[i], phys_page, FALSE); - vm_page_pages++; - pages_initialized++; +#if CONFIG_SECLUDED_MEMORY + /* + * Figure out how much secluded memory to have before we start + * release pages to free lists. + * The default, if specified nowhere else, is no secluded mem. + */ + secluded_mem_mb = 0; + if (max_mem > 1 * 1024 * 1024 * 1024) { + /* default to 90MB for devices with > 1GB of RAM */ + secluded_mem_mb = 90; } - vm_pages_count = pages_initialized; + /* override with value from device tree, if provided */ + PE_get_default("kern.secluded_mem_mb", + &secluded_mem_mb, sizeof(secluded_mem_mb)); + /* override with value from boot-args, if provided */ + PE_parse_boot_argn("secluded_mem_mb", + &secluded_mem_mb, + sizeof(secluded_mem_mb)); + + vm_page_secluded_target = (unsigned int) + ((secluded_mem_mb * 1024ULL * 1024ULL) / PAGE_SIZE); + PE_parse_boot_argn("secluded_for_iokit", + &secluded_for_iokit, + sizeof(secluded_for_iokit)); + PE_parse_boot_argn("secluded_for_apps", + &secluded_for_apps, + sizeof(secluded_for_apps)); + PE_parse_boot_argn("secluded_for_filecache", + &secluded_for_filecache, + sizeof(secluded_for_filecache)); +#if 11 + PE_parse_boot_argn("secluded_for_fbdp", + &secluded_for_fbdp, + sizeof(secluded_for_fbdp)); +#endif /* - * Check if we want to initialize pages to a known value + * On small devices, allow a large app to effectively suppress + * secluded memory until it exits. */ - fill = 0; /* Assume no fill */ - if (PE_parse_boot_argn("fill", &fillval, sizeof (fillval))) fill = 1; /* Set fill */ - - // -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(&vm_pages[i - 1]); + if (max_mem <= 1 * 1024 * 1024 * 1024 && vm_page_secluded_target != 0) { + /* + * Get an amount from boot-args, else use 500MB. + * 500MB was chosen from a Peace daemon tentpole test which used munch + * to induce jetsam thrashing of false idle daemons. + */ + int secluded_shutoff_mb; + if (PE_parse_boot_argn("secluded_shutoff_mb", &secluded_shutoff_mb, + sizeof(secluded_shutoff_mb))) { + secluded_shutoff_trigger = (uint64_t)secluded_shutoff_mb * 1024 * 1024; + } else { + secluded_shutoff_trigger = 500 * 1024 * 1024; + } + + if (secluded_shutoff_trigger != 0) { + secluded_suppression_init(); } } - else - // debug code remove- + +#endif /* CONFIG_SECLUDED_MEMORY */ + +#if defined(__x86_64__) /* - * Release pages in reverse order so that physical pages - * initially get allocated in ascending addresses. This keeps - * the devices (which must address physical memory) happy if - * they require several consecutive pages. + * Decide how much memory we delay freeing at boot time. */ - for (i = pages_initialized; i > 0; 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(&vm_pages[i - 1]); + uint32_t delay_above_gb; + if (!PE_parse_boot_argn("delay_above_gb", &delay_above_gb, sizeof(delay_above_gb))) { + delay_above_gb = DEFAULT_DELAY_ABOVE_PHYS_GB; } -#if 0 - { - vm_page_t xx, xxo, xxl; - int i, j, k, l; - - j = 0; /* (BRINGUP) */ - xxl = 0; - - 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); - } - } + if (delay_above_gb == 0) { + delay_above_pnum = PPNUM_MAX; + } else { + delay_above_pnum = delay_above_gb * (1024 * 1024 * 1024 / PAGE_SIZE); + } - xxl = xx; - } + /* make sure we have sane breathing room: 1G above low memory */ + if (delay_above_pnum <= max_valid_low_ppnum) { + delay_above_pnum = max_valid_low_ppnum + ((1024 * 1024 * 1024) >> PAGE_SHIFT); + } + + if (delay_above_pnum < PPNUM_MAX) { + printf("pmap_startup() delaying init/free of page nums > 0x%x\n", delay_above_pnum); + } + +#endif /* defined(__x86_64__) */ + + /* + * Initialize and release the page frames. + */ + kernel_debug_string_early("Initialize and free the page frames"); + + vm_page_array_beginning_addr = &vm_pages[0]; + vm_page_array_ending_addr = &vm_pages[npages]; /* used by ptr packing/unpacking code */ + + vm_delayed_count = 0; + + absolutetime_to_nanoseconds(mach_absolute_time(), &start_ns); + vm_pages_count = 0; + for (i = 0; i < npages; i++) { + /* Did we run out of pages? */ + if (!pmap_next_page(&phys_page)) { + break; + } + + if (phys_page < max_valid_low_ppnum) { + ++low_page_count; + } + + /* Are we at high enough pages to delay the rest? */ + if (low_page_count > vm_lopage_free_limit && phys_page > delay_above_pnum) { + vm_delayed_count = pmap_free_pages(); + break; + } + +#if defined(__arm__) || defined(__arm64__) + if (i == 0) { + vm_first_phys_ppnum = phys_page; + patch_low_glo_vm_page_info((void *)vm_page_array_beginning_addr, + (void *)vm_page_array_ending_addr, vm_first_phys_ppnum); + } + assert((i + vm_first_phys_ppnum) == phys_page); +#endif + + ++vm_pages_count; + vm_page_init(&vm_pages[i], phys_page, FALSE); + if (fill) { + fillPage(phys_page, fillval); } - - if(j != vm_page_free_count) { /* (BRINGUP) */ - panic("pmap_startup: vm_page_free_count does not match, calc = %d, vm_page_free_count = %08X\n", j, vm_page_free_count); + if (vm_himemory_mode) { + vm_page_release_startup(&vm_pages[i]); } } + vm_page_pages = vm_pages_count; /* used to report to user space */ + + if (!vm_himemory_mode) { + do { + vm_page_release_startup(&vm_pages[--i]); + } while (i != 0); + } + + absolutetime_to_nanoseconds(mach_absolute_time(), &now_ns); + printf("pmap_startup() init/release time: %lld microsec\n", (now_ns - start_ns) / NSEC_PER_USEC); + printf("pmap_startup() delayed init/release of %d pages\n", vm_delayed_count); + +#if defined(__LP64__) + + if ((vm_page_t)(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_t)(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 + VM_CHECK_MEMORYSTATUS; /* - * We have to re-align virtual_space_start, - * because pmap_steal_memory has been using it. + * We have to re-align virtual_space_start, + * because pmap_steal_memory has been using it. */ - virtual_space_start = round_page(virtual_space_start); - *startp = virtual_space_start; *endp = virtual_space_end; } -#endif /* MACHINE_PAGES */ +#endif /* MACHINE_PAGES */ /* - * Routine: vm_page_module_init - * Purpose: - * Second initialization pass, to be done after - * the basic VM system is ready. + * Create the zone that represents the vm_pages[] array. Nothing ever allocates + * or frees to this zone. It's just here for reporting purposes via zprint command. + * This needs to be done after all initially delayed pages are put on the free lists. + */ +static void +vm_page_module_init_delayed(void) +{ + uint64_t vm_page_zone_pages, vm_page_array_zone_data_size; + + vm_page_array_zone = zinit((vm_size_t) sizeof(struct vm_page), + 0, PAGE_SIZE, "vm pages array"); + + zone_change(vm_page_array_zone, Z_CALLERACCT, FALSE); + zone_change(vm_page_array_zone, Z_EXPAND, FALSE); + zone_change(vm_page_array_zone, Z_EXHAUST, TRUE); + zone_change(vm_page_array_zone, Z_FOREIGN, TRUE); + zone_change(vm_page_array_zone, Z_GZALLOC_EXEMPT, TRUE); + + /* + * Reflect size and usage information for vm_pages[]. + */ + vm_page_array_zone->count = vm_pages_count; + vm_page_array_zone->countfree = (int)(vm_page_array_ending_addr - &vm_pages[vm_pages_count]); + vm_page_array_zone->sum_count = vm_pages_count; + vm_page_array_zone_data_size = (uintptr_t)((void *)vm_page_array_ending_addr - (void *)vm_pages); + vm_page_array_zone->cur_size = vm_page_array_zone_data_size; + vm_page_zone_pages = ((round_page(vm_page_array_zone_data_size)) / PAGE_SIZE); + OSAddAtomic64(vm_page_zone_pages, &(vm_page_array_zone->page_count)); + /* since zone accounts for these, take them out of stolen */ + VM_PAGE_MOVE_STOLEN(vm_page_zone_pages); +} + +/* + * Create the vm_pages zone. This is used for the vm_page structures for the pages + * that are scavanged from other boot time usages by ml_static_mfree(). As such, + * this needs to happen in early VM bootstrap. */ void vm_page_module_init(void) { - vm_page_zone = zinit((vm_size_t) sizeof(struct vm_page), - 0, PAGE_SIZE, "vm pages"); + vm_size_t vm_page_with_ppnum_size; + + /* + * Since the pointers to elements in this zone will be packed, they + * must have appropriate size. Not strictly what sizeof() reports. + */ + vm_page_with_ppnum_size = + (sizeof(struct vm_page_with_ppnum) + (VM_PACKED_POINTER_ALIGNMENT - 1)) & + ~(VM_PACKED_POINTER_ALIGNMENT - 1); -#if ZONE_DEBUG - zone_debug_disable(vm_page_zone); -#endif /* ZONE_DEBUG */ + vm_page_zone = zinit(vm_page_with_ppnum_size, 0, PAGE_SIZE, "vm pages"); + 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; - - lck_mtx_init(&vm_page_alloc_lock, &vm_page_lck_grp_alloc, &vm_page_lck_attr); + zone_change(vm_page_zone, Z_GZALLOC_EXEMPT, TRUE); + zone_change(vm_page_zone, Z_ALIGNMENT_REQUIRED, TRUE); } /* @@ -938,20 +1602,24 @@ vm_page_create( ppnum_t start, ppnum_t end) { - ppnum_t phys_page; - vm_page_t m; + ppnum_t phys_page; + vm_page_t m; for (phys_page = start; - phys_page < end; - phys_page++) { - while ((m = (vm_page_t) vm_page_grab_fictitious()) - == VM_PAGE_NULL) + phys_page < end; + phys_page++) { + 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, FALSE); + m->vmp_fictitious = FALSE; pmap_clear_noencrypt(phys_page); + + lck_mtx_lock(&vm_page_queue_free_lock); vm_page_pages++; - vm_page_release(m); + lck_mtx_unlock(&vm_page_queue_free_lock); + vm_page_release(m, FALSE); } } @@ -977,64 +1645,99 @@ vm_page_create( */ void vm_page_insert( - vm_page_t mem, - vm_object_t object, - 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, FALSE, TRUE); + 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, - boolean_t queues_lock_held, - boolean_t insert_in_hash) + 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) { - vm_page_bucket_t *bucket; - lck_spin_t *bucket_lock; - int hash_id; + vm_page_bucket_t *bucket; + lck_spin_t *bucket_lock; + int hash_id; + task_t owner; + int ledger_idx_volatile; + int ledger_idx_nonvolatile; + int ledger_idx_volatile_compressed; + int ledger_idx_nonvolatile_compressed; + boolean_t do_footprint; + + XPR(XPR_VM_PAGE, + "vm_page_insert, object 0x%X offset 0x%X page 0x%X\n", + 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); +#endif - XPR(XPR_VM_PAGE, - "vm_page_insert, object 0x%X offset 0x%X page 0x%X\n", - object, offset, mem, 0,0); + assert(page_aligned(offset)); - VM_PAGE_CHECK(mem); + assert(!VM_PAGE_WIRED(mem) || mem->vmp_private || mem->vmp_fictitious || (tag != VM_KERN_MEMORY_NONE)); - if (object == vm_submap_object) { - /* the vm_submap_object is only a placeholder for submaps */ - panic("vm_page_insert(vm_submap_object,0x%llx)\n", offset); - } + /* 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 */ - + LCK_MTX_ASSERT(&vm_page_queue_lock, + queues_lock_held ? LCK_MTX_ASSERT_OWNED + : LCK_MTX_ASSERT_NOTOWNED); + + if (queues_lock_held == FALSE) { + assert(!VM_PAGE_PAGEABLE(mem)); + } + if (insert_in_hash == TRUE) { -#if DEBUG - if (mem->tabled || mem->object != VM_OBJECT_NULL) +#if DEBUG || VM_PAGE_CHECK_BUCKETS + if (mem->vmp_tabled || mem->vmp_object) { 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); + "already in (obj=%p,off=0x%llx)", + mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset); + } #endif - assert(!object->internal || offset < object->size); - - /* only insert "pageout" pages into "pageout" objects, - * and normal pages into normal objects */ - assert(object->pageout == mem->pageout); + if (object->internal && (offset >= object->vo_size)) { + panic("vm_page_insert_internal: (page=%p,obj=%p,off=0x%llx,size=0x%llx) inserted at offset past object bounds", + mem, object, offset, object->vo_size); + } assert(vm_page_lookup(object, offset) == VM_PAGE_NULL); - + /* * Record the object/offset pair in this page */ - mem->object = object; - mem->offset = offset; + mem->vmp_object = VM_PAGE_PACK_OBJECT(object); + mem->vmp_offset = offset; + +#if CONFIG_SECLUDED_MEMORY + if (object->eligible_for_secluded) { + vm_page_secluded.eligible_for_secluded++; + } +#endif /* CONFIG_SECLUDED_MEMORY */ /* * Insert it into the object_object/offset hash table @@ -1042,24 +1745,37 @@ vm_page_insert_internal( 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); - mem->next = bucket->pages; - bucket->pages = mem; + lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket); + + mem->vmp_next_m = bucket->page_list; + bucket->page_list = VM_PAGE_PACK_PTR(mem); + assert(mem == (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list))); + #if MACH_PAGE_HASH_STATS - if (++bucket->cur_count > bucket->hi_count) + if (++bucket->cur_count > bucket->hi_count) { bucket->hi_count = bucket->cur_count; + } #endif /* MACH_PAGE_HASH_STATS */ - + mem->vmp_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); - mem->tabled = TRUE; + vm_page_queue_enter(&object->memq, mem, vmp_listq); + object->memq_hint = mem; + mem->vmp_tabled = TRUE; /* * Show that the object has one more resident page. @@ -1067,20 +1783,86 @@ vm_page_insert_internal( object->resident_page_count++; if (VM_PAGE_WIRED(mem)) { - object->wired_page_count++; + assert(mem->vmp_wire_count > 0); + VM_OBJECT_WIRED_PAGE_UPDATE_START(object); + VM_OBJECT_WIRED_PAGE_ADD(object, mem); + VM_OBJECT_WIRED_PAGE_UPDATE_END(object, tag); } assert(object->resident_page_count >= object->wired_page_count); - assert(!mem->reusable); + 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->vmp_reusable); + if (object->all_reusable) { + OSAddAtomic(+1, &vm_page_stats_reusable.reusable_count); + } + + if (object->purgable == VM_PURGABLE_DENY && + !object->vo_ledger_tag) { + owner = TASK_NULL; + } else { + owner = VM_OBJECT_OWNER(object); + vm_object_ledger_tag_ledgers(object, + &ledger_idx_volatile, + &ledger_idx_nonvolatile, + &ledger_idx_volatile_compressed, + &ledger_idx_nonvolatile_compressed, + &do_footprint); + } + if (owner && + (object->purgable == VM_PURGABLE_NONVOLATILE || + object->purgable == VM_PURGABLE_DENY || + VM_PAGE_WIRED(mem))) { + if (delayed_ledger_update) { + *delayed_ledger_update += PAGE_SIZE; + } else { + /* more non-volatile bytes */ + ledger_credit(owner->ledger, + ledger_idx_nonvolatile, + PAGE_SIZE); + if (do_footprint) { + /* 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, + ledger_idx_volatile, + PAGE_SIZE); + } if (object->purgable == VM_PURGABLE_VOLATILE) { if (VM_PAGE_WIRED(mem)) { - OSAddAtomic(1, &vm_page_purgeable_wired_count); + OSAddAtomic(+1, &vm_page_purgeable_wired_count); } else { - OSAddAtomic(1, &vm_page_purgeable_count); + OSAddAtomic(+1, &vm_page_purgeable_count); } } else if (object->purgable == VM_PURGABLE_EMPTY && - mem->throttled) { + mem->vmp_q_state == VM_PAGE_ON_THROTTLED_Q) { /* * This page belongs to a purged VM object but hasn't * been purged (because it was "busy"). @@ -1089,12 +1871,33 @@ vm_page_insert_internal( * queue, so that it can eventually be reclaimed, instead * of lingering in the "empty" object. */ - if (queues_lock_held == FALSE) + if (queues_lock_held == FALSE) { vm_page_lockspin_queues(); + } vm_page_deactivate(mem); - if (queues_lock_held == FALSE) + 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 */ } /* @@ -1107,30 +1910,39 @@ vm_page_insert_internal( */ void vm_page_replace( - 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_bucket_t *bucket; - vm_page_t found_m = VM_PAGE_NULL; - lck_spin_t *bucket_lock; - int hash_id; + 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); +#endif vm_object_lock_assert_exclusive(object); -#if DEBUG - if (mem->tabled || mem->object != VM_OBJECT_NULL) +#if DEBUG || VM_PAGE_CHECK_BUCKETS + if (mem->vmp_tabled || mem->vmp_object) { 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); + "already in (obj=%p,off=0x%llx)", + mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset); + } #endif + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); + + assert(!VM_PAGE_PAGEABLE(mem)); + /* * Record the object/offset pair in this page */ - - mem->object = object; - mem->offset = offset; + mem->vmp_object = VM_PAGE_PACK_OBJECT(object); + mem->vmp_offset = offset; /* * Insert it into the object_object/offset hash table, @@ -1141,45 +1953,51 @@ vm_page_replace( bucket = &vm_page_buckets[hash_id]; bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; - lck_spin_lock(bucket_lock); + lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket); - if (bucket->pages) { - vm_page_t *mp = &bucket->pages; - vm_page_t m = *mp; + if (bucket->page_list) { + vm_page_packed_t *mp = &bucket->page_list; + vm_page_t m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp)); do { - if (m->object == object && m->offset == offset) { + /* + * compare packed object pointers + */ + if (m->vmp_object == mem->vmp_object && m->vmp_offset == offset) { /* * Remove old page from hash list */ - *mp = m->next; + *mp = m->vmp_next_m; + m->vmp_hashed = FALSE; + m->vmp_next_m = VM_PAGE_PACK_PTR(NULL); found_m = m; break; } - mp = &m->next; - } while ((m = *mp)); + mp = &m->vmp_next_m; + } while ((m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp)))); - mem->next = bucket->pages; + mem->vmp_next_m = bucket->page_list; } else { - mem->next = VM_PAGE_NULL; + mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL); } /* * insert new page at head of hash list */ - bucket->pages = mem; + bucket->page_list = VM_PAGE_PACK_PTR(mem); + mem->vmp_hashed = TRUE; lck_spin_unlock(bucket_lock); 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, FALSE, FALSE); + vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, FALSE, FALSE, FALSE, NULL); } /* @@ -1193,86 +2011,168 @@ vm_page_replace( void vm_page_remove( - vm_page_t mem, - boolean_t remove_from_hash) + vm_page_t mem, + boolean_t remove_from_hash) { vm_page_bucket_t *bucket; - vm_page_t this; - lck_spin_t *bucket_lock; - int hash_id; - - XPR(XPR_VM_PAGE, - "vm_page_remove, object 0x%X offset 0x%X page 0x%X\n", - mem->object, mem->offset, - mem, 0,0); - - vm_object_lock_assert_exclusive(mem->object); - assert(mem->tabled); - assert(!mem->cleaning); + vm_page_t this; + lck_spin_t *bucket_lock; + int hash_id; + task_t owner; + vm_object_t m_object; + int ledger_idx_volatile; + int ledger_idx_nonvolatile; + int ledger_idx_volatile_compressed; + int ledger_idx_nonvolatile_compressed; + int do_footprint; + + m_object = VM_PAGE_OBJECT(mem); + + XPR(XPR_VM_PAGE, + "vm_page_remove, object 0x%X offset 0x%X page 0x%X\n", + m_object, mem->vmp_offset, + mem, 0, 0); + + vm_object_lock_assert_exclusive(m_object); + assert(mem->vmp_tabled); + assert(!mem->vmp_cleaning); + assert(!mem->vmp_laundry); + + if (VM_PAGE_PAGEABLE(mem)) { + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + } +#if 0 + /* + * 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); + hash_id = vm_page_hash(m_object, mem->vmp_offset); bucket = &vm_page_buckets[hash_id]; bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; - lck_spin_lock(bucket_lock); + lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket); - if ((this = bucket->pages) == mem) { + if ((this = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list))) == mem) { /* optimize for common case */ - bucket->pages = mem->next; + bucket->page_list = mem->vmp_next_m; } else { - vm_page_t *prev; + vm_page_packed_t *prev; - for (prev = &this->next; - (this = *prev) != mem; - prev = &this->next) + for (prev = &this->vmp_next_m; + (this = (vm_page_t)(VM_PAGE_UNPACK_PTR(*prev))) != mem; + prev = &this->vmp_next_m) { continue; - *prev = this->next; + } + *prev = this->vmp_next_m; } #if MACH_PAGE_HASH_STATS bucket->cur_count--; #endif /* MACH_PAGE_HASH_STATS */ - + mem->vmp_hashed = FALSE; + this->vmp_next_m = VM_PAGE_PACK_PTR(NULL); 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--; + assert(m_object->resident_page_count > 0); + m_object->resident_page_count--; + + if (m_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->vmp_xpmapped) { + assert(vm_page_xpmapped_external_count); + OSAddAtomic(-1, &vm_page_xpmapped_external_count); + } + } + if (!m_object->internal && + m_object->cached_list.next && + m_object->cached_list.prev) { + if (m_object->resident_page_count == 0) { + vm_object_cache_remove(m_object); + } + } + if (VM_PAGE_WIRED(mem)) { - assert(mem->object->wired_page_count > 0); - mem->object->wired_page_count--; - } - 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; + assert(mem->vmp_wire_count > 0); + VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); + VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem); + VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag); + } + assert(m_object->resident_page_count >= + m_object->wired_page_count); + if (mem->vmp_reusable) { + assert(m_object->reusable_page_count > 0); + m_object->reusable_page_count--; + assert(m_object->reusable_page_count <= + m_object->resident_page_count); + mem->vmp_reusable = FALSE; OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count); vm_page_stats_reusable.reused_remove++; - } else if (mem->object->all_reusable) { + } else if (m_object->all_reusable) { OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count); vm_page_stats_reusable.reused_remove++; } - if (mem->object->purgable == VM_PURGABLE_VOLATILE) { + if (m_object->purgable == VM_PURGABLE_DENY && + !m_object->vo_ledger_tag) { + owner = TASK_NULL; + } else { + owner = VM_OBJECT_OWNER(m_object); + vm_object_ledger_tag_ledgers(m_object, + &ledger_idx_volatile, + &ledger_idx_nonvolatile, + &ledger_idx_volatile_compressed, + &ledger_idx_nonvolatile_compressed, + &do_footprint); + } + if (owner && + (m_object->purgable == VM_PURGABLE_NONVOLATILE || + m_object->purgable == VM_PURGABLE_DENY || + VM_PAGE_WIRED(mem))) { + /* less non-volatile bytes */ + ledger_debit(owner->ledger, + ledger_idx_nonvolatile, + PAGE_SIZE); + if (do_footprint) { + /* less footprint */ + ledger_debit(owner->ledger, + task_ledgers.phys_footprint, + PAGE_SIZE); + } + } else if (owner && + (m_object->purgable == VM_PURGABLE_VOLATILE || + m_object->purgable == VM_PURGABLE_EMPTY)) { + assert(!VM_PAGE_WIRED(mem)); + /* less volatile bytes */ + ledger_debit(owner->ledger, + ledger_idx_volatile, + PAGE_SIZE); + } + if (m_object->purgable == VM_PURGABLE_VOLATILE) { if (VM_PAGE_WIRED(mem)) { assert(vm_page_purgeable_wired_count > 0); OSAddAtomic(-1, &vm_page_purgeable_wired_count); @@ -1281,9 +2181,14 @@ vm_page_remove( OSAddAtomic(-1, &vm_page_purgeable_count); } } - mem->tabled = FALSE; - mem->object = VM_OBJECT_NULL; - mem->offset = (vm_object_offset_t) -1; + + if (m_object->set_cache_attr == TRUE) { + pmap_set_cache_attributes(VM_PAGE_GET_PHYS_PAGE(mem), 0); + } + + mem->vmp_tabled = FALSE; + mem->vmp_object = 0; + mem->vmp_offset = (vm_object_offset_t) -1; } @@ -1296,60 +2201,120 @@ vm_page_remove( * 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; -unsigned long vm_page_lookup_bucket_NULL = 0; -unsigned long vm_page_lookup_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"); + } + + vm_page_queue_iterate(&object->memq, cur_page, vmp_listq) { + if (cur_page->vmp_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( - vm_object_t object, - vm_object_offset_t offset) + vm_object_t object, + vm_object_offset_t offset) { - vm_page_t mem; + vm_page_t mem; vm_page_bucket_t *bucket; - queue_entry_t qe; - lck_spin_t *bucket_lock; - int hash_id; + vm_page_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); + assert(VM_PAGE_OBJECT(mem) == object); - if (mem->offset == offset) { - vm_page_lookup_hint++; + if (mem->vmp_offset == offset) { +#if DEBUG_VM_PAGE_LOOKUP + OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint); +#endif return mem; } - qe = queue_next(&mem->listq); + qe = (vm_page_queue_entry_t)vm_page_queue_next(&mem->vmp_listq); - if (! queue_end(&object->memq, qe)) { - vm_page_t next_page; + if (!vm_page_queue_end(&object->memq, qe)) { + vm_page_t next_page; - next_page = (vm_page_t) qe; - assert(next_page->object == object); + next_page = (vm_page_t)((uintptr_t)qe); + assert(VM_PAGE_OBJECT(next_page) == object); - if (next_page->offset == offset) { - vm_page_lookup_hint_next++; + if (next_page->vmp_offset == offset) { object->memq_hint = next_page; /* new hint */ +#if DEBUG_VM_PAGE_LOOKUP + OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_next); +#endif return next_page; } } - qe = queue_prev(&mem->listq); + qe = (vm_page_queue_entry_t)vm_page_queue_prev(&mem->vmp_listq); - if (! queue_end(&object->memq, qe)) { + if (!vm_page_queue_end(&object->memq, qe)) { vm_page_t prev_page; - prev_page = (vm_page_t) qe; - assert(prev_page->object == object); + prev_page = (vm_page_t)((uintptr_t)qe); + assert(VM_PAGE_OBJECT(prev_page) == object); - if (prev_page->offset == offset) { - vm_page_lookup_hint_prev++; + if (prev_page->vmp_offset == offset) { object->memq_hint = prev_page; /* new hint */ +#if DEBUG_VM_PAGE_LOOKUP + OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_prev); +#endif return prev_page; } } @@ -1365,35 +2330,84 @@ vm_page_lookup( * 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 + * at outside the scope of the hash bucket lock... this is a * really cheap optimiztion to avoid taking the lock */ - if (bucket->pages == VM_PAGE_NULL) { - vm_page_lookup_bucket_NULL++; - - return (VM_PAGE_NULL); + if (!bucket->page_list) { +#if DEBUG_VM_PAGE_LOOKUP + OSAddAtomic64(1, &vm_page_lookup_stats.vpl_bucket_NULL); +#endif + return VM_PAGE_NULL; } - bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; - lck_spin_lock(bucket_lock); +#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)vm_page_queue_first(&object->memq); - for (mem = bucket->pages; mem != VM_PAGE_NULL; mem = mem->next) { - VM_PAGE_CHECK(mem); - if ((mem->object == object) && (mem->offset == offset)) - break; + while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) { + if (mem->vmp_offset == offset) { + break; + } + + mem = (vm_page_t)vm_page_queue_next(&mem->vmp_listq); + } + if (vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) { + mem = NULL; + } + } else { + vm_page_object_t packed_object; + + packed_object = VM_PAGE_PACK_OBJECT(object); + + bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; + + lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket); + + for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); + mem != VM_PAGE_NULL; + mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m))) { +#if 0 + /* + * we don't hold the page queue lock + * so this check isn't safe to make + */ + VM_PAGE_CHECK(mem); +#endif + if ((mem->vmp_object == packed_object) && (mem->vmp_offset == offset)) { + break; + } + } + lck_spin_unlock(bucket_lock); } - lck_spin_unlock(bucket_lock); +#if DEBUG_VM_PAGE_LOOKUP + elapsed = mach_absolute_time() - start; + + 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); + } 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; - } else - vm_page_lookup_miss++; + OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit); + } else { + OSAddAtomic64(1, &vm_page_lookup_stats.vpl_miss); + } +#endif + if (mem != VM_PAGE_NULL) { + assert(VM_PAGE_OBJECT(mem) == object); - return(mem); + object->memq_hint = mem; + } + return mem; } @@ -1407,49 +2421,69 @@ vm_page_lookup( */ void vm_page_rename( - register vm_page_t mem, - register vm_object_t new_object, - vm_object_offset_t new_offset, - boolean_t encrypted_ok) + vm_page_t mem, + vm_object_t new_object, + vm_object_offset_t new_offset) { - assert(mem->object != new_object); + boolean_t internal_to_external, external_to_internal; + vm_tag_t tag; + vm_object_t m_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, 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 info - * should remain consistent. The caller (vm_object_do_collapse()) - * sets "encrypted_ok" in this case. - */ - if (!encrypted_ok && mem->encrypted) { - panic("vm_page_rename: page %p is encrypted\n", mem); - } + m_object = VM_PAGE_OBJECT(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); + assert(m_object != new_object); + assert(m_object); + + 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 + * Changes to mem->vmp_object require the page lock because * the pageout daemon uses that lock to get the object. */ vm_page_lockspin_queues(); - vm_page_remove(mem, TRUE); - vm_page_insert_internal(mem, new_object, new_offset, TRUE, TRUE); - - vm_page_unlock_queues(); -} + internal_to_external = FALSE; + external_to_internal = FALSE; -/* - * vm_page_init: + if (mem->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q) { + /* + * 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, FALSE); + vm_page_activate(mem); + } + if (VM_PAGE_PAGEABLE(mem)) { + if (m_object->internal && !new_object->internal) { + internal_to_external = TRUE; + } + if (!m_object->internal && new_object->internal) { + external_to_internal = TRUE; + } + } + + tag = m_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_unlock_queues(); +} + +/* + * vm_page_init: * * Initialize the fields in a new page. * This takes a structure with random values and initializes it @@ -1457,15 +2491,59 @@ vm_page_rename( */ void vm_page_init( - vm_page_t mem, - ppnum_t phys_page, - boolean_t lopage) + vm_page_t mem, + ppnum_t phys_page, + boolean_t lopage) { + uint_t i; + uintptr_t *p; + assert(phys_page); - *mem = vm_page_template; - mem->phys_page = phys_page; - mem->lopage = lopage; +#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 /* DEBUG */ + + /* + * Initialize the fields of the vm_page. If adding any new fields to vm_page, + * try to use initial values which match 0. This minimizes the number of writes + * needed for boot-time initialization. + * + * Kernel bzero() isn't an inline yet, so do it by hand for performance. + */ + assert(VM_PAGE_NOT_ON_Q == 0); + assert(sizeof(*mem) % sizeof(uintptr_t) == 0); + for (p = (uintptr_t *)(void *)mem, i = sizeof(*mem) / sizeof(uintptr_t); i != 0; --i) { + *p++ = 0; + } + mem->vmp_offset = (vm_object_offset_t)-1; + mem->vmp_busy = TRUE; + mem->vmp_lopage = lopage; + + VM_PAGE_SET_PHYS_PAGE(mem, 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 } /* @@ -1474,25 +2552,26 @@ vm_page_init( * Remove a fictitious page from the free list. * Returns VM_PAGE_NULL if there are no free pages. */ -int c_vm_page_grab_fictitious = 0; -int c_vm_page_release_fictitious = 0; -int c_vm_page_more_fictitious = 0; - -extern vm_page_t vm_page_grab_fictitious_common(ppnum_t phys_addr); +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_common( ppnum_t phys_addr) { - register vm_page_t m; + vm_page_t m; - m = (vm_page_t)zget(vm_page_zone); - if (m) { + if ((m = (vm_page_t)zget(vm_page_zone))) { vm_page_init(m, phys_addr, FALSE); - m->fictitious = TRUE; + m->vmp_fictitious = TRUE; + + c_vm_page_grab_fictitious++; + } else { + c_vm_page_grab_fictitious_failed++; } - c_vm_page_grab_fictitious++; return m; } @@ -1502,41 +2581,49 @@ vm_page_grab_fictitious(void) return vm_page_grab_fictitious_common(vm_page_fictitious_addr); } +int vm_guard_count; + + vm_page_t vm_page_grab_guard(void) { - return vm_page_grab_fictitious_common(vm_page_guard_addr); + vm_page_t page; + page = vm_page_grab_fictitious_common(vm_page_guard_addr); + if (page) { + OSAddAtomic(1, &vm_guard_count); + } + return page; } + /* * 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 || - m->phys_page == vm_page_guard_addr); + assert((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || (m->vmp_q_state == VM_PAGE_IS_WIRED)); + assert(m->vmp_fictitious); + assert(VM_PAGE_GET_PHYS_PAGE(m) == vm_page_fictitious_addr || + VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr); + + + if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) { + OSAddAtomic(-1, &vm_guard_count); + } 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 @@ -1550,23 +2637,14 @@ vm_page_release_fictitious( * 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) +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++; @@ -1601,11 +2679,11 @@ void vm_page_more_fictitious(void) } retval = kernel_memory_allocate(zone_map, - &addr, PAGE_SIZE, VM_PROT_ALL, - KMA_KOBJECT|KMA_NOPAGEWAIT); - if (retval != KERN_SUCCESS) { + &addr, PAGE_SIZE, 0, + 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. */ @@ -1613,18 +2691,9 @@ void vm_page_more_fictitious(void) 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, FALSE); - m->fictitious = TRUE; - m++; - } - zcram(vm_page_zone, (void *) addr, PAGE_SIZE); + + zcram(vm_page_zone, addr, PAGE_SIZE); + lck_mtx_unlock(&vm_page_alloc_lock); } @@ -1640,137 +2709,366 @@ int 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; } - +boolean_t vm_darkwake_mode = FALSE; /* - * this is an interface to support bring-up of drivers - * on platforms with physical memory > 4G... + * vm_update_darkwake_mode(): + * + * Tells the VM that the system is in / out of darkwake. + * + * Today, the VM only lowers/raises the background queue target + * so as to favor consuming more/less background pages when + * darwake is ON/OFF. + * + * We might need to do more things in the future. */ -int vm_himemory_mode = 0; +void +vm_update_darkwake_mode(boolean_t darkwake_mode) +{ + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); + + vm_page_lockspin_queues(); + + if (vm_darkwake_mode == darkwake_mode) { + /* + * No change. + */ + vm_page_unlock_queues(); + return; + } + + vm_darkwake_mode = darkwake_mode; + + if (vm_darkwake_mode == TRUE) { +#if CONFIG_BACKGROUND_QUEUE + + /* save background target to restore later */ + vm_page_background_target_snapshot = vm_page_background_target; + + /* target is set to 0...no protection for background pages */ + vm_page_background_target = 0; + +#endif /* CONFIG_BACKGROUND_QUEUE */ + } else if (vm_darkwake_mode == FALSE) { +#if CONFIG_BACKGROUND_QUEUE + + if (vm_page_background_target_snapshot) { + vm_page_background_target = vm_page_background_target_snapshot; + } +#endif /* CONFIG_BACKGROUND_QUEUE */ + } + vm_page_unlock_queues(); +} + +#if CONFIG_BACKGROUND_QUEUE + +void +vm_page_update_background_state(vm_page_t mem) +{ + if (vm_page_background_mode == VM_PAGE_BG_DISABLED) { + return; + } + + if (mem->vmp_in_background == FALSE) { + return; + } + + task_t my_task = current_task(); + + if (my_task) { + if (task_get_darkwake_mode(my_task)) { + return; + } + } + +#if BACKGROUNDQ_BASED_ON_QOS + if (proc_get_effective_thread_policy(current_thread(), TASK_POLICY_QOS) <= THREAD_QOS_LEGACY) { + return; + } +#else + if (my_task) { + if (proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG)) { + return; + } + } +#endif + vm_page_lockspin_queues(); + + mem->vmp_in_background = FALSE; + vm_page_background_promoted_count++; + + vm_page_remove_from_backgroundq(mem); + + vm_page_unlock_queues(); +} + + +void +vm_page_assign_background_state(vm_page_t mem) +{ + if (vm_page_background_mode == VM_PAGE_BG_DISABLED) { + return; + } + + task_t my_task = current_task(); + + if (my_task) { + if (task_get_darkwake_mode(my_task)) { + mem->vmp_in_background = TRUE; + return; + } + } + +#if BACKGROUNDQ_BASED_ON_QOS + if (proc_get_effective_thread_policy(current_thread(), TASK_POLICY_QOS) <= THREAD_QOS_LEGACY) { + mem->vmp_in_background = TRUE; + } else { + mem->vmp_in_background = FALSE; + } +#else + if (my_task) { + mem->vmp_in_background = proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG); + } +#endif +} + + +void +vm_page_remove_from_backgroundq( + vm_page_t mem) +{ + vm_object_t m_object; + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + + if (mem->vmp_on_backgroundq) { + vm_page_queue_remove(&vm_page_queue_background, mem, vmp_backgroundq); + + mem->vmp_backgroundq.next = 0; + mem->vmp_backgroundq.prev = 0; + mem->vmp_on_backgroundq = FALSE; + + vm_page_background_count--; + + m_object = VM_PAGE_OBJECT(mem); + + if (m_object->internal) { + vm_page_background_internal_count--; + } else { + vm_page_background_external_count--; + } + } else { + assert(VM_PAGE_UNPACK_PTR(mem->vmp_backgroundq.next) == (uintptr_t)NULL && + VM_PAGE_UNPACK_PTR(mem->vmp_backgroundq.prev) == (uintptr_t)NULL); + } +} + + +void +vm_page_add_to_backgroundq( + vm_page_t mem, + boolean_t first) +{ + vm_object_t m_object; + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + + if (vm_page_background_mode == VM_PAGE_BG_DISABLED) { + return; + } + + if (mem->vmp_on_backgroundq == FALSE) { + m_object = VM_PAGE_OBJECT(mem); + + if (vm_page_background_exclude_external && !m_object->internal) { + return; + } + + if (first == TRUE) { + vm_page_queue_enter_first(&vm_page_queue_background, mem, vmp_backgroundq); + } else { + vm_page_queue_enter(&vm_page_queue_background, mem, vmp_backgroundq); + } + mem->vmp_on_backgroundq = TRUE; + + vm_page_background_count++; + + if (m_object->internal) { + vm_page_background_internal_count++; + } else { + vm_page_background_external_count++; + } + } +} + +#endif /* CONFIG_BACKGROUND_QUEUE */ + +/* + * This can be switched to FALSE to help debug drivers + * that are having problems with memory > 4G. + */ +boolean_t vm_himemory_mode = TRUE; /* * 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; +unsigned int vm_lopages_allocated_q = 0; +unsigned int vm_lopages_allocated_cpm_success = 0; +unsigned int vm_lopages_allocated_cpm_failed = 0; +vm_page_queue_head_t vm_lopage_queue_free __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); vm_page_t vm_page_grablo(void) { - vm_page_t mem; + vm_page_t mem; - if (vm_lopage_needed == FALSE) - return (vm_page_grab()); + 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); + if (!vm_page_queue_empty(&vm_lopage_queue_free)) { + vm_page_queue_remove_first(&vm_lopage_queue_free, mem, vmp_pageq); assert(vm_lopage_free_count); + assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q); + mem->vmp_q_state = VM_PAGE_NOT_ON_Q; - vm_lopage_free_count--; + vm_lopage_free_count--; vm_lopages_allocated_q++; - if (vm_lopage_free_count < vm_lopage_lowater) + if (vm_lopage_free_count < vm_lopage_lowater) { vm_lopage_refill = TRUE; + } lck_mtx_unlock(&vm_page_queue_free_lock); + +#if CONFIG_BACKGROUND_QUEUE + vm_page_assign_background_state(mem); +#endif } else { lck_mtx_unlock(&vm_page_queue_free_lock); - if (cpm_allocate(PAGE_SIZE, &mem, atop(0xffffffff), 0, FALSE, KMA_LOMEM) != KERN_SUCCESS) { - + if (cpm_allocate(PAGE_SIZE, &mem, atop(PPNUM_MAX), 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); + return VM_PAGE_NULL; } - mem->busy = TRUE; + assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); + + mem->vmp_busy = TRUE; vm_page_lockspin_queues(); - - mem->gobbled = FALSE; + + mem->vmp_gobbled = FALSE; vm_page_gobble_count--; vm_page_wire_count--; vm_lopages_allocated_cpm_success++; vm_page_unlock_queues(); } - assert(mem->gobbled); - assert(mem->busy); - assert(!mem->free); - assert(!mem->pmapped); - assert(!mem->wpmapped); + assert(mem->vmp_busy); + assert(!mem->vmp_pmapped); + assert(!mem->vmp_wpmapped); + assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem))); + + VM_PAGE_ZERO_PAGEQ_ENTRY(mem); - mem->pageq.next = NULL; - mem->pageq.prev = NULL; + disable_preemption(); + PROCESSOR_DATA(current_processor(), page_grab_count) += 1; + VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, 0, 1, 0, 0); + enable_preemption(); - return (mem); + return mem; } + /* * vm_page_grab: * * 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... + * 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... + * 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 + * 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 + * 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 + * 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. */ -#define COLOR_GROUPS_TO_STEAL 4 +#if CONFIG_SECLUDED_MEMORY +vm_page_t vm_page_grab_secluded(void); +#endif /* CONFIG_SECLUDED_MEMORY */ +static inline void +vm_page_grab_diags(void); vm_page_t -vm_page_grab( void ) +vm_page_grab(void) { - vm_page_t mem; + return vm_page_grab_options(VM_PAGE_GRAB_OPTIONS_NONE); +} + +#if HIBERNATION +boolean_t hibernate_rebuild_needed = FALSE; +#endif /* HIBERNATION */ +vm_page_t +vm_page_grab_options( + int grab_options) +{ + 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; - mem->pageq.next = NULL; - - enable_preemption(); - - 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->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q); + +#if HIBERNATION + if (hibernate_rebuild_needed) { + panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__); + } +#endif /* HIBERNATION */ + vm_page_grab_diags(); + PROCESSOR_DATA(current_processor(), page_grab_count) += 1; + PROCESSOR_DATA(current_processor(), free_pages) = mem->vmp_snext; + VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0); + + enable_preemption(); + VM_PAGE_ZERO_PAGEQ_ENTRY(mem); + mem->vmp_q_state = VM_PAGE_NOT_ON_Q; + + assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0); + assert(mem->vmp_tabled == FALSE); + assert(mem->vmp_object == 0); + assert(!mem->vmp_laundry); + ASSERT_PMAP_FREE(mem); + assert(mem->vmp_busy); + assert(!mem->vmp_pmapped); + assert(!mem->vmp_wpmapped); + assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem))); + +#if CONFIG_BACKGROUND_QUEUE + vm_page_assign_background_state(mem); +#endif return mem; } enable_preemption(); @@ -1780,17 +3078,27 @@ return_page_from_cpu_list: * 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); + vm_page_wire_count); } - 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); + vm_page_gobble_count); + } +#endif + + /* + * If free count is low and we have delayed pages from early boot, + * get one of those instead. + */ + if (__improbable(vm_delayed_count > 0 && + vm_page_free_count <= vm_page_free_target && + (mem = vm_get_delayed_page(grab_options)) != NULL)) { + return mem; } lck_mtx_lock_spin(&vm_page_queue_free_lock); @@ -1801,21 +3109,49 @@ return_page_from_cpu_list: */ if ((vm_page_free_count < vm_page_free_reserved) && !(current_thread()->options & TH_OPT_VMPRIV)) { + /* no page for us in the free queue... */ lck_mtx_unlock(&vm_page_queue_free_lock); mem = VM_PAGE_NULL; - } - else { - vm_page_t head; - vm_page_t tail; - unsigned int pages_to_steal; - unsigned int color; - while ( vm_page_free_count == 0 ) { +#if CONFIG_SECLUDED_MEMORY + /* ... but can we try and grab from the secluded queue? */ + if (vm_page_secluded_count > 0 && + ((grab_options & VM_PAGE_GRAB_SECLUDED) || + task_can_use_secluded_mem(current_task(), TRUE))) { + mem = vm_page_grab_secluded(); + if (grab_options & VM_PAGE_GRAB_SECLUDED) { + vm_page_secluded.grab_for_iokit++; + if (mem) { + vm_page_secluded.grab_for_iokit_success++; + } + } + if (mem) { + VM_CHECK_MEMORYSTATUS; + + disable_preemption(); + vm_page_grab_diags(); + PROCESSOR_DATA(current_processor(), page_grab_count) += 1; + VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0); + enable_preemption(); + + return mem; + } + } +#else /* CONFIG_SECLUDED_MEMORY */ + (void) grab_options; +#endif /* CONFIG_SECLUDED_MEMORY */ + } else { + vm_page_t head; + vm_page_t tail; + unsigned int pages_to_steal; + unsigned int color; + unsigned int clump_end, sub_count; + while (vm_page_free_count == 0) { lck_mtx_unlock(&vm_page_queue_free_lock); /* * must be a privileged thread to be - * in this state since a non-privileged + * in this state since a non-privileged * thread would have bailed if we were * under the vm_page_free_reserved mark */ @@ -1828,71 +3164,108 @@ return_page_from_cpu_list: 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 { - pages_to_steal = COLOR_GROUPS_TO_STEAL * vm_colors; - - if (pages_to_steal > (vm_page_free_count - vm_page_free_reserved)) - pages_to_steal = (vm_page_free_count - vm_page_free_reserved); + 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; + clump_end = sub_count = 0; + while (pages_to_steal--) { - if (--vm_page_free_count < vm_page_free_count_minimum) - vm_page_free_count_minimum = vm_page_free_count; - - 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; + while (vm_page_queue_empty(&vm_page_queue_free[color].qhead)) { + color = (color + 1) & vm_color_mask; + } +#if defined(__x86_64__) + vm_page_queue_remove_first_with_clump(&vm_page_queue_free[color].qhead, + mem, clump_end); +#else + vm_page_queue_remove_first(&vm_page_queue_free[color].qhead, + mem, vmp_pageq); +#endif + assert(mem->vmp_q_state == VM_PAGE_ON_FREE_Q); + + VM_PAGE_ZERO_PAGEQ_ENTRY(mem); + +#if defined(__arm__) || defined(__arm64__) color = (color + 1) & vm_color_mask; +#else + +#if DEVELOPMENT || DEBUG + + sub_count++; + if (clump_end) { + vm_clump_update_stats(sub_count); + sub_count = 0; + color = (color + 1) & vm_color_mask; + } +#else + if (clump_end) { + color = (color + 1) & vm_color_mask; + } + +#endif /* if DEVELOPMENT || DEBUG */ - if (head == NULL) +#endif /* if defined(__arm__) || defined(__arm64__) */ + + if (head == NULL) { head = mem; - else - tail->pageq.next = (queue_t)mem; - tail = mem; - - 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); - 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); - } - PROCESSOR_DATA(current_processor(), free_pages) = head->pageq.next; + } else { + tail->vmp_snext = mem; + } + tail = mem; + + assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0); + assert(mem->vmp_tabled == FALSE); + assert(mem->vmp_object == 0); + assert(!mem->vmp_laundry); + + mem->vmp_q_state = VM_PAGE_ON_FREE_LOCAL_Q; + + ASSERT_PMAP_FREE(mem); + assert(mem->vmp_busy); + assert(!mem->vmp_pmapped); + assert(!mem->vmp_wpmapped); + assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem))); + } +#if defined (__x86_64__) && (DEVELOPMENT || DEBUG) + vm_clump_update_stats(sub_count); +#endif + lck_mtx_unlock(&vm_page_queue_free_lock); + +#if HIBERNATION + if (hibernate_rebuild_needed) { + panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__); + } +#endif /* HIBERNATION */ + PROCESSOR_DATA(current_processor(), free_pages) = head->vmp_snext; PROCESSOR_DATA(current_processor(), start_color) = color; /* * satisfy this request */ - PROCESSOR_DATA(current_processor(), page_grab_count) += 1; + vm_page_grab_diags(); + PROCESSOR_DATA(current_processor(), page_grab_count) += 1; + VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0); mem = head; - mem->pageq.next = NULL; + assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q); - lck_mtx_unlock(&vm_page_queue_free_lock); + VM_PAGE_ZERO_PAGEQ_ENTRY(mem); + mem->vmp_q_state = VM_PAGE_NOT_ON_Q; enable_preemption(); } @@ -1906,105 +3279,280 @@ return_page_from_cpu_list: * We don't have the counts locked ... if they change a little, * it doesn't really matter. */ - 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); - -#if CONFIG_EMBEDDED - { - int percent_avail; - - /* - * Decide if we need to poke the memorystatus notification thread. - */ - percent_avail = - (vm_page_active_count + vm_page_inactive_count + - vm_page_speculative_count + vm_page_free_count + - (IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 / - atop_64(max_mem); - if (percent_avail <= (kern_memorystatus_level - 5)) { - kern_memorystatus_level = percent_avail; - thread_wakeup((event_t)&kern_memorystatus_wakeup); - } + if (vm_page_free_count < vm_page_free_min) { + thread_wakeup((event_t) &vm_page_free_wanted); } -#endif -// dbgLog(mem->phys_page, vm_page_free_count, vm_page_wire_count, 4); /* (TEST/DEBUG) */ + VM_CHECK_MEMORYSTATUS; + + if (mem) { +// dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 4); /* (TEST/DEBUG) */ +#if CONFIG_BACKGROUND_QUEUE + vm_page_assign_background_state(mem); +#endif + } return mem; } -/* - * vm_page_release: - * - * Return a page to the free list. - */ - -void -vm_page_release( - register vm_page_t mem) +#if CONFIG_SECLUDED_MEMORY +vm_page_t +vm_page_grab_secluded(void) { - unsigned int color; - int need_wakeup = 0; - int need_priv_wakeup = 0; -#if 0 - unsigned int pindex; - phys_entry *physent; + vm_page_t mem; + vm_object_t object; + int refmod_state; - 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); + if (vm_page_secluded_count == 0) { + /* no secluded pages to grab... */ + return VM_PAGE_NULL; } - physent->ppLink = physent->ppLink | ppN; /* (BRINGUP) */ -#endif - assert(!mem->private && !mem->fictitious); + + /* secluded queue is protected by the VM page queue lock */ + vm_page_lock_queues(); + + if (vm_page_secluded_count == 0) { + /* no secluded pages to grab... */ + vm_page_unlock_queues(); + return VM_PAGE_NULL; + } + +#if 00 + /* can we grab from the secluded queue? */ + if (vm_page_secluded_count > vm_page_secluded_target || + (vm_page_secluded_count > 0 && + task_can_use_secluded_mem(current_task(), TRUE))) { + /* OK */ + } else { + /* can't grab from secluded queue... */ + vm_page_unlock_queues(); + return VM_PAGE_NULL; + } +#endif + + /* we can grab a page from secluded queue! */ + assert((vm_page_secluded_count_free + + vm_page_secluded_count_inuse) == + vm_page_secluded_count); + if (current_task()->task_can_use_secluded_mem) { + assert(num_tasks_can_use_secluded_mem > 0); + } + assert(!vm_page_queue_empty(&vm_page_queue_secluded)); + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + mem = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded); + assert(mem->vmp_q_state == VM_PAGE_ON_SECLUDED_Q); + vm_page_queues_remove(mem, TRUE); + + object = VM_PAGE_OBJECT(mem); + + assert(!mem->vmp_fictitious); + assert(!VM_PAGE_WIRED(mem)); + if (object == VM_OBJECT_NULL) { + /* free for grab! */ + vm_page_unlock_queues(); + vm_page_secluded.grab_success_free++; + + assert(mem->vmp_busy); + assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); + assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL); + assert(mem->vmp_pageq.next == 0); + assert(mem->vmp_pageq.prev == 0); + assert(mem->vmp_listq.next == 0); + assert(mem->vmp_listq.prev == 0); +#if CONFIG_BACKGROUND_QUEUE + assert(mem->vmp_on_backgroundq == 0); + assert(mem->vmp_backgroundq.next == 0); + assert(mem->vmp_backgroundq.prev == 0); +#endif /* CONFIG_BACKGROUND_QUEUE */ + return mem; + } + + assert(!object->internal); +// vm_page_pageable_external_count--; + + if (!vm_object_lock_try(object)) { +// printf("SECLUDED: page %p: object %p locked\n", mem, object); + vm_page_secluded.grab_failure_locked++; +reactivate_secluded_page: + vm_page_activate(mem); + vm_page_unlock_queues(); + return VM_PAGE_NULL; + } + if (mem->vmp_busy || + mem->vmp_cleaning || + mem->vmp_laundry) { + /* can't steal page in this state... */ + vm_object_unlock(object); + vm_page_secluded.grab_failure_state++; + goto reactivate_secluded_page; + } + + mem->vmp_busy = TRUE; + refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(mem)); + if (refmod_state & VM_MEM_REFERENCED) { + mem->vmp_reference = TRUE; + } + if (refmod_state & VM_MEM_MODIFIED) { + SET_PAGE_DIRTY(mem, FALSE); + } + if (mem->vmp_dirty || mem->vmp_precious) { + /* can't grab a dirty page; re-activate */ +// printf("SECLUDED: dirty page %p\n", mem); + PAGE_WAKEUP_DONE(mem); + vm_page_secluded.grab_failure_dirty++; + vm_object_unlock(object); + goto reactivate_secluded_page; + } + if (mem->vmp_reference) { + /* it's been used but we do need to grab a page... */ + } + + vm_page_unlock_queues(); + + /* finish what vm_page_free() would have done... */ + vm_page_free_prepare_object(mem, TRUE); + vm_object_unlock(object); + object = VM_OBJECT_NULL; if (vm_page_free_verify) { - assert(pmap_verify_free(mem->phys_page)); + ASSERT_PMAP_FREE(mem); } -// dbgLog(mem->phys_page, vm_page_free_count, vm_page_wire_count, 5); /* (TEST/DEBUG) */ + pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); + vm_page_secluded.grab_success_other++; + + assert(mem->vmp_busy); + assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); + assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL); + assert(mem->vmp_pageq.next == 0); + assert(mem->vmp_pageq.prev == 0); + assert(mem->vmp_listq.next == 0); + assert(mem->vmp_listq.prev == 0); +#if CONFIG_BACKGROUND_QUEUE + assert(mem->vmp_on_backgroundq == 0); + assert(mem->vmp_backgroundq.next == 0); + assert(mem->vmp_backgroundq.prev == 0); +#endif /* CONFIG_BACKGROUND_QUEUE */ + return mem; +} +#endif /* CONFIG_SECLUDED_MEMORY */ + + +static inline void +vm_page_grab_diags() +{ +#if DEVELOPMENT || DEBUG + task_t task = current_task(); + if (task == NULL) { + return; + } + + ledger_credit(task->ledger, task_ledgers.pages_grabbed, 1); +#endif /* DEVELOPMENT || DEBUG */ +} + +/* + * vm_page_release: + * + * Return a page to the free list. + */ + +void +vm_page_release( + vm_page_t mem, + boolean_t page_queues_locked) +{ + unsigned int color; + int need_wakeup = 0; + int need_priv_wakeup = 0; +#if CONFIG_SECLUDED_MEMORY + int need_secluded_wakeup = 0; +#endif /* CONFIG_SECLUDED_MEMORY */ + + if (page_queues_locked) { + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + } else { + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); + } + + assert(!mem->vmp_private && !mem->vmp_fictitious); + if (vm_page_free_verify) { + ASSERT_PMAP_FREE(mem); + } +// dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 5); /* (TEST/DEBUG) */ + + pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); lck_mtx_lock_spin(&vm_page_queue_free_lock); -#if DEBUG - if (mem->free) - panic("vm_page_release"); + + assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); + assert(mem->vmp_busy); + assert(!mem->vmp_laundry); + assert(mem->vmp_object == 0); + assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); + assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0); +#if CONFIG_BACKGROUND_QUEUE + assert(mem->vmp_backgroundq.next == 0 && + mem->vmp_backgroundq.prev == 0 && + mem->vmp_on_backgroundq == FALSE); #endif - assert(mem->busy); - assert(!mem->laundry); - assert(mem->object == VM_OBJECT_NULL); - assert(mem->pageq.next == NULL && - mem->pageq.prev == NULL); - assert(mem->listq.next == NULL && - mem->listq.prev == NULL); - - if ((mem->lopage || vm_lopage_refill == TRUE) && + if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) && vm_lopage_free_count < vm_lopage_free_limit && - mem->phys_page < max_valid_low_ppnum) { - /* + VM_PAGE_GET_PHYS_PAGE(mem) < 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_page_queue_enter_first(&vm_lopage_queue_free, mem, vmp_pageq); vm_lopage_free_count++; - if (vm_lopage_free_count >= vm_lopage_free_limit) + if (vm_lopage_free_count >= vm_lopage_free_limit) { vm_lopage_refill = FALSE; + } - mem->lopage = TRUE; - } else { - mem->lopage = FALSE; - mem->free = TRUE; + mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q; + mem->vmp_lopage = TRUE; +#if CONFIG_SECLUDED_MEMORY + } else if (vm_page_free_count > vm_page_free_reserved && + vm_page_secluded_count < vm_page_secluded_target && + num_tasks_can_use_secluded_mem == 0) { + /* + * XXX FBDP TODO: also avoid refilling secluded queue + * when some IOKit objects are already grabbing from it... + */ + if (!page_queues_locked) { + if (!vm_page_trylock_queues()) { + /* take locks in right order */ + lck_mtx_unlock(&vm_page_queue_free_lock); + vm_page_lock_queues(); + lck_mtx_lock_spin(&vm_page_queue_free_lock); + } + } + mem->vmp_lopage = FALSE; + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + vm_page_queue_enter_first(&vm_page_queue_secluded, mem, vmp_pageq); + mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q; + vm_page_secluded_count++; + vm_page_secluded_count_free++; + if (!page_queues_locked) { + vm_page_unlock_queues(); + } + LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_OWNED); + if (vm_page_free_wanted_secluded > 0) { + vm_page_free_wanted_secluded--; + need_secluded_wakeup = 1; + } +#endif /* CONFIG_SECLUDED_MEMORY */ + } else { + mem->vmp_lopage = FALSE; + mem->vmp_q_state = VM_PAGE_ON_FREE_Q; - color = mem->phys_page & vm_color_mask; - queue_enter_first(&vm_page_queue_free[color], - mem, - vm_page_t, - pageq); + color = VM_PAGE_GET_COLOR(mem); +#if defined(__x86_64__) + vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem); +#else + vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq); +#endif vm_page_free_count++; /* * Check if we should wake up someone waiting for page. @@ -2028,40 +3576,82 @@ vm_page_release( assert(vm_page_free_count > 0); if (vm_page_free_wanted_privileged > 0) { - vm_page_free_wanted_privileged--; + vm_page_free_wanted_privileged--; need_priv_wakeup = 1; +#if CONFIG_SECLUDED_MEMORY + } else if (vm_page_free_wanted_secluded > 0 && + vm_page_free_count > vm_page_free_reserved) { + vm_page_free_wanted_secluded--; + need_secluded_wakeup = 1; +#endif /* CONFIG_SECLUDED_MEMORY */ } else if (vm_page_free_wanted > 0 && - vm_page_free_count > vm_page_free_reserved) { - vm_page_free_wanted--; + vm_page_free_count > vm_page_free_reserved) { + vm_page_free_wanted--; need_wakeup = 1; } } + vm_pageout_vminfo.vm_page_pages_freed++; + + VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, 1, 0, 0, 0); + lck_mtx_unlock(&vm_page_queue_free_lock); - if (need_priv_wakeup) + if (need_priv_wakeup) { thread_wakeup_one((event_t) &vm_page_free_wanted_privileged); - else if (need_wakeup) + } +#if CONFIG_SECLUDED_MEMORY + else if (need_secluded_wakeup) { + thread_wakeup_one((event_t) &vm_page_free_wanted_secluded); + } +#endif /* CONFIG_SECLUDED_MEMORY */ + else if (need_wakeup) { thread_wakeup_one((event_t) &vm_page_free_count); + } -#if CONFIG_EMBEDDED - { - int percent_avail; + VM_CHECK_MEMORYSTATUS; +} - /* - * Decide if we need to poke the memorystatus notification thread. - * Locking is not a big issue, as only a single thread delivers these. - */ - percent_avail = - (vm_page_active_count + vm_page_inactive_count + - vm_page_speculative_count + vm_page_free_count + - (IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 / - atop_64(max_mem); - if (percent_avail >= (kern_memorystatus_level + 5)) { - kern_memorystatus_level = percent_avail; - thread_wakeup((event_t)&kern_memorystatus_wakeup); - } +/* + * 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( + vm_page_t mem) +{ + vm_page_queue_t queue_free; + + if (vm_lopage_free_count < vm_lopage_free_limit && + VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) { + mem->vmp_lopage = TRUE; + mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q; + vm_lopage_free_count++; + queue_free = &vm_lopage_queue_free; +#if CONFIG_SECLUDED_MEMORY + } else if (vm_page_secluded_count < vm_page_secluded_target) { + mem->vmp_lopage = FALSE; + mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q; + vm_page_secluded_count++; + vm_page_secluded_count_free++; + queue_free = &vm_page_queue_secluded; +#endif /* CONFIG_SECLUDED_MEMORY */ + } else { + mem->vmp_lopage = FALSE; + mem->vmp_q_state = VM_PAGE_ON_FREE_Q; + vm_page_free_count++; + queue_free = &vm_page_queue_free[VM_PAGE_GET_COLOR(mem)].qhead; } + if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) { +#if defined(__x86_64__) + vm_page_queue_enter_clump(queue_free, mem); +#else + vm_page_queue_enter(queue_free, mem, vmp_pageq); #endif + } else { + vm_page_queue_enter_first(queue_free, mem, vmp_pageq); + } } /* @@ -2077,7 +3667,7 @@ vm_page_release( boolean_t vm_page_wait( - int interruptible ) + int interruptible ) { /* * We can't use vm_page_free_reserved to make this @@ -2086,9 +3676,9 @@ vm_page_wait( * succeeds, the second fails. After the first page is freed, * a call to vm_page_wait must really block. */ - kern_return_t wait_result; - int need_wakeup = 0; - int is_privileged = current_thread()->options & TH_OPT_VMPRIV; + 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); @@ -2096,31 +3686,66 @@ vm_page_wait( lck_mtx_unlock(&vm_page_queue_free_lock); return TRUE; } - if (vm_page_free_count < vm_page_free_target) { - 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); - } + if (vm_page_free_count >= vm_page_free_target) { lck_mtx_unlock(&vm_page_queue_free_lock); - counter(c_vm_page_wait_block++); + return TRUE; + } - if (need_wakeup) - thread_wakeup((event_t)&vm_page_free_wanted); + 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); +#if CONFIG_SECLUDED_MEMORY + } else if (secluded_for_apps && + task_can_use_secluded_mem(current_task(), FALSE)) { +#if 00 + /* XXX FBDP: need pageq lock for this... */ + /* XXX FBDP: might wait even if pages available, */ + /* XXX FBDP: hopefully not for too long... */ + if (vm_page_secluded_count > 0) { + lck_mtx_unlock(&vm_page_queue_free_lock); + return TRUE; + } +#endif + if (vm_page_free_wanted_secluded++ == 0) { + need_wakeup = 1; + } + wait_result = assert_wait( + (event_t)&vm_page_free_wanted_secluded, + interruptible); +#endif /* CONFIG_SECLUDED_MEMORY */ + } 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 (wait_result == THREAD_WAITING) - wait_result = thread_block(THREAD_CONTINUE_NULL); + if (need_wakeup) { + thread_wakeup((event_t)&vm_page_free_wanted); + } - return(wait_result == THREAD_AWAKENED); - } else { - lck_mtx_unlock(&vm_page_queue_free_lock); - return TRUE; + if (wait_result == THREAD_WAITING) { + VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, VM_PAGE_WAIT_BLOCK, DBG_FUNC_START, + vm_page_free_wanted_privileged, + vm_page_free_wanted, +#if CONFIG_SECLUDED_MEMORY + vm_page_free_wanted_secluded, +#else /* CONFIG_SECLUDED_MEMORY */ + 0, +#endif /* CONFIG_SECLUDED_MEMORY */ + 0); + wait_result = thread_block(THREAD_CONTINUE_NULL); + VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, + VM_PAGE_WAIT_BLOCK, DBG_FUNC_END, 0, 0, 0, 0); } + + return wait_result == THREAD_AWAKENED; } /* @@ -2134,78 +3759,69 @@ vm_page_wait( vm_page_t vm_page_alloc( - vm_object_t object, - vm_object_offset_t offset) -{ - register vm_page_t mem; - - vm_object_lock_assert_exclusive(object); - mem = vm_page_grab(); - if (mem == VM_PAGE_NULL) - return VM_PAGE_NULL; - - vm_page_insert(mem, object, offset); - - return(mem); -} - -vm_page_t -vm_page_alloclo( - vm_object_t object, - vm_object_offset_t offset) + vm_object_t object, + vm_object_offset_t offset) { - register vm_page_t mem; + vm_page_t mem; + int grab_options; vm_object_lock_assert_exclusive(object); - mem = vm_page_grablo(); - if (mem == VM_PAGE_NULL) + grab_options = 0; +#if CONFIG_SECLUDED_MEMORY + if (object->can_grab_secluded) { + grab_options |= VM_PAGE_GRAB_SECLUDED; + } +#endif /* CONFIG_SECLUDED_MEMORY */ + mem = vm_page_grab_options(grab_options); + if (mem == VM_PAGE_NULL) { return VM_PAGE_NULL; + } vm_page_insert(mem, object, offset); - return(mem); + return mem; } - /* * vm_page_alloc_guard: - * - * Allocate a fictitious page which will be used + * + * 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) + vm_object_t object, + vm_object_offset_t offset) { - register vm_page_t mem; + vm_page_t mem; vm_object_lock_assert_exclusive(object); mem = vm_page_grab_guard(); - if (mem == VM_PAGE_NULL) + if (mem == VM_PAGE_NULL) { return VM_PAGE_NULL; + } vm_page_insert(mem, object, offset); - return(mem); + return mem; } -counter(unsigned int c_laundry_pages_freed = 0;) +counter(unsigned int c_laundry_pages_freed = 0; ) /* - * 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. */ static void vm_page_free_prepare( - register vm_page_t mem) + vm_page_t mem) { vm_page_free_prepare_queues(mem); vm_page_free_prepare_object(mem, TRUE); @@ -2214,48 +3830,100 @@ vm_page_free_prepare( void vm_page_free_prepare_queues( - vm_page_t mem) + vm_page_t mem) { + vm_object_t m_object; + VM_PAGE_CHECK(mem); - assert(!mem->free); - assert(!mem->cleaning); - assert(!mem->pageout); -#if 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->object) { - vm_object_lock_assert_exclusive(mem->object); - } - if (mem->laundry) { + assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q); + assert(!mem->vmp_cleaning); + m_object = VM_PAGE_OBJECT(mem); + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + if (m_object) { + vm_object_lock_assert_exclusive(m_object); + } + if (mem->vmp_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_throttle_up() before removing - * the page from its VM object, so that we can find out on - * which pageout queue the page is on. + * 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_throttle_up(mem); + vm_pageout_steal_laundry(mem, TRUE); counter(++c_laundry_pages_freed); } - VM_PAGE_QUEUES_REMOVE(mem); /* clears local/active/inactive/throttled/speculative */ + + vm_page_queues_remove(mem, TRUE); if (VM_PAGE_WIRED(mem)) { - if (mem->object) { - assert(mem->object->wired_page_count > 0); - mem->object->wired_page_count--; - assert(mem->object->resident_page_count >= - mem->object->wired_page_count); + assert(mem->vmp_wire_count > 0); + + if (m_object) { + VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); + VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem); + VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag); + + assert(m_object->resident_page_count >= + m_object->wired_page_count); + + if (m_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 ((m_object->purgable == VM_PURGABLE_VOLATILE || + m_object->purgable == VM_PURGABLE_EMPTY) && + m_object->vo_owner != TASK_NULL) { + task_t owner; + int ledger_idx_volatile; + int ledger_idx_nonvolatile; + int ledger_idx_volatile_compressed; + int ledger_idx_nonvolatile_compressed; + boolean_t do_footprint; + + owner = VM_OBJECT_OWNER(m_object); + vm_object_ledger_tag_ledgers( + m_object, + &ledger_idx_volatile, + &ledger_idx_nonvolatile, + &ledger_idx_volatile_compressed, + &ledger_idx_nonvolatile_compressed, + &do_footprint); + /* + * 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, + ledger_idx_nonvolatile, + PAGE_SIZE); + if (do_footprint) { + /* ... and "phys_footprint" */ + ledger_debit(owner->ledger, + task_ledgers.phys_footprint, + PAGE_SIZE); + } + /* one more "volatile" */ + ledger_credit(owner->ledger, + ledger_idx_volatile, + PAGE_SIZE); + } } - if (!mem->private && !mem->fictitious) + if (!mem->vmp_private && !mem->vmp_fictitious) { vm_page_wire_count--; - mem->wire_count = 0; - assert(!mem->gobbled); - } else if (mem->gobbled) { - if (!mem->private && !mem->fictitious) + } + + mem->vmp_q_state = VM_PAGE_NOT_ON_Q; + mem->vmp_wire_count = 0; + assert(!mem->vmp_gobbled); + } else if (mem->vmp_gobbled) { + if (!mem->vmp_private && !mem->vmp_fictitious) { vm_page_wire_count--; + } vm_page_gobble_count--; } } @@ -2263,62 +3931,62 @@ vm_page_free_prepare_queues( void vm_page_free_prepare_object( - vm_page_t mem, - boolean_t remove_from_hash) -{ - if (mem->object) { - vm_object_lock_assert_exclusive(mem->object); - } - - if (mem->tabled) - vm_page_remove(mem, remove_from_hash); /* clears tabled, object, offset */ - - PAGE_WAKEUP(mem); /* clears wanted */ - - if (mem->private) { - mem->private = FALSE; - mem->fictitious = TRUE; - mem->phys_page = vm_page_fictitious_addr; - } - if (mem->fictitious) { - /* Some of these may be unnecessary */ - mem->gobbled = FALSE; - mem->busy = TRUE; - mem->absent = FALSE; - mem->error = FALSE; - mem->dirty = FALSE; - mem->precious = FALSE; - mem->reference = FALSE; - mem->encrypted = FALSE; - mem->encrypted_cleaning = FALSE; - mem->pmapped = FALSE; - mem->wpmapped = FALSE; - mem->reusable = FALSE; - } else { - if (mem->zero_fill == TRUE) - VM_ZF_COUNT_DECR(); - vm_page_init(mem, mem->phys_page, mem->lopage); + vm_page_t mem, + boolean_t remove_from_hash) +{ + if (mem->vmp_tabled) { + vm_page_remove(mem, remove_from_hash); /* clears tabled, object, offset */ + } + PAGE_WAKEUP(mem); /* clears wanted */ + + if (mem->vmp_private) { + mem->vmp_private = FALSE; + mem->vmp_fictitious = TRUE; + VM_PAGE_SET_PHYS_PAGE(mem, vm_page_fictitious_addr); + } + if (!mem->vmp_fictitious) { + assert(mem->vmp_pageq.next == 0); + assert(mem->vmp_pageq.prev == 0); + assert(mem->vmp_listq.next == 0); + assert(mem->vmp_listq.prev == 0); +#if CONFIG_BACKGROUND_QUEUE + assert(mem->vmp_backgroundq.next == 0); + assert(mem->vmp_backgroundq.prev == 0); +#endif /* CONFIG_BACKGROUND_QUEUE */ + assert(mem->vmp_next_m == 0); + ASSERT_PMAP_FREE(mem); + vm_page_init(mem, VM_PAGE_GET_PHYS_PAGE(mem), mem->vmp_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_t mem) { vm_page_free_prepare(mem); - if (mem->fictitious) { + + if (mem->vmp_fictitious) { vm_page_release_fictitious(mem); } else { - vm_page_release(mem); + vm_page_release(mem, + TRUE); /* page queues are locked */ } } void vm_page_free_unlocked( - vm_page_t mem, - boolean_t remove_from_hash) + vm_page_t mem, + boolean_t remove_from_hash) { vm_page_lockspin_queues(); vm_page_free_prepare_queues(mem); @@ -2326,217 +3994,227 @@ vm_page_free_unlocked( vm_page_free_prepare_object(mem, remove_from_hash); - if (mem->fictitious) { + if (mem->vmp_fictitious) { vm_page_release_fictitious(mem); } else { - vm_page_release(mem); + vm_page_release(mem, FALSE); /* page queues are not locked */ } } + /* * 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. We sort the incoming pages into n lists, one for - * each color. + * per page. + * + * The VM page queues lock (vm_page_queue_lock) should NOT be held. + * The VM page free queues lock (vm_page_queue_free_lock) should NOT be held. */ void vm_page_free_list( - vm_page_t mem, - boolean_t prepare_object) -{ - vm_page_t nxt; - int pg_count = 0; - int color; - int inuse_list_head = -1; - - queue_head_t free_list[MAX_COLORS]; - int inuse[MAX_COLORS]; - - for (color = 0; color < (signed) vm_colors; color++) { - queue_init(&free_list[color]); - } - - while (mem) { - 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); + vm_page_t freeq, + boolean_t prepare_object) +{ + vm_page_t mem; + vm_page_t nxt; + vm_page_t local_freeq; + int pg_count; - nxt = (vm_page_t)(mem->pageq.next); - - if (prepare_object == TRUE) - vm_page_free_prepare_object(mem, TRUE); + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); + LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_NOTOWNED); - if (vm_page_free_verify && !mem->fictitious && !mem->private) { - assert(pmap_verify_free(mem->phys_page)); - } - assert(mem->busy); + while (freeq) { + pg_count = 0; + local_freeq = VM_PAGE_NULL; + mem = freeq; - if (!mem->fictitious) { - 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 { + /* + * 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) { + assert((mem->vmp_q_state == VM_PAGE_NOT_ON_Q) || + (mem->vmp_q_state == VM_PAGE_IS_WIRED)); +#if CONFIG_BACKGROUND_QUEUE + assert(mem->vmp_backgroundq.next == 0 && + mem->vmp_backgroundq.prev == 0 && + mem->vmp_on_backgroundq == FALSE); +#endif + nxt = mem->vmp_snext; + mem->vmp_snext = NULL; + assert(mem->vmp_pageq.prev == 0); - /* - * 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... - */ - color = mem->phys_page & vm_color_mask; - if (queue_empty(&free_list[color])) { - inuse[color] = inuse_list_head; - inuse_list_head = color; + if (vm_page_free_verify && !mem->vmp_fictitious && !mem->vmp_private) { + ASSERT_PMAP_FREE(mem); + } + if (prepare_object == TRUE) { + vm_page_free_prepare_object(mem, TRUE); + } + + if (!mem->vmp_fictitious) { + assert(mem->vmp_busy); + + if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) && + vm_lopage_free_count < vm_lopage_free_limit && + VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) { + vm_page_release(mem, FALSE); /* page queues are not locked */ +#if CONFIG_SECLUDED_MEMORY + } else if (vm_page_secluded_count < vm_page_secluded_target && + num_tasks_can_use_secluded_mem == 0) { + vm_page_release(mem, + FALSE); /* page queues are not locked */ +#endif /* CONFIG_SECLUDED_MEMORY */ + } 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->vmp_snext = local_freeq; + local_freeq = mem; + pg_count++; + + pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); } - queue_enter_first(&free_list[color], - mem, - vm_page_t, - pageq); - pg_count++; + } else { + assert(VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_fictitious_addr || + VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_guard_addr); + vm_page_release_fictitious(mem); } - } 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 (pg_count) { - 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); + freeq = mem; - color = inuse_list_head; - - while( color != -1 ) { - vm_page_t first, last; - vm_page_t first_free; + if ((mem = local_freeq)) { + unsigned int avail_free_count; + unsigned int need_wakeup = 0; + unsigned int need_priv_wakeup = 0; +#if CONFIG_SECLUDED_MEMORY + unsigned int need_wakeup_secluded = 0; +#endif /* CONFIG_SECLUDED_MEMORY */ - /* - * Now that we hold the vm_page_queue_free lock, - * it's safe to mark all pages in our local queue - * as "free"... - */ - queue_iterate(&free_list[color], - mem, - vm_page_t, - pageq) { - assert(!mem->free); - assert(mem->busy); - mem->free = TRUE; + lck_mtx_lock_spin(&vm_page_queue_free_lock); + + while (mem) { + int color; + + nxt = mem->vmp_snext; + + assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); + assert(mem->vmp_busy); + mem->vmp_lopage = FALSE; + mem->vmp_q_state = VM_PAGE_ON_FREE_Q; + + color = VM_PAGE_GET_COLOR(mem); +#if defined(__x86_64__) + vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem); +#else + vm_page_queue_enter(&vm_page_queue_free[color].qhead, + mem, vmp_pageq); +#endif + mem = nxt; } + vm_pageout_vminfo.vm_page_pages_freed += pg_count; + vm_page_free_count += pg_count; + avail_free_count = vm_page_free_count; - /* - * ... and insert our local queue at the head of - * the global free queue. - */ - first = (vm_page_t) queue_first(&free_list[color]); - last = (vm_page_t) queue_last(&free_list[color]); - first_free = (vm_page_t) queue_first(&vm_page_queue_free[color]); - if (queue_empty(&vm_page_queue_free[color])) { - queue_last(&vm_page_queue_free[color]) = - (queue_entry_t) last; - } else { - queue_prev(&first_free->pageq) = - (queue_entry_t) last; - } - queue_first(&vm_page_queue_free[color]) = - (queue_entry_t) first; - queue_prev(&first->pageq) = - (queue_entry_t) &vm_page_queue_free[color]; - queue_next(&last->pageq) = - (queue_entry_t) first_free; - - /* next color */ - color = inuse[color]; - } - - vm_page_free_count += pg_count; - avail_free_count = vm_page_free_count; - - if (vm_page_free_wanted_privileged > 0 && - avail_free_count > 0) { - 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; + VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, pg_count, 0, 0, 0); + + if (vm_page_free_wanted_privileged > 0 && avail_free_count > 0) { + 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; + avail_free_count -= vm_page_free_wanted_privileged; + vm_page_free_wanted_privileged = 0; + } } - } +#if CONFIG_SECLUDED_MEMORY + if (vm_page_free_wanted_secluded > 0 && + avail_free_count > vm_page_free_reserved) { + unsigned int available_pages; + available_pages = (avail_free_count - + vm_page_free_reserved); + if (available_pages < + vm_page_free_wanted_secluded) { + need_wakeup_secluded = available_pages; + vm_page_free_wanted_secluded -= + available_pages; + avail_free_count -= available_pages; + } else { + need_wakeup_secluded = + vm_page_free_wanted_secluded; + avail_free_count -= + vm_page_free_wanted_secluded; + vm_page_free_wanted_secluded = 0; + } + } +#endif /* CONFIG_SECLUDED_MEMORY */ + if (vm_page_free_wanted > 0 && avail_free_count > vm_page_free_reserved) { + unsigned int available_pages; - if (vm_page_free_wanted > 0 && - avail_free_count > vm_page_free_reserved) { - unsigned int available_pages; + available_pages = avail_free_count - vm_page_free_reserved; - available_pages = (avail_free_count - - vm_page_free_reserved); + 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) { - need_wakeup = 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 CONFIG_SECLUDED_MEMORY + if (need_wakeup_secluded != 0 && + vm_page_free_wanted_secluded == 0) { + thread_wakeup((event_t) + &vm_page_free_wanted_secluded); } else { - need_wakeup = available_pages; - vm_page_free_wanted -= available_pages; + for (; + need_wakeup_secluded != 0; + need_wakeup_secluded--) { + thread_wakeup_one( + (event_t) + &vm_page_free_wanted_secluded); + } + } +#endif /* CONFIG_SECLUDED_MEMORY */ + 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 { + for (; need_wakeup != 0; need_wakeup--) { + /* + * Wake up one waiter per page we just released. + */ + thread_wakeup_one((event_t) &vm_page_free_count); + } } - } - lck_mtx_unlock(&vm_page_queue_free_lock); - - 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 for (; need_wakeup != 0; need_wakeup--) { - /* - * Wake up one waiter per page we just released. - */ - thread_wakeup_one((event_t) &vm_page_free_count); - } -#if CONFIG_EMBEDDED - { - int percent_avail; - /* - * Decide if we need to poke the memorystatus notification thread. - */ - percent_avail = - (vm_page_active_count + vm_page_inactive_count + - vm_page_speculative_count + vm_page_free_count + - (IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 / - atop_64(max_mem); - if (percent_avail >= (kern_memorystatus_level + 5)) { - kern_memorystatus_level = percent_avail; - thread_wakeup((event_t)&kern_memorystatus_wakeup); - } + VM_CHECK_MEMORYSTATUS; } -#endif } } @@ -2550,16 +4228,23 @@ vm_page_free_list( * * The page's object and the page queues must be locked. */ + + void vm_page_wire( - register vm_page_t mem) + vm_page_t mem, + vm_tag_t tag, + boolean_t check_memorystatus) { + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(mem); -// dbgLog(current_thread(), mem->offset, mem->object, 1); /* (TEST/DEBUG) */ +// dbgLog(current_thread(), mem->vmp_offset, m_object, 1); /* (TEST/DEBUG) */ VM_PAGE_CHECK(mem); - if (mem->object) { - vm_object_lock_assert_exclusive(mem->object); + if (m_object) { + vm_object_lock_assert_exclusive(m_object); } else { /* * In theory, the page should be in an object before it @@ -2571,109 +4256,103 @@ vm_page_wire( * that page and update it at the same time. */ } -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); -#endif - if ( !VM_PAGE_WIRED(mem)) { - VM_PAGE_QUEUES_REMOVE(mem); - - if (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) { + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + if (!VM_PAGE_WIRED(mem)) { + if (mem->vmp_laundry) { + vm_pageout_steal_laundry(mem, TRUE); + } + + vm_page_queues_remove(mem, TRUE); + + assert(mem->vmp_wire_count == 0); + mem->vmp_q_state = VM_PAGE_IS_WIRED; + + if (m_object) { + VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); + VM_OBJECT_WIRED_PAGE_ADD(m_object, mem); + VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, tag); + + assert(m_object->resident_page_count >= + m_object->wired_page_count); + if (m_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->all_reusable) { + if ((m_object->purgable == VM_PURGABLE_VOLATILE || + m_object->purgable == VM_PURGABLE_EMPTY) && + m_object->vo_owner != TASK_NULL) { + task_t owner; + int ledger_idx_volatile; + int ledger_idx_nonvolatile; + int ledger_idx_volatile_compressed; + int ledger_idx_nonvolatile_compressed; + boolean_t do_footprint; + + owner = VM_OBJECT_OWNER(m_object); + vm_object_ledger_tag_ledgers( + m_object, + &ledger_idx_volatile, + &ledger_idx_nonvolatile, + &ledger_idx_volatile_compressed, + &ledger_idx_nonvolatile_compressed, + &do_footprint); + /* less volatile bytes */ + ledger_debit(owner->ledger, + ledger_idx_volatile, + PAGE_SIZE); + /* more not-quite-volatile bytes */ + ledger_credit(owner->ledger, + ledger_idx_nonvolatile, + PAGE_SIZE); + if (do_footprint) { + /* more footprint */ + ledger_credit(owner->ledger, + task_ledgers.phys_footprint, + PAGE_SIZE); + } + } + if (m_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) { + } else if (mem->vmp_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); + vm_object_reuse_pages(m_object, + mem->vmp_offset, + mem->vmp_offset + PAGE_SIZE_64, + FALSE); } } - assert(!mem->reusable); + assert(!mem->vmp_reusable); - if (!mem->private && !mem->fictitious && !mem->gobbled) + if (!mem->vmp_private && !mem->vmp_fictitious && !mem->vmp_gobbled) { vm_page_wire_count++; - if (mem->gobbled) + } + if (mem->vmp_gobbled) { vm_page_gobble_count--; - mem->gobbled = FALSE; - if (mem->zero_fill == TRUE) { - mem->zero_fill = FALSE; - VM_ZF_COUNT_DECR(); } -#if CONFIG_EMBEDDED - { - int percent_avail; + mem->vmp_gobbled = FALSE; - /* - * Decide if we need to poke the memorystatus notification thread. - */ - percent_avail = - (vm_page_active_count + vm_page_inactive_count + - vm_page_speculative_count + vm_page_free_count + - (IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 / - atop_64(max_mem); - if (percent_avail <= (kern_memorystatus_level - 5)) { - kern_memorystatus_level = percent_avail; - thread_wakeup((event_t)&kern_memorystatus_wakeup); - } + if (check_memorystatus == TRUE) { + VM_CHECK_MEMORYSTATUS; } -#endif - /* - * ENCRYPTED SWAP: - * The page could be encrypted, but - * We don't have to decrypt it here - * because we don't guarantee that the - * data is actually valid at this point. - * The page will get decrypted in - * vm_fault_wire() if needed. - */ } - assert(!mem->gobbled); - mem->wire_count++; + assert(!mem->vmp_gobbled); + assert(mem->vmp_q_state == VM_PAGE_IS_WIRED); + mem->vmp_wire_count++; + if (__improbable(mem->vmp_wire_count == 0)) { + panic("vm_page_wire(%p): wire_count overflow", mem); + } VM_PAGE_CHECK(mem); } -/* - * 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_lockspin_queues(); - VM_PAGE_CHECK(mem); - - assert(!mem->gobbled); - assert( !VM_PAGE_WIRED(mem)); - - if (!mem->gobbled && !VM_PAGE_WIRED(mem)) { - if (!mem->private && !mem->fictitious) - vm_page_wire_count++; - } - vm_page_gobble_count++; - mem->gobbled = TRUE; - vm_page_unlock_queues(); -} - /* * vm_page_unwire: * @@ -2684,60 +4363,84 @@ vm_page_gobble( */ void vm_page_unwire( - vm_page_t mem, - boolean_t queueit) + vm_page_t mem, + boolean_t queueit) { + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(mem); -// dbgLog(current_thread(), mem->offset, mem->object, 0); /* (TEST/DEBUG) */ +// dbgLog(current_thread(), mem->vmp_offset, m_object, 0); /* (TEST/DEBUG) */ VM_PAGE_CHECK(mem); assert(VM_PAGE_WIRED(mem)); - assert(mem->object != VM_OBJECT_NULL); -#if DEBUG - 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->object->wired_page_count > 0); - mem->object->wired_page_count--; - assert(mem->object->resident_page_count >= - mem->object->wired_page_count); - if (mem->object->purgable == VM_PURGABLE_VOLATILE) { + assert(mem->vmp_wire_count > 0); + assert(!mem->vmp_gobbled); + assert(m_object != VM_OBJECT_NULL); + vm_object_lock_assert_exclusive(m_object); + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + if (--mem->vmp_wire_count == 0) { + mem->vmp_q_state = VM_PAGE_NOT_ON_Q; + + VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); + VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem); + VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag); + if (!mem->vmp_private && !mem->vmp_fictitious) { + vm_page_wire_count--; + } + + assert(m_object->resident_page_count >= + m_object->wired_page_count); + if (m_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); } - assert(!mem->laundry); - assert(mem->object != kernel_object); - assert(mem->pageq.next == NULL && mem->pageq.prev == NULL); + if ((m_object->purgable == VM_PURGABLE_VOLATILE || + m_object->purgable == VM_PURGABLE_EMPTY) && + m_object->vo_owner != TASK_NULL) { + task_t owner; + int ledger_idx_volatile; + int ledger_idx_nonvolatile; + int ledger_idx_volatile_compressed; + int ledger_idx_nonvolatile_compressed; + boolean_t do_footprint; + + owner = VM_OBJECT_OWNER(m_object); + vm_object_ledger_tag_ledgers( + m_object, + &ledger_idx_volatile, + &ledger_idx_nonvolatile, + &ledger_idx_volatile_compressed, + &ledger_idx_nonvolatile_compressed, + &do_footprint); + /* more volatile bytes */ + ledger_credit(owner->ledger, + ledger_idx_volatile, + PAGE_SIZE); + /* less not-quite-volatile bytes */ + ledger_debit(owner->ledger, + ledger_idx_nonvolatile, + PAGE_SIZE); + if (do_footprint) { + /* less footprint */ + ledger_debit(owner->ledger, + task_ledgers.phys_footprint, + PAGE_SIZE); + } + } + assert(m_object != kernel_object); + assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); if (queueit == TRUE) { - if (mem->object->purgable == VM_PURGABLE_EMPTY) { + if (m_object->purgable == VM_PURGABLE_EMPTY) { vm_page_deactivate(mem); } else { vm_page_activate(mem); } } -#if CONFIG_EMBEDDED - { - int percent_avail; - /* - * Decide if we need to poke the memorystatus notification thread. - */ - percent_avail = - (vm_page_active_count + vm_page_inactive_count + - vm_page_speculative_count + vm_page_free_count + - (IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 / - atop_64(max_mem); - if (percent_avail >= (kern_memorystatus_level + 5)) { - kern_memorystatus_level = percent_avail; - thread_wakeup((event_t)&kern_memorystatus_wakeup); - } - } -#endif + VM_CHECK_MEMORYSTATUS; } VM_PAGE_CHECK(mem); } @@ -2753,7 +4456,7 @@ vm_page_unwire( */ void vm_page_deactivate( - vm_page_t m) + vm_page_t m) { vm_page_deactivate_internal(m, TRUE); } @@ -2761,18 +4464,19 @@ vm_page_deactivate( void vm_page_deactivate_internal( - vm_page_t m, - boolean_t clear_hw_reference) + vm_page_t m, + boolean_t clear_hw_reference) { + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(m); VM_PAGE_CHECK(m); - assert(m->object != kernel_object); - assert(m->phys_page != vm_page_guard_addr); + assert(m_object != kernel_object); + assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); -// dbgLog(m->phys_page, vm_page_free_count, vm_page_wire_count, 6); /* (TEST/DEBUG) */ -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); -#endif +// dbgLog(VM_PAGE_GET_PHYS_PAGE(m), vm_page_free_count, vm_page_wire_count, 6); /* (TEST/DEBUG) */ + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); /* * This page is no longer very interesting. If it was * interesting (active or inactive/referenced), then we @@ -2780,65 +4484,129 @@ vm_page_deactivate_internal( * inactive queue. Note wired pages should not have * their reference bit cleared. */ + assert( !(m->vmp_absent && !m->vmp_unusual)); - if (m->absent && !m->unusual) - panic("vm_page_deactivate: %p absent", m); - - if (m->gobbled) { /* can this happen? */ + if (m->vmp_gobbled) { /* can this happen? */ assert( !VM_PAGE_WIRED(m)); - if (!m->private && !m->fictitious) + if (!m->vmp_private && !m->vmp_fictitious) { vm_page_wire_count--; + } vm_page_gobble_count--; - m->gobbled = FALSE; + m->vmp_gobbled = FALSE; } - if (m->private || (VM_PAGE_WIRED(m))) + /* + * 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->vmp_laundry || m->vmp_private || m->vmp_fictitious || + (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || + (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) || + VM_PAGE_WIRED(m)) { return; + } + if (!m->vmp_absent && clear_hw_reference == TRUE) { + pmap_clear_reference(VM_PAGE_GET_PHYS_PAGE(m)); + } - if (!m->fictitious && !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); + m->vmp_reference = FALSE; + m->vmp_no_cache = FALSE; - assert(!m->laundry); - assert(m->pageq.next == NULL && m->pageq.prev == NULL); + if (!VM_PAGE_INACTIVE(m)) { + vm_page_queues_remove(m, FALSE); - if (!IP_VALID(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; + if (!VM_DYNAMIC_PAGING_ENABLED() && + m->vmp_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); + vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq); + m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q; vm_page_throttled_count++; } else { - if (!m->fictitious && m->object->named && m->object->ref_count == 1) { - vm_page_speculate(m, FALSE); + if (m_object->named && m_object->ref_count == 1) { + vm_page_speculate(m, FALSE); #if DEVELOPMENT || DEBUG vm_page_speculative_recreated++; #endif - return; } else { - if (m->zero_fill) { - queue_enter(&vm_page_queue_zf, m, vm_page_t, pageq); - vm_zf_queue_count++; - } else { - queue_enter(&vm_page_queue_inactive, m, vm_page_t, pageq); - } - } - m->inactive = TRUE; - if (!m->fictitious) { - vm_page_inactive_count++; - token_new_pagecount++; + 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) +{ + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(m); + + assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + assert( !(m->vmp_absent && !m->vmp_unusual)); + + if (VM_PAGE_WIRED(m)) { + return; + } + + if (m->vmp_gobbled) { + if (!m->vmp_private && !m->vmp_fictitious) { + vm_page_wire_count--; + } + vm_page_gobble_count--; + m->vmp_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->vmp_laundry || m->vmp_private || m->vmp_fictitious || + (m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) || + (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { + return; + } + vm_page_queues_remove(m, FALSE); + + vm_page_check_pageable_safe(m); + vm_page_queue_enter(&vm_page_queue_cleaned, m, vmp_pageq); + m->vmp_q_state = VM_PAGE_ON_INACTIVE_CLEANED_Q; + vm_page_cleaned_count++; + + vm_page_inactive_count++; + if (m_object->internal) { + vm_page_pageable_internal_count++; + } else { + vm_page_pageable_external_count++; + } +#if CONFIG_BACKGROUND_QUEUE + if (m->vmp_in_background) { + vm_page_add_to_backgroundq(m, TRUE); + } +#endif + VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1); +} + /* * vm_page_activate: * @@ -2849,61 +4617,83 @@ vm_page_deactivate_internal( void vm_page_activate( - register vm_page_t m) + vm_page_t m) { + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(m); + VM_PAGE_CHECK(m); -#ifdef FIXME_4778297 - assert(m->object != kernel_object); -#endif - assert(m->phys_page != vm_page_guard_addr); -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); +#ifdef FIXME_4778297 + assert(m_object != kernel_object); #endif + assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + assert( !(m->vmp_absent && !m->vmp_unusual)); - if (m->absent && !m->unusual) - panic("vm_page_activate: %p absent", m); - - if (m->gobbled) { + if (m->vmp_gobbled) { assert( !VM_PAGE_WIRED(m)); - if (!m->private && !m->fictitious) + if (!m->vmp_private && !m->vmp_fictitious) { vm_page_wire_count--; + } vm_page_gobble_count--; - m->gobbled = FALSE; + m->vmp_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->vmp_laundry || m->vmp_private || m->vmp_fictitious || + (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || + (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { return; + } #if DEBUG - if (m->active) - panic("vm_page_activate: already active"); + if (m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q) { + panic("vm_page_activate: already active"); + } #endif - if (m->speculative) { + if (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) { 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)) { - assert(!m->laundry); - assert(m->pageq.next == NULL && m->pageq.prev == NULL); - if (!IP_VALID(memory_manager_default) && - !m->fictitious && 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_queues_remove(m, FALSE); + + if (!VM_PAGE_WIRED(m)) { + vm_page_check_pageable_safe(m); + if (!VM_DYNAMIC_PAGING_ENABLED() && + m->vmp_dirty && m_object->internal && + (m_object->purgable == VM_PURGABLE_DENY || + m_object->purgable == VM_PURGABLE_NONVOLATILE || + m_object->purgable == VM_PURGABLE_VOLATILE)) { + vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq); + m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q; vm_page_throttled_count++; } else { - queue_enter(&vm_page_queue_active, m, vm_page_t, pageq); - m->active = TRUE; - if (!m->fictitious) - vm_page_active_count++; +#if CONFIG_SECLUDED_MEMORY + if (secluded_for_filecache && + vm_page_secluded_target != 0 && + num_tasks_can_use_secluded_mem == 0 && + m_object->eligible_for_secluded) { + vm_page_queue_enter(&vm_page_queue_secluded, m, vmp_pageq); + m->vmp_q_state = VM_PAGE_ON_SECLUDED_Q; + vm_page_secluded_count++; + vm_page_secluded_count_inuse++; + assert(!m_object->internal); +// vm_page_pageable_external_count++; + } else +#endif /* CONFIG_SECLUDED_MEMORY */ + vm_page_enqueue_active(m, FALSE); } - m->reference = TRUE; - m->no_cache = FALSE; + m->vmp_reference = TRUE; + m->vmp_no_cache = FALSE; } VM_PAGE_CHECK(m); } @@ -2918,78 +4708,95 @@ vm_page_activate( */ void vm_page_speculate( - vm_page_t m, - boolean_t new) + vm_page_t m, + boolean_t new) { - struct vm_speculative_age_q *aq; + struct vm_speculative_age_q *aq; + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(m); VM_PAGE_CHECK(m); - assert(m->object != kernel_object); - assert(m->phys_page != vm_page_guard_addr); -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); -#endif + vm_page_check_pageable_safe(m); + + assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + assert( !(m->vmp_absent && !m->vmp_unusual)); + assert(m_object->internal == FALSE); - if (m->absent && !m->unusual) - panic("vm_page_speculate: %p absent", m); + /* + * 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->vmp_laundry || m->vmp_private || m->vmp_fictitious || + (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || + (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { + return; + } - VM_PAGE_QUEUES_REMOVE(m); + vm_page_queues_remove(m, FALSE); - if ( !VM_PAGE_WIRED(m)) { - mach_timespec_t ts; + if (!VM_PAGE_WIRED(m)) { + mach_timespec_t ts; clock_sec_t sec; clock_nsec_t nsec; - clock_get_system_nanotime(&sec, &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; - + aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000; + aq->age_ts.tv_nsec = (vm_pageout_state.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++; - 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 > 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; + 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; + 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; + if (!vm_page_queue_empty(&aq->age_q)) { + vm_page_speculate_ageit(aq); + } + aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000; + aq->age_ts.tv_nsec = (vm_pageout_state.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_enqueue_tail(&aq->age_q, &m->vmp_pageq); + m->vmp_q_state = VM_PAGE_ON_SPECULATIVE_Q; vm_page_speculative_count++; + vm_page_pageable_external_count++; if (new == TRUE) { - m->object->pages_created++; + vm_object_lock_assert_exclusive(m_object); + + m_object->pages_created++; #if DEVELOPMENT || DEBUG vm_page_speculative_created++; #endif @@ -3008,95 +4815,127 @@ vm_page_speculate( void vm_page_speculate_ageit(struct vm_speculative_age_q *aq) { - struct vm_speculative_age_q *sq; - vm_page_t t; + 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; + if (vm_page_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; + t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.next); + t->vmp_pageq.prev = VM_PAGE_PACK_PTR(&sq->age_q); + + t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev); + t->vmp_pageq.next = VM_PAGE_PACK_PTR(&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)VM_PAGE_UNPACK_PTR(sq->age_q.prev); + t->vmp_pageq.next = aq->age_q.next; - t = (vm_page_t)aq->age_q.prev; - t->pageq.next = &sq->age_q; + t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.next); + t->vmp_pageq.prev = sq->age_q.prev; + + t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.prev); + t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q); sq->age_q.prev = aq->age_q.prev; } - queue_init(&aq->age_q); + vm_page_queue_init(&aq->age_q); } void vm_page_lru( - vm_page_t m) + vm_page_t m) { VM_PAGE_CHECK(m); - assert(m->object != kernel_object); - assert(m->phys_page != vm_page_guard_addr); + assert(VM_PAGE_OBJECT(m) != kernel_object); + assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); -#endif - if (m->active || m->reference) - return; + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); - if (m->private || (VM_PAGE_WIRED(m))) - return; + if (m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q) { + /* + * we don't need to do all the other work that + * vm_page_queues_remove and vm_page_enqueue_inactive + * bring along for the ride + */ + assert(!m->vmp_laundry); + assert(!m->vmp_private); + + m->vmp_no_cache = FALSE; - m->no_cache = FALSE; + vm_page_queue_remove(&vm_page_queue_inactive, m, vmp_pageq); + vm_page_queue_enter(&vm_page_queue_inactive, m, vmp_pageq); - VM_PAGE_QUEUES_REMOVE(m); + return; + } + /* + * 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->vmp_laundry || m->vmp_private || + (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || + (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) || + VM_PAGE_WIRED(m)) { + return; + } - assert(!m->laundry); - assert(m->pageq.next == NULL && m->pageq.prev == NULL); + m->vmp_no_cache = FALSE; - queue_enter(&vm_page_queue_inactive, m, vm_page_t, pageq); - m->inactive = TRUE; + vm_page_queues_remove(m, FALSE); - vm_page_inactive_count++; - token_new_pagecount++; + 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; + 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; + vm_object_t m_object; + + if (!VM_DYNAMIC_PAGING_ENABLED()) { + return; + } extra_active_count = 0; + extra_internal_count = 0; + extra_external_count = 0; vm_page_lock_queues(); - if (! queue_empty(&vm_page_queue_throttled)) { + if (!vm_page_queue_empty(&vm_page_queue_throttled)) { /* * Switch "throttled" pages to "active". */ - queue_iterate(&vm_page_queue_throttled, m, vm_page_t, pageq) { + vm_page_queue_iterate(&vm_page_queue_throttled, m, vmp_pageq) { VM_PAGE_CHECK(m); - assert(m->throttled); - assert(!m->active); - assert(!m->inactive); - assert(!m->speculative); - assert(!VM_PAGE_WIRED(m)); - if (!m->fictitious) { - extra_active_count++; - } - m->throttled = FALSE; - m->active = TRUE; + assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q); + + m_object = VM_PAGE_OBJECT(m); + + extra_active_count++; + if (m_object->internal) { + extra_internal_count++; + } else { + extra_external_count++; + } + + m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q; VM_PAGE_CHECK(m); +#if CONFIG_BACKGROUND_QUEUE + if (m->vmp_in_background) { + vm_page_add_to_backgroundq(m, FALSE); + } +#endif } /* @@ -3105,30 +4944,32 @@ vm_page_reactivate_all_throttled(void) * 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; + first_throttled = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled); + last_throttled = (vm_page_t) vm_page_queue_last(&vm_page_queue_throttled); + first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); + if (vm_page_queue_empty(&vm_page_queue_active)) { + vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled); } else { - queue_prev(&first_active->pageq) = (queue_entry_t) last_throttled; + first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(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; + vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_throttled); + first_throttled->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active); + last_throttled->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active); #if DEBUG printf("reactivated %d throttled pages\n", vm_page_throttled_count); #endif - queue_init(&vm_page_queue_throttled); + vm_page_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)); + assert(vm_page_queue_empty(&vm_page_queue_throttled)); vm_page_unlock_queues(); } @@ -3143,23 +4984,26 @@ vm_page_reactivate_all_throttled(void) 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; + 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) + 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()) + if (!vm_page_trylockspin_queues()) { return; - } else + } + } else { vm_page_lockspin_queues(); + } VPL_LOCK(&lq->vpl_lock); } @@ -3167,58 +5011,65 @@ vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks) /* * Switch "local" pages to "active". */ - assert(!queue_empty(&lq->vpl_queue)); + assert(!vm_page_queue_empty(&lq->vpl_queue)); - queue_iterate(&lq->vpl_queue, m, vm_page_t, pageq) { + vm_page_queue_iterate(&lq->vpl_queue, m, vmp_pageq) { VM_PAGE_CHECK(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) + vm_page_check_pageable_safe(m); + assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q); + assert(!m->vmp_fictitious); + + if (m->vmp_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); + } + m->vmp_local_id = 0; + m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q; + VM_PAGE_CHECK(m); +#if CONFIG_BACKGROUND_QUEUE + if (m->vmp_in_background) { + vm_page_add_to_backgroundq(m, FALSE); + } +#endif count++; } - if (count != lq->vpl_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); + first_local = (vm_page_t) vm_page_queue_first(&lq->vpl_queue); + last_local = (vm_page_t) vm_page_queue_last(&lq->vpl_queue); + first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); - if (queue_empty(&vm_page_queue_active)) { - queue_last(&vm_page_queue_active) = (queue_entry_t) last_local; + if (vm_page_queue_empty(&vm_page_queue_active)) { + vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local); } else { - queue_prev(&first_active->pageq) = (queue_entry_t) last_local; + first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(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; + vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_local); + first_local->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active); + last_local->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active); - queue_init(&lq->vpl_queue); + vm_page_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)); + assert(vm_page_queue_empty(&lq->vpl_queue)); if (nolocks == FALSE) { VPL_UNLOCK(&lq->vpl_lock); + + vm_page_balance_inactive(count / 4); vm_page_unlock_queues(); } } @@ -3228,38 +5079,44 @@ vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks) * * 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 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); + pmap_zero_part_page(VM_PAGE_GET_PHYS_PAGE(m), m_pa, len); #else + vm_page_t tmp; while (1) { - tmp = vm_page_grab(); + tmp = vm_page_grab(); if (tmp == VM_PAGE_NULL) { vm_page_wait(THREAD_UNINT); continue; } - break; + break; } vm_page_zero_fill(tmp); - if(m_pa != 0) { + if (m_pa != 0) { vm_page_part_copy(m, 0, tmp, 0, m_pa); } - if((m_pa + len) < PAGE_SIZE) { - vm_page_part_copy(m, m_pa + len, tmp, - m_pa + len, PAGE_SIZE - (m_pa + len)); + if ((m_pa + len) < PAGE_SIZE) { + vm_page_part_copy(m, m_pa + len, tmp, + m_pa + len, PAGE_SIZE - (m_pa + len)); } - vm_page_copy(tmp,m); - VM_PAGE_FREE(tmp); + vm_page_copy(tmp, m); + VM_PAGE_FREE(tmp); #endif - } /* @@ -3269,16 +5126,21 @@ vm_page_part_zero_fill( */ void vm_page_zero_fill( - vm_page_t m) + vm_page_t m) { - XPR(XPR_VM_PAGE, - "vm_page_zero_fill, object 0x%X offset 0x%X page 0x%X\n", - m->object, m->offset, m, 0,0); - + XPR(XPR_VM_PAGE, + "vm_page_zero_fill, object 0x%X offset 0x%X page 0x%X\n", + VM_PAGE_OBJECT(m), m->vmp_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); +// dbgTrace(0xAEAEAEAE, VM_PAGE_GET_PHYS_PAGE(m), 0); /* (BRINGUP) */ + pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(m)); } /* @@ -3289,27 +5151,28 @@ vm_page_zero_fill( void vm_page_part_copy( - vm_page_t src_m, - vm_offset_t src_pa, - vm_page_t dst_m, - vm_offset_t dst_pa, - vm_size_t len) + vm_page_t src_m, + vm_offset_t src_pa, + vm_page_t dst_m, + 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); - - pmap_copy_part_page(src_m->phys_page, src_pa, - dst_m->phys_page, dst_pa, len); +#endif + pmap_copy_part_page(VM_PAGE_GET_PHYS_PAGE(src_m), src_pa, + VM_PAGE_GET_PHYS_PAGE(dst_m), dst_pa, len); } /* * vm_page_copy: * * Copy one page to another - * - * ENCRYPTED SWAP: - * The source page should not be encrypted. The caller should - * make sure the page is decrypted first, if necessary. */ int vm_page_copy_cs_validations = 0; @@ -3317,31 +5180,30 @@ int vm_page_copy_cs_tainted = 0; void vm_page_copy( - vm_page_t src_m, - vm_page_t dest_m) + vm_page_t src_m, + vm_page_t dest_m) { - XPR(XPR_VM_PAGE, - "vm_page_copy, object 0x%X offset 0x%X to object 0x%X offset 0x%X\n", - src_m->object, src_m->offset, - dest_m->object, dest_m->offset, - 0); + vm_object_t src_m_object; - VM_PAGE_CHECK(src_m); - VM_PAGE_CHECK(dest_m); + src_m_object = VM_PAGE_OBJECT(src_m); + XPR(XPR_VM_PAGE, + "vm_page_copy, object 0x%X offset 0x%X to object 0x%X offset 0x%X\n", + src_m_object, src_m->vmp_offset, + VM_PAGE_OBJECT(dest_m), dest_m->vmp_offset, + 0); +#if 0 /* - * ENCRYPTED SWAP: - * The source page should not be encrypted at this point. - * The destination page will therefore not contain encrypted - * data after the copy. + * we don't hold the page queue lock + * so this check isn't safe to make */ - if (src_m->encrypted) { - panic("vm_page_copy: source page %p is encrypted\n", src_m); - } - dest_m->encrypted = FALSE; + VM_PAGE_CHECK(src_m); + VM_PAGE_CHECK(dest_m); +#endif + vm_object_lock_assert_held(src_m_object); - if (src_m->object != VM_OBJECT_NULL && - src_m->object->code_signed) { + 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 @@ -3350,75 +5212,75 @@ vm_page_copy( */ vm_page_copy_cs_validations++; vm_page_validate_cs(src_m); +#if DEVELOPMENT || DEBUG + DTRACE_VM4(codesigned_copy, + vm_object_t, src_m_object, + vm_object_offset_t, src_m->vmp_offset, + int, src_m->vmp_cs_validated, + int, src_m->vmp_cs_tainted); +#endif /* DEVELOPMENT || DEBUG */ } + /* * 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) { + dest_m->vmp_cs_tainted = src_m->vmp_cs_tainted; + if (dest_m->vmp_cs_tainted) { vm_page_copy_cs_tainted++; } - - pmap_copy_page(src_m->phys_page, dest_m->phys_page); + dest_m->vmp_error = src_m->vmp_error; /* sliding src_m might have failed... */ + pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(src_m), VM_PAGE_GET_PHYS_PAGE(dest_m)); } #if MACH_ASSERT static void _vm_page_print( - vm_page_t p) + 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", p->next); - 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(" pageq: next=%p prev=%p\n", + (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.next), + (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.prev)); + printf(" listq: next=%p prev=%p\n", + (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.next)), + (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.prev))); + printf(" next=%p\n", (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m))); + printf(" object=%p offset=0x%llx\n", VM_PAGE_OBJECT(p), p->vmp_offset); + printf(" wire_count=%u\n", p->vmp_wire_count); + printf(" q_state=%u\n", p->vmp_q_state); + + printf(" %slaundry, %sref, %sgobbled, %sprivate\n", + (p->vmp_laundry ? "" : "!"), + (p->vmp_reference ? "" : "!"), + (p->vmp_gobbled ? "" : "!"), + (p->vmp_private ? "" : "!")); 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(" %slist_req_pending, %sdump_cleaning, %scs_validated, %scs_tainted, %sno_cache\n", - (p->list_req_pending ? "" : "!"), - (p->dump_cleaning ? "" : "!"), - (p->cs_validated ? "" : "!"), - (p->cs_tainted ? "" : "!"), - (p->no_cache ? "" : "!")); - printf(" %szero_fill\n", - (p->zero_fill ? "" : "!")); - - printf("phys_page=0x%x\n", p->phys_page); + (p->vmp_busy ? "" : "!"), + (p->vmp_wanted ? "" : "!"), + (p->vmp_tabled ? "" : "!"), + (p->vmp_fictitious ? "" : "!"), + (p->vmp_pmapped ? "" : "!"), + (p->vmp_wpmapped ? "" : "!")); + printf(" %sfree_when_done, %sabsent, %serror, %sdirty, %scleaning, %sprecious, %sclustered\n", + (p->vmp_free_when_done ? "" : "!"), + (p->vmp_absent ? "" : "!"), + (p->vmp_error ? "" : "!"), + (p->vmp_dirty ? "" : "!"), + (p->vmp_cleaning ? "" : "!"), + (p->vmp_precious ? "" : "!"), + (p->vmp_clustered ? "" : "!")); + printf(" %soverwriting, %srestart, %sunusual\n", + (p->vmp_overwriting ? "" : "!"), + (p->vmp_restart ? "" : "!"), + (p->vmp_unusual ? "" : "!")); + printf(" %scs_validated, %scs_tainted, %scs_nx, %sno_cache\n", + (p->vmp_cs_validated ? "" : "!"), + (p->vmp_cs_tainted ? "" : "!"), + (p->vmp_cs_nx ? "" : "!"), + (p->vmp_no_cache ? "" : "!")); + + printf("phys_page=0x%x\n", VM_PAGE_GET_PHYS_PAGE(p)); } /* @@ -3427,28 +5289,28 @@ _vm_page_print( */ static int vm_page_verify_contiguous( - vm_page_t pages, - unsigned int npages) + vm_page_t pages, + unsigned int npages) { - register vm_page_t m; - unsigned int page_count; - vm_offset_t prev_addr; + vm_page_t m; + unsigned int page_count; + vm_offset_t prev_addr; - prev_addr = pages->phys_page; + prev_addr = VM_PAGE_GET_PHYS_PAGE(pages); page_count = 1; for (m = NEXT_PAGE(pages); m != VM_PAGE_NULL; m = NEXT_PAGE(m)) { - if (m->phys_page != prev_addr + 1) { + if (VM_PAGE_GET_PHYS_PAGE(m) != prev_addr + 1) { 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\n", pages, page_count); + m, (long)prev_addr, VM_PAGE_GET_PHYS_PAGE(m)); + 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; + prev_addr = VM_PAGE_GET_PHYS_PAGE(m); ++page_count; } if (page_count != npages) { printf("pages %p actual count 0x%x but requested 0x%x\n", - pages, page_count, npages); + pages, page_count, npages); panic("vm_page_verify_contiguous: count error"); } return 1; @@ -3458,40 +5320,54 @@ vm_page_verify_contiguous( /* * Check the free lists for proper length etc. */ +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_queue_head_t *vm_page_queue, + unsigned int color, + vm_page_t look_for_page, + boolean_t expect_page) { - unsigned int npages; - vm_page_t m; - vm_page_t prev_m; - boolean_t found_page; + unsigned int npages; + vm_page_t m; + vm_page_t prev_m; + boolean_t found_page; + + if (!vm_page_verify_this_free_list_enabled) { + return 0; + } found_page = FALSE; npages = 0; - prev_m = (vm_page_t) vm_page_queue; - queue_iterate(vm_page_queue, - m, - vm_page_t, - pageq) { + prev_m = (vm_page_t)((uintptr_t)vm_page_queue); + + vm_page_queue_iterate(vm_page_queue, m, vmp_pageq) { if (m == look_for_page) { found_page = TRUE; } - if ((vm_page_t) m->pageq.prev != prev_m) + if ((vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_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->free ) - panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not free\n", - color, npages, m); - if ( ! m->busy ) + color, npages, m, (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev), prev_m); + } + if (!m->vmp_busy) { panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not busy\n", - color, npages, m); - if ( color != (unsigned int) -1 && (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); + color, npages, m); + } + if (color != (unsigned int) -1) { + if (VM_PAGE_GET_COLOR(m) != color) { + panic("vm_page_verify_free_list(color=%u, npages=%u): page %p wrong color %u instead of %u\n", + color, npages, m, VM_PAGE_GET_COLOR(m), color); + } + if (m->vmp_q_state != VM_PAGE_ON_FREE_Q) { + panic("vm_page_verify_free_list(color=%u, npages=%u): page %p - expecting q_state == VM_PAGE_ON_FREE_Q, found %d\n", + color, npages, m, m->vmp_q_state); + } + } else { + if (m->vmp_q_state != VM_PAGE_ON_FREE_LOCAL_Q) { + panic("vm_page_verify_free_list(npages=%u): local page %p - expecting q_state == VM_PAGE_ON_FREE_LOCAL_Q, found %d\n", + npages, m, m->vmp_q_state); + } + } ++npages; prev_m = m; } @@ -3500,78 +5376,83 @@ vm_page_verify_free_list( 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); + color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page)); _vm_page_print(look_for_page); for (other_color = 0; - other_color < vm_colors; - other_color++) { - if (other_color == color) + 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); + } + vm_page_verify_free_list(&vm_page_queue_free[other_color].qhead, + other_color, look_for_page, FALSE); } - if (color != (unsigned int) -1) { + if (color == (unsigned int) -1) { vm_page_verify_free_list(&vm_lopage_queue_free, - (unsigned int) -1, look_for_page, FALSE); + (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); + color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page)); } } return npages; } -static boolean_t vm_page_verify_free_lists_enabled = FALSE; +static boolean_t vm_page_verify_all_free_lists_enabled = FALSE; static void vm_page_verify_free_lists( void ) { - unsigned int color, npages, nlopages; + unsigned int color, npages, nlopages; + boolean_t toggle = TRUE; - if (! vm_page_verify_free_lists_enabled) + if (!vm_page_verify_all_free_lists_enabled) { return; + } npages = 0; lck_mtx_lock(&vm_page_queue_free_lock); - for( color = 0; color < vm_colors; color++ ) { - npages += vm_page_verify_free_list(&vm_page_queue_free[color], - color, VM_PAGE_NULL, FALSE); + 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].qhead, + 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) + (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); - lck_mtx_unlock(&vm_page_queue_free_lock); -} - -void -vm_page_queues_assert( - vm_page_t mem, - int val) -{ - 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); + "npages %u free_count %d nlopages %u lo_free_count %u", + npages, vm_page_free_count, nlopages, vm_lopage_free_count); } - if (VM_PAGE_WIRED(mem)) { - assert(!mem->active); - assert(!mem->inactive); - assert(!mem->speculative); - assert(!mem->throttled); + + if (toggle == TRUE) { + vm_page_verify_this_free_list_enabled = FALSE; } + + lck_mtx_unlock(&vm_page_queue_free_lock); } -#endif /* MACH_ASSERT */ +#endif /* MACH_ASSERT */ + + +extern boolean_t(*volatile consider_buffer_cache_collect)(int); /* * CONTIGUOUS PAGE ALLOCATION @@ -3582,27 +5463,27 @@ vm_page_queues_assert( * 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 + * 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 + * 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 + * 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 + * 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 + * 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 + * 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. @@ -3616,17 +5497,17 @@ vm_page_queues_assert( * Algorithm: */ -#define MAX_CONSIDERED_BEFORE_YIELD 1000 +#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 +#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 @@ -3634,40 +5515,45 @@ vm_page_queues_assert( */ #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; +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, + 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; + boolean_t wire, + int flags) +{ + vm_page_t m = NULL; + ppnum_t prevcontaddr = 0; + ppnum_t start_pnum = 0; + unsigned int npages = 0, considered = 0, scanned = 0; + unsigned int page_idx = 0, start_idx = 0, last_idx = 0, orig_last_idx = 0; + unsigned int idx_last_contig_page_found = 0; + int free_considered = 0, free_available = 0; + int substitute_needed = 0; + boolean_t wrapped, zone_gc_called = FALSE; + kern_return_t kr; #if DEBUG - clock_sec_t tv_start_sec, tv_end_sec; - clock_usec_t tv_start_usec, tv_end_usec; -#endif -#if MACH_ASSERT - int yielded = 0; - int dumped_run = 0; - int stolen_pages = 0; + clock_sec_t tv_start_sec = 0, tv_end_sec = 0; + clock_usec_t tv_start_usec = 0, tv_end_usec = 0; #endif - if (contig_pages == 0) + 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(); @@ -3675,6 +5561,15 @@ vm_page_find_contiguous( #if DEBUG clock_get_system_microtime(&tv_start_sec, &tv_start_usec); #endif + PAGE_REPLACEMENT_ALLOWED(TRUE); + + /* + * If there are still delayed pages, try to free up some that match. + */ + if (__improbable(vm_delayed_count != 0 && contig_pages != 0)) { + vm_free_delayed_pages_contig(contig_pages, max_pnum, pnum_mask); + } + vm_page_lock_queues(); lck_mtx_lock(&vm_page_queue_free_lock); @@ -3685,18 +5580,19 @@ vm_page_find_contiguous( free_available = vm_page_free_count - vm_page_free_reserved; wrapped = FALSE; - - if(flags & KMA_LOMEM) + + if (flags & KMA_LOMEM) { idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx; - else + } 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++) { + npages < contig_pages && page_idx < vm_pages_count; + page_idx++) { retry: if (wrapped && npages == 0 && @@ -3711,24 +5607,21 @@ retry: scanned++; m = &vm_pages[page_idx]; - assert(!m->fictitious); - assert(!m->private); + assert(!m->vmp_fictitious); + assert(!m->vmp_private); - if (max_pnum && m->phys_page > max_pnum) { + if (max_pnum && VM_PAGE_GET_PHYS_PAGE(m) > max_pnum) { /* no more low pages... */ break; } - if (!npages && ((m->phys_page & pnum_mask) != 0)) { + if (!npages & ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0)) { /* * not aligned */ RESET_STATE_OF_RUN(); - - } else if (VM_PAGE_WIRED(m) || m->gobbled || - m->encrypted || m->encrypted_cleaning || m->cs_validated || m->cs_tainted || - m->error || m->absent || m->pageout_queue || m->laundry || m->wanted || m->precious || - m->cleaning || m->overwriting || m->restart || m->unusual || m->list_req_pending || - m->pageout) { + } else if (VM_PAGE_WIRED(m) || m->vmp_gobbled || + m->vmp_laundry || m->vmp_wanted || + m->vmp_cleaning || m->vmp_overwriting || m->vmp_free_when_done) { /* * page is in a transient state * or a state we don't want to deal @@ -3736,18 +5629,22 @@ retry: * means starting a new run */ RESET_STATE_OF_RUN(); - - } else if (!m->free && !m->active && !m->inactive && !m->speculative && !m->throttled) { + } else if ((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || + (m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q) || + (m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q) || + (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { /* - * page needs to be on one of our queues + * page needs to be on one of our queues (other then the pageout or special free queues) + * or it needs to belong to the compressor pool (which is now indicated + * by vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR and falls out + * from the check for VM_PAGE_NOT_ON_Q) * 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)) { + } else if ((m->vmp_q_state != VM_PAGE_ON_FREE_Q) && (!m->vmp_tabled || m->vmp_busy)) { /* * pages on the free list are always 'busy' * so we couldn't test for 'busy' in the check @@ -3755,29 +5652,28 @@ retry: * '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... + * 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) { + if (VM_PAGE_GET_PHYS_PAGE(m) != prevcontaddr + 1) { + if ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0) { RESET_STATE_OF_RUN(); goto did_consider; } else { npages = 1; start_idx = page_idx; - start_pnum = m->phys_page; + start_pnum = VM_PAGE_GET_PHYS_PAGE(m); } } else { npages++; } - prevcontaddr = m->phys_page; - + prevcontaddr = VM_PAGE_GET_PHYS_PAGE(m); + VM_PAGE_CHECK(m); - if (m->free) { + if (m->vmp_q_state == VM_PAGE_ON_FREE_Q) { free_considered++; } else { /* @@ -3790,7 +5686,7 @@ retry: * into a substitute page. */ #if VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL - if (m->pmapped || m->dirty) { + if (m->vmp_pmapped || m->vmp_dirty || m->vmp_precious) { substitute_needed++; } #else @@ -3798,7 +5694,7 @@ retry: #endif } - if ((free_considered + substitute_needed) > free_available) { + if ((free_considered + substitute_needed) > free_available) { /* * if we let this run continue * we will end up dropping the vm_page_free_count @@ -3825,12 +5721,15 @@ retry: } 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); @@ -3841,9 +5740,9 @@ did_consider: */ free_available = vm_page_free_count - vm_page_free_reserved; considered = 0; -#if MACH_ASSERT + yielded++; -#endif + goto retry; } considered++; @@ -3858,10 +5757,11 @@ did_consider: * Start again from the very first page. */ RESET_STATE_OF_RUN(); - if( flags & KMA_LOMEM) + if (flags & KMA_LOMEM) { idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx = 0; - else + } else { idx_last_contig_page_found = vm_page_find_contiguous_last_idx = 0; + } last_idx = 0; page_idx = last_idx; wrapped = TRUE; @@ -3869,13 +5769,13 @@ did_consider: } 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; - + 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; @@ -3883,60 +5783,49 @@ did_consider: /* * 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 + * 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); + assert(m1->vmp_q_state == VM_PAGE_ON_FREE_Q); #endif - if (m1->free) { + if (m1->vmp_q_state == VM_PAGE_ON_FREE_Q) { unsigned int color; - color = m1->phys_page & vm_color_mask; + color = VM_PAGE_GET_COLOR(m1); #if MACH_ASSERT - vm_page_verify_free_list(&vm_page_queue_free[color], - color, m1, TRUE); + vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, m1, TRUE); #endif - queue_remove(&vm_page_queue_free[color], - m1, - vm_page_t, - pageq); - m1->pageq.next = NULL; - m1->pageq.prev = NULL; + vm_page_queue_remove(&vm_page_queue_free[color].qhead, m1, vmp_pageq); + + VM_PAGE_ZERO_PAGEQ_ENTRY(m1); #if MACH_ASSERT - vm_page_verify_free_list(&vm_page_queue_free[color], - color, VM_PAGE_NULL, FALSE); + vm_page_verify_free_list(&vm_page_queue_free[color].qhead, 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); + m1->vmp_q_state = VM_PAGE_NOT_ON_Q; + assert(m1->vmp_busy); vm_page_free_count--; } } - /* - * adjust global freelist counts - */ - if (vm_page_free_count < vm_page_free_count_minimum) - vm_page_free_count_minimum = vm_page_free_count; - - if( flags & KMA_LOMEM) + if (flags & KMA_LOMEM) { vm_page_lomem_find_contiguous_last_idx = page_idx; - else + } 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 @@ -3956,38 +5845,43 @@ did_consider: */ m1 = &vm_pages[cur_idx--]; - assert(!m1->free); - if (m1->object == VM_OBJECT_NULL) { + if (m1->vmp_object == 0) { /* * 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); + assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q); + assert(m1->vmp_offset == (vm_object_offset_t) -1); + assert(m1->vmp_busy); + assert(!m1->vmp_wanted); + assert(!m1->vmp_laundry); } else { vm_object_t object; + int refmod; + boolean_t disconnected, reusable; - if (abort_run == TRUE) + if (abort_run == TRUE) { continue; + } + + assert(m1->vmp_q_state != VM_PAGE_NOT_ON_Q); - object = m1->object; + object = VM_PAGE_OBJECT(m1); if (object != locked_object) { if (locked_object) { vm_object_unlock(locked_object); locked_object = VM_OBJECT_NULL; } - if (vm_object_lock_try(object)) + if (vm_object_lock_try(object)) { locked_object = object; + } } - if (locked_object == VM_OBJECT_NULL || - (VM_PAGE_WIRED(m1) || m1->gobbled || - m1->encrypted || m1->encrypted_cleaning || m1->cs_validated || m1->cs_tainted || - m1->error || m1->absent || m1->pageout_queue || m1->laundry || m1->wanted || m1->precious || - m1->cleaning || m1->overwriting || m1->restart || m1->unusual || m1->list_req_pending || m1->busy)) { - + if (locked_object == VM_OBJECT_NULL || + (VM_PAGE_WIRED(m1) || m1->vmp_gobbled || + m1->vmp_laundry || m1->vmp_wanted || + m1->vmp_cleaning || m1->vmp_overwriting || m1->vmp_free_when_done || m1->vmp_busy) || + (m1->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { if (locked_object) { vm_object_unlock(locked_object); locked_object = VM_OBJECT_NULL; @@ -3996,11 +5890,34 @@ did_consider: abort_run = TRUE; continue; } - if (m1->pmapped || m1->dirty) { - int refmod; + + disconnected = FALSE; + reusable = FALSE; + + if ((m1->vmp_reusable || + object->all_reusable) && + (m1->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q) && + !m1->vmp_dirty && + !m1->vmp_reference) { + /* reusable page... */ + refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1)); + disconnected = TRUE; + if (refmod == 0) { + /* + * ... not reused: can steal + * without relocating contents. + */ + reusable = TRUE; + } + } + + if ((m1->vmp_pmapped && + !reusable) || + m1->vmp_dirty || + m1->vmp_precious) { vm_object_offset_t offset; - m2 = vm_page_grab(); + m2 = vm_page_grab_options(VM_PAGE_GRAB_Q_LOCK_HELD); if (m2 == VM_PAGE_NULL) { if (locked_object) { @@ -4011,29 +5928,68 @@ did_consider: abort_run = TRUE; continue; } - if (m1->pmapped) - refmod = pmap_disconnect(m1->phys_page); - else - refmod = 0; - vm_page_copy(m1, m2); - - m2->reference = m1->reference; - m2->dirty = m1->dirty; - - if (refmod & VM_MEM_REFERENCED) - m2->reference = TRUE; - if (refmod & VM_MEM_MODIFIED) - m2->dirty = TRUE; - offset = m1->offset; + if (!disconnected) { + if (m1->vmp_pmapped) { + refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1)); + } else { + refmod = 0; + } + } + + /* copy the page's contents */ + pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(m1), VM_PAGE_GET_PHYS_PAGE(m2)); + /* copy the page's state */ + assert(!VM_PAGE_WIRED(m1)); + assert(m1->vmp_q_state != VM_PAGE_ON_FREE_Q); + assert(m1->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q); + assert(!m1->vmp_laundry); + m2->vmp_reference = m1->vmp_reference; + assert(!m1->vmp_gobbled); + assert(!m1->vmp_private); + m2->vmp_no_cache = m1->vmp_no_cache; + m2->vmp_xpmapped = 0; + assert(!m1->vmp_busy); + assert(!m1->vmp_wanted); + assert(!m1->vmp_fictitious); + m2->vmp_pmapped = m1->vmp_pmapped; /* should flush cache ? */ + m2->vmp_wpmapped = m1->vmp_wpmapped; + assert(!m1->vmp_free_when_done); + m2->vmp_absent = m1->vmp_absent; + m2->vmp_error = m1->vmp_error; + m2->vmp_dirty = m1->vmp_dirty; + assert(!m1->vmp_cleaning); + m2->vmp_precious = m1->vmp_precious; + m2->vmp_clustered = m1->vmp_clustered; + assert(!m1->vmp_overwriting); + m2->vmp_restart = m1->vmp_restart; + m2->vmp_unusual = m1->vmp_unusual; + m2->vmp_cs_validated = m1->vmp_cs_validated; + m2->vmp_cs_tainted = m1->vmp_cs_tainted; + m2->vmp_cs_nx = m1->vmp_cs_nx; /* - * 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 + * 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". */ - vm_page_free_prepare(m1); + // m2->vmp_reusable = m1->vmp_reusable; + assert(!m2->vmp_reusable); + + // assert(!m1->vmp_lopage); + + if (m1->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) { + m2->vmp_q_state = VM_PAGE_USED_BY_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->vmp_written_by_kernel = TRUE; /* * make sure we clear the ref/mod state @@ -4041,20 +5997,52 @@ did_consider: * inheriting state from the last time * this page was used... */ - pmap_clear_refmod(m2->phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED); + pmap_clear_refmod(VM_PAGE_GET_PHYS_PAGE(m2), VM_MEM_MODIFIED | VM_MEM_REFERENCED); + + if (refmod & VM_MEM_REFERENCED) { + m2->vmp_reference = TRUE; + } + if (refmod & VM_MEM_MODIFIED) { + SET_PAGE_DIRTY(m2, TRUE); + } + offset = m1->vmp_offset; + /* - * now put the substitute page on the object + * 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_insert_internal(m2, locked_object, offset, TRUE, TRUE); + vm_page_free_prepare(m1); - if (m2->reference) - vm_page_activate(m2); - else - vm_page_deactivate(m2); + /* + * 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); - PAGE_WAKEUP_DONE(m2); + if (m2->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) { + m2->vmp_pmapped = TRUE; + m2->vmp_wpmapped = TRUE; + PMAP_ENTER(kernel_pmap, m2->vmp_offset, m2, + VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, 0, TRUE, kr); + + assert(kr == KERN_SUCCESS); + + compressed_pages++; + } else { + if (m2->vmp_reference) { + vm_page_activate(m2); + } else { + vm_page_deactivate(m2); + } + } + PAGE_WAKEUP_DONE(m2); } else { + assert(m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR); + /* * completely cleans up the state * of the page so that it is ready @@ -4064,12 +6052,14 @@ did_consider: */ vm_page_free_prepare(m1); } -#if MACH_ASSERT + stolen_pages++; -#endif } - m1->pageq.next = (queue_entry_t) m; - m1->pageq.prev = NULL; +#if CONFIG_BACKGROUND_QUEUE + vm_page_assign_background_state(m1); +#endif + VM_PAGE_ZERO_PAGEQ_ENTRY(m1); + m1->vmp_snext = m; m = m1; } if (locked_object) { @@ -4078,12 +6068,6 @@ did_consider: } if (abort_run == TRUE) { - if (m != VM_PAGE_NULL) { - vm_page_free_list(m, FALSE); - } -#if MACH_ASSERT - dumped_run++; -#endif /* * want the index of the last * page in this run that was @@ -4093,54 +6077,78 @@ did_consider: */ page_idx = tmp_start_idx + 2; if (page_idx >= vm_pages_count) { - if (wrapped) + if (wrapped) { + if (m != VM_PAGE_NULL) { + vm_page_unlock_queues(); + vm_page_free_list(m, FALSE); + vm_page_lock_queues(); + m = VM_PAGE_NULL; + } + dumped_run++; 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) + + if (flags & KMA_LOMEM) { idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx = page_idx; - else + } else { idx_last_contig_page_found = vm_page_find_contiguous_last_idx = page_idx; - + } + last_idx = page_idx; - + + if (m != VM_PAGE_NULL) { + vm_page_unlock_queues(); + vm_page_free_list(m, FALSE); + vm_page_lock_queues(); + m = VM_PAGE_NULL; + } + dumped_run++; + lck_mtx_lock(&vm_page_queue_free_lock); /* - * reset our free page limit since we - * dropped the lock protecting the vm_page_free_queue - */ + * 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)) { + assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q); + assert(m1->vmp_wire_count == 0); - if (wire == TRUE) - m1->wire_count++; - else - m1->gobbled = TRUE; + if (wire == TRUE) { + m1->vmp_wire_count++; + m1->vmp_q_state = VM_PAGE_IS_WIRED; + } else { + m1->vmp_gobbled = TRUE; + } } - if (wire == FALSE) + 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)); + assert(vm_page_verify_contiguous(m, npages)); } done_scanning: + PAGE_REPLACEMENT_ALLOWED(FALSE); + vm_page_unlock_queues(); #if DEBUG @@ -4157,16 +6165,33 @@ done_scanning: 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\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); + 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(FALSE); + + 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; } @@ -4175,18 +6200,19 @@ done_scanning: */ 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_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; + vm_page_t pages; + unsigned int npages; - if (size % page_size != 0) + if (size % PAGE_SIZE != 0) { return KERN_INVALID_ARGUMENT; + } npages = (unsigned int) (size / PAGE_SIZE); if (npages != size / PAGE_SIZE) { @@ -4201,34 +6227,18 @@ cpm_allocate( */ pages = vm_page_find_contiguous(npages, max_pnum, pnum_mask, wire, flags); - if (pages == VM_PAGE_NULL) + 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))) + if (vm_page_free_count < vm_page_free_min) { thread_wakeup((event_t) &vm_page_free_wanted); - -#if CONFIG_EMBEDDED - { - int percent_avail; - - /* - * Decide if we need to poke the memorystatus notification thread. - */ - percent_avail = - (vm_page_active_count + vm_page_inactive_count + - vm_page_speculative_count + vm_page_free_count + - (IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 / - atop_64(max_mem); - if (percent_avail <= (kern_memorystatus_level - 5)) { - kern_memorystatus_level = percent_avail; - thread_wakeup((event_t)&kern_memorystatus_wakeup); } - } -#endif + + VM_CHECK_MEMORYSTATUS; + /* * The CPM pages should now be available and * ordered by ascending physical address. @@ -4238,83 +6248,251 @@ cpm_allocate( *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->vmp_pageq.next == 0 && m->vmp_pageq.prev == 0); + /* + * Add this page to our list of reclaimed pages, + * to be freed later. + */ + m->vmp_snext = 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->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) { + 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(VM_PAGE_GET_PHYS_PAGE(m)); + + 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_DEBUG(vm_pageout_enqueued_cleaned, 1); + VM_PAGEOUT_DEBUG(vm_pageout_cleaned_reactivated, 1); + + if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) { + vm_page_activate(m); + } + } else { + m->vmp_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->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) { + vm_page_queues_remove(m, TRUE); + } + } + if (dwp->dw_mask & DW_set_reference) { + m->vmp_reference = TRUE; + } else if (dwp->dw_mask & DW_clear_reference) { + m->vmp_reference = FALSE; + } + + if (dwp->dw_mask & DW_move_page) { + if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) { + vm_page_queues_remove(m, FALSE); + + assert(VM_PAGE_OBJECT(m) != kernel_object); + + vm_page_enqueue_inactive(m, FALSE); + } + } + if (dwp->dw_mask & DW_clear_busy) { + m->vmp_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, + int page_count, + int flags, vm_page_t *list) { - vm_page_t lo_page_list = VM_PAGE_NULL; - vm_page_t mem; - int i; + vm_page_t lo_page_list = VM_PAGE_NULL; + vm_page_t mem; + int i; - if ( !(flags & KMA_LOMEM)) + 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) + if (lo_page_list) { vm_page_free_list(lo_page_list, FALSE); + } *list = VM_PAGE_NULL; - return (KERN_RESOURCE_SHORTAGE); + return KERN_RESOURCE_SHORTAGE; } - mem->pageq.next = (queue_entry_t) lo_page_list; + mem->vmp_snext = lo_page_list; lo_page_list = mem; } *list = lo_page_list; - return (KERN_SUCCESS); + return KERN_SUCCESS; } void vm_page_set_offset(vm_page_t page, vm_object_offset_t offset) { - page->offset = offset; + page->vmp_offset = offset; } vm_page_t vm_page_get_next(vm_page_t page) { - return ((vm_page_t) page->pageq.next); + return page->vmp_snext; } vm_object_offset_t vm_page_get_offset(vm_page_t page) { - return (page->offset); + return page->vmp_offset; } ppnum_t vm_page_get_phys_page(vm_page_t page) { - return (page->phys_page); + return VM_PAGE_GET_PHYS_PAGE(page); } - - + + /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ #if HIBERNATION static vm_page_t hibernate_gobble_queue; -extern boolean_t (* volatile consider_buffer_cache_collect)(int); - static int hibernate_drain_pageout_queue(struct vm_pageout_queue *); -static int hibernate_flush_dirty_pages(void); -static int hibernate_flush_queue(queue_head_t *, int); -static void hibernate_dirty_page(vm_page_t); +static int hibernate_flush_dirty_pages(int); +static int hibernate_flush_queue(vm_page_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; @@ -4323,6 +6501,7 @@ struct hibernate_statistics { 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; @@ -4337,36 +6516,50 @@ struct hibernate_statistics { 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; + wait_result_t wait_result; vm_page_lock_queues(); - while (q->pgo_laundry) { - + while (!vm_page_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); + 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) { + if (wait_result == THREAD_TIMED_OUT && !vm_page_queue_empty(&q->pgo_pending)) { hibernate_stats.hibernate_drain_timeout++; - return (1); + + if (q == &vm_pageout_queue_external) { + return 0; + } + + return 1; } vm_page_lock_queues(); @@ -4374,74 +6567,35 @@ hibernate_drain_pageout_queue(struct vm_pageout_queue *q) } vm_page_unlock_queues(); - return (0); + return 0; } -static void -hibernate_dirty_page(vm_page_t m) -{ - vm_object_t object = m->object; - struct vm_pageout_queue *q; - -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); -#endif - vm_object_lock_assert_exclusive(object); - - /* - * protect the object from collapse - - * locking in the object's paging_offset. - */ - vm_object_paging_begin(object); - - m->list_req_pending = TRUE; - m->cleaning = TRUE; - m->busy = TRUE; - - if (object->internal == TRUE) - q = &vm_pageout_queue_internal; - else - q = &vm_pageout_queue_external; - - /* - * pgo_laundry count is tied to the laundry bit - */ - m->laundry = TRUE; - q->pgo_laundry++; - m->pageout_queue = TRUE; - queue_enter(&q->pgo_pending, m, vm_page_t, pageq); - - if (q->pgo_idle == TRUE) { - q->pgo_idle = FALSE; - thread_wakeup((event_t) &q->pgo_pending); - } -} +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); - +hibernate_flush_queue(vm_page_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; + + KDBG(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_START, + VM_KERNEL_UNSLIDE_OR_PERM(q), qcount); + iq = &vm_pageout_queue_internal; eq = &vm_pageout_queue_external; vm_page_lock_queues(); - while (qcount && !queue_empty(q)) { - + while (qcount && !vm_page_queue_empty(q)) { if (current_run++ == 1000) { if (hibernate_should_abort()) { retval = 1; @@ -4450,8 +6604,8 @@ hibernate_flush_queue(queue_head_t *q, int qcount) current_run = 0; } - m = (vm_page_t) queue_first(q); - m_object = m->object; + m = (vm_page_t) vm_page_queue_first(q); + m_object = VM_PAGE_OBJECT(m); /* * check to see if we currently are working @@ -4459,13 +6613,13 @@ hibernate_flush_queue(queue_head_t *q, int qcount) * already got the lock */ if (m_object != l_object) { - /* - * the object associated with candidate page is + /* + * 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); + if (l_object != NULL) { + vm_object_unlock(l_object); l_object = NULL; } /* @@ -4473,16 +6627,14 @@ hibernate_flush_queue(queue_head_t *q, int qcount) * 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... + * run... */ - if ( !vm_object_lock_try_scan(m_object)) { - + if (!vm_object_lock_try_scan(m_object)) { if (try_failed_count > 20) { hibernate_stats.hibernate_queue_nolock++; goto reenter_pg_on_q; } - vm_pageout_scan_wants_object = m_object; vm_page_unlock_queues(); mutex_pause(try_failed_count++); @@ -4492,24 +6644,21 @@ hibernate_flush_queue(queue_head_t *q, int qcount) continue; } else { l_object = m_object; - vm_pageout_scan_wants_object = VM_OBJECT_NULL; } } - if ( !m_object->alive || m->encrypted_cleaning || m->cleaning || m->busy || m->absent || m->error) { + if (!m_object->alive || m->vmp_cleaning || m->vmp_laundry || m->vmp_busy || m->vmp_absent || m->vmp_error) { /* * page is not to be cleaned * put it back on the head of its queue */ - if (m->cleaning) + if (m->vmp_cleaning) { hibernate_stats.hibernate_skipped_cleaning++; - else + } else { hibernate_stats.hibernate_skipped_transient++; + } goto reenter_pg_on_q; } - if ( !m_object->pager_initialized && m_object->pager_created) - 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) { /* @@ -4519,83 +6668,112 @@ hibernate_flush_queue(queue_head_t *q, int qcount) goto reenter_pg_on_q; } } - if ( !m->dirty && m->pmapped) { - refmod_state = pmap_get_refmod(m->phys_page); + if (!m->vmp_dirty && m->vmp_pmapped) { + refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m)); - if ((refmod_state & VM_MEM_MODIFIED)) - m->dirty = TRUE; - } else + if ((refmod_state & VM_MEM_MODIFIED)) { + SET_PAGE_DIRTY(m, FALSE); + } + } else { refmod_state = 0; + } - if ( !m->dirty) { + if (!m->vmp_dirty) { /* * page is not to be cleaned * put it back on the head of its queue */ - if (m->precious) + if (m->vmp_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)) + if (VM_PAGE_Q_THROTTLED(iq)) { tq = iq; - } else if (VM_PAGE_Q_THROTTLED(eq)) + } + } else if (VM_PAGE_Q_THROTTLED(eq)) { tq = eq; + } if (tq != NULL) { - wait_result_t wait_result; - int wait_count = 5; + wait_result_t wait_result; + int wait_count = 5; - if (l_object != NULL) { - vm_object_unlock(l_object); + if (l_object != NULL) { + vm_object_unlock(l_object); l_object = NULL; } - vm_pageout_scan_wants_object = VM_OBJECT_NULL; - - tq->pgo_throttled = TRUE; while (retval == 0) { + tq->pgo_throttled = TRUE; - assert_wait_timeout((event_t) &tq->pgo_laundry, THREAD_INTERRUPTIBLE, 1000, 1000*NSEC_PER_USEC); + assert_wait_timeout((event_t) &tq->pgo_laundry, THREAD_INTERRUPTIBLE, 1000, 1000 * NSEC_PER_USEC); - vm_page_unlock_queues(); + vm_page_unlock_queues(); - wait_result = thread_block(THREAD_CONTINUE_NULL); + wait_result = thread_block(THREAD_CONTINUE_NULL); vm_page_lock_queues(); - if (hibernate_should_abort()) + if (wait_result != THREAD_TIMED_OUT) { + break; + } + if (!VM_PAGE_Q_THROTTLED(tq)) { + break; + } + + if (hibernate_should_abort()) { retval = 1; + } - if (wait_result != THREAD_TIMED_OUT) - break; - if (--wait_count == 0) { - hibernate_stats.hibernate_throttle_timeout++; - retval = 1; - } + hibernate_stats.hibernate_throttle_timeout++; + + if (tq == eq) { + hibernate_skip_external = TRUE; + break; + } + retval = 1; + } } - if (retval) + if (retval) { break; + } hibernate_stats.hibernate_throttled++; continue; } - VM_PAGE_QUEUES_REMOVE(m); + /* + * 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 + */ + vm_page_queues_remove(m, TRUE); + + if (m_object->internal == TRUE) { + pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(m), PMAP_OPTIONS_COMPRESSOR, NULL); + } - hibernate_dirty_page(m); + vm_pageout_cluster(m); 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); + vm_page_queue_remove(q, m, vmp_pageq); + vm_page_queue_enter(q, m, vmp_pageq); hibernate_stats.hibernate_reentered_on_q++; next_pg: @@ -4608,720 +6786,2688 @@ next_pg: vm_object_unlock(l_object); l_object = NULL; } - vm_pageout_scan_wants_object = VM_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); + return retval; } static int -hibernate_flush_dirty_pages() +hibernate_flush_dirty_pages(int pass) { - struct vm_speculative_age_q *aq; - uint32_t i; - - bzero(&hibernate_stats, sizeof(struct hibernate_statistics)); + struct vm_speculative_age_q *aq; + uint32_t i; if (vm_page_local_q) { - for (i = 0; i < vm_page_local_q_count; i++) + 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; + int qcount; + vm_page_t m; aq = &vm_page_queue_speculative[i]; - if (queue_empty(&aq->age_q)) + if (vm_page_queue_empty(&aq->age_q)) { continue; + } qcount = 0; vm_page_lockspin_queues(); - queue_iterate(&aq->age_q, - m, - vm_page_t, - pageq) - { + vm_page_queue_iterate(&aq->age_q, m, vmp_pageq) { qcount++; } vm_page_unlock_queues(); if (qcount) { - if (hibernate_flush_queue(&aq->age_q, qcount)) - return (1); + 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; + } + /* XXX FBDP TODO: flush secluded queue */ + 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 (pass == 1) { + vm_compressor_record_warmup_start(); + } + + if (hibernate_flush_queue(&vm_page_queue_active, vm_page_active_count)) { + if (pass == 1) { + vm_compressor_record_warmup_end(); + } + return 1; + } + if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) { + if (pass == 1) { + vm_compressor_record_warmup_end(); } + return 1; + } + if (pass == 1) { + vm_compressor_record_warmup_end(); + } + + if (hibernate_skip_external == FALSE && hibernate_drain_pageout_queue(&vm_pageout_queue_external)) { + return 1; } - if (hibernate_flush_queue(&vm_page_queue_active, vm_page_active_count)) - return (1); - if (hibernate_flush_queue(&vm_page_queue_inactive, vm_page_inactive_count - vm_zf_queue_count)) - return (1); - if (hibernate_flush_queue(&vm_page_queue_zf, vm_zf_queue_count)) - return (1); - if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) - return (1); - return (hibernate_drain_pageout_queue(&vm_pageout_queue_external)); + return 0; } -extern void IOSleep(unsigned int); -extern int sync_internal(void); +void +hibernate_reset_stats() +{ + bzero(&hibernate_stats, sizeof(struct hibernate_statistics)); +} + int hibernate_flush_memory() { - int retval; + int retval; + + assert(VM_CONFIG_COMPRESSOR_IS_PRESENT); KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_START, vm_page_free_count, 0, 0, 0, 0); - IOSleep(2 * 1000); + hibernate_cleaning_in_progress = TRUE; + hibernate_skip_external = FALSE; + + if ((retval = hibernate_flush_dirty_pages(1)) == 0) { + 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, 3) | DBG_FUNC_NONE, vm_page_free_count, 0, 0, 0, 0); + KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_END, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0); - if ((retval = hibernate_flush_dirty_pages()) == 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; - KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_START, vm_page_wire_count, 0, 0, 0, 0); - - sync_internal(); (void)(*consider_buffer_cache_collect)(1); - consider_zone_gc(1); + consider_zone_gc(FALSE); + + 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, vm_page_wire_count, 0, 0, 0, 0); + KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_END, orig_wire_count - vm_page_wire_count, 0, 0, 0, 0); } } - KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_END, vm_page_free_count, hibernate_stats.hibernate_found_dirty, retval, 0, 0); - - 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) queue_nolock(%d)\n", - hibernate_stats.hibernate_skipped_cleaning, - hibernate_stats.hibernate_skipped_transient, - hibernate_stats.hibernate_skipped_precious, - 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); -} + hibernate_cleaning_in_progress = FALSE; -static void -hibernate_page_list_zero(hibernate_page_list_t *list) -{ - uint32_t bank; - hibernate_bitmap_t * bitmap; + KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_END, vm_page_free_count, hibernate_stats.hibernate_found_dirty, retval, 0, 0); - bitmap = &list->bank_bitmap[0]; - for (bank = 0; bank < list->bank_count; bank++) - { - uint32_t last_bit; + if (retval) { + HIBLOG("hibernate_flush_memory() failed to finish - vm_page_compressor_count(%d)\n", VM_PAGE_COMPRESSOR_COUNT); + } - 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]; - } + 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; } -void -hibernate_gobble_pages(uint32_t gobble_count, uint32_t free_page_time) + +static void +hibernate_page_list_zero(hibernate_page_list_t *list) { - uint32_t i; - vm_page_t m; - uint64_t start, end, timeout, nsec; - clock_interval_to_deadline(free_page_time, 1000 * 1000 /*ms*/, &timeout); - clock_get_uptime(&start); + 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); + } - for (i = 0; i < gobble_count; i++) - { - while (VM_PAGE_NULL == (m = vm_page_grab())) - { - clock_get_uptime(&end); - if (end >= timeout) - break; - VM_PAGE_WAIT(); + bitmap = (hibernate_bitmap_t *) &bitmap->bitmap[bitmap->bitmapwords]; } - if (!m) - break; - m->busy = FALSE; - vm_page_gobble(m); - - m->pageq.next = (queue_entry_t) hibernate_gobble_queue; - hibernate_gobble_queue = m; - } - - clock_get_uptime(&end); - absolutetime_to_nanoseconds(end - start, &nsec); - HIBLOG("Gobbled %d pages, time: %qd ms\n", i, nsec / 1000000ULL); } void hibernate_free_gobble_pages(void) { - vm_page_t m, next; - uint32_t count = 0; + vm_page_t m, next; + uint32_t count = 0; + + m = (vm_page_t) hibernate_gobble_queue; + while (m) { + next = m->vmp_snext; + vm_page_free(m); + count++; + m = next; + } + hibernate_gobble_queue = VM_PAGE_NULL; - 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); + if (count) { + HIBLOG("Freed %d pages\n", count); + } } -static boolean_t -hibernate_consider_discard(vm_page_t m) +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)) { - hibernate_stats.cd_lock_failed++; - break; - } - object = m->object; + vm_object_t object = NULL; + int refmod_state; + boolean_t discard = FALSE; - if (VM_PAGE_WIRED(m)) { - hibernate_stats.cd_found_wired++; - break; - } - if (m->precious) { - hibernate_stats.cd_found_precious++; - break; - } - if (m->busy || !object->alive) { - /* - * Somebody is playing with this page. - */ - hibernate_stats.cd_found_busy++; - break; - } - if (m->absent || m->unusual || m->error) { - /* - * If it's unusual in anyway, ignore it - */ - hibernate_stats.cd_found_unusual++; - break; - } - if (m->cleaning) { - hibernate_stats.cd_found_cleaning++; - break; - } - if (m->laundry || m->list_req_pending) { - 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) - m->dirty = TRUE; - } - - /* - * 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) - hibernate_stats.cd_found_dirty++; - } - while (FALSE); + do{ + if (m->vmp_private) { + panic("hibernate_consider_discard: private"); + } - if (object) - vm_object_unlock(object); + object = VM_PAGE_OBJECT(m); - return (discard); + if (!vm_object_lock_try(object)) { + object = NULL; + if (!preflight) { + hibernate_stats.cd_lock_failed++; + } + break; + } + if (VM_PAGE_WIRED(m)) { + if (!preflight) { + hibernate_stats.cd_found_wired++; + } + break; + } + if (m->vmp_precious) { + if (!preflight) { + hibernate_stats.cd_found_precious++; + } + break; + } + if (m->vmp_busy || !object->alive) { + /* + * Somebody is playing with this page. + */ + if (!preflight) { + hibernate_stats.cd_found_busy++; + } + break; + } + if (m->vmp_absent || m->vmp_unusual || m->vmp_error) { + /* + * If it's unusual in anyway, ignore it + */ + if (!preflight) { + hibernate_stats.cd_found_unusual++; + } + break; + } + if (m->vmp_cleaning) { + if (!preflight) { + hibernate_stats.cd_found_cleaning++; + } + break; + } + if (m->vmp_laundry) { + if (!preflight) { + hibernate_stats.cd_found_laundry++; + } + break; + } + if (!m->vmp_dirty) { + refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m)); + + if (refmod_state & VM_MEM_REFERENCED) { + m->vmp_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->vmp_dirty) + || (VM_PURGABLE_VOLATILE == object->purgable) + || (VM_PURGABLE_EMPTY == object->purgable); + + + if (discard == FALSE) { + if (!preflight) { + hibernate_stats.cd_found_dirty++; + } + } else if (m->vmp_xpmapped && m->vmp_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) +{ + vm_object_t m_object; + + if (m->vmp_absent || m->vmp_unusual || m->vmp_error) { + /* + * If it's unusual in anyway, ignore + */ + return; + } + + m_object = VM_PAGE_OBJECT(m); + +#if MACH_ASSERT || DEBUG + 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->vmp_pmapped == TRUE) { + __unused int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); + } + + if (m->vmp_laundry) { + panic("hibernate_discard_page(%p) laundry", m); + } + if (m->vmp_private) { + panic("hibernate_discard_page(%p) private", m); + } + if (m->vmp_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); + } + vm_object_lock_assert_exclusive(m_object); + 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(m_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); + lck_mtx_lock(&vm_purgeable_queue_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_purgeable_queue_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) { + assert(hibernate_vm_locks_are_safe()); + 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); + } + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + + hibernation_vmqueues_inspection = TRUE; + + m = (vm_page_t) hibernate_gobble_queue; + while (m) { + pages--; + count_wire--; + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + m = m->vmp_snext; + } + + 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 = m->vmp_snext) { + assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q); + + pages--; + count_wire--; + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + + hibernate_stats.cd_local_free++; + hibernate_stats.cd_total_free++; + } + } + } + } + + for (i = 0; i < vm_colors; i++) { + vm_page_queue_iterate(&vm_page_queue_free[i].qhead, m, vmp_pageq) { + assert(m->vmp_q_state == VM_PAGE_ON_FREE_Q); + + pages--; + count_wire--; + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + + hibernate_stats.cd_total_free++; + } + } + } + + vm_page_queue_iterate(&vm_lopage_queue_free, m, vmp_pageq) { + assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q); + + pages--; + count_wire--; + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + + hibernate_stats.cd_total_free++; + } + } + + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled); + while (m && !vm_page_queue_end(&vm_page_queue_throttled, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q); + + next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + discard = FALSE; + if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) + && hibernate_consider_discard(m, preflight)) { + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + count_discard_inactive++; + discard = discard_all; + } else { + count_throttled++; + } + count_wire--; + if (!preflight) { + hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + + if (discard) { + hibernate_discard_page(m); + } + m = next; + } + + m = (vm_page_t)vm_page_queue_first(&vm_page_queue_anonymous); + while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q); + + next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + discard = FALSE; + if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) && + hibernate_consider_discard(m, preflight)) { + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (m->vmp_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, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (discard) { + hibernate_discard_page(m); + } + m = next; + } + + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned); + while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q); + + next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + discard = FALSE; + if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) && + hibernate_consider_discard(m, preflight)) { + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (m->vmp_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, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (discard) { + hibernate_discard_page(m); + } + m = next; + } + + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); + while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q); + + next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + discard = FALSE; + if ((kIOHibernateModeDiscardCleanActive & gIOHibernateMode) && + hibernate_consider_discard(m, preflight)) { + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (m->vmp_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, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (discard) { + hibernate_discard_page(m); + } + m = next; + } + + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive); + while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q); + + next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + discard = FALSE; + if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) && + hibernate_consider_discard(m, preflight)) { + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (m->vmp_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, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (discard) { + hibernate_discard_page(m); + } + m = next; + } + /* XXX FBDP TODO: secluded queue */ + + for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) { + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q); + while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) { + assertf(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q, + "Bad page: %p (0x%x:0x%x) on queue %d has state: %d (Discard: %d, Preflight: %d)", + m, m->vmp_pageq.next, m->vmp_pageq.prev, i, m->vmp_q_state, discard, preflight); + + next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + discard = FALSE; + if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) && + hibernate_consider_discard(m, preflight)) { + if (!preflight) { + hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + count_discard_speculative++; + discard = discard_all; + } else { + count_speculative++; + } + count_wire--; + if (!preflight) { + hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + if (discard) { + hibernate_discard_page(m); + } + m = next; + } + } + + vm_page_queue_iterate(&compressor_object->memq, m, vmp_listq) { + assert(m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR); + + count_compressor++; + count_wire--; + if (!preflight) { + hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); + } + } + + if (preflight == FALSE && discard_all == TRUE) { + KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_START); + + 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; + + KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_END); + } + + 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; + } + + hibernation_vmqueues_inspection = FALSE; + +#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) vm_page_queue_first(&vm_page_queue_anonymous); + while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q); + + next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) { + if (m->vmp_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) vm_page_queue_first(&vm_page_queue_speculative[i].age_q); + while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q); + + next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) { + count_discard_speculative++; + hibernate_discard_page(m); + } + m = next; + } + } + + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive); + while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q); + + next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) { + if (m->vmp_dirty) { + count_discard_purgeable++; + } else { + count_discard_inactive++; + } + hibernate_discard_page(m); + } + m = next; + } + /* XXX FBDP TODO: secluded queue */ + + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); + while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q); + + next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) { + if (m->vmp_dirty) { + count_discard_purgeable++; + } else { + count_discard_active++; + } + hibernate_discard_page(m); + } + m = next; + } + + m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned); + while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) { + assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q); + + next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); + if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) { + if (m->vmp_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; +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_PAGE_GET_PHYS_PAGE(&vm_pages[i]) != 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_PAGE_GET_PHYS_PAGE(&vm_pages[i]); + } + next_ppnum_in_run = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) + 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; + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(mem); + + assert(mem->vmp_hashed); + assert(m_object); + assert(mem->vmp_offset != (vm_object_offset_t) -1); + + /* + * Insert it into the object_object/offset hash table + */ + hash_id = vm_page_hash(m_object, mem->vmp_offset); + bucket = &vm_page_buckets[hash_id]; + + mem->vmp_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->vmp_lopage = FALSE; + mem->vmp_q_state = VM_PAGE_ON_FREE_Q; + + color = VM_PAGE_GET_COLOR(mem); +#if defined(__x86_64__) + vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem); +#else + vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq); +#endif + vm_page_free_count++; + + sindx++; + } +} + + +extern void hibernate_rebuild_pmap_structs(void); + +void +hibernate_rebuild_vm_structs(void) +{ + int i, cindx, sindx, eindx; + vm_page_t mem, tmem, mem_next; + AbsoluteTime startTime, endTime; + uint64_t nsec; + + if (hibernate_rebuild_needed == FALSE) { + return; + } + + KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_START); + HIBLOG("hibernate_rebuild started\n"); + + 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; + + /* + * Mark all the vm_pages[] that have not been initialized yet as being + * transient. This is needed to ensure that buddy page search is corrrect. + * Without this random data in these vm_pages[] can trip the buddy search + */ + for (i = hibernate_teardown_last_valid_compact_indx + 1; i < eindx; ++i) { + vm_pages[i].vmp_q_state = VM_PAGE_NOT_ON_Q; + } + + for (cindx = hibernate_teardown_last_valid_compact_indx; cindx >= 0; cindx--) { + mem = &vm_pages[cindx]; + assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q); + /* + * hibernate_teardown_vm_structs leaves the location where + * this vm_page_t must be located in "next". + */ + tmem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m)); + mem->vmp_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->vmp_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); + + eindx = sindx; + } + if (sindx) { + hibernate_free_range(0, sindx); + } + + assert(vm_page_free_count == hibernate_teardown_vm_page_free_count); + + /* + * 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_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m)); + + mem->vmp_next_m = 0; + hibernate_hash_insert_page(mem); + } + hibernate_rebuild_hash_list = NULL; + + 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; + + KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_END); +} + + +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) +{ + 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 (hibernate_should_abort()) { + return 0; + } + + hibernate_rebuild_needed = TRUE; + + 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); + + for (i = 0; i < vm_page_bucket_count; i++) { + bucket = &vm_page_buckets[i]; + + for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); mem != VM_PAGE_NULL; mem = mem_next) { + assert(mem->vmp_hashed); + + mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m)); + + if (mem < &vm_pages[0] || mem >= &vm_pages[vm_pages_count]) { + mem->vmp_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; + + hibernate_teardown_vm_page_free_count = vm_page_free_count; + + compact_target_indx = 0; + + for (i = 0; i < vm_pages_count; i++) { + mem = &vm_pages[i]; + + if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) { + unsigned int color; + + assert(mem->vmp_busy); + assert(!mem->vmp_lopage); + + color = VM_PAGE_GET_COLOR(mem); + + vm_page_queue_remove(&vm_page_queue_free[color].qhead, mem, vmp_pageq); + + VM_PAGE_ZERO_PAGEQ_ENTRY(mem); + + vm_page_free_count--; + + hibernate_teardown_found_free_pages++; + + if (vm_pages[compact_target_indx].vmp_q_state != VM_PAGE_ON_FREE_Q) { + compact_target_indx = i; + } + } else { + /* + * 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 + */ + mem->vmp_next_m = VM_PAGE_PACK_PTR(mem); + + if (vm_pages[compact_target_indx].vmp_q_state == VM_PAGE_ON_FREE_Q) { + /* + * 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->vmp_q_state = VM_PAGE_ON_FREE_Q; + + hibernate_teardown_last_valid_compact_indx = compact_target_indx; + compact_target_indx++; + } else { + hibernate_teardown_last_valid_compact_indx = i; + } + } + } + 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; + + hibernate_teardown_pmap_structs(&start_of_unneeded, &end_of_unneeded); + + 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); + + return mark_as_unneeded_pages; +} + + +#endif /* HIBERNATION */ + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ + +#include +#if MACH_VM_DEBUG + +#include +#include + +/* + * Routine: vm_page_info + * Purpose: + * Return information about the global VP table. + * Fills the buffer with as much information as possible + * and returns the desired size of the buffer. + * Conditions: + * Nothing locked. The caller should provide + * possibly-pageable memory. + */ + +unsigned int +vm_page_info( + hash_info_bucket_t *info, + unsigned int count) +{ + unsigned int i; + lck_spin_t *bucket_lock; + + if (vm_page_bucket_count < count) { + count = vm_page_bucket_count; + } + + for (i = 0; i < count; i++) { + vm_page_bucket_t *bucket = &vm_page_buckets[i]; + unsigned int bucket_count = 0; + vm_page_t m; + + bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK]; + lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket); + + for (m = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); + m != VM_PAGE_NULL; + m = (vm_page_t)(VM_PAGE_UNPACK_PTR(m->vmp_next_m))) { + bucket_count++; + } + + lck_spin_unlock(bucket_lock); + + /* don't touch pageable memory while holding locks */ + info[i].hib_count = bucket_count; + } + + return vm_page_bucket_count; +} +#endif /* MACH_VM_DEBUG */ + +#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++) { + vm_object_t p_object; + + bucket = &vm_page_buckets[i]; + if (!bucket->page_list) { + continue; + } + + bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK]; + lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket); + p = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); + + while (p != VM_PAGE_NULL) { + p_object = VM_PAGE_OBJECT(p); + + if (!p->vmp_hashed) { + panic("BUCKET_CHECK: page %p (%p,0x%llx) " + "hash %d in bucket %d at %p " + "is not hashed\n", + p, p_object, p->vmp_offset, + p_hash, i, bucket); + } + p_hash = vm_page_hash(p_object, p->vmp_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->vmp_offset, + p_hash); + } + p = (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m)); + } + lck_spin_unlock(bucket_lock); + } + +// printf("BUCKET_CHECK: checked buckets\n"); +} +#endif /* VM_PAGE_BUCKETS_CHECK */ + +/* + * '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. + */ + +#if CONFIG_BACKGROUND_QUEUE +void +vm_page_queues_remove(vm_page_t mem, boolean_t remove_from_backgroundq) +#else +void +vm_page_queues_remove(vm_page_t mem, boolean_t __unused remove_from_backgroundq) +#endif +{ + boolean_t was_pageable = TRUE; + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(mem); + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + + if (mem->vmp_q_state == VM_PAGE_NOT_ON_Q) { + assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); +#if CONFIG_BACKGROUND_QUEUE + if (remove_from_backgroundq == TRUE) { + vm_page_remove_from_backgroundq(mem); + } + if (mem->vmp_on_backgroundq) { + assert(mem->vmp_backgroundq.next != 0); + assert(mem->vmp_backgroundq.prev != 0); + } else { + assert(mem->vmp_backgroundq.next == 0); + assert(mem->vmp_backgroundq.prev == 0); + } +#endif /* CONFIG_BACKGROUND_QUEUE */ + return; + } + + if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) { + assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); +#if CONFIG_BACKGROUND_QUEUE + assert(mem->vmp_backgroundq.next == 0 && + mem->vmp_backgroundq.prev == 0 && + mem->vmp_on_backgroundq == FALSE); +#endif + return; + } + if (mem->vmp_q_state == VM_PAGE_IS_WIRED) { + /* + * might put these guys on a list for debugging purposes + * if we do, we'll need to remove this assert + */ + assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); +#if CONFIG_BACKGROUND_QUEUE + assert(mem->vmp_backgroundq.next == 0 && + mem->vmp_backgroundq.prev == 0 && + mem->vmp_on_backgroundq == FALSE); +#endif + return; + } + + assert(m_object != compressor_object); + assert(m_object != kernel_object); + assert(m_object != vm_submap_object); + assert(!mem->vmp_fictitious); + + switch (mem->vmp_q_state) { + case VM_PAGE_ON_ACTIVE_LOCAL_Q: + { + struct vpl *lq; + + lq = &vm_page_local_q[mem->vmp_local_id].vpl_un.vpl; + VPL_LOCK(&lq->vpl_lock); + vm_page_queue_remove(&lq->vpl_queue, mem, vmp_pageq); + mem->vmp_local_id = 0; + lq->vpl_count--; + if (m_object->internal) { + lq->vpl_internal_count--; + } else { + lq->vpl_external_count--; + } + VPL_UNLOCK(&lq->vpl_lock); + was_pageable = FALSE; + break; + } + case VM_PAGE_ON_ACTIVE_Q: + { + vm_page_queue_remove(&vm_page_queue_active, mem, vmp_pageq); + vm_page_active_count--; + break; + } + + case VM_PAGE_ON_INACTIVE_INTERNAL_Q: + { + assert(m_object->internal == TRUE); + + vm_page_inactive_count--; + vm_page_queue_remove(&vm_page_queue_anonymous, mem, vmp_pageq); + vm_page_anonymous_count--; + + vm_purgeable_q_advance_all(); + vm_page_balance_inactive(3); + break; + } + + case VM_PAGE_ON_INACTIVE_EXTERNAL_Q: + { + assert(m_object->internal == FALSE); + + vm_page_inactive_count--; + vm_page_queue_remove(&vm_page_queue_inactive, mem, vmp_pageq); + vm_purgeable_q_advance_all(); + vm_page_balance_inactive(3); + break; + } + + case VM_PAGE_ON_INACTIVE_CLEANED_Q: + { + assert(m_object->internal == FALSE); + + vm_page_inactive_count--; + vm_page_queue_remove(&vm_page_queue_cleaned, mem, vmp_pageq); + vm_page_cleaned_count--; + vm_page_balance_inactive(3); + break; + } + + case VM_PAGE_ON_THROTTLED_Q: + { + assert(m_object->internal == TRUE); + + vm_page_queue_remove(&vm_page_queue_throttled, mem, vmp_pageq); + vm_page_throttled_count--; + was_pageable = FALSE; + break; + } + + case VM_PAGE_ON_SPECULATIVE_Q: + { + assert(m_object->internal == FALSE); + + vm_page_remque(&mem->vmp_pageq); + vm_page_speculative_count--; + vm_page_balance_inactive(3); + break; + } + +#if CONFIG_SECLUDED_MEMORY + case VM_PAGE_ON_SECLUDED_Q: + { + vm_page_queue_remove(&vm_page_queue_secluded, mem, vmp_pageq); + vm_page_secluded_count--; + if (m_object == VM_OBJECT_NULL) { + vm_page_secluded_count_free--; + was_pageable = FALSE; + } else { + assert(!m_object->internal); + vm_page_secluded_count_inuse--; + was_pageable = FALSE; +// was_pageable = TRUE; + } + break; + } +#endif /* CONFIG_SECLUDED_MEMORY */ + + default: + { + /* + * if (mem->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) + * 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 + * + * we also don't expect to encounter VM_PAGE_ON_FREE_Q, VM_PAGE_ON_FREE_LOCAL_Q, VM_PAGE_ON_FREE_LOPAGE_Q + * or any of the undefined states + */ + panic("vm_page_queues_remove - bad page q_state (%p, %d)\n", mem, mem->vmp_q_state); + break; + } + } + VM_PAGE_ZERO_PAGEQ_ENTRY(mem); + mem->vmp_q_state = VM_PAGE_NOT_ON_Q; + +#if CONFIG_BACKGROUND_QUEUE + if (remove_from_backgroundq == TRUE) { + vm_page_remove_from_backgroundq(mem); + } +#endif + if (was_pageable) { + if (m_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 = VM_PAGE_OBJECT(page); + if (page == __object->memq_hint) { + vm_page_t __new_hint; + vm_page_queue_entry_t __qe; + __qe = (vm_page_queue_entry_t)vm_page_queue_next(&page->vmp_listq); + if (vm_page_queue_end(&__object->memq, __qe)) { + __qe = (vm_page_queue_entry_t)vm_page_queue_prev(&page->vmp_listq); + if (vm_page_queue_end(&__object->memq, __qe)) { + __qe = NULL; + } + } + __new_hint = (vm_page_t)((uintptr_t) __qe); + __object->memq_hint = __new_hint; + } + vm_page_queue_remove(&__object->memq, page, vmp_listq); +#if CONFIG_SECLUDED_MEMORY + if (__object->eligible_for_secluded) { + vm_page_secluded.eligible_for_secluded--; + } +#endif /* CONFIG_SECLUDED_MEMORY */ +} + +void +vm_page_enqueue_inactive(vm_page_t mem, boolean_t first) +{ + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(mem); + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + assert(!mem->vmp_fictitious); + assert(!mem->vmp_laundry); + assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); + vm_page_check_pageable_safe(mem); + + if (m_object->internal) { + mem->vmp_q_state = VM_PAGE_ON_INACTIVE_INTERNAL_Q; + + if (first == TRUE) { + vm_page_queue_enter_first(&vm_page_queue_anonymous, mem, vmp_pageq); + } else { + vm_page_queue_enter(&vm_page_queue_anonymous, mem, vmp_pageq); + } + + vm_page_anonymous_count++; + vm_page_pageable_internal_count++; + } else { + mem->vmp_q_state = VM_PAGE_ON_INACTIVE_EXTERNAL_Q; + + if (first == TRUE) { + vm_page_queue_enter_first(&vm_page_queue_inactive, mem, vmp_pageq); + } else { + vm_page_queue_enter(&vm_page_queue_inactive, mem, vmp_pageq); + } + + vm_page_pageable_external_count++; + } + vm_page_inactive_count++; + token_new_pagecount++; + +#if CONFIG_BACKGROUND_QUEUE + if (mem->vmp_in_background) { + vm_page_add_to_backgroundq(mem, FALSE); + } +#endif +} + +void +vm_page_enqueue_active(vm_page_t mem, boolean_t first) +{ + vm_object_t m_object; + + m_object = VM_PAGE_OBJECT(mem); + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); + assert(!mem->vmp_fictitious); + assert(!mem->vmp_laundry); + assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); + vm_page_check_pageable_safe(mem); + + mem->vmp_q_state = VM_PAGE_ON_ACTIVE_Q; + if (first == TRUE) { + vm_page_queue_enter_first(&vm_page_queue_active, mem, vmp_pageq); + } else { + vm_page_queue_enter(&vm_page_queue_active, mem, vmp_pageq); + } + vm_page_active_count++; + + if (m_object->internal) { + vm_page_pageable_internal_count++; + } else { + vm_page_pageable_external_count++; + } + +#if CONFIG_BACKGROUND_QUEUE + if (mem->vmp_in_background) { + vm_page_add_to_backgroundq(mem, FALSE); + } +#endif + vm_page_balance_inactive(3); +} + +/* + * Pages from special kernel objects shouldn't + * be placed on pageable queues. + */ +void +vm_page_check_pageable_safe(vm_page_t page) +{ + vm_object_t page_object; + + page_object = VM_PAGE_OBJECT(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 + +#define KA_SIZE(namelen, subtotalscount) \ + (sizeof(struct vm_allocation_site) + (namelen) + 1 + ((subtotalscount) * sizeof(struct vm_allocation_total))) + +#define KA_NAME(alloc) \ + ((char *)(&(alloc)->subtotals[(alloc->subtotalscount)])) + +#define KA_NAME_LEN(alloc) \ + (VM_TAG_NAME_LEN_MAX & (alloc->flags >> VM_TAG_NAME_LEN_SHIFT)) + +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; + kern_allocation_name_t name; + + cthread = current_thread(); + if (__improbable(cthread == NULL)) { + return VM_KERN_MEMORY_OSFMK; + } + + if ((name = thread_get_kernel_state(cthread)->allocation_name)) { + if (!name->tag) { + vm_tag_alloc(name); + } + return name->tag; + } + + 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_builtinkmod_text_end) && (retaddr >= vm_kernel_builtinkmod_text)) + || (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[VM_MAX_TAG_VALUE / 64]; + +void +vm_tag_alloc_locked(vm_allocation_site_t * site, vm_allocation_site_t ** releasesiteP) +{ + vm_tag_t tag; + uint64_t avail; + uint32_t idx; + vm_allocation_site_t * prev; + + 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 >= ARRAY_COUNT(free_tag_bits)) { + for (idx = 0; idx < ARRAY_COUNT(vm_allocation_sites); idx++) { + prev = vm_allocation_sites[idx]; + if (!prev) { + continue; + } + if (!KA_NAME_LEN(prev)) { + continue; + } + if (!prev->tag) { + continue; + } + if (prev->total) { + continue; + } + if (1 != prev->refcount) { + continue; + } + + assert(idx == prev->tag); + tag = idx; + prev->tag = VM_KERN_MEMORY_NONE; + *releasesiteP = prev; + break; + } + if (idx >= ARRAY_COUNT(vm_allocation_sites)) { + tag = VM_KERN_MEMORY_ANY; + } + break; + } + } + site->tag = tag; + + OSAddAtomic16(1, &site->refcount); + + if (VM_KERN_MEMORY_ANY != tag) { + vm_allocation_sites[tag] = site; + } + + if (tag > vm_allocation_tag_highest) { + vm_allocation_tag_highest = tag; + } +} + +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); + } + + for (tag = VM_KERN_MEMORY_ANY + 1; tag < VM_MAX_TAG_VALUE; tag++) { + vm_tag_free_locked(tag); + } } +vm_tag_t +vm_tag_alloc(vm_allocation_site_t * site) +{ + vm_tag_t tag; + vm_allocation_site_t * releasesite; -static void -hibernate_discard_page(vm_page_t m) + if (VM_TAG_BT & site->flags) { + tag = vm_tag_bt(); + if (VM_KERN_MEMORY_NONE != tag) { + return tag; + } + } + + if (!site->tag) { + releasesite = NULL; + lck_spin_lock(&vm_allocation_sites_lock); + vm_tag_alloc_locked(site, &releasesite); + lck_spin_unlock(&vm_allocation_sites_lock); + if (releasesite) { + kern_allocation_name_release(releasesite); + } + } + + return site->tag; +} + +void +vm_tag_update_size(vm_tag_t tag, int64_t delta) { - if (m->absent || m->unusual || m->error) - /* - * If it's unusual in anyway, ignore - */ - return; - - 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); - /* No need to lock page queue for token delete, hibernate_vm_unlock() - makes sure these locks are uncontended before sleep */ - vm_purgeable_token_delete_first(old_queue); - m->object->purgable = VM_PURGABLE_EMPTY; - } - - vm_page_free(m); + vm_allocation_site_t * allocation; + uint64_t prior; + + assert(VM_KERN_MEMORY_NONE != tag); + assert(tag < VM_MAX_TAG_VALUE); + + allocation = vm_allocation_sites[tag]; + assert(allocation); + + if (delta < 0) { + assertf(allocation->total >= ((uint64_t)-delta), "tag %d, site %p", tag, allocation); + } + prior = OSAddAtomic64(delta, &allocation->total); + +#if DEBUG || DEVELOPMENT + + uint64_t new, peak; + new = prior + delta; + do{ + peak = allocation->peak; + if (new <= peak) { + break; + } + }while (!OSCompareAndSwap64(peak, new, &allocation->peak)); + +#endif /* DEBUG || DEVELOPMENT */ + + if (tag < VM_KERN_MEMORY_FIRST_DYNAMIC) { + return; + } + + if (!prior && !allocation->tag) { + vm_tag_alloc(allocation); + } } -/* - 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 +kern_allocation_update_size(kern_allocation_name_t allocation, int64_t delta) +{ + uint64_t prior; + + if (delta < 0) { + assertf(allocation->total >= ((uint64_t)-delta), "name %p", allocation); + } + prior = OSAddAtomic64(delta, &allocation->total); + +#if DEBUG || DEVELOPMENT + + uint64_t new, peak; + new = prior + delta; + do{ + peak = allocation->peak; + if (new <= peak) { + break; + } + }while (!OSCompareAndSwap64(peak, new, &allocation->peak)); + +#endif /* DEBUG || DEVELOPMENT */ + + if (!prior && !allocation->tag) { + vm_tag_alloc(allocation); + } +} + +#if VM_MAX_TAG_ZONES void -hibernate_page_list_setall(hibernate_page_list_t * page_list, - hibernate_page_list_t * page_list_wired, - uint32_t * pagesOut) -{ - uint64_t start, end, nsec; - vm_page_t m; - uint32_t pages = page_list->page_count; - uint32_t count_zf = 0, count_throttled = 0; - uint32_t count_inactive = 0, count_active = 0, count_speculative = 0; - uint32_t count_wire = pages; - uint32_t count_discard_active = 0; - uint32_t count_discard_inactive = 0; - uint32_t count_discard_purgeable = 0; - uint32_t count_discard_speculative = 0; - uint32_t i; - uint32_t bank; - hibernate_bitmap_t * bitmap; - hibernate_bitmap_t * bitmap_wired; - - - HIBLOG("hibernate_page_list_setall start %p, %p\n", page_list, page_list_wired); - - KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_START, count_wire, 0, 0, 0, 0); - - clock_get_uptime(&start); - - hibernate_page_list_zero(page_list); - hibernate_page_list_zero(page_list_wired); - - 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, TRUE); - } - - m = (vm_page_t) hibernate_gobble_queue; - while(m) - { - pages--; - count_wire--; - 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; - } -#ifndef PPC - 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); +vm_allocation_zones_init(void) +{ + kern_return_t ret; + vm_offset_t addr; + vm_size_t size; + + size = VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *) + + 2 * VM_MAX_TAG_ZONES * sizeof(vm_allocation_zone_total_t); + + ret = kernel_memory_allocate(kernel_map, + &addr, round_page(size), 0, + KMA_ZERO, VM_KERN_MEMORY_DIAG); + assert(KERN_SUCCESS == ret); + + vm_allocation_zone_totals = (vm_allocation_zone_total_t **) addr; + addr += VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *); + + // prepopulate VM_KERN_MEMORY_DIAG & VM_KERN_MEMORY_KALLOC so allocations + // in vm_tag_update_zone_size() won't recurse + vm_allocation_zone_totals[VM_KERN_MEMORY_DIAG] = (vm_allocation_zone_total_t *) addr; + addr += VM_MAX_TAG_ZONES * sizeof(vm_allocation_zone_total_t); + vm_allocation_zone_totals[VM_KERN_MEMORY_KALLOC] = (vm_allocation_zone_total_t *) addr; +} + +void +vm_tag_will_update_zone(vm_tag_t tag, uint32_t zidx) +{ + vm_allocation_zone_total_t * zone; - hibernate_stats.cd_local_free++; - hibernate_stats.cd_total_free++; - } + assert(VM_KERN_MEMORY_NONE != tag); + assert(tag < VM_MAX_TAG_VALUE); + + if (zidx >= VM_MAX_TAG_ZONES) { + return; + } + + zone = vm_allocation_zone_totals[tag]; + if (!zone) { + zone = kalloc_tag(VM_MAX_TAG_ZONES * sizeof(*zone), VM_KERN_MEMORY_DIAG); + if (!zone) { + return; + } + bzero(zone, VM_MAX_TAG_ZONES * sizeof(*zone)); + if (!OSCompareAndSwapPtr(NULL, zone, &vm_allocation_zone_totals[tag])) { + kfree(zone, VM_MAX_TAG_ZONES * sizeof(*zone)); + } } - } -#endif - for( i = 0; i < vm_colors; i++ ) - { - queue_iterate(&vm_page_queue_free[i], - m, - vm_page_t, - pageq) - { - 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_total_free++; - } - } - - queue_iterate(&vm_lopage_queue_free, - m, - vm_page_t, - pageq) - { - 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_total_free++; - } - - queue_iterate( &vm_page_queue_throttled, - m, - vm_page_t, - pageq ) - { - if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) - && hibernate_consider_discard(m)) - { - hibernate_page_bitset(page_list, TRUE, m->phys_page); - count_discard_inactive++; - } - else - count_throttled++; - count_wire--; - hibernate_page_bitset(page_list_wired, TRUE, m->phys_page); - } - - queue_iterate( &vm_page_queue_zf, - m, - vm_page_t, - pageq ) - { - if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) - && hibernate_consider_discard(m)) - { - hibernate_page_bitset(page_list, TRUE, m->phys_page); - if (m->dirty) - count_discard_purgeable++; - else - count_discard_inactive++; - } - else - count_zf++; - count_wire--; - hibernate_page_bitset(page_list_wired, TRUE, m->phys_page); - } - - queue_iterate( &vm_page_queue_inactive, - m, - vm_page_t, - pageq ) - { - if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) - && hibernate_consider_discard(m)) - { - hibernate_page_bitset(page_list, TRUE, m->phys_page); - if (m->dirty) - count_discard_purgeable++; - else - count_discard_inactive++; - } - else - count_inactive++; - count_wire--; - hibernate_page_bitset(page_list_wired, TRUE, m->phys_page); - } - - for( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ ) - { - queue_iterate(&vm_page_queue_speculative[i].age_q, - m, - vm_page_t, - pageq) - { - if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) - && hibernate_consider_discard(m)) - { - hibernate_page_bitset(page_list, TRUE, m->phys_page); - count_discard_speculative++; - } - else - count_speculative++; - count_wire--; - hibernate_page_bitset(page_list_wired, TRUE, m->phys_page); - } - } - - queue_iterate( &vm_page_queue_active, - m, - vm_page_t, - pageq ) - { - if ((kIOHibernateModeDiscardCleanActive & gIOHibernateMode) - && hibernate_consider_discard(m)) - { - hibernate_page_bitset(page_list, TRUE, m->phys_page); - if (m->dirty) - count_discard_purgeable++; - else - count_discard_active++; - } - else - count_active++; - count_wire--; - hibernate_page_bitset(page_list_wired, TRUE, m->phys_page); - } - - // 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, &pages); - - hibernate_stats.cd_count_wire = count_wire; - hibernate_stats.cd_discarded = count_discard_active + count_discard_inactive + count_discard_purgeable + count_discard_speculative; - - 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, spec %d, zf %d, throt %d, could discard act %d inact %d purgeable %d spec %d\n", - pages, count_wire, count_active, count_inactive, count_speculative, count_zf, count_throttled, - count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative); - - *pagesOut = pages - count_discard_active - count_discard_inactive - count_discard_purgeable - count_discard_speculative; - - 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) +vm_tag_update_zone_size(vm_tag_t tag, uint32_t zidx, int64_t delta, int64_t dwaste) { - 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_speculative = 0; - - clock_get_uptime(&start); - - m = (vm_page_t) queue_first(&vm_page_queue_zf); - while (m && !queue_end(&vm_page_queue_zf, (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; - } - - 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\n", - nsec / 1000000ULL, - count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative); + vm_allocation_zone_total_t * zone; + uint32_t new; + + assert(VM_KERN_MEMORY_NONE != tag); + assert(tag < VM_MAX_TAG_VALUE); + + if (zidx >= VM_MAX_TAG_ZONES) { + return; + } + + zone = vm_allocation_zone_totals[tag]; + assert(zone); + zone += zidx; + + /* the zone is locked */ + if (delta < 0) { + assertf(zone->total >= ((uint64_t)-delta), "zidx %d, tag %d, %p", zidx, tag, zone); + zone->total += delta; + } else { + zone->total += delta; + if (zone->total > zone->peak) { + zone->peak = zone->total; + } + if (dwaste) { + new = zone->waste; + if (zone->wastediv < 65536) { + zone->wastediv++; + } else { + new -= (new >> 16); + } + __assert_only bool ov = os_add_overflow(new, dwaste, &new); + assert(!ov); + zone->waste = new; + } + } } -#endif /* HIBERNATION */ +#endif /* VM_MAX_TAG_ZONES */ -/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ +void +kern_allocation_update_subtotal(kern_allocation_name_t allocation, uint32_t subtag, int64_t delta) +{ + kern_allocation_name_t other; + struct vm_allocation_total * total; + uint32_t subidx; + + subidx = 0; + assert(VM_KERN_MEMORY_NONE != subtag); + for (; subidx < allocation->subtotalscount; subidx++) { + if (VM_KERN_MEMORY_NONE == allocation->subtotals[subidx].tag) { + allocation->subtotals[subidx].tag = subtag; + break; + } + if (subtag == allocation->subtotals[subidx].tag) { + break; + } + } + assert(subidx < allocation->subtotalscount); + if (subidx >= allocation->subtotalscount) { + return; + } -#include -#if MACH_VM_DEBUG + total = &allocation->subtotals[subidx]; + other = vm_allocation_sites[subtag]; + assert(other); -#include -#include + if (delta < 0) { + assertf(total->total >= ((uint64_t)-delta), "name %p", allocation); + OSAddAtomic64(delta, &total->total); + assertf(other->mapped >= ((uint64_t)-delta), "other %p", other); + OSAddAtomic64(delta, &other->mapped); + } else { + OSAddAtomic64(delta, &other->mapped); + OSAddAtomic64(delta, &total->total); + } +} -/* - * Routine: vm_page_info - * Purpose: - * Return information about the global VP table. - * Fills the buffer with as much information as possible - * and returns the desired size of the buffer. - * Conditions: - * Nothing locked. The caller should provide - * possibly-pageable memory. - */ +const char * +kern_allocation_get_name(kern_allocation_name_t allocation) +{ + return KA_NAME(allocation); +} -unsigned int -vm_page_info( - hash_info_bucket_t *info, - unsigned int count) +kern_allocation_name_t +kern_allocation_name_allocate(const char * name, uint32_t subtotalscount) { - unsigned int i; - lck_spin_t *bucket_lock; + uint32_t namelen; - if (vm_page_bucket_count < count) - count = vm_page_bucket_count; + namelen = (uint32_t) strnlen(name, MACH_MEMORY_INFO_NAME_MAX_LEN - 1); - for (i = 0; i < count; i++) { - vm_page_bucket_t *bucket = &vm_page_buckets[i]; - unsigned int bucket_count = 0; - vm_page_t m; + kern_allocation_name_t allocation; + allocation = kalloc(KA_SIZE(namelen, subtotalscount)); + bzero(allocation, KA_SIZE(namelen, subtotalscount)); - bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK]; - lck_spin_lock(bucket_lock); + allocation->refcount = 1; + allocation->subtotalscount = subtotalscount; + allocation->flags = (namelen << VM_TAG_NAME_LEN_SHIFT); + strlcpy(KA_NAME(allocation), name, namelen + 1); - for (m = bucket->pages; m != VM_PAGE_NULL; m = m->next) - bucket_count++; + return allocation; +} - lck_spin_unlock(bucket_lock); +void +kern_allocation_name_release(kern_allocation_name_t allocation) +{ + assert(allocation->refcount > 0); + if (1 == OSAddAtomic16(-1, &allocation->refcount)) { + kfree(allocation, KA_SIZE(KA_NAME_LEN(allocation), allocation->subtotalscount)); + } +} - /* don't touch pageable memory while holding locks */ - info[i].hib_count = bucket_count; +vm_tag_t +kern_allocation_name_get_vm_tag(kern_allocation_name_t allocation) +{ + return vm_tag_alloc(allocation); +} + +#if !VM_TAG_ACTIVE_UPDATE +static void +vm_page_count_object(mach_memory_info_t * info, unsigned int __unused num_info, vm_object_t object) +{ + if (!object->wired_page_count) { + return; + } + if (object != kernel_object) { + assert(object->wire_tag < num_info); + info[object->wire_tag].size += ptoa_64(object->wired_page_count); + } +} + +typedef void (*vm_page_iterate_proc)(mach_memory_info_t * info, + unsigned int num_info, vm_object_t object); + +static void +vm_page_iterate_purgeable_objects(mach_memory_info_t * info, unsigned int num_info, + 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(info, num_info, object); } +} - return vm_page_bucket_count; +static void +vm_page_iterate_objects(mach_memory_info_t * info, unsigned int num_info, + vm_page_iterate_proc proc) +{ + vm_object_t object; + + lck_spin_lock_grp(&vm_objects_wired_lock, &vm_page_lck_grp_bucket); + queue_iterate(&vm_objects_wired, + object, + vm_object_t, + wired_objq) + { + proc(info, num_info, object); + } + lck_spin_unlock(&vm_objects_wired_lock); +} +#endif /* ! VM_TAG_ACTIVE_UPDATE */ + +static uint64_t +process_account(mach_memory_info_t * info, unsigned int num_info, uint64_t zones_collectable_bytes, boolean_t iterated) +{ + size_t namelen; + unsigned int idx, count, nextinfo; + vm_allocation_site_t * site; + lck_spin_lock(&vm_allocation_sites_lock); + + for (idx = 0; idx <= vm_allocation_tag_highest; idx++) { + site = vm_allocation_sites[idx]; + if (!site) { + continue; + } + info[idx].mapped = site->mapped; + info[idx].tag = site->tag; + if (!iterated) { + info[idx].size = site->total; +#if DEBUG || DEVELOPMENT + info[idx].peak = site->peak; +#endif /* DEBUG || DEVELOPMENT */ + } else { + if (!site->subtotalscount && (site->total != info[idx].size)) { + printf("tag mismatch[%d] 0x%qx, iter 0x%qx\n", idx, site->total, info[idx].size); + info[idx].size = site->total; + } + } + info[idx].flags |= VM_KERN_SITE_WIRED; + if (idx < VM_KERN_MEMORY_FIRST_DYNAMIC) { + info[idx].site = idx; + info[idx].flags |= VM_KERN_SITE_TAG; + if (VM_KERN_MEMORY_ZONE == idx) { + info[idx].flags |= VM_KERN_SITE_HIDE; + info[idx].flags &= ~VM_KERN_SITE_WIRED; + info[idx].collectable_bytes = zones_collectable_bytes; + } + } else if ((namelen = (VM_TAG_NAME_LEN_MAX & (site->flags >> VM_TAG_NAME_LEN_SHIFT)))) { + info[idx].site = 0; + info[idx].flags |= VM_KERN_SITE_NAMED; + if (namelen > sizeof(info[idx].name)) { + namelen = sizeof(info[idx].name); + } + strncpy(&info[idx].name[0], KA_NAME(site), namelen); + } else if (VM_TAG_KMOD & site->flags) { + info[idx].site = OSKextGetKmodIDForSite(site, NULL, 0); + info[idx].flags |= VM_KERN_SITE_KMOD; + } else { + info[idx].site = VM_KERNEL_UNSLIDE(site); + info[idx].flags |= VM_KERN_SITE_KERNEL; + } + } + + nextinfo = (vm_allocation_tag_highest + 1); + count = nextinfo; + if (count >= num_info) { + count = num_info; + } + + for (idx = 0; idx < count; idx++) { + site = vm_allocation_sites[idx]; + if (!site) { + continue; + } +#if VM_MAX_TAG_ZONES + vm_allocation_zone_total_t * zone; + unsigned int zidx; + vm_size_t elem_size; + + if (vm_allocation_zone_totals + && (zone = vm_allocation_zone_totals[idx]) + && (nextinfo < num_info)) { + for (zidx = 0; zidx < VM_MAX_TAG_ZONES; zidx++) { + if (!zone[zidx].peak) { + continue; + } + info[nextinfo] = info[idx]; + info[nextinfo].zone = zone_index_from_tag_index(zidx, &elem_size); + info[nextinfo].flags &= ~VM_KERN_SITE_WIRED; + info[nextinfo].flags |= VM_KERN_SITE_ZONE; + info[nextinfo].size = zone[zidx].total; + info[nextinfo].peak = zone[zidx].peak; + info[nextinfo].mapped = 0; + if (zone[zidx].wastediv) { + info[nextinfo].collectable_bytes = ((zone[zidx].waste * zone[zidx].total / elem_size) / zone[zidx].wastediv); + } + nextinfo++; + } + } +#endif /* VM_MAX_TAG_ZONES */ + if (site->subtotalscount) { + uint64_t mapped, mapcost, take; + uint32_t sub; + vm_tag_t alloctag; + + info[idx].size = site->total; + mapped = info[idx].size; + info[idx].mapped = mapped; + mapcost = 0; + for (sub = 0; sub < site->subtotalscount; sub++) { + alloctag = site->subtotals[sub].tag; + assert(alloctag < num_info); + if (info[alloctag].name[0]) { + continue; + } + take = site->subtotals[sub].total; + if (take > info[alloctag].size) { + take = info[alloctag].size; + } + if (take > mapped) { + take = mapped; + } + info[alloctag].mapped -= take; + info[alloctag].size -= take; + mapped -= take; + mapcost += take; + } + info[idx].size = mapcost; + } + } + lck_spin_unlock(&vm_allocation_sites_lock); + + return 0; +} + +uint32_t +vm_page_diagnose_estimate(void) +{ + vm_allocation_site_t * site; + uint32_t count; + uint32_t idx; + + lck_spin_lock(&vm_allocation_sites_lock); + for (count = idx = 0; idx < VM_MAX_TAG_VALUE; idx++) { + site = vm_allocation_sites[idx]; + if (!site) { + continue; + } + count++; +#if VM_MAX_TAG_ZONES + if (vm_allocation_zone_totals) { + vm_allocation_zone_total_t * zone; + zone = vm_allocation_zone_totals[idx]; + if (!zone) { + continue; + } + for (uint32_t zidx = 0; zidx < VM_MAX_TAG_ZONES; zidx++) { + if (zone[zidx].peak) { + count++; + } + } + } +#endif + } + lck_spin_unlock(&vm_allocation_sites_lock); + + /* some slop for new tags created */ + count += 8; + count += VM_KERN_COUNTER_COUNT; + + return count; +} + + +kern_return_t +vm_page_diagnose(mach_memory_info_t * info, unsigned int num_info, uint64_t zones_collectable_bytes) +{ + uint64_t wired_size; + uint64_t wired_managed_size; + uint64_t wired_reserved_size; + uint64_t booter_size; + boolean_t iterate; + mach_memory_info_t * counts; + + bzero(info, num_info * sizeof(mach_memory_info_t)); + + if (!vm_page_wire_count_initial) { + return KERN_ABORTED; + } + +#if CONFIG_EMBEDDED + wired_size = ptoa_64(vm_page_wire_count); + wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count); +#else + 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); +#endif + wired_managed_size = ptoa_64(vm_page_wire_count - vm_page_wire_count_initial); + + booter_size = ml_get_booter_memory_size(); + wired_size += booter_size; + + assert(num_info >= VM_KERN_COUNTER_COUNT); + num_info -= VM_KERN_COUNTER_COUNT; + counts = &info[num_info]; + +#define SET_COUNT(xcount, xsize, xflags) \ + counts[xcount].tag = VM_MAX_TAG_VALUE + xcount; \ + counts[xcount].site = (xcount); \ + counts[xcount].size = (xsize); \ + counts[xcount].mapped = (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); + SET_COUNT(VM_KERN_COUNT_WIRED_BOOT, ptoa_64(vm_page_wire_count_on_boot), 0); + SET_COUNT(VM_KERN_COUNT_BOOT_STOLEN, booter_size, VM_KERN_SITE_WIRED); + +#define SET_MAP(xcount, xsize, xfree, xlargest) \ + counts[xcount].site = (xcount); \ + counts[xcount].size = (xsize); \ + counts[xcount].mapped = (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); + + iterate = !VM_TAG_ACTIVE_UPDATE; + if (iterate) { + 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; + +#if !VM_TAG_ACTIVE_UPDATE + vm_page_iterate_objects(info, num_info, &vm_page_count_object); +#endif /* ! VM_TAG_ACTIVE_UPDATE */ + + 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) != VM_KERN_MEMORY_NONE); + assert(VME_ALIAS(entry) < num_info); + info[VME_ALIAS(entry)].size += ptoa_64(count); + } + } + while (map && (entry == vm_map_last_entry(map))) { + vm_map_unlock(map); + if (!stackIdx) { + map = NULL; + } else { + --stackIdx; + map = maps[stackIdx]; + entry = entries[stackIdx]; + } + } + } + } + } + + process_account(info, num_info, zones_collectable_bytes, iterate); + + return KERN_SUCCESS; +} + +#if DEBUG || DEVELOPMENT + +kern_return_t +vm_kern_allocation_info(uintptr_t addr, vm_size_t * size, vm_tag_t * tag, vm_size_t * zone_size) +{ + kern_return_t ret; + vm_size_t zsize; + vm_map_t map; + vm_map_entry_t entry; + + zsize = zone_element_info((void *) addr, tag); + if (zsize) { + *zone_size = *size = zsize; + return KERN_SUCCESS; + } + + *zone_size = 0; + ret = KERN_INVALID_ADDRESS; + for (map = kernel_map; map;) { + vm_map_lock(map); + if (!vm_map_lookup_entry(map, addr, &entry)) { + break; + } + if (entry->is_sub_map) { + if (map != kernel_map) { + break; + } + map = VME_SUBMAP(entry); + continue; + } + if (entry->vme_start != addr) { + break; + } + *tag = VME_ALIAS(entry); + *size = (entry->vme_end - addr); + ret = KERN_SUCCESS; + break; + } + if (map != kernel_map) { + vm_map_unlock(map); + } + vm_map_unlock(kernel_map); + + return ret; } -#endif /* MACH_VM_DEBUG */ -#include -#if MACH_KDB +#endif /* DEBUG || DEVELOPMENT */ + +uint32_t +vm_tag_get_kext(vm_tag_t tag, char * name, vm_size_t namelen) +{ + vm_allocation_site_t * site; + uint32_t kmodId; + + kmodId = 0; + lck_spin_lock(&vm_allocation_sites_lock); + if ((site = vm_allocation_sites[tag])) { + if (VM_TAG_KMOD & site->flags) { + kmodId = OSKextGetKmodIDForSite(site, name, namelen); + } + } + lck_spin_unlock(&vm_allocation_sites_lock); + + return kmodId; +} -#include -#include -#define printf kdbprintf +#if CONFIG_SECLUDED_MEMORY /* - * Routine: vm_page_print [exported] + * Note that there's no locking around other accesses to vm_page_secluded_target. + * That should be OK, since these are the only place where it can be changed after + * initialization. Other users (like vm_pageout) may see the wrong value briefly, + * but will eventually get the correct value. This brief mismatch is OK as pageout + * and page freeing will auto-adjust the vm_page_secluded_count to match the target + * over time. */ +unsigned int vm_page_secluded_suppress_cnt = 0; +unsigned int vm_page_secluded_save_target; + + +lck_grp_attr_t secluded_suppress_slock_grp_attr; +lck_grp_t secluded_suppress_slock_grp; +lck_attr_t secluded_suppress_slock_attr; +lck_spin_t secluded_suppress_slock; + +void +secluded_suppression_init(void) +{ + lck_grp_attr_setdefault(&secluded_suppress_slock_grp_attr); + lck_grp_init(&secluded_suppress_slock_grp, + "secluded_suppress_slock", &secluded_suppress_slock_grp_attr); + lck_attr_setdefault(&secluded_suppress_slock_attr); + lck_spin_init(&secluded_suppress_slock, + &secluded_suppress_slock_grp, &secluded_suppress_slock_attr); +} + +void +start_secluded_suppression(task_t task) +{ + if (task->task_suppressed_secluded) { + return; + } + lck_spin_lock(&secluded_suppress_slock); + if (!task->task_suppressed_secluded && vm_page_secluded_suppress_cnt++ == 0) { + task->task_suppressed_secluded = TRUE; + vm_page_secluded_save_target = vm_page_secluded_target; + vm_page_secluded_target = 0; + } + lck_spin_unlock(&secluded_suppress_slock); +} + void -vm_page_print( - db_addr_t db_addr) -{ - vm_page_t p; - - p = (vm_page_t) (long) db_addr; - - iprintf("page 0x%x\n", p); - - db_indent += 2; - - iprintf("object=0x%x", p->object); - printf(", offset=0x%x", p->offset); - printf(", wire_count=%d", p->wire_count); - - iprintf("%slocal, %sinactive, %sactive, %sthrottled, %sgobbled, %slaundry, %sfree, %sref, %sencrypted\n", - (p->local ? "" : "!"), - (p->inactive ? "" : "!"), - (p->active ? "" : "!"), - (p->throttled ? "" : "!"), - (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("%soverwriting, %srestart, %sunusual\n", - (p->overwriting ? "" : "!"), - (p->restart ? "" : "!"), - (p->unusual ? "" : "!")); - - iprintf("phys_page=0x%x", p->phys_page); - - db_indent -= 2; -} -#endif /* MACH_KDB */ +stop_secluded_suppression(task_t task) +{ + lck_spin_lock(&secluded_suppress_slock); + if (task->task_suppressed_secluded && --vm_page_secluded_suppress_cnt == 0) { + task->task_suppressed_secluded = FALSE; + vm_page_secluded_target = vm_page_secluded_save_target; + } + lck_spin_unlock(&secluded_suppress_slock); +} + +#endif /* CONFIG_SECLUDED_MEMORY */