/*
- * Copyright (c) 2000-2009 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <debug.h>
#include <mach_pagemap.h>
#include <mach_cluster_stats.h>
-#include <advisory_pageout.h>
#include <mach/mach_types.h>
#include <mach/memory_object.h>
#include <vm/vm_shared_region.h>
#include <vm/vm_compressor.h>
+#if CONFIG_PHANTOM_CACHE
+#include <vm/vm_phantom_cache.h>
+#endif
/*
* ENCRYPTED SWAP:
*/
#include <libkern/OSDebug.h>
#endif
+extern void m_drain(void);
+
+#if VM_PRESSURE_EVENTS
+extern unsigned int memorystatus_available_pages;
+extern unsigned int memorystatus_available_pages_pressure;
+extern unsigned int memorystatus_available_pages_critical;
+extern unsigned int memorystatus_frozen_count;
+extern unsigned int memorystatus_suspended_count;
+
extern vm_pressure_level_t memorystatus_vm_pressure_level;
int memorystatus_purge_on_warning = 2;
int memorystatus_purge_on_urgent = 5;
int memorystatus_purge_on_critical = 8;
-#if VM_PRESSURE_EVENTS
void vm_pressure_response(void);
boolean_t vm_pressure_thread_running = FALSE;
extern void consider_vm_pressure_events(void);
-#endif
+
+#define MEMORYSTATUS_SUSPENDED_THRESHOLD 4
+#endif /* VM_PRESSURE_EVENTS */
+
boolean_t vm_pressure_changed = FALSE;
#ifndef VM_PAGEOUT_BURST_ACTIVE_THROTTLE /* maximum iterations of the active queue to move pages to inactive */
#endif /* VM_PAGEOUT_LAUNDRY_MAX */
#ifndef VM_PAGEOUT_BURST_WAIT
-#define VM_PAGEOUT_BURST_WAIT 30 /* milliseconds */
+#define VM_PAGEOUT_BURST_WAIT 10 /* milliseconds */
#endif /* VM_PAGEOUT_BURST_WAIT */
#ifndef VM_PAGEOUT_EMPTY_WAIT
#define VM_PAGE_FREE_MIN(free) (10 + (free) / 100)
#endif /* VM_PAGE_FREE_MIN */
-#define VM_PAGE_FREE_RESERVED_LIMIT 100
-#define VM_PAGE_FREE_MIN_LIMIT 1500
-#define VM_PAGE_FREE_TARGET_LIMIT 2000
-
+#define VM_PAGE_FREE_RESERVED_LIMIT 1700
+#define VM_PAGE_FREE_MIN_LIMIT 3500
+#define VM_PAGE_FREE_TARGET_LIMIT 4000
/*
* When vm_page_free_count falls below vm_page_free_reserved,
#if VM_PRESSURE_EVENTS
void vm_pressure_thread(void);
+
+boolean_t VM_PRESSURE_NORMAL_TO_WARNING(void);
+boolean_t VM_PRESSURE_WARNING_TO_CRITICAL(void);
+
+boolean_t VM_PRESSURE_WARNING_TO_NORMAL(void);
+boolean_t VM_PRESSURE_CRITICAL_TO_WARNING(void);
#endif
static void vm_pageout_garbage_collect(int);
static void vm_pageout_iothread_continue(struct vm_pageout_queue *);
*/
extern void vm_pageout_io_throttle(void);
-
-#if LATENCY_JETSAM
-boolean_t jlp_init = FALSE;
-uint64_t jlp_time = 0, jlp_current = 0;
-struct vm_page jetsam_latency_page[NUM_OF_JETSAM_LATENCY_TOKENS];
-unsigned int latency_jetsam_wakeup = 0;
-#endif /* LATENCY_JETSAM */
-
/*
* Page States: Used below to maintain the page state
* before it's removed from it's Q. This saved state
int page_prev_state = 0;
int cache_evict_throttle = 0;
uint32_t vm_pageout_inactive_external_forced_reactivate_limit = 0;
+ int force_purge = 0;
+
+#if VM_PRESSURE_EVENTS
vm_pressure_level_t pressure_level;
+#endif /* VM_PRESSURE_EVENTS */
VM_DEBUG_EVENT(vm_pageout_scan, VM_PAGEOUT_SCAN, DBG_FUNC_START,
vm_pageout_speculative_clean, vm_pageout_inactive_clean,
vm_pageout_inactive_dirty_internal, vm_pageout_inactive_dirty_external);
-#if LATENCY_JETSAM
- if (jlp_init == FALSE) {
- int i=0;
- vm_page_t jlp;
- for(; i < NUM_OF_JETSAM_LATENCY_TOKENS; i++) {
- jlp = &jetsam_latency_page[i];
- jlp->fictitious = TRUE;
- jlp->offset = 0;
-
- }
- jlp = &jetsam_latency_page[0];
- queue_enter(&vm_page_queue_active, jlp, vm_page_t, pageq);
- jlp->active = TRUE;
-
- jlp->offset = mach_absolute_time();
- jlp_time = jlp->offset;
- jlp_current++;
- jlp_init = TRUE;
- }
-#endif /* LATENCY_JETSAM */
-
flow_control.state = FCS_IDLE;
iq = &vm_pageout_queue_internal;
eq = &vm_pageout_queue_external;
DTRACE_VM2(scan, int, 1, (uint64_t *), NULL);
-#if LATENCY_JETSAM
- if (m->fictitious) {
- const uint32_t FREE_TARGET_MULTIPLIER = 2;
-
- uint64_t now = mach_absolute_time();
- uint64_t delta = now - m->offset;
- clock_sec_t jl_secs = 0;
- clock_usec_t jl_usecs = 0;
- boolean_t issue_jetsam = FALSE;
-
- absolutetime_to_microtime(delta, &jl_secs, &jl_usecs);
- jl_usecs += jl_secs * USEC_PER_SEC;
+ /*
+ * by not passing in a pmap_flush_context we will forgo any TLB flushing, local or otherwise...
+ *
+ * a TLB flush isn't really needed here since at worst we'll miss the reference bit being
+ * updated in the PTE if a remote processor still has this mapping cached in its TLB when the
+ * new reference happens. If no futher references happen on the page after that remote TLB flushes
+ * we'll see a clean, non-referenced page when it eventually gets pulled out of the inactive queue
+ * by pageout_scan, which is just fine since the last reference would have happened quite far
+ * in the past (TLB caches don't hang around for very long), and of course could just as easily
+ * have happened before we moved the page
+ */
+ pmap_clear_refmod_options(m->phys_page, VM_MEM_REFERENCED, PMAP_OPTIONS_NOFLUSH, (void *)NULL);
- /* Jetsam only if the token hasn't aged sufficiently and the free count is close to the target (avoiding spurious triggers) */
- if ((jl_usecs <= JETSAM_AGE_NOTIFY_CRITICAL) && (vm_page_free_count < (FREE_TARGET_MULTIPLIER * vm_page_free_target))) {
- issue_jetsam = TRUE;
- }
-
- VM_DEBUG_EVENT(vm_pageout_page_token, VM_PAGEOUT_PAGE_TOKEN, DBG_FUNC_NONE,
- vm_page_active_count, vm_page_inactive_count, vm_page_free_count, jl_usecs);
-
- m->offset = 0;
- queue_remove(&vm_page_queue_active, m, vm_page_t, pageq);
- queue_enter(&vm_page_queue_active, m, vm_page_t, pageq);
+ /*
+ * The page might be absent or busy,
+ * but vm_page_deactivate can handle that.
+ * FALSE indicates that we don't want a H/W clear reference
+ */
+ vm_page_deactivate_internal(m, FALSE);
- m->offset = now;
- jlp_time = now;
-
- if (issue_jetsam) {
- vm_page_unlock_queues();
+ if (delayed_unlock++ > delayed_unlock_limit) {
- if (local_freeq) {
- vm_page_free_list(local_freeq, TRUE);
- local_freeq = NULL;
- local_freed = 0;
- }
-
- VM_DEBUG_EVENT(vm_pageout_jetsam, VM_PAGEOUT_JETSAM, DBG_FUNC_START,
- vm_page_active_count, vm_page_inactive_count, vm_page_free_count, 0);
-
- assert_wait_timeout(&latency_jetsam_wakeup, THREAD_INTERRUPTIBLE, 10 /* msecs */, 1000*NSEC_PER_USEC);
- /* Kill the top process asynchronously */
- memorystatus_kill_on_VM_page_shortage(TRUE);
- thread_block(THREAD_CONTINUE_NULL);
+ if (local_freeq) {
+ vm_page_unlock_queues();
- VM_DEBUG_EVENT(vm_pageout_jetsam, VM_PAGEOUT_JETSAM, DBG_FUNC_END, 0, 0, 0, 0);
+ VM_DEBUG_EVENT(vm_pageout_freelist, VM_PAGEOUT_FREELIST, DBG_FUNC_START,
+ vm_page_free_count, local_freed, delayed_unlock_limit, 1);
+
+ vm_page_free_list(local_freeq, TRUE);
+
+ VM_DEBUG_EVENT(vm_pageout_freelist, VM_PAGEOUT_FREELIST, DBG_FUNC_END,
+ vm_page_free_count, 0, 0, 1);
+ local_freeq = NULL;
+ local_freed = 0;
vm_page_lock_queues();
+ } else {
+ lck_mtx_yield(&vm_page_queue_lock);
}
- } else {
-#endif /* LATENCY_JETSAM */
- /*
- * by not passing in a pmap_flush_context we will forgo any TLB flushing, local or otherwise...
- *
- * a TLB flush isn't really needed here since at worst we'll miss the reference bit being
- * updated in the PTE if a remote processor still has this mapping cached in its TLB when the
- * new reference happens. If no futher references happen on the page after that remote TLB flushes
- * we'll see a clean, non-referenced page when it eventually gets pulled out of the inactive queue
- * by pageout_scan, which is just fine since the last reference would have happened quite far
- * in the past (TLB caches don't hang around for very long), and of course could just as easily
- * have happened before we moved the page
- */
- pmap_clear_refmod_options(m->phys_page, VM_MEM_REFERENCED, PMAP_OPTIONS_NOFLUSH, (void *)NULL);
+
+ delayed_unlock = 1;
/*
- * The page might be absent or busy,
- * but vm_page_deactivate can handle that.
- * FALSE indicates that we don't want a H/W clear reference
+ * continue the while loop processing
+ * the active queue... need to hold
+ * the page queues lock
*/
- vm_page_deactivate_internal(m, FALSE);
-
- if (delayed_unlock++ > delayed_unlock_limit) {
-
- if (local_freeq) {
- vm_page_unlock_queues();
-
- VM_DEBUG_EVENT(vm_pageout_freelist, VM_PAGEOUT_FREELIST, DBG_FUNC_START,
- vm_page_free_count, local_freed, delayed_unlock_limit, 1);
-
- vm_page_free_list(local_freeq, TRUE);
-
- VM_DEBUG_EVENT(vm_pageout_freelist, VM_PAGEOUT_FREELIST, DBG_FUNC_END,
- vm_page_free_count, 0, 0, 1);
-
- local_freeq = NULL;
- local_freed = 0;
- vm_page_lock_queues();
- } else
- lck_mtx_yield(&vm_page_queue_lock);
-
- delayed_unlock = 1;
-
- /*
- * continue the while loop processing
- * the active queue... need to hold
- * the page queues lock
- */
- }
-#if LATENCY_JETSAM
}
-#endif /* LATENCY_JETSAM */
}
VM_DEBUG_EVENT(vm_pageout_balance, VM_PAGEOUT_BALANCE, DBG_FUNC_END,
* If the purge succeeds, go back to the top and reevalute
* the new memory situation.
*/
- pressure_level = memorystatus_vm_pressure_level;
+
assert (available_for_purge>=0);
+ force_purge = 0; /* no force-purging */
- if (available_for_purge
- || pressure_level > kVMPressureNormal
- ) {
- int force_purge;
-
- if (object != NULL) {
- vm_object_unlock(object);
- object = NULL;
- }
+#if VM_PRESSURE_EVENTS
+ pressure_level = memorystatus_vm_pressure_level;
- VM_DEBUG_EVENT(vm_pageout_purgeone, VM_PAGEOUT_PURGEONE, DBG_FUNC_START, vm_page_free_count, 0, 0, 0);
- memoryshot(VM_PAGEOUT_PURGEONE, DBG_FUNC_START);
+ if (pressure_level > kVMPressureNormal) {
- force_purge = 0; /* no force-purging */
if (pressure_level >= kVMPressureCritical) {
force_purge = memorystatus_purge_on_critical;
} else if (pressure_level >= kVMPressureUrgent) {
force_purge = memorystatus_purge_on_urgent;
} else if (pressure_level >= kVMPressureWarning) {
force_purge = memorystatus_purge_on_warning;
- } else {
- force_purge = 0;
}
- if (vm_purgeable_object_purge_one(force_purge)) {
+ }
+#endif /* VM_PRESSURE_EVENTS */
+
+ if (available_for_purge || force_purge) {
+
+ if (object != NULL) {
+ vm_object_unlock(object);
+ object = NULL;
+ }
+
+ memoryshot(VM_PAGEOUT_PURGEONE, DBG_FUNC_START);
+
+ VM_DEBUG_EVENT(vm_pageout_purgeone, VM_PAGEOUT_PURGEONE, DBG_FUNC_START, vm_page_free_count, 0, 0, 0);
+ if (vm_purgeable_object_purge_one(force_purge, C_DONT_BLOCK)) {
VM_DEBUG_EVENT(vm_pageout_purgeone, VM_PAGEOUT_PURGEONE, DBG_FUNC_END, vm_page_free_count, 0, 0, 0);
memoryshot(VM_PAGEOUT_PURGEONE, DBG_FUNC_END);
VM_DEBUG_EVENT(vm_pageout_purgeone, VM_PAGEOUT_PURGEONE, DBG_FUNC_END, 0, 0, 0, -1);
memoryshot(VM_PAGEOUT_PURGEONE, DBG_FUNC_END);
}
+
if (queue_empty(&sq->age_q) && vm_page_speculative_count) {
/*
* try to pull pages from the aging bins...
}
vm_pageout_scan_wants_object = VM_OBJECT_NULL;
+ vm_page_unlock_queues();
+
if (local_freeq) {
- vm_page_unlock_queues();
VM_DEBUG_EVENT(vm_pageout_freelist, VM_PAGEOUT_FREELIST, DBG_FUNC_START,
vm_page_free_count, local_freed, delayed_unlock_limit, 3);
local_freeq = NULL;
local_freed = 0;
- vm_page_lock_queues();
+ }
+ if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE)
+ vm_consider_waking_compactor_swapper();
- if (flow_control.state == FCS_DELAYED &&
- !VM_PAGE_Q_THROTTLED(iq)) {
- flow_control.state = FCS_IDLE;
- goto consider_inactive;
- }
+ vm_page_lock_queues();
+
+ if (flow_control.state == FCS_DELAYED &&
+ !VM_PAGE_Q_THROTTLED(iq)) {
+ flow_control.state = FCS_IDLE;
+ goto consider_inactive;
}
if (vm_page_free_count >= vm_page_free_target) {
* The most eligible pages are ones we paged in speculatively,
* but which have not yet been touched.
*/
- if (!queue_empty(&sq->age_q) ) {
+ if (!queue_empty(&sq->age_q) && force_anonymous == FALSE) {
m = (vm_page_t) queue_first(&sq->age_q);
page_prev_state = PAGE_STATE_SPECULATIVE;
anons_grabbed = 0;
}
- if (grab_anonymous == TRUE && vm_compression_available() == FALSE)
- grab_anonymous = FALSE;
-
if (grab_anonymous == FALSE || anons_grabbed >= ANONS_GRABBED_LIMIT || queue_empty(&vm_page_queue_anonymous)) {
if ( !queue_empty(&vm_page_queue_inactive) ) {
vm_page_lock_queues();
delayed_unlock = 1;
+ force_anonymous = FALSE;
+
if ((vm_page_inactive_count + vm_page_speculative_count) < vm_page_inactive_target)
goto Restart;
+ if (!queue_empty(&sq->age_q))
+ goto Restart;
+
panic("vm_pageout: no victim");
/* NOTREACHED */
if (page_prev_state != PAGE_STATE_SPECULATIVE)
vm_pageout_stats[vm_pageout_stat_now].reclaimed++;
+ inactive_burst_count = 0;
goto done_with_inactivepage;
}
/*
goto reclaim_page;
}
- if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ if (COMPRESSED_PAGER_IS_ACTIVE) {
/*
* With the VM compressor, the cost of
* reclaiming a page is much lower (no I/O),
VM_PAGEOUT_SCAN_HANDLE_REUSABLE_PAGE(m);
}
- if (m->reference && !m->no_cache) {
+ if (!m->no_cache &&
+ (m->reference ||
+ (m->xpmapped && !object->internal && (vm_page_xpmapped_external_count < (vm_page_external_count / 4))))) {
/*
* The page we pulled off the inactive list has
* been referenced. It is possible for other
*/
vm_page_activate(m);
VM_STAT_INCR(reactivations);
+ inactive_burst_count = 0;
}
if (page_prev_state == PAGE_STATE_CLEAN)
vm_pageout_scan_reclaimed_throttled++;
+ inactive_burst_count = 0;
goto done_with_inactivepage;
}
if (inactive_throttled == TRUE) {
queue_enter(&vm_page_queue_active, m, vm_page_t, pageq);
m->active = TRUE;
vm_page_active_count++;
- if (m->object->internal) {
- vm_page_pageable_internal_count++;
- } else {
- vm_page_pageable_external_count++;
- }
+ vm_page_pageable_external_count++;
vm_pageout_adjust_io_throttles(iq, eq, FALSE);
#else /* CONFIG_MEMORYSTATUS && CONFIG_JETSAM */
force_anonymous = TRUE;
#endif
+ inactive_burst_count = 0;
goto done_with_inactivepage;
} else {
if (page_prev_state == PAGE_STATE_SPECULATIVE)
*/
if (m->pmapped == TRUE) {
- if (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE || object->internal == FALSE) {
+ if (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE || object->internal == FALSE) {
+ /*
+ * Don't count this page as going into the compressor if any of these are true:
+ * 1) We have the dynamic pager i.e. no compressed pager
+ * 2) Freezer enabled device with a freezer file to hold the app data i.e. no compressed pager
+ * 3) Freezer enabled device with compressed pager backend (exclusive use) i.e. most of the VM system
+ (including vm_pageout_scan) has no knowledge of the compressor
+ * 4) This page belongs to a file and hence will not be sent into the compressor
+ */
+
refmod_state = pmap_disconnect_options(m->phys_page, 0, NULL);
} else {
refmod_state = pmap_disconnect_options(m->phys_page, PMAP_OPTIONS_COMPRESSOR, NULL);
else if (page_prev_state == PAGE_STATE_CLEAN)
vm_pageout_cleaned_reclaimed++;
- if (m->was_dirty) {
- /* page on clean queue used to be dirty; we should increment the vm_stat pageout count here */
- VM_STAT_INCR(pageouts);
- DTRACE_VM2(pgout, int, 1, (uint64_t *), NULL);
- }
vm_pageout_inactive_clean++;
}
/*
* OK, at this point we have found a page we are going to free.
*/
+#if CONFIG_PHANTOM_CACHE
+ if (!object->internal)
+ vm_phantom_cache_add_ghost(m);
+#endif
goto reclaim_page;
}
if (inactive_throttled == TRUE)
goto throttle_inactive;
-#if VM_PRESSURE_EVENTS
+#if VM_PRESSURE_EVENTS
+#if CONFIG_JETSAM
+
+ /*
+ * If Jetsam is enabled, then the sending
+ * of memory pressure notifications is handled
+ * from the same thread that takes care of high-water
+ * and other jetsams i.e. the memorystatus_thread.
+ */
+
+#else /* CONFIG_JETSAM */
+
vm_pressure_response();
+
+#endif /* CONFIG_JETSAM */
#endif /* VM_PRESSURE_EVENTS */
/*
done_with_inactivepage:
- inactive_burst_count = 0;
if (delayed_unlock++ > delayed_unlock_limit || try_failed == TRUE) {
+ boolean_t need_delay = TRUE;
if (object != NULL) {
vm_pageout_scan_wants_object = VM_OBJECT_NULL;
vm_object_unlock(object);
object = NULL;
}
+ vm_page_unlock_queues();
+
if (local_freeq) {
- vm_page_unlock_queues();
VM_DEBUG_EVENT(vm_pageout_freelist, VM_PAGEOUT_FREELIST, DBG_FUNC_START,
vm_page_free_count, local_freed, delayed_unlock_limit, 4);
local_freeq = NULL;
local_freed = 0;
- vm_page_lock_queues();
- } else
+ need_delay = FALSE;
+ }
+ if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ vm_consider_waking_compactor_swapper();
+ need_delay = FALSE;
+ }
+ vm_page_lock_queues();
+
+ if (need_delay == TRUE)
lck_mtx_yield(&vm_page_queue_lock);
delayed_unlock = 1;
}
vm_pageout_considered_page++;
- if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE)
- vm_consider_waking_compactor_swapper();
-
/*
* back to top of pageout scan loop
*/
int local_freed = 0;
int local_batch_size;
kern_return_t retval;
+ int compressed_count_delta;
KERNEL_DEBUG(0xe040000c | DBG_FUNC_END, 0, 0, 0, 0, 0);
} else
lck_mtx_unlock(&vm_page_queue_free_lock);
}
- retval = vm_compressor_pager_put(pager, m->offset + object->paging_offset, m->phys_page, &cq->current_chead, cq->scratch_buf);
+
+ assert(object->activity_in_progress > 0);
+
+ retval = vm_compressor_pager_put(
+ pager,
+ m->offset + object->paging_offset,
+ m->phys_page,
+ &cq->current_chead,
+ cq->scratch_buf,
+ &compressed_count_delta);
vm_object_lock(object);
+ assert(object->activity_in_progress > 0);
+
+ assert(m->object == object);
+
+ vm_compressor_pager_count(pager,
+ compressed_count_delta,
+ FALSE, /* shared_lock */
+ object);
+
m->laundry = FALSE;
m->pageout = FALSE;
if (retval == KERN_SUCCESS) {
+ /*
+ * If the object is purgeable, its owner's
+ * purgeable ledgers will be updated in
+ * vm_page_remove() but the page still
+ * contributes to the owner's memory footprint,
+ * so account for it as such.
+ */
+ if (object->purgable != VM_PURGABLE_DENY &&
+ object->vo_purgeable_owner != NULL) {
+ /* one more compressed purgeable page */
+ vm_purgeable_compressed_update(object,
+ +1);
+ }
vm_page_compressions_failing = FALSE;
}
}
-
extern boolean_t memorystatus_manual_testing_on;
extern unsigned int memorystatus_level;
-
#if VM_PRESSURE_EVENTS
+boolean_t vm_pressure_events_enabled = FALSE;
+
void
vm_pressure_response(void)
{
-
vm_pressure_level_t old_level = kVMPressureNormal;
int new_level = -1;
- uint64_t available_memory = (((uint64_t) AVAILABLE_NON_COMPRESSED_MEMORY) * 100);
+ uint64_t available_memory = 0;
+
+ if (vm_pressure_events_enabled == FALSE)
+ return;
+
+
+ available_memory = (((uint64_t) AVAILABLE_NON_COMPRESSED_MEMORY) * 100);
+
memorystatus_level = (unsigned int) (available_memory / atop_64(max_mem));
if (new_level != -1) {
memorystatus_vm_pressure_level = (vm_pressure_level_t) new_level;
- if (old_level != new_level) {
+ if ((memorystatus_vm_pressure_level != kVMPressureNormal) || (old_level != new_level)) {
if (vm_pressure_thread_running == FALSE) {
thread_wakeup(&vm_pressure_thread);
}
- thread_wakeup(&vm_pressure_changed);
+
+ if (old_level != new_level) {
+ thread_wakeup(&vm_pressure_changed);
+ }
}
}
mach_vm_pressure_level_monitor(__unused boolean_t wait_for_pressure, __unused unsigned int *pressure_level) {
#if !VM_PRESSURE_EVENTS
-
+
return KERN_FAILURE;
#else /* VM_PRESSURE_EVENTS */
#if VM_PRESSURE_EVENTS
void
vm_pressure_thread(void) {
- static boolean_t set_up_thread = FALSE;
+ static boolean_t thread_initialized = FALSE;
- if (set_up_thread) {
+ if (thread_initialized == TRUE) {
vm_pressure_thread_running = TRUE;
consider_vm_pressure_events();
vm_pressure_thread_running = FALSE;
}
- set_up_thread = TRUE;
+ thread_initialized = TRUE;
assert_wait((event_t) &vm_pressure_thread, THREAD_UNINT);
thread_block((thread_continue_t)vm_pressure_thread);
}
stack_collect();
consider_machine_collect();
+ m_drain();
do {
if (consider_buffer_cache_collect != NULL) {
}
+void vm_pageout_reinit_tuneables(void);
+
+void
+vm_pageout_reinit_tuneables(void)
+{
+ vm_page_filecache_min = (uint32_t) (max_mem / PAGE_SIZE) / 15;
+
+ if (vm_page_filecache_min < VM_PAGE_FILECACHE_MIN)
+ vm_page_filecache_min = VM_PAGE_FILECACHE_MIN;
+
+ vm_compressor_minorcompact_threshold_divisor = 18;
+ vm_compressor_majorcompact_threshold_divisor = 22;
+ vm_compressor_unthrottle_threshold_divisor = 32;
+}
+
+
#if VM_PAGE_BUCKETS_CHECK
#if VM_PAGE_FAKE_BUCKETS
extern vm_map_offset_t vm_page_fake_buckets_start, vm_page_fake_buckets_end;
#endif /* VM_PAGE_FAKE_BUCKETS */
#endif /* VM_PAGE_BUCKETS_CHECK */
+#define FBDP_TEST_COLLAPSE_COMPRESSOR 0
+#if FBDP_TEST_COLLAPSE_COMPRESSOR
+extern boolean_t vm_object_collapse_compressor_allowed;
+#include <IOKit/IOLib.h>
+#endif /* FBDP_TEST_COLLAPSE_COMPRESSOR */
+
+#define FBDP_TEST_WIRE_AND_EXTRACT 0
+#if FBDP_TEST_WIRE_AND_EXTRACT
+extern ledger_template_t task_ledger_template;
+#include <mach/mach_vm.h>
+extern ppnum_t vm_map_get_phys_page(vm_map_t map,
+ vm_offset_t offset);
+#endif /* FBDP_TEST_WIRE_AND_EXTRACT */
+
void
vm_pageout(void)
{
if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE)
vm_compressor_pager_init();
+#if VM_PRESSURE_EVENTS
+ vm_pressure_events_enabled = TRUE;
+#endif /* VM_PRESSURE_EVENTS */
+
+#if CONFIG_PHANTOM_CACHE
+ vm_phantom_cache_init();
+#endif
#if VM_PAGE_BUCKETS_CHECK
#if VM_PAGE_FAKE_BUCKETS
printf("**** DEBUG: protecting fake buckets [0x%llx:0x%llx]\n",
- vm_page_fake_buckets_start, vm_page_fake_buckets_end);
+ (uint64_t) vm_page_fake_buckets_start,
+ (uint64_t) vm_page_fake_buckets_end);
pmap_protect(kernel_pmap,
vm_page_fake_buckets_start,
vm_page_fake_buckets_end,
#endif /* VM_PAGE_FAKE_BUCKETS */
#endif /* VM_PAGE_BUCKETS_CHECK */
+#if VM_OBJECT_TRACKING
+ vm_object_tracking_init();
+#endif /* VM_OBJECT_TRACKING */
+
+
+#if FBDP_TEST_COLLAPSE_COMPRESSOR
+ vm_object_size_t backing_size, top_size;
+ vm_object_t backing_object, top_object;
+ vm_map_offset_t backing_offset, top_offset;
+ unsigned char *backing_address, *top_address;
+ kern_return_t kr;
+
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR:\n");
+
+ /* create backing object */
+ backing_size = 15 * PAGE_SIZE;
+ backing_object = vm_object_allocate(backing_size);
+ assert(backing_object != VM_OBJECT_NULL);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: created backing object %p\n",
+ backing_object);
+ /* map backing object */
+ backing_offset = 0;
+ kr = vm_map_enter(kernel_map, &backing_offset, backing_size, 0,
+ VM_FLAGS_ANYWHERE, backing_object, 0, FALSE,
+ VM_PROT_DEFAULT, VM_PROT_DEFAULT, VM_INHERIT_DEFAULT);
+ assert(kr == KERN_SUCCESS);
+ backing_address = (unsigned char *) backing_offset;
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "mapped backing object %p at 0x%llx\n",
+ backing_object, (uint64_t) backing_offset);
+ /* populate with pages to be compressed in backing object */
+ backing_address[0x1*PAGE_SIZE] = 0xB1;
+ backing_address[0x4*PAGE_SIZE] = 0xB4;
+ backing_address[0x7*PAGE_SIZE] = 0xB7;
+ backing_address[0xa*PAGE_SIZE] = 0xBA;
+ backing_address[0xd*PAGE_SIZE] = 0xBD;
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "populated pages to be compressed in "
+ "backing_object %p\n", backing_object);
+ /* compress backing object */
+ vm_object_pageout(backing_object);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: compressing backing_object %p\n",
+ backing_object);
+ /* wait for all the pages to be gone */
+ while (*(volatile int *)&backing_object->resident_page_count != 0)
+ IODelay(10);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: backing_object %p compressed\n",
+ backing_object);
+ /* populate with pages to be resident in backing object */
+ backing_address[0x0*PAGE_SIZE] = 0xB0;
+ backing_address[0x3*PAGE_SIZE] = 0xB3;
+ backing_address[0x6*PAGE_SIZE] = 0xB6;
+ backing_address[0x9*PAGE_SIZE] = 0xB9;
+ backing_address[0xc*PAGE_SIZE] = 0xBC;
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "populated pages to be resident in "
+ "backing_object %p\n", backing_object);
+ /* leave the other pages absent */
+ /* mess with the paging_offset of the backing_object */
+ assert(backing_object->paging_offset == 0);
+ backing_object->paging_offset = 0x3000;
+
+ /* create top object */
+ top_size = 9 * PAGE_SIZE;
+ top_object = vm_object_allocate(top_size);
+ assert(top_object != VM_OBJECT_NULL);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: created top object %p\n",
+ top_object);
+ /* map top object */
+ top_offset = 0;
+ kr = vm_map_enter(kernel_map, &top_offset, top_size, 0,
+ VM_FLAGS_ANYWHERE, top_object, 0, FALSE,
+ VM_PROT_DEFAULT, VM_PROT_DEFAULT, VM_INHERIT_DEFAULT);
+ assert(kr == KERN_SUCCESS);
+ top_address = (unsigned char *) top_offset;
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "mapped top object %p at 0x%llx\n",
+ top_object, (uint64_t) top_offset);
+ /* populate with pages to be compressed in top object */
+ top_address[0x3*PAGE_SIZE] = 0xA3;
+ top_address[0x4*PAGE_SIZE] = 0xA4;
+ top_address[0x5*PAGE_SIZE] = 0xA5;
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "populated pages to be compressed in "
+ "top_object %p\n", top_object);
+ /* compress top object */
+ vm_object_pageout(top_object);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: compressing top_object %p\n",
+ top_object);
+ /* wait for all the pages to be gone */
+ while (top_object->resident_page_count != 0);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: top_object %p compressed\n",
+ top_object);
+ /* populate with pages to be resident in top object */
+ top_address[0x0*PAGE_SIZE] = 0xA0;
+ top_address[0x1*PAGE_SIZE] = 0xA1;
+ top_address[0x2*PAGE_SIZE] = 0xA2;
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "populated pages to be resident in "
+ "top_object %p\n", top_object);
+ /* leave the other pages absent */
+
+ /* link the 2 objects */
+ vm_object_reference(backing_object);
+ top_object->shadow = backing_object;
+ top_object->vo_shadow_offset = 0x3000;
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: linked %p and %p\n",
+ top_object, backing_object);
+
+ /* unmap backing object */
+ vm_map_remove(kernel_map,
+ backing_offset,
+ backing_offset + backing_size,
+ 0);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "unmapped backing_object %p [0x%llx:0x%llx]\n",
+ backing_object,
+ (uint64_t) backing_offset,
+ (uint64_t) (backing_offset + backing_size));
+
+ /* collapse */
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: collapsing %p\n", top_object);
+ vm_object_lock(top_object);
+ vm_object_collapse(top_object, 0, FALSE);
+ vm_object_unlock(top_object);
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: collapsed %p\n", top_object);
+
+ /* did it work? */
+ if (top_object->shadow != VM_OBJECT_NULL) {
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: not collapsed\n");
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: FAIL\n");
+ if (vm_object_collapse_compressor_allowed) {
+ panic("FBDP_TEST_COLLAPSE_COMPRESSOR: FAIL\n");
+ }
+ } else {
+ /* check the contents of the mapping */
+ unsigned char expect[9] =
+ { 0xA0, 0xA1, 0xA2, /* resident in top */
+ 0xA3, 0xA4, 0xA5, /* compressed in top */
+ 0xB9, /* resident in backing + shadow_offset */
+ 0xBD, /* compressed in backing + shadow_offset + paging_offset */
+ 0x00 }; /* absent in both */
+ unsigned char actual[9];
+ unsigned int i, errors;
+
+ errors = 0;
+ for (i = 0; i < sizeof (actual); i++) {
+ actual[i] = (unsigned char) top_address[i*PAGE_SIZE];
+ if (actual[i] != expect[i]) {
+ errors++;
+ }
+ }
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: "
+ "actual [%x %x %x %x %x %x %x %x %x] "
+ "expect [%x %x %x %x %x %x %x %x %x] "
+ "%d errors\n",
+ actual[0], actual[1], actual[2], actual[3],
+ actual[4], actual[5], actual[6], actual[7],
+ actual[8],
+ expect[0], expect[1], expect[2], expect[3],
+ expect[4], expect[5], expect[6], expect[7],
+ expect[8],
+ errors);
+ if (errors) {
+ panic("FBDP_TEST_COLLAPSE_COMPRESSOR: FAIL\n");
+ } else {
+ printf("FBDP_TEST_COLLAPSE_COMPRESSOR: PASS\n");
+ }
+ }
+#endif /* FBDP_TEST_COLLAPSE_COMPRESSOR */
+
+#if FBDP_TEST_WIRE_AND_EXTRACT
+ ledger_t ledger;
+ vm_map_t user_map, wire_map;
+ mach_vm_address_t user_addr, wire_addr;
+ mach_vm_size_t user_size, wire_size;
+ mach_vm_offset_t cur_offset;
+ vm_prot_t cur_prot, max_prot;
+ ppnum_t user_ppnum, wire_ppnum;
+ kern_return_t kr;
+
+ ledger = ledger_instantiate(task_ledger_template,
+ LEDGER_CREATE_ACTIVE_ENTRIES);
+ user_map = vm_map_create(pmap_create(ledger, 0, TRUE),
+ 0x100000000ULL,
+ 0x200000000ULL,
+ TRUE);
+ wire_map = vm_map_create(NULL,
+ 0x100000000ULL,
+ 0x200000000ULL,
+ TRUE);
+ user_addr = 0;
+ user_size = 0x10000;
+ kr = mach_vm_allocate(user_map,
+ &user_addr,
+ user_size,
+ VM_FLAGS_ANYWHERE);
+ assert(kr == KERN_SUCCESS);
+ wire_addr = 0;
+ wire_size = user_size;
+ kr = mach_vm_remap(wire_map,
+ &wire_addr,
+ wire_size,
+ 0,
+ VM_FLAGS_ANYWHERE,
+ user_map,
+ user_addr,
+ FALSE,
+ &cur_prot,
+ &max_prot,
+ VM_INHERIT_NONE);
+ assert(kr == KERN_SUCCESS);
+ for (cur_offset = 0;
+ cur_offset < wire_size;
+ cur_offset += PAGE_SIZE) {
+ kr = vm_map_wire_and_extract(wire_map,
+ wire_addr + cur_offset,
+ VM_PROT_DEFAULT,
+ TRUE,
+ &wire_ppnum);
+ assert(kr == KERN_SUCCESS);
+ user_ppnum = vm_map_get_phys_page(user_map,
+ user_addr + cur_offset);
+ printf("FBDP_TEST_WIRE_AND_EXTRACT: kr=0x%x "
+ "user[%p:0x%llx:0x%x] wire[%p:0x%llx:0x%x]\n",
+ kr,
+ user_map, user_addr + cur_offset, user_ppnum,
+ wire_map, wire_addr + cur_offset, wire_ppnum);
+ if (kr != KERN_SUCCESS ||
+ wire_ppnum == 0 ||
+ wire_ppnum != user_ppnum) {
+ panic("FBDP_TEST_WIRE_AND_EXTRACT: FAIL\n");
+ }
+ }
+ cur_offset -= PAGE_SIZE;
+ kr = vm_map_wire_and_extract(wire_map,
+ wire_addr + cur_offset,
+ VM_PROT_DEFAULT,
+ TRUE,
+ &wire_ppnum);
+ assert(kr == KERN_SUCCESS);
+ printf("FBDP_TEST_WIRE_AND_EXTRACT: re-wire kr=0x%x "
+ "user[%p:0x%llx:0x%x] wire[%p:0x%llx:0x%x]\n",
+ kr,
+ user_map, user_addr + cur_offset, user_ppnum,
+ wire_map, wire_addr + cur_offset, wire_ppnum);
+ if (kr != KERN_SUCCESS ||
+ wire_ppnum == 0 ||
+ wire_ppnum != user_ppnum) {
+ panic("FBDP_TEST_WIRE_AND_EXTRACT: FAIL\n");
+ }
+
+ printf("FBDP_TEST_WIRE_AND_EXTRACT: PASS\n");
+#endif /* FBDP_TEST_WIRE_AND_EXTRACT */
+
+
vm_pageout_continue();
/*
return result;
}
+#if CONFIG_IOSCHED
+/*
+ * To support I/O Expedite for compressed files we mark the upls with special flags.
+ * The way decmpfs works is that we create a big upl which marks all the pages needed to
+ * represent the compressed file as busy. We tag this upl with the flag UPL_DECMP_REQ. Decmpfs
+ * then issues smaller I/Os for compressed I/Os, deflates them and puts the data into the pages
+ * being held in the big original UPL. We mark each of these smaller UPLs with the flag
+ * UPL_DECMP_REAL_IO. Any outstanding real I/O UPL is tracked by the big req upl using the
+ * decmp_io_upl field (in the upl structure). This link is protected in the forward direction
+ * by the req upl lock (the reverse link doesnt need synch. since we never inspect this link
+ * unless the real I/O upl is being destroyed).
+ */
+
+
+static void
+upl_set_decmp_info(upl_t upl, upl_t src_upl)
+{
+ assert((src_upl->flags & UPL_DECMP_REQ) != 0);
+
+ upl_lock(src_upl);
+ if (src_upl->decmp_io_upl) {
+ /*
+ * If there is already an alive real I/O UPL, ignore this new UPL.
+ * This case should rarely happen and even if it does, it just means
+ * that we might issue a spurious expedite which the driver is expected
+ * to handle.
+ */
+ upl_unlock(src_upl);
+ return;
+ }
+ src_upl->decmp_io_upl = (void *)upl;
+ src_upl->ref_count++;
+ upl_unlock(src_upl);
+
+ upl->flags |= UPL_DECMP_REAL_IO;
+ upl->decmp_io_upl = (void *)src_upl;
+
+}
+#endif /* CONFIG_IOSCHED */
+
+#if UPL_DEBUG
+int upl_debug_enabled = 1;
+#else
+int upl_debug_enabled = 0;
+#endif
static upl_t
upl_create(int type, int flags, upl_size_t size)
upl->highest_page = 0;
upl_lock_init(upl);
upl->vector_upl = NULL;
+#if CONFIG_IOSCHED
+ if (type & UPL_CREATE_IO_TRACKING) {
+ upl->upl_priority = proc_get_effective_thread_policy(current_thread(), TASK_POLICY_IO);
+ }
+
+ upl->upl_reprio_info = 0;
+ upl->decmp_io_upl = 0;
+ if ((type & UPL_CREATE_INTERNAL) && (type & UPL_CREATE_EXPEDITE_SUP)) {
+ /* Only support expedite on internal UPLs */
+ thread_t curthread = current_thread();
+ upl->upl_reprio_info = (uint64_t *)kalloc(sizeof(uint64_t) * atop(size));
+ bzero(upl->upl_reprio_info, (sizeof(uint64_t) * atop(size)));
+ upl->flags |= UPL_EXPEDITE_SUPPORTED;
+ if (curthread->decmp_upl != NULL)
+ upl_set_decmp_info(upl, curthread->decmp_upl);
+ }
+#endif
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if ((type & UPL_CREATE_IO_TRACKING) || upl_debug_enabled) {
+ upl->upl_creator = current_thread();
+ upl->uplq.next = 0;
+ upl->uplq.prev = 0;
+ upl->flags |= UPL_TRACKED_BY_OBJECT;
+ }
+#endif
+
#if UPL_DEBUG
upl->ubc_alias1 = 0;
upl->ubc_alias2 = 0;
- upl->upl_creator = current_thread();
upl->upl_state = 0;
upl->upl_commit_index = 0;
bzero(&upl->upl_commit_records[0], sizeof(upl->upl_commit_records));
- upl->uplq.next = 0;
- upl->uplq.prev = 0;
-
(void) OSBacktrace(&upl->upl_create_retaddr[0], UPL_DEBUG_STACK_FRAMES);
#endif /* UPL_DEBUG */
panic("upl(%p) ext_ref_count", upl);
}
-#if UPL_DEBUG
- if ( !(upl->flags & UPL_VECTOR)) {
+#if CONFIG_IOSCHED
+ if ((upl->flags & UPL_DECMP_REAL_IO) && upl->decmp_io_upl) {
+ upl_t src_upl;
+ src_upl = upl->decmp_io_upl;
+ assert((src_upl->flags & UPL_DECMP_REQ) != 0);
+ upl_lock(src_upl);
+ src_upl->decmp_io_upl = NULL;
+ upl_unlock(src_upl);
+ upl_deallocate(src_upl);
+ }
+#endif /* CONFIG_IOSCHED */
+
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if ((upl->flags & UPL_TRACKED_BY_OBJECT) && !(upl->flags & UPL_VECTOR)) {
vm_object_t object;
if (upl->flags & UPL_SHADOWED) {
} else {
object = upl->map_object;
}
+
vm_object_lock(object);
queue_remove(&object->uplq, upl, upl_t, uplq);
vm_object_activity_end(object);
vm_object_collapse(object, 0, TRUE);
vm_object_unlock(object);
}
-#endif /* UPL_DEBUG */
+#endif
/*
* drop a reference on the map_object whether or
* not a pageout object is inserted
upl_lock_destroy(upl);
upl->vector_upl = (vector_upl_t) 0xfeedbeef;
+#if CONFIG_IOSCHED
+ if (upl->flags & UPL_EXPEDITE_SUPPORTED)
+ kfree(upl->upl_reprio_info, sizeof(uint64_t) * (size/PAGE_SIZE));
+#endif
+
if (upl->flags & UPL_INTERNAL) {
kfree(upl,
sizeof(struct upl) +
void
upl_deallocate(upl_t upl)
{
+ upl_lock(upl);
if (--upl->ref_count == 0) {
if(vector_upl_is_valid(upl))
vector_upl_deallocate(upl);
+ upl_unlock(upl);
upl_destroy(upl);
}
+ else
+ upl_unlock(upl);
+}
+
+#if CONFIG_IOSCHED
+void
+upl_mark_decmp(upl_t upl)
+{
+ if (upl->flags & UPL_TRACKED_BY_OBJECT) {
+ upl->flags |= UPL_DECMP_REQ;
+ upl->upl_creator->decmp_upl = (void *)upl;
+ }
+}
+
+void
+upl_unmark_decmp(upl_t upl)
+{
+ if(upl && (upl->flags & UPL_DECMP_REQ)) {
+ upl->upl_creator->decmp_upl = NULL;
+ }
+}
+
+#endif /* CONFIG_IOSCHED */
+
+#define VM_PAGE_Q_BACKING_UP(q) \
+ ((q)->pgo_laundry >= (((q)->pgo_maxlaundry * 8) / 10))
+
+boolean_t must_throttle_writes(void);
+
+boolean_t
+must_throttle_writes()
+{
+ if (VM_PAGE_Q_BACKING_UP(&vm_pageout_queue_external) &&
+ vm_page_pageable_external_count > (AVAILABLE_NON_COMPRESSED_MEMORY * 6) / 10)
+ return (TRUE);
+
+ return (FALSE);
}
+
#if DEVELOPMENT || DEBUG
/*/*
* Statistics about UPL enforcement of copy-on-write obligations.
struct vm_page_delayed_work *dwp;
int dw_count;
int dw_limit;
+ int io_tracking_flag = 0;
if (cntrl_flags & ~UPL_VALID_FLAGS) {
/*
panic("vm_object_upl_request: contiguous object specified\n");
- if ((size / PAGE_SIZE) > MAX_UPL_SIZE)
- size = MAX_UPL_SIZE * PAGE_SIZE;
+ if (size > MAX_UPL_SIZE_BYTES)
+ size = MAX_UPL_SIZE_BYTES;
if ( (cntrl_flags & UPL_SET_INTERNAL) && page_list_count != NULL)
- *page_list_count = MAX_UPL_SIZE;
+ *page_list_count = MAX_UPL_SIZE_BYTES >> PAGE_SHIFT;
+
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if (object->io_tracking || upl_debug_enabled)
+ io_tracking_flag |= UPL_CREATE_IO_TRACKING;
+#endif
+#if CONFIG_IOSCHED
+ if (object->io_tracking)
+ io_tracking_flag |= UPL_CREATE_EXPEDITE_SUP;
+#endif
if (cntrl_flags & UPL_SET_INTERNAL) {
if (cntrl_flags & UPL_SET_LITE) {
- upl = upl_create(UPL_CREATE_INTERNAL | UPL_CREATE_LITE, 0, size);
+ upl = upl_create(UPL_CREATE_INTERNAL | UPL_CREATE_LITE | io_tracking_flag, 0, size);
user_page_list = (upl_page_info_t *) (((uintptr_t)upl) + sizeof(struct upl));
lite_list = (wpl_array_t)
lite_list = NULL;
}
} else {
- upl = upl_create(UPL_CREATE_INTERNAL, 0, size);
+ upl = upl_create(UPL_CREATE_INTERNAL | io_tracking_flag, 0, size);
user_page_list = (upl_page_info_t *) (((uintptr_t)upl) + sizeof(struct upl));
if (size == 0) {
} else {
if (cntrl_flags & UPL_SET_LITE) {
- upl = upl_create(UPL_CREATE_EXTERNAL | UPL_CREATE_LITE, 0, size);
+ upl = upl_create(UPL_CREATE_EXTERNAL | UPL_CREATE_LITE | io_tracking_flag, 0, size);
lite_list = (wpl_array_t) (((uintptr_t)upl) + sizeof(struct upl));
if (size == 0) {
lite_list = NULL;
}
} else {
- upl = upl_create(UPL_CREATE_EXTERNAL, 0, size);
+ upl = upl_create(UPL_CREATE_EXTERNAL | io_tracking_flag, 0, size);
}
}
*upl_ptr = upl;
upl->size = size;
upl->offset = offset + object->paging_offset;
-#if UPL_DEBUG
- vm_object_activity_begin(object);
- queue_enter(&object->uplq, upl, upl_t, uplq);
-#endif /* UPL_DEBUG */
-
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if (object->io_tracking || upl_debug_enabled) {
+ vm_object_activity_begin(object);
+ queue_enter(&object->uplq, upl, upl_t, uplq);
+ }
+#endif
if ((cntrl_flags & UPL_WILL_MODIFY) && object->copy != VM_OBJECT_NULL) {
/*
* Honor copy-on-write obligations
dw_limit = DELAYED_WORK_LIMIT(DEFAULT_DELAYED_WORK_LIMIT);
if (vm_page_free_count > (vm_page_free_target + size_in_pages) ||
- object->resident_page_count < (MAX_UPL_SIZE * 2))
+ object->resident_page_count < ((MAX_UPL_SIZE_BYTES * 2) >> PAGE_SHIFT))
object->scan_collisions = 0;
+ if ((cntrl_flags & UPL_WILL_MODIFY) && must_throttle_writes() == TRUE) {
+ boolean_t isSSD = FALSE;
+
+ vnode_pager_get_isSSD(object->pager, &isSSD);
+ vm_object_unlock(object);
+
+ OSAddAtomic(size_in_pages, &vm_upl_wait_for_pages);
+
+ if (isSSD == TRUE)
+ delay(1000 * size_in_pages);
+ else
+ delay(5000 * size_in_pages);
+ OSAddAtomic(-size_in_pages, &vm_upl_wait_for_pages);
+
+ vm_object_lock(object);
+ }
+
while (xfer_size) {
dwp->dw_mask = 0;
* speculative list
*/
dst_page->clustered = TRUE;
+
+ if ( !(cntrl_flags & UPL_FILE_IO))
+ VM_STAT_INCR(pageins);
}
}
/*
* update clustered and speculative state
*
*/
- VM_PAGE_CONSUME_CLUSTERED(dst_page);
+ if (dst_page->clustered)
+ VM_PAGE_CONSUME_CLUSTERED(dst_page);
}
try_next_page:
if (dwp->dw_mask) {
}
if (entry->object.vm_object == VM_OBJECT_NULL || !entry->object.vm_object->phys_contiguous) {
- if ((*upl_size/PAGE_SIZE) > MAX_UPL_SIZE)
- *upl_size = MAX_UPL_SIZE * PAGE_SIZE;
+ if (*upl_size > MAX_UPL_SIZE_BYTES)
+ *upl_size = MAX_UPL_SIZE_BYTES;
}
/*
* Create an object if necessary.
vm_map_unlock_read(map);
return KERN_PROTECTION_FAILURE;
}
+ }
- local_object = entry->object.vm_object;
- if (vm_map_entry_should_cow_for_true_share(entry) &&
- local_object->vo_size > *upl_size &&
- *upl_size != 0) {
- vm_prot_t prot;
+ local_object = entry->object.vm_object;
+ if (vm_map_entry_should_cow_for_true_share(entry) &&
+ local_object->vo_size > *upl_size &&
+ *upl_size != 0) {
+ vm_prot_t prot;
- /*
- * Set up the targeted range for copy-on-write to avoid
- * applying true_share/copy_delay to the entire object.
- */
+ /*
+ * Set up the targeted range for copy-on-write to avoid
+ * applying true_share/copy_delay to the entire object.
+ */
- if (vm_map_lock_read_to_write(map)) {
- goto REDISCOVER_ENTRY;
- }
+ if (vm_map_lock_read_to_write(map)) {
+ goto REDISCOVER_ENTRY;
+ }
- vm_map_clip_start(map,
- entry,
- vm_map_trunc_page(offset,
- VM_MAP_PAGE_MASK(map)));
- vm_map_clip_end(map,
- entry,
- vm_map_round_page(offset + *upl_size,
- VM_MAP_PAGE_MASK(map)));
- prot = entry->protection & ~VM_PROT_WRITE;
- if (override_nx(map, entry->alias) && prot)
- prot |= VM_PROT_EXECUTE;
- vm_object_pmap_protect(local_object,
- entry->offset,
- entry->vme_end - entry->vme_start,
- ((entry->is_shared || map->mapped_in_other_pmaps)
- ? PMAP_NULL
- : map->pmap),
- entry->vme_start,
- prot);
- entry->needs_copy = TRUE;
-
- vm_map_lock_write_to_read(map);
- }
-
- if (entry->needs_copy) {
- /*
- * Honor copy-on-write for COPY_SYMMETRIC
- * strategy.
- */
- vm_map_t local_map;
- vm_object_t object;
- vm_object_offset_t new_offset;
- vm_prot_t prot;
- boolean_t wired;
- vm_map_version_t version;
- vm_map_t real_map;
-
- local_map = map;
-
- if (vm_map_lookup_locked(&local_map,
- offset, VM_PROT_WRITE,
- OBJECT_LOCK_EXCLUSIVE,
- &version, &object,
- &new_offset, &prot, &wired,
- NULL,
- &real_map) != KERN_SUCCESS) {
- vm_map_unlock_read(local_map);
- return KERN_FAILURE;
+ vm_map_clip_start(map,
+ entry,
+ vm_map_trunc_page(offset,
+ VM_MAP_PAGE_MASK(map)));
+ vm_map_clip_end(map,
+ entry,
+ vm_map_round_page(offset + *upl_size,
+ VM_MAP_PAGE_MASK(map)));
+ if ((entry->vme_end - offset) < *upl_size) {
+ *upl_size = (upl_size_t) (entry->vme_end - offset);
+ assert(*upl_size == entry->vme_end - offset);
+ }
+
+ prot = entry->protection & ~VM_PROT_WRITE;
+ if (override_nx(map, entry->alias) && prot)
+ prot |= VM_PROT_EXECUTE;
+ vm_object_pmap_protect(local_object,
+ entry->offset,
+ entry->vme_end - entry->vme_start,
+ ((entry->is_shared || map->mapped_in_other_pmaps)
+ ? PMAP_NULL
+ : map->pmap),
+ entry->vme_start,
+ prot);
+ entry->needs_copy = TRUE;
+
+ vm_map_lock_write_to_read(map);
+ }
+
+ if (entry->needs_copy) {
+ /*
+ * Honor copy-on-write for COPY_SYMMETRIC
+ * strategy.
+ */
+ vm_map_t local_map;
+ vm_object_t object;
+ vm_object_offset_t new_offset;
+ vm_prot_t prot;
+ boolean_t wired;
+ vm_map_version_t version;
+ vm_map_t real_map;
+ vm_prot_t fault_type;
+
+ local_map = map;
+
+ if (caller_flags & UPL_COPYOUT_FROM) {
+ fault_type = VM_PROT_READ | VM_PROT_COPY;
+ vm_counters.create_upl_extra_cow++;
+ vm_counters.create_upl_extra_cow_pages += (entry->vme_end - entry->vme_start) / PAGE_SIZE;
+ } else {
+ fault_type = VM_PROT_WRITE;
+ }
+ if (vm_map_lookup_locked(&local_map,
+ offset, fault_type,
+ OBJECT_LOCK_EXCLUSIVE,
+ &version, &object,
+ &new_offset, &prot, &wired,
+ NULL,
+ &real_map) != KERN_SUCCESS) {
+ if (fault_type == VM_PROT_WRITE) {
+ vm_counters.create_upl_lookup_failure_write++;
+ } else {
+ vm_counters.create_upl_lookup_failure_copy++;
}
- if (real_map != map)
- vm_map_unlock(real_map);
vm_map_unlock_read(local_map);
+ return KERN_FAILURE;
+ }
+ if (real_map != map)
+ vm_map_unlock(real_map);
+ vm_map_unlock_read(local_map);
- vm_object_unlock(object);
+ vm_object_unlock(object);
- goto REDISCOVER_ENTRY;
- }
+ goto REDISCOVER_ENTRY;
}
+
if (sync_cow_data) {
if (entry->object.vm_object->shadow || entry->object.vm_object->copy) {
local_object = entry->object.vm_object;
* but only in kernel space. If this was on a user map,
* we'd have to set the wpmapped bit. */
/* m->wpmapped = TRUE; */
- assert(map==kernel_map);
+ assert(map->pmap == kernel_pmap);
- PMAP_ENTER(map->pmap, addr, m, VM_PROT_ALL, VM_PROT_NONE, 0, TRUE);
+ PMAP_ENTER(map->pmap, addr, m, VM_PROT_DEFAULT, VM_PROT_NONE, 0, TRUE);
}
offset += PAGE_SIZE_64;
}
return KERN_FAILURE;
}
-extern int panic_on_cs_killed;
kern_return_t
upl_commit_range(
upl_t upl,
upl_t vector_upl = NULL;
boolean_t should_be_throttled = FALSE;
+ vm_page_t nxt_page = VM_PAGE_NULL;
+ int fast_path_possible = 0;
+ int fast_path_full_commit = 0;
+ int throttle_page = 0;
+ int unwired_count = 0;
+ int local_queue_count = 0;
+ queue_head_t local_queue;
+
*empty = FALSE;
if (upl == UPL_NULL)
dw_count = 0;
dw_limit = DELAYED_WORK_LIMIT(DEFAULT_DELAYED_WORK_LIMIT);
+ if ((upl->flags & UPL_IO_WIRE) &&
+ !(flags & UPL_COMMIT_FREE_ABSENT) &&
+ !isVectorUPL &&
+ shadow_object->purgable != VM_PURGABLE_VOLATILE &&
+ shadow_object->purgable != VM_PURGABLE_EMPTY) {
+
+ if (!queue_empty(&shadow_object->memq)) {
+ queue_init(&local_queue);
+ if (size == shadow_object->vo_size) {
+ nxt_page = (vm_page_t)queue_first(&shadow_object->memq);
+ fast_path_full_commit = 1;
+ }
+ fast_path_possible = 1;
+
+ if (!VM_DYNAMIC_PAGING_ENABLED(memory_manager_default) && shadow_object->internal &&
+ (shadow_object->purgable == VM_PURGABLE_DENY ||
+ shadow_object->purgable == VM_PURGABLE_NONVOLATILE ||
+ shadow_object->purgable == VM_PURGABLE_VOLATILE)) {
+ throttle_page = 1;
+ }
+ }
+ }
+
while (xfer_size) {
vm_page_t t, m;
if (upl->flags & UPL_LITE) {
unsigned int pg_num;
+ if (nxt_page != VM_PAGE_NULL) {
+ m = nxt_page;
+ nxt_page = (vm_page_t)queue_next(&nxt_page->listq);
+ target_offset = m->offset;
+ }
pg_num = (unsigned int) (target_offset/PAGE_SIZE);
assert(pg_num == target_offset/PAGE_SIZE);
if (lite_list[pg_num>>5] & (1 << (pg_num & 31))) {
lite_list[pg_num>>5] &= ~(1 << (pg_num & 31));
- if (!(upl->flags & UPL_KERNEL_OBJECT))
+ if (!(upl->flags & UPL_KERNEL_OBJECT) && m == VM_PAGE_NULL)
m = vm_page_lookup(shadow_object, target_offset + (upl->offset - shadow_object->paging_offset));
- }
+ } else
+ m = NULL;
}
if (upl->flags & UPL_SHADOWED) {
if ((t = vm_page_lookup(object, target_offset)) != VM_PAGE_NULL) {
VM_PAGE_FREE(t);
- if (m == VM_PAGE_NULL)
+ if (!(upl->flags & UPL_KERNEL_OBJECT) && m == VM_PAGE_NULL)
m = vm_page_lookup(shadow_object, target_offset + object->vo_shadow_offset);
}
}
- if ((upl->flags & UPL_KERNEL_OBJECT) || m == VM_PAGE_NULL)
+ if (m == VM_PAGE_NULL)
goto commit_next_page;
if (m->compressor) {
m->cs_tainted = page_list[entry].cs_tainted;
}
if (flags & UPL_COMMIT_WRITTEN_BY_KERNEL)
- m->written_by_kernel = TRUE;
+ m->written_by_kernel = TRUE;
if (upl->flags & UPL_IO_WIRE) {
* so it will need to be
* re-validated.
*/
- if (panic_on_cs_killed &&
- m->slid) {
+ if (m->slid) {
panic("upl_commit_range(%p): page %p was slid\n",
upl, m);
}
}
clear_refmod |= VM_MEM_MODIFIED;
}
- if (flags & UPL_COMMIT_INACTIVATE) {
- dwp->dw_mask |= DW_vm_page_deactivate_internal;
- clear_refmod |= VM_MEM_REFERENCED;
- }
if (upl->flags & UPL_ACCESS_BLOCKED) {
/*
* We blocked access to the pages in this UPL.
*/
dwp->dw_mask |= (DW_clear_busy | DW_PAGE_WAKEUP);
}
- if (m->absent) {
- if (flags & UPL_COMMIT_FREE_ABSENT)
- dwp->dw_mask |= DW_vm_page_free;
- else {
+ if (fast_path_possible) {
+ assert(m->object->purgable != VM_PURGABLE_EMPTY);
+ assert(m->object->purgable != VM_PURGABLE_VOLATILE);
+ if (m->absent) {
+ assert(m->wire_count == 0);
+ assert(m->busy);
+
m->absent = FALSE;
dwp->dw_mask |= (DW_clear_busy | DW_PAGE_WAKEUP);
+ } else {
+ if (m->wire_count == 0)
+ panic("wire_count == 0, m = %p, obj = %p\n", m, shadow_object);
+
+ /*
+ * XXX FBDP need to update some other
+ * counters here (purgeable_wired_count)
+ * (ledgers), ...
+ */
+ assert(m->wire_count);
+ m->wire_count--;
- if ( !(dwp->dw_mask & DW_vm_page_deactivate_internal))
- dwp->dw_mask |= DW_vm_page_activate;
+ if (m->wire_count == 0)
+ unwired_count++;
}
- } else
- dwp->dw_mask |= DW_vm_page_unwire;
+ if (m->wire_count == 0) {
+ queue_enter(&local_queue, m, vm_page_t, pageq);
+ local_queue_count++;
+ if (throttle_page) {
+ m->throttled = TRUE;
+ } else {
+ if (flags & UPL_COMMIT_INACTIVATE)
+ m->inactive = TRUE;
+ else
+ m->active = TRUE;
+ }
+ }
+ } else {
+ if (flags & UPL_COMMIT_INACTIVATE) {
+ dwp->dw_mask |= DW_vm_page_deactivate_internal;
+ clear_refmod |= VM_MEM_REFERENCED;
+ }
+ if (m->absent) {
+ if (flags & UPL_COMMIT_FREE_ABSENT)
+ dwp->dw_mask |= DW_vm_page_free;
+ else {
+ m->absent = FALSE;
+ dwp->dw_mask |= (DW_clear_busy | DW_PAGE_WAKEUP);
+
+ if ( !(dwp->dw_mask & DW_vm_page_deactivate_internal))
+ dwp->dw_mask |= DW_vm_page_activate;
+ }
+ } else
+ dwp->dw_mask |= DW_vm_page_unwire;
+ }
goto commit_next_page;
}
assert(!m->compressor);
* so it will need to be
* re-validated.
*/
- if (panic_on_cs_killed &&
- m->slid) {
+ if (m->slid) {
panic("upl_commit_range(%p): page %p was slid\n",
upl, m);
}
* the (COPY_OUT_FROM == FALSE) request_page_list case
*/
if (m->busy) {
+#if CONFIG_PHANTOM_CACHE
+ if (m->absent && !m->object->internal)
+ dwp->dw_mask |= DW_vm_phantom_cache_update;
+#endif
m->absent = FALSE;
dwp->dw_mask |= DW_clear_busy;
if (hibernate_cleaning_in_progress == FALSE && !m->dirty && (upl->flags & UPL_PAGEOUT)) {
pgpgout_count++;
- /* this page used to be dirty; now it's on the clean queue. */
- m->was_dirty = TRUE;
+ VM_STAT_INCR(pageouts);
+ DTRACE_VM2(pgout, int, 1, (uint64_t *), NULL);
dwp->dw_mask |= DW_enqueue_cleaned;
vm_pageout_enqueued_cleaned_from_inactive_dirty++;
*/
dwp->dw_mask |= DW_clear_busy;
}
-
/*
* Wakeup any thread waiting for the page to be un-cleaning.
*/
if (dw_count)
vm_page_do_delayed_work(shadow_object, &dw_array[0], dw_count);
+ if (fast_path_possible) {
+
+ assert(shadow_object->purgable != VM_PURGABLE_VOLATILE);
+ assert(shadow_object->purgable != VM_PURGABLE_EMPTY);
+
+ if (local_queue_count || unwired_count) {
+
+ if (local_queue_count) {
+ vm_page_t first_local, last_local;
+ vm_page_t first_target;
+ queue_head_t *target_queue;
+
+ if (throttle_page)
+ target_queue = &vm_page_queue_throttled;
+ else {
+ if (flags & UPL_COMMIT_INACTIVATE) {
+ if (shadow_object->internal)
+ target_queue = &vm_page_queue_anonymous;
+ else
+ target_queue = &vm_page_queue_inactive;
+ } else
+ target_queue = &vm_page_queue_active;
+ }
+ /*
+ * Transfer the entire local queue to a regular LRU page queues.
+ */
+ first_local = (vm_page_t) queue_first(&local_queue);
+ last_local = (vm_page_t) queue_last(&local_queue);
+
+ vm_page_lockspin_queues();
+
+ first_target = (vm_page_t) queue_first(target_queue);
+
+ if (queue_empty(target_queue))
+ queue_last(target_queue) = (queue_entry_t) last_local;
+ else
+ queue_prev(&first_target->pageq) = (queue_entry_t) last_local;
+
+ queue_first(target_queue) = (queue_entry_t) first_local;
+ queue_prev(&first_local->pageq) = (queue_entry_t) target_queue;
+ queue_next(&last_local->pageq) = (queue_entry_t) first_target;
+
+ /*
+ * Adjust the global page counts.
+ */
+ if (throttle_page) {
+ vm_page_throttled_count += local_queue_count;
+ } else {
+ if (flags & UPL_COMMIT_INACTIVATE) {
+ if (shadow_object->internal)
+ vm_page_anonymous_count += local_queue_count;
+ vm_page_inactive_count += local_queue_count;
+
+ token_new_pagecount += local_queue_count;
+ } else
+ vm_page_active_count += local_queue_count;
+
+ if (shadow_object->internal)
+ vm_page_pageable_internal_count += local_queue_count;
+ else
+ vm_page_pageable_external_count += local_queue_count;
+ }
+ } else {
+ vm_page_lockspin_queues();
+ }
+ if (unwired_count) {
+ vm_page_wire_count -= unwired_count;
+ VM_CHECK_MEMORYSTATUS;
+ }
+ vm_page_unlock_queues();
+
+ shadow_object->wired_page_count -= unwired_count;
+ }
+ }
occupied = 1;
if (upl->flags & UPL_DEVICE_MEMORY) {
int pg_num;
int i;
- pg_num = upl->size/PAGE_SIZE;
- pg_num = (pg_num + 31) >> 5;
occupied = 0;
- for (i = 0; i < pg_num; i++) {
- if (lite_list[i] != 0) {
- occupied = 1;
- break;
+ if (!fast_path_full_commit) {
+ pg_num = upl->size/PAGE_SIZE;
+ pg_num = (pg_num + 31) >> 5;
+
+ for (i = 0; i < pg_num; i++) {
+ if (lite_list[i] != 0) {
+ occupied = 1;
+ break;
+ }
}
}
} else {
return upl_commit_range(upl, 0, upl->size, 0, page_list, count, &empty);
}
+
+void
+iopl_valid_data(
+ upl_t upl)
+{
+ vm_object_t object;
+ vm_offset_t offset;
+ vm_page_t m, nxt_page = VM_PAGE_NULL;
+ upl_size_t size;
+ int wired_count = 0;
+
+ if (upl == NULL)
+ panic("iopl_valid_data: NULL upl");
+ if (vector_upl_is_valid(upl))
+ panic("iopl_valid_data: vector upl");
+ if ((upl->flags & (UPL_DEVICE_MEMORY|UPL_SHADOWED|UPL_ACCESS_BLOCKED|UPL_IO_WIRE|UPL_INTERNAL)) != UPL_IO_WIRE)
+ panic("iopl_valid_data: unsupported upl, flags = %x", upl->flags);
+
+ object = upl->map_object;
+
+ if (object == kernel_object || object == compressor_object)
+ panic("iopl_valid_data: object == kernel or compressor");
+
+ if (object->purgable == VM_PURGABLE_VOLATILE)
+ panic("iopl_valid_data: object == VM_PURGABLE_VOLATILE");
+
+ size = upl->size;
+
+ vm_object_lock(object);
+
+ if (object->vo_size == size && object->resident_page_count == (size / PAGE_SIZE))
+ nxt_page = (vm_page_t)queue_first(&object->memq);
+ else
+ offset = 0 + upl->offset - object->paging_offset;
+
+ while (size) {
+
+ if (nxt_page != VM_PAGE_NULL) {
+ m = nxt_page;
+ nxt_page = (vm_page_t)queue_next(&nxt_page->listq);
+ } else {
+ m = vm_page_lookup(object, offset);
+ offset += PAGE_SIZE;
+
+ if (m == VM_PAGE_NULL)
+ panic("iopl_valid_data: missing expected page at offset %lx", (long)offset);
+ }
+ if (m->busy) {
+ if (!m->absent)
+ panic("iopl_valid_data: busy page w/o absent");
+
+ if (m->pageq.next || m->pageq.prev)
+ panic("iopl_valid_data: busy+absent page on page queue");
+
+ m->absent = FALSE;
+ m->dirty = TRUE;
+ m->wire_count++;
+ wired_count++;
+
+ PAGE_WAKEUP_DONE(m);
+ }
+ size -= PAGE_SIZE;
+ }
+ if (wired_count) {
+ object->wired_page_count += wired_count;
+
+ vm_page_lockspin_queues();
+ vm_page_wire_count += wired_count;
+ vm_page_unlock_queues();
+ }
+ vm_object_unlock(object);
+}
+
+
+
+
void
vm_object_set_pmap_cache_attr(
vm_object_t object,
int dw_limit;
int dw_index;
boolean_t caller_lookup;
+ int io_tracking_flag = 0;
+ int interruptible;
+
+ boolean_t set_cache_attr_needed = FALSE;
+ boolean_t free_wired_pages = FALSE;
+ int fast_path_possible = 0;
+
if (cntrl_flags & ~UPL_VALID_FLAGS) {
/*
else
prot = VM_PROT_READ | VM_PROT_WRITE;
- if (((size/PAGE_SIZE) > MAX_UPL_SIZE) && !object->phys_contiguous)
- size = MAX_UPL_SIZE * PAGE_SIZE;
-
- if (cntrl_flags & UPL_SET_INTERNAL) {
- if (page_list_count != NULL)
- *page_list_count = MAX_UPL_SIZE;
- }
- if (((cntrl_flags & UPL_SET_INTERNAL) && !(object->phys_contiguous)) &&
- ((page_list_count != NULL) && (*page_list_count != 0) && *page_list_count < (size/page_size)))
- return KERN_INVALID_ARGUMENT;
-
if ((!object->internal) && (object->paging_offset != 0))
panic("vm_object_iopl_request: external object with non-zero paging offset\n");
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if ((object->io_tracking && object != kernel_object) || upl_debug_enabled)
+ io_tracking_flag |= UPL_CREATE_IO_TRACKING;
+#endif
+
+#if CONFIG_IOSCHED
+ if (object->io_tracking) {
+ /* Check if we're dealing with the kernel object. We do not support expedite on kernel object UPLs */
+ if (object != kernel_object)
+ io_tracking_flag |= UPL_CREATE_EXPEDITE_SUP;
+ }
+#endif
if (object->phys_contiguous)
psize = PAGE_SIZE;
psize = size;
if (cntrl_flags & UPL_SET_INTERNAL) {
- upl = upl_create(UPL_CREATE_INTERNAL | UPL_CREATE_LITE, UPL_IO_WIRE, psize);
+ upl = upl_create(UPL_CREATE_INTERNAL | UPL_CREATE_LITE | io_tracking_flag, UPL_IO_WIRE, psize);
user_page_list = (upl_page_info_t *) (((uintptr_t)upl) + sizeof(struct upl));
lite_list = (wpl_array_t) (((uintptr_t)user_page_list) +
lite_list = NULL;
}
} else {
- upl = upl_create(UPL_CREATE_LITE, UPL_IO_WIRE, psize);
+ upl = upl_create(UPL_CREATE_LITE | io_tracking_flag, UPL_IO_WIRE, psize);
lite_list = (wpl_array_t) (((uintptr_t)upl) + sizeof(struct upl));
if (size == 0) {
upl->flags |= UPL_ACCESS_BLOCKED;
}
- if (object->phys_contiguous) {
-#if UPL_DEBUG
+ if (!(cntrl_flags & (UPL_NEED_32BIT_ADDR | UPL_BLOCK_ACCESS)) &&
+ object->purgable != VM_PURGABLE_VOLATILE &&
+ object->purgable != VM_PURGABLE_EMPTY &&
+ object->copy == NULL &&
+ size == object->vo_size &&
+ offset == 0 &&
+ object->resident_page_count == 0 &&
+ object->shadow == NULL &&
+ object->pager == NULL)
+ {
+ fast_path_possible = 1;
+ set_cache_attr_needed = TRUE;
+ }
+
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if (upl->flags & UPL_TRACKED_BY_OBJECT) {
vm_object_activity_begin(object);
queue_enter(&object->uplq, upl, upl_t, uplq);
-#endif /* UPL_DEBUG */
+ }
+#endif
+
+ if (object->phys_contiguous) {
if (upl->flags & UPL_ACCESS_BLOCKED) {
assert(!object->blocked_access);
/*
* Protect user space from future COW operations
*/
+#if VM_OBJECT_TRACKING_OP_TRUESHARE
+ if (!object->true_share &&
+ vm_object_tracking_inited) {
+ void *bt[VM_OBJECT_TRACKING_BTDEPTH];
+ int num = 0;
+
+ num = OSBacktrace(bt,
+ VM_OBJECT_TRACKING_BTDEPTH);
+ btlog_add_entry(vm_object_tracking_btlog,
+ object,
+ VM_OBJECT_TRACKING_OP_TRUESHARE,
+ bt,
+ num);
+ }
+#endif /* VM_OBJECT_TRACKING_OP_TRUESHARE */
+
object->true_share = TRUE;
if (object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC)
object->copy_strategy = MEMORY_OBJECT_COPY_DELAY;
}
-
-#if UPL_DEBUG
- vm_object_activity_begin(object);
- queue_enter(&object->uplq, upl, upl_t, uplq);
-#endif /* UPL_DEBUG */
if (!(cntrl_flags & UPL_COPYOUT_FROM) &&
object->copy != VM_OBJECT_NULL) {
iopl_cow_pages += size >> PAGE_SHIFT;
#endif
}
-
+ if (cntrl_flags & UPL_SET_INTERRUPTIBLE)
+ interruptible = THREAD_ABORTSAFE;
+ else
+ interruptible = THREAD_UNINT;
entry = 0;
xfer_size = size;
dst_offset = offset;
+ dw_count = 0;
+
+ if (fast_path_possible) {
+ int wired_count = 0;
+
+ while (xfer_size) {
+
+ while ( (dst_page = vm_page_grab()) == VM_PAGE_NULL) {
+ OSAddAtomic(size_in_pages, &vm_upl_wait_for_pages);
+
+ VM_DEBUG_EVENT(vm_iopl_page_wait, VM_IOPL_PAGE_WAIT, DBG_FUNC_START, vm_upl_wait_for_pages, 0, 0, 0);
+
+ if (vm_page_wait(interruptible) == FALSE) {
+ /*
+ * interrupted case
+ */
+ OSAddAtomic(-size_in_pages, &vm_upl_wait_for_pages);
+
+ VM_DEBUG_EVENT(vm_iopl_page_wait, VM_IOPL_PAGE_WAIT, DBG_FUNC_END, vm_upl_wait_for_pages, 0, 0, -1);
+
+ if (wired_count) {
+ vm_page_lockspin_queues();
+ vm_page_wire_count += wired_count;
+ vm_page_unlock_queues();
+
+ free_wired_pages = TRUE;
+ }
+ ret = MACH_SEND_INTERRUPTED;
+
+ goto return_err;
+ }
+ OSAddAtomic(-size_in_pages, &vm_upl_wait_for_pages);
+
+ VM_DEBUG_EVENT(vm_iopl_page_wait, VM_IOPL_PAGE_WAIT, DBG_FUNC_END, vm_upl_wait_for_pages, 0, 0, 0);
+ }
+ if (no_zero_fill == FALSE)
+ vm_page_zero_fill(dst_page);
+ else
+ dst_page->absent = TRUE;
+
+ dst_page->reference = TRUE;
+
+ if (!(cntrl_flags & UPL_COPYOUT_FROM)) {
+ SET_PAGE_DIRTY(dst_page, FALSE);
+ }
+ if (dst_page->absent == FALSE) {
+ assert(object->purgable != VM_PURGABLE_VOLATILE);
+ assert(object->purgable != VM_PURGABLE_EMPTY);
+ dst_page->wire_count++;
+ wired_count++;
+
+ PAGE_WAKEUP_DONE(dst_page);
+ }
+ vm_page_insert_internal(dst_page, object, dst_offset, FALSE, TRUE, TRUE);
+
+ lite_list[entry>>5] |= 1 << (entry & 31);
+
+ if (dst_page->phys_page > upl->highest_page)
+ upl->highest_page = dst_page->phys_page;
+
+ if (user_page_list) {
+ user_page_list[entry].phys_addr = dst_page->phys_page;
+ user_page_list[entry].absent = dst_page->absent;
+ user_page_list[entry].dirty = dst_page->dirty;
+ user_page_list[entry].precious = FALSE;
+ user_page_list[entry].pageout = FALSE;
+ user_page_list[entry].device = FALSE;
+ user_page_list[entry].needed = FALSE;
+ user_page_list[entry].speculative = FALSE;
+ user_page_list[entry].cs_validated = FALSE;
+ user_page_list[entry].cs_tainted = FALSE;
+ }
+ entry++;
+ dst_offset += PAGE_SIZE_64;
+ xfer_size -= PAGE_SIZE;
+ size_in_pages--;
+ }
+ if (wired_count) {
+ vm_page_lockspin_queues();
+ vm_page_wire_count += wired_count;
+ vm_page_unlock_queues();
+ }
+ goto finish;
+ }
fault_info.behavior = VM_BEHAVIOR_SEQUENTIAL;
fault_info.user_tag = 0;
fault_info.stealth = FALSE;
fault_info.io_sync = FALSE;
fault_info.cs_bypass = FALSE;
- fault_info.mark_zf_absent = (0 == (cntrl_flags & UPL_NOZEROFILLIO));
+ fault_info.mark_zf_absent = TRUE;
+ fault_info.interruptible = interruptible;
+ fault_info.batch_pmap_op = TRUE;
dwp = &dw_array[0];
- dw_count = 0;
dw_limit = DELAYED_WORK_LIMIT(DEFAULT_DELAYED_WORK_LIMIT);
while (xfer_size) {
ret = KERN_MEMORY_ERROR;
goto return_err;
}
+ set_cache_attr_needed = TRUE;
/*
* We just looked up the page and the result remains valid
do {
vm_page_t top_page;
kern_return_t error_code;
- int interruptible;
-
- if (cntrl_flags & UPL_SET_INTERRUPTIBLE)
- interruptible = THREAD_ABORTSAFE;
- else
- interruptible = THREAD_UNINT;
- fault_info.interruptible = interruptible;
fault_info.cluster_size = xfer_size;
- fault_info.batch_pmap_op = TRUE;
vm_object_paging_begin(object);
* update clustered and speculative state
*
*/
- VM_PAGE_CONSUME_CLUSTERED(dst_page);
+ if (dst_page->clustered)
+ VM_PAGE_CONSUME_CLUSTERED(dst_page);
}
entry++;
dst_offset += PAGE_SIZE_64;
xfer_size -= PAGE_SIZE;
+ size_in_pages--;
if (dwp->dw_mask) {
VM_PAGE_ADD_DELAYED_WORK(dwp, dst_page, dw_count);
if (dw_count)
vm_page_do_delayed_work(object, &dw_array[0], dw_count);
- vm_object_set_pmap_cache_attr(object, user_page_list, entry, TRUE);
+finish:
+ if (user_page_list && set_cache_attr_needed == TRUE)
+ vm_object_set_pmap_cache_attr(object, user_page_list, entry, TRUE);
if (page_list_count != NULL) {
if (upl->flags & UPL_INTERNAL)
}
vm_page_lock_queues();
- if (dst_page->absent) {
+ if (dst_page->absent || free_wired_pages == TRUE) {
vm_page_free(dst_page);
need_unwire = FALSE;
* Make each UPL point to the correct VM object, i.e. the
* object holding the pages that the UPL refers to...
*/
-#if UPL_DEBUG
- queue_remove(&object1->uplq, upl1, upl_t, uplq);
- queue_remove(&object2->uplq, upl2, upl_t, uplq);
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if ((upl1->flags & UPL_TRACKED_BY_OBJECT) || (upl2->flags & UPL_TRACKED_BY_OBJECT)) {
+ vm_object_lock(object1);
+ vm_object_lock(object2);
+ }
+ if (upl1->flags & UPL_TRACKED_BY_OBJECT)
+ queue_remove(&object1->uplq, upl1, upl_t, uplq);
+ if (upl2->flags & UPL_TRACKED_BY_OBJECT)
+ queue_remove(&object2->uplq, upl2, upl_t, uplq);
#endif
upl1->map_object = object2;
upl2->map_object = object1;
-#if UPL_DEBUG
- queue_enter(&object1->uplq, upl2, upl_t, uplq);
- queue_enter(&object2->uplq, upl1, upl_t, uplq);
+
+#if CONFIG_IOSCHED || UPL_DEBUG
+ if (upl1->flags & UPL_TRACKED_BY_OBJECT)
+ queue_enter(&object2->uplq, upl1, upl_t, uplq);
+ if (upl2->flags & UPL_TRACKED_BY_OBJECT)
+ queue_enter(&object1->uplq, upl2, upl_t, uplq);
+ if ((upl1->flags & UPL_TRACKED_BY_OBJECT) || (upl2->flags & UPL_TRACKED_BY_OBJECT)) {
+ vm_object_unlock(object2);
+ vm_object_unlock(object1);
+ }
#endif
}
*/
vm_paging_page_waiter_total++;
vm_paging_page_waiter++;
- thread_sleep_fast_usimple_lock(&vm_paging_page_waiter,
- &vm_paging_lock,
- THREAD_UNINT);
+ kr = assert_wait((event_t)&vm_paging_page_waiter, THREAD_UNINT);
+ if (kr == THREAD_WAITING) {
+ simple_unlock(&vm_paging_lock);
+ kr = thread_block(THREAD_CONTINUE_NULL);
+ simple_lock(&vm_paging_lock);
+ }
vm_paging_page_waiter--;
/* ... and try again */
}
}
}
-#if CRYPTO
+#if ENCRYPTED_SWAP
/*
* Encryption data.
* "iv" is the "initial vector". Ideally, we want to
goto process_upl_to_encrypt;
}
-#else /* CRYPTO */
+#else /* ENCRYPTED_SWAP */
void
upl_encrypt(
__unused upl_t upl,
{
}
-#endif /* CRYPTO */
+#endif /* ENCRYPTED_SWAP */
/*
* page->object must be locked
upl_unlock(upl);
}
+#if CONFIG_IOSCHED
+void
+upl_set_blkno(
+ upl_t upl,
+ vm_offset_t upl_offset,
+ int io_size,
+ int64_t blkno)
+{
+ int i,j;
+ if ((upl->flags & UPL_EXPEDITE_SUPPORTED) == 0)
+ return;
+
+ assert(upl->upl_reprio_info != 0);
+ for(i = (int)(upl_offset / PAGE_SIZE), j = 0; j < io_size; i++, j += PAGE_SIZE) {
+ UPL_SET_REPRIO_INFO(upl, i, blkno, io_size);
+ }
+}
+#endif
+
boolean_t
vm_page_is_slideable(vm_page_t m)
{
assert(!page->slid);
assert(page->object->object_slid);
- pageIndex = (uint32_t)((page->offset - page->object->vo_slide_info->start)/PAGE_SIZE);
- kr = vm_shared_region_slide_page(page->object->vo_slide_info, kernel_vaddr, pageIndex);
+ /* on some platforms this is an extern int, on others it's a cpp macro */
+ __unreachable_ok_push
+ /* TODO: Consider this */
+ if (!TEST_PAGE_SIZE_4K) {
+ for (int i = 0; i < 4; i++) {
+ pageIndex = (uint32_t)((page->offset - page->object->vo_slide_info->start)/0x1000);
+ kr = vm_shared_region_slide_page(page->object->vo_slide_info, kernel_vaddr + (0x1000*i), pageIndex + i);
+ }
+ } else {
+ pageIndex = (uint32_t)((page->offset - page->object->vo_slide_info->start)/PAGE_SIZE);
+ kr = vm_shared_region_slide_page(page->object->vo_slide_info, kernel_vaddr, pageIndex);
+ }
+ __unreachable_ok_pop
+
vm_page_slide_counter++;
/*
return(UPL_PHYS_PAGE(upl, index));
}
-
void
vm_countdirtypages(void)
{
return KERN_SUCCESS;
}
#endif /* UPL_DEBUG */
+
+#if VM_PRESSURE_EVENTS
+/*
+ * Upward trajectory.
+ */
+extern boolean_t vm_compressor_low_on_space(void);
+
+boolean_t
+VM_PRESSURE_NORMAL_TO_WARNING(void) {
+
+ if (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPLESS) {
+
+ /* Available pages below our threshold */
+ if (memorystatus_available_pages < memorystatus_available_pages_pressure) {
+ /* No frozen processes to kill */
+ if (memorystatus_frozen_count == 0) {
+ /* Not enough suspended processes available. */
+ if (memorystatus_suspended_count < MEMORYSTATUS_SUSPENDED_THRESHOLD) {
+ return TRUE;
+ }
+ }
+ }
+ return FALSE;
+
+ } else {
+ return ((AVAILABLE_NON_COMPRESSED_MEMORY < VM_PAGE_COMPRESSOR_COMPACT_THRESHOLD) ? 1 : 0);
+ }
+}
+
+boolean_t
+VM_PRESSURE_WARNING_TO_CRITICAL(void) {
+
+ if (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPLESS) {
+ /* Available pages below our threshold */
+ if (memorystatus_available_pages < memorystatus_available_pages_critical) {
+ return TRUE;
+ }
+ return FALSE;
+ } else {
+ return (vm_compressor_low_on_space() || (AVAILABLE_NON_COMPRESSED_MEMORY < ((12 * VM_PAGE_COMPRESSOR_SWAP_UNTHROTTLE_THRESHOLD) / 10)) ? 1 : 0);
+ }
+}
+
+/*
+ * Downward trajectory.
+ */
+boolean_t
+VM_PRESSURE_WARNING_TO_NORMAL(void) {
+
+ if (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPLESS) {
+ /* Available pages above our threshold */
+ unsigned int target_threshold = memorystatus_available_pages_pressure + ((15 * memorystatus_available_pages_pressure) / 100);
+ if (memorystatus_available_pages > target_threshold) {
+ return TRUE;
+ }
+ return FALSE;
+ } else {
+ return ((AVAILABLE_NON_COMPRESSED_MEMORY > ((12 * VM_PAGE_COMPRESSOR_COMPACT_THRESHOLD) / 10)) ? 1 : 0);
+ }
+}
+
+boolean_t
+VM_PRESSURE_CRITICAL_TO_WARNING(void) {
+
+ if (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPLESS) {
+ /* Available pages above our threshold */
+ unsigned int target_threshold = memorystatus_available_pages_critical + ((15 * memorystatus_available_pages_critical) / 100);
+ if (memorystatus_available_pages > target_threshold) {
+ return TRUE;
+ }
+ return FALSE;
+ } else {
+ return ((AVAILABLE_NON_COMPRESSED_MEMORY > ((14 * VM_PAGE_COMPRESSOR_SWAP_UNTHROTTLE_THRESHOLD) / 10)) ? 1 : 0);
+ }
+}
+#endif /* VM_PRESSURE_EVENTS */
+
projects / apple / xnu.git / blobdiff