4 * Copyright (c) 1991 Regents of the University of California.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
35 * $FreeBSD: src/sys/vm/vm_page.c,v 1.147.2.18 2002/03/10 05:03:19 alc Exp $
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
54 * Carnegie Mellon requests users of this software to return to
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
65 * Resident memory management module. The module manipulates 'VM pages'.
66 * A VM page is the core building block for memory management.
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/malloc.h>
73 #include <sys/vmmeter.h>
74 #include <sys/vnode.h>
75 #include <sys/kernel.h>
78 #include <vm/vm_param.h>
80 #include <vm/vm_kern.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_pageout.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_extern.h>
88 #include <vm/swap_pager.h>
90 #include <machine/inttypes.h>
91 #include <machine/md_var.h>
93 #include <vm/vm_page2.h>
94 #include <sys/spinlock2.h>
96 #define VMACTION_HSIZE 256
97 #define VMACTION_HMASK (VMACTION_HSIZE - 1)
99 static void vm_page_queue_init(void);
100 static void vm_page_free_wakeup(void);
101 static vm_page_t vm_page_select_cache(u_short pg_color);
102 static vm_page_t _vm_page_list_find2(int basequeue, int index);
103 static void _vm_page_deactivate_locked(vm_page_t m, int athead);
106 * Array of tailq lists
108 __cachealign struct vpgqueues vm_page_queues[PQ_COUNT];
110 LIST_HEAD(vm_page_action_list, vm_page_action);
111 struct vm_page_action_list action_list[VMACTION_HSIZE];
112 static volatile int vm_pages_waiting;
115 RB_GENERATE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare,
116 vm_pindex_t, pindex);
119 vm_page_queue_init(void)
123 for (i = 0; i < PQ_L2_SIZE; i++)
124 vm_page_queues[PQ_FREE+i].cnt = &vmstats.v_free_count;
125 for (i = 0; i < PQ_L2_SIZE; i++)
126 vm_page_queues[PQ_CACHE+i].cnt = &vmstats.v_cache_count;
127 for (i = 0; i < PQ_L2_SIZE; i++)
128 vm_page_queues[PQ_INACTIVE+i].cnt = &vmstats.v_inactive_count;
129 for (i = 0; i < PQ_L2_SIZE; i++)
130 vm_page_queues[PQ_ACTIVE+i].cnt = &vmstats.v_active_count;
131 for (i = 0; i < PQ_L2_SIZE; i++)
132 vm_page_queues[PQ_HOLD+i].cnt = &vmstats.v_active_count;
133 /* PQ_NONE has no queue */
135 for (i = 0; i < PQ_COUNT; i++) {
136 TAILQ_INIT(&vm_page_queues[i].pl);
137 spin_init(&vm_page_queues[i].spin);
140 for (i = 0; i < VMACTION_HSIZE; i++)
141 LIST_INIT(&action_list[i]);
145 * note: place in initialized data section? Is this necessary?
148 int vm_page_array_size = 0;
149 int vm_page_zero_count = 0;
150 vm_page_t vm_page_array = 0;
155 * Sets the page size, perhaps based upon the memory size.
156 * Must be called before any use of page-size dependent functions.
159 vm_set_page_size(void)
161 if (vmstats.v_page_size == 0)
162 vmstats.v_page_size = PAGE_SIZE;
163 if (((vmstats.v_page_size - 1) & vmstats.v_page_size) != 0)
164 panic("vm_set_page_size: page size not a power of two");
170 * Add a new page to the freelist for use by the system. New pages
171 * are added to both the head and tail of the associated free page
172 * queue in a bottom-up fashion, so both zero'd and non-zero'd page
173 * requests pull 'recent' adds (higher physical addresses) first.
175 * Beware that the page zeroing daemon will also be running soon after
176 * boot, moving pages from the head to the tail of the PQ_FREE queues.
178 * Must be called in a critical section.
181 vm_add_new_page(vm_paddr_t pa)
183 struct vpgqueues *vpq;
186 m = PHYS_TO_VM_PAGE(pa);
189 m->pc = (pa >> PAGE_SHIFT) & PQ_L2_MASK;
192 * Twist for cpu localization in addition to page coloring, so
193 * different cpus selecting by m->queue get different page colors.
195 m->pc ^= ((pa >> PAGE_SHIFT) / PQ_L2_SIZE) & PQ_L2_MASK;
196 m->pc ^= ((pa >> PAGE_SHIFT) / (PQ_L2_SIZE * PQ_L2_SIZE)) & PQ_L2_MASK;
198 m->queue = m->pc + PQ_FREE;
199 KKASSERT(m->dirty == 0);
201 atomic_add_int(&vmstats.v_page_count, 1);
202 atomic_add_int(&vmstats.v_free_count, 1);
203 vpq = &vm_page_queues[m->queue];
204 if ((vpq->flipflop & 15) == 0) {
205 pmap_zero_page(VM_PAGE_TO_PHYS(m));
207 TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
208 atomic_add_int(&vm_page_zero_count, 1);
210 TAILQ_INSERT_HEAD(&vpq->pl, m, pageq);
221 * Initializes the resident memory module.
223 * Preallocates memory for critical VM structures and arrays prior to
224 * kernel_map becoming available.
226 * Memory is allocated from (virtual2_start, virtual2_end) if available,
227 * otherwise memory is allocated from (virtual_start, virtual_end).
229 * On x86-64 (virtual_start, virtual_end) is only 2GB and may not be
230 * large enough to hold vm_page_array & other structures for machines with
231 * large amounts of ram, so we want to use virtual2* when available.
234 vm_page_startup(void)
236 vm_offset_t vaddr = virtual2_start ? virtual2_start : virtual_start;
239 vm_paddr_t page_range;
246 vm_paddr_t biggestone, biggestsize;
253 vaddr = round_page(vaddr);
255 for (i = 0; phys_avail[i + 1]; i += 2) {
256 phys_avail[i] = round_page64(phys_avail[i]);
257 phys_avail[i + 1] = trunc_page64(phys_avail[i + 1]);
260 for (i = 0; phys_avail[i + 1]; i += 2) {
261 vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
263 if (size > biggestsize) {
271 end = phys_avail[biggestone+1];
272 end = trunc_page(end);
275 * Initialize the queue headers for the free queue, the active queue
276 * and the inactive queue.
279 vm_page_queue_init();
281 #if !defined(_KERNEL_VIRTUAL)
283 * VKERNELs don't support minidumps and as such don't need
286 * Allocate a bitmap to indicate that a random physical page
287 * needs to be included in a minidump.
289 * The amd64 port needs this to indicate which direct map pages
290 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
292 * However, i386 still needs this workspace internally within the
293 * minidump code. In theory, they are not needed on i386, but are
294 * included should the sf_buf code decide to use them.
296 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
297 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
298 end -= vm_page_dump_size;
299 vm_page_dump = (void *)pmap_map(&vaddr, end, end + vm_page_dump_size,
300 VM_PROT_READ | VM_PROT_WRITE);
301 bzero((void *)vm_page_dump, vm_page_dump_size);
305 * Compute the number of pages of memory that will be available for
306 * use (taking into account the overhead of a page structure per
309 first_page = phys_avail[0] / PAGE_SIZE;
310 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE - first_page;
311 npages = (total - (page_range * sizeof(struct vm_page))) / PAGE_SIZE;
314 * Initialize the mem entry structures now, and put them in the free
317 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
318 mapped = pmap_map(&vaddr, new_end, end,
319 VM_PROT_READ | VM_PROT_WRITE);
320 vm_page_array = (vm_page_t)mapped;
322 #if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
324 * since pmap_map on amd64 returns stuff out of a direct-map region,
325 * we have to manually add these pages to the minidump tracking so
326 * that they can be dumped, including the vm_page_array.
328 for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
333 * Clear all of the page structures
335 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
336 vm_page_array_size = page_range;
339 * Construct the free queue(s) in ascending order (by physical
340 * address) so that the first 16MB of physical memory is allocated
341 * last rather than first. On large-memory machines, this avoids
342 * the exhaustion of low physical memory before isa_dmainit has run.
344 vmstats.v_page_count = 0;
345 vmstats.v_free_count = 0;
346 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) {
351 last_pa = phys_avail[i + 1];
352 while (pa < last_pa && npages-- > 0) {
358 virtual2_start = vaddr;
360 virtual_start = vaddr;
364 * Scan comparison function for Red-Black tree scans. An inclusive
365 * (start,end) is expected. Other fields are not used.
368 rb_vm_page_scancmp(struct vm_page *p, void *data)
370 struct rb_vm_page_scan_info *info = data;
372 if (p->pindex < info->start_pindex)
374 if (p->pindex > info->end_pindex)
380 rb_vm_page_compare(struct vm_page *p1, struct vm_page *p2)
382 if (p1->pindex < p2->pindex)
384 if (p1->pindex > p2->pindex)
390 * Each page queue has its own spin lock, which is fairly optimal for
391 * allocating and freeing pages at least.
393 * The caller must hold the vm_page_spin_lock() before locking a vm_page's
394 * queue spinlock via this function. Also note that m->queue cannot change
395 * unless both the page and queue are locked.
399 _vm_page_queue_spin_lock(vm_page_t m)
404 if (queue != PQ_NONE) {
405 spin_lock(&vm_page_queues[queue].spin);
406 KKASSERT(queue == m->queue);
412 _vm_page_queue_spin_unlock(vm_page_t m)
418 if (queue != PQ_NONE)
419 spin_unlock(&vm_page_queues[queue].spin);
424 _vm_page_queues_spin_lock(u_short queue)
427 if (queue != PQ_NONE)
428 spin_lock(&vm_page_queues[queue].spin);
434 _vm_page_queues_spin_unlock(u_short queue)
437 if (queue != PQ_NONE)
438 spin_unlock(&vm_page_queues[queue].spin);
442 vm_page_queue_spin_lock(vm_page_t m)
444 _vm_page_queue_spin_lock(m);
448 vm_page_queues_spin_lock(u_short queue)
450 _vm_page_queues_spin_lock(queue);
454 vm_page_queue_spin_unlock(vm_page_t m)
456 _vm_page_queue_spin_unlock(m);
460 vm_page_queues_spin_unlock(u_short queue)
462 _vm_page_queues_spin_unlock(queue);
466 * This locks the specified vm_page and its queue in the proper order
467 * (page first, then queue). The queue may change so the caller must
472 _vm_page_and_queue_spin_lock(vm_page_t m)
474 vm_page_spin_lock(m);
475 _vm_page_queue_spin_lock(m);
480 _vm_page_and_queue_spin_unlock(vm_page_t m)
482 _vm_page_queues_spin_unlock(m->queue);
483 vm_page_spin_unlock(m);
487 vm_page_and_queue_spin_unlock(vm_page_t m)
489 _vm_page_and_queue_spin_unlock(m);
493 vm_page_and_queue_spin_lock(vm_page_t m)
495 _vm_page_and_queue_spin_lock(m);
499 * Helper function removes vm_page from its current queue.
500 * Returns the base queue the page used to be on.
502 * The vm_page and the queue must be spinlocked.
503 * This function will unlock the queue but leave the page spinlocked.
505 static __inline u_short
506 _vm_page_rem_queue_spinlocked(vm_page_t m)
508 struct vpgqueues *pq;
512 if (queue != PQ_NONE) {
513 pq = &vm_page_queues[queue];
514 TAILQ_REMOVE(&pq->pl, m, pageq);
515 atomic_add_int(pq->cnt, -1);
518 vm_page_queues_spin_unlock(queue);
519 if ((queue - m->pc) == PQ_FREE && (m->flags & PG_ZERO))
520 atomic_subtract_int(&vm_page_zero_count, 1);
521 if ((queue - m->pc) == PQ_CACHE || (queue - m->pc) == PQ_FREE)
522 return (queue - m->pc);
528 * Helper function places the vm_page on the specified queue.
530 * The vm_page must be spinlocked.
531 * This function will return with both the page and the queue locked.
534 _vm_page_add_queue_spinlocked(vm_page_t m, u_short queue, int athead)
536 struct vpgqueues *pq;
538 KKASSERT(m->queue == PQ_NONE);
540 if (queue != PQ_NONE) {
541 vm_page_queues_spin_lock(queue);
542 pq = &vm_page_queues[queue];
544 atomic_add_int(pq->cnt, 1);
548 * Put zero'd pages on the end ( where we look for zero'd pages
549 * first ) and non-zerod pages at the head.
551 if (queue - m->pc == PQ_FREE) {
552 if (m->flags & PG_ZERO) {
553 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
554 atomic_add_int(&vm_page_zero_count, 1);
556 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
559 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
561 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
563 /* leave the queue spinlocked */
568 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
569 * m->busy is zero. Returns TRUE if it had to sleep, FALSE if we
570 * did not. Only one sleep call will be made before returning.
572 * This function does NOT busy the page and on return the page is not
573 * guaranteed to be available.
576 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
584 if ((flags & PG_BUSY) == 0 &&
585 (also_m_busy == 0 || (flags & PG_SBUSY) == 0)) {
588 tsleep_interlock(m, 0);
589 if (atomic_cmpset_int(&m->flags, flags,
590 flags | PG_WANTED | PG_REFERENCED)) {
591 tsleep(m, PINTERLOCKED, msg, 0);
598 * Wait until PG_BUSY can be set, then set it. If also_m_busy is TRUE we
599 * also wait for m->busy to become 0 before setting PG_BUSY.
602 VM_PAGE_DEBUG_EXT(vm_page_busy_wait)(vm_page_t m,
603 int also_m_busy, const char *msg
611 if (flags & PG_BUSY) {
612 tsleep_interlock(m, 0);
613 if (atomic_cmpset_int(&m->flags, flags,
614 flags | PG_WANTED | PG_REFERENCED)) {
615 tsleep(m, PINTERLOCKED, msg, 0);
617 } else if (also_m_busy && (flags & PG_SBUSY)) {
618 tsleep_interlock(m, 0);
619 if (atomic_cmpset_int(&m->flags, flags,
620 flags | PG_WANTED | PG_REFERENCED)) {
621 tsleep(m, PINTERLOCKED, msg, 0);
624 if (atomic_cmpset_int(&m->flags, flags,
628 m->busy_line = lineno;
637 * Attempt to set PG_BUSY. If also_m_busy is TRUE we only succeed if m->busy
640 * Returns non-zero on failure.
643 VM_PAGE_DEBUG_EXT(vm_page_busy_try)(vm_page_t m, int also_m_busy
653 if (also_m_busy && (flags & PG_SBUSY))
655 if (atomic_cmpset_int(&m->flags, flags, flags | PG_BUSY)) {
658 m->busy_line = lineno;
666 * Clear the PG_BUSY flag and return non-zero to indicate to the caller
667 * that a wakeup() should be performed.
669 * The vm_page must be spinlocked and will remain spinlocked on return.
670 * The related queue must NOT be spinlocked (which could deadlock us).
676 _vm_page_wakeup(vm_page_t m)
683 if (atomic_cmpset_int(&m->flags, flags,
684 flags & ~(PG_BUSY | PG_WANTED))) {
688 return(flags & PG_WANTED);
692 * Clear the PG_BUSY flag and wakeup anyone waiting for the page. This
693 * is typically the last call you make on a page before moving onto
697 vm_page_wakeup(vm_page_t m)
699 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
700 vm_page_spin_lock(m);
701 if (_vm_page_wakeup(m)) {
702 vm_page_spin_unlock(m);
705 vm_page_spin_unlock(m);
710 * Holding a page keeps it from being reused. Other parts of the system
711 * can still disassociate the page from its current object and free it, or
712 * perform read or write I/O on it and/or otherwise manipulate the page,
713 * but if the page is held the VM system will leave the page and its data
714 * intact and not reuse the page for other purposes until the last hold
715 * reference is released. (see vm_page_wire() if you want to prevent the
716 * page from being disassociated from its object too).
718 * The caller must still validate the contents of the page and, if necessary,
719 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
720 * before manipulating the page.
722 * XXX get vm_page_spin_lock() here and move FREE->HOLD if necessary
725 vm_page_hold(vm_page_t m)
727 vm_page_spin_lock(m);
728 atomic_add_int(&m->hold_count, 1);
729 if (m->queue - m->pc == PQ_FREE) {
730 _vm_page_queue_spin_lock(m);
731 _vm_page_rem_queue_spinlocked(m);
732 _vm_page_add_queue_spinlocked(m, PQ_HOLD + m->pc, 0);
733 _vm_page_queue_spin_unlock(m);
735 vm_page_spin_unlock(m);
739 * The opposite of vm_page_hold(). A page can be freed while being held,
740 * which places it on the PQ_HOLD queue. If we are able to busy the page
741 * after the hold count drops to zero we will move the page to the
742 * appropriate PQ_FREE queue by calling vm_page_free_toq().
745 vm_page_unhold(vm_page_t m)
747 vm_page_spin_lock(m);
748 atomic_add_int(&m->hold_count, -1);
749 if (m->hold_count == 0 && m->queue - m->pc == PQ_HOLD) {
750 _vm_page_queue_spin_lock(m);
751 _vm_page_rem_queue_spinlocked(m);
752 _vm_page_add_queue_spinlocked(m, PQ_FREE + m->pc, 0);
753 _vm_page_queue_spin_unlock(m);
755 vm_page_spin_unlock(m);
759 * Inserts the given vm_page into the object and object list.
761 * The pagetables are not updated but will presumably fault the page
762 * in if necessary, or if a kernel page the caller will at some point
763 * enter the page into the kernel's pmap. We are not allowed to block
764 * here so we *can't* do this anyway.
766 * This routine may not block.
767 * This routine must be called with the vm_object held.
768 * This routine must be called with a critical section held.
770 * This routine returns TRUE if the page was inserted into the object
771 * successfully, and FALSE if the page already exists in the object.
774 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
776 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
777 if (m->object != NULL)
778 panic("vm_page_insert: already inserted");
780 object->generation++;
783 * Record the object/offset pair in this page and add the
784 * pv_list_count of the page to the object.
786 * The vm_page spin lock is required for interactions with the pmap.
788 vm_page_spin_lock(m);
791 if (vm_page_rb_tree_RB_INSERT(&object->rb_memq, m)) {
794 vm_page_spin_unlock(m);
797 object->resident_page_count++;
798 /* atomic_add_int(&object->agg_pv_list_count, m->md.pv_list_count); */
799 vm_page_spin_unlock(m);
802 * Since we are inserting a new and possibly dirty page,
803 * update the object's OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY flags.
805 if ((m->valid & m->dirty) || (m->flags & PG_WRITEABLE))
806 vm_object_set_writeable_dirty(object);
809 * Checks for a swap assignment and sets PG_SWAPPED if appropriate.
811 swap_pager_page_inserted(m);
816 * Removes the given vm_page_t from the (object,index) table
818 * The underlying pmap entry (if any) is NOT removed here.
819 * This routine may not block.
821 * The page must be BUSY and will remain BUSY on return.
822 * No other requirements.
824 * NOTE: FreeBSD side effect was to unbusy the page on return. We leave
828 vm_page_remove(vm_page_t m)
832 if (m->object == NULL) {
836 if ((m->flags & PG_BUSY) == 0)
837 panic("vm_page_remove: page not busy");
841 vm_object_hold(object);
844 * Remove the page from the object and update the object.
846 * The vm_page spin lock is required for interactions with the pmap.
848 vm_page_spin_lock(m);
849 vm_page_rb_tree_RB_REMOVE(&object->rb_memq, m);
850 object->resident_page_count--;
851 /* atomic_add_int(&object->agg_pv_list_count, -m->md.pv_list_count); */
853 vm_page_spin_unlock(m);
855 object->generation++;
857 vm_object_drop(object);
861 * Locate and return the page at (object, pindex), or NULL if the
862 * page could not be found.
864 * The caller must hold the vm_object token.
867 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
872 * Search the hash table for this object/offset pair
874 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
875 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
876 KKASSERT(m == NULL || (m->object == object && m->pindex == pindex));
881 VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_wait)(struct vm_object *object,
883 int also_m_busy, const char *msg
889 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
890 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
892 KKASSERT(m->object == object && m->pindex == pindex);
895 if (flags & PG_BUSY) {
896 tsleep_interlock(m, 0);
897 if (atomic_cmpset_int(&m->flags, flags,
898 flags | PG_WANTED | PG_REFERENCED)) {
899 tsleep(m, PINTERLOCKED, msg, 0);
900 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq,
903 } else if (also_m_busy && (flags & PG_SBUSY)) {
904 tsleep_interlock(m, 0);
905 if (atomic_cmpset_int(&m->flags, flags,
906 flags | PG_WANTED | PG_REFERENCED)) {
907 tsleep(m, PINTERLOCKED, msg, 0);
908 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq,
911 } else if (atomic_cmpset_int(&m->flags, flags,
915 m->busy_line = lineno;
924 * Attempt to lookup and busy a page.
926 * Returns NULL if the page could not be found
928 * Returns a vm_page and error == TRUE if the page exists but could not
931 * Returns a vm_page and error == FALSE on success.
934 VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_try)(struct vm_object *object,
936 int also_m_busy, int *errorp
942 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
943 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
946 KKASSERT(m->object == object && m->pindex == pindex);
949 if (flags & PG_BUSY) {
953 if (also_m_busy && (flags & PG_SBUSY)) {
957 if (atomic_cmpset_int(&m->flags, flags, flags | PG_BUSY)) {
960 m->busy_line = lineno;
969 * Caller must hold the related vm_object
972 vm_page_next(vm_page_t m)
976 next = vm_page_rb_tree_RB_NEXT(m);
977 if (next && next->pindex != m->pindex + 1)
985 * Move the given vm_page from its current object to the specified
986 * target object/offset. The page must be busy and will remain so
989 * new_object must be held.
990 * This routine might block. XXX ?
992 * NOTE: Swap associated with the page must be invalidated by the move. We
993 * have to do this for several reasons: (1) we aren't freeing the
994 * page, (2) we are dirtying the page, (3) the VM system is probably
995 * moving the page from object A to B, and will then later move
996 * the backing store from A to B and we can't have a conflict.
998 * NOTE: We *always* dirty the page. It is necessary both for the
999 * fact that we moved it, and because we may be invalidating
1000 * swap. If the page is on the cache, we have to deactivate it
1001 * or vm_page_dirty() will panic. Dirty pages are not allowed
1005 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1007 KKASSERT(m->flags & PG_BUSY);
1008 ASSERT_LWKT_TOKEN_HELD(vm_object_token(new_object));
1010 ASSERT_LWKT_TOKEN_HELD(vm_object_token(m->object));
1013 if (vm_page_insert(m, new_object, new_pindex) == FALSE) {
1014 panic("vm_page_rename: target exists (%p,%"PRIu64")",
1015 new_object, new_pindex);
1017 if (m->queue - m->pc == PQ_CACHE)
1018 vm_page_deactivate(m);
1023 * vm_page_unqueue() without any wakeup. This routine is used when a page
1024 * is being moved between queues or otherwise is to remain BUSYied by the
1027 * This routine may not block.
1030 vm_page_unqueue_nowakeup(vm_page_t m)
1032 vm_page_and_queue_spin_lock(m);
1033 (void)_vm_page_rem_queue_spinlocked(m);
1034 vm_page_spin_unlock(m);
1038 * vm_page_unqueue() - Remove a page from its queue, wakeup the pagedemon
1041 * This routine may not block.
1044 vm_page_unqueue(vm_page_t m)
1048 vm_page_and_queue_spin_lock(m);
1049 queue = _vm_page_rem_queue_spinlocked(m);
1050 if (queue == PQ_FREE || queue == PQ_CACHE) {
1051 vm_page_spin_unlock(m);
1052 pagedaemon_wakeup();
1054 vm_page_spin_unlock(m);
1059 * vm_page_list_find()
1061 * Find a page on the specified queue with color optimization.
1063 * The page coloring optimization attempts to locate a page that does
1064 * not overload other nearby pages in the object in the cpu's L1 or L2
1065 * caches. We need this optimization because cpu caches tend to be
1066 * physical caches, while object spaces tend to be virtual.
1068 * On MP systems each PQ_FREE and PQ_CACHE color queue has its own spinlock
1069 * and the algorithm is adjusted to localize allocations on a per-core basis.
1070 * This is done by 'twisting' the colors.
1072 * The page is returned spinlocked and removed from its queue (it will
1073 * be on PQ_NONE), or NULL. The page is not PG_BUSY'd. The caller
1074 * is responsible for dealing with the busy-page case (usually by
1075 * deactivating the page and looping).
1077 * NOTE: This routine is carefully inlined. A non-inlined version
1078 * is available for outside callers but the only critical path is
1079 * from within this source file.
1081 * NOTE: This routine assumes that the vm_pages found in PQ_CACHE and PQ_FREE
1082 * represent stable storage, allowing us to order our locks vm_page
1083 * first, then queue.
1087 _vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
1093 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
1095 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
1097 m = _vm_page_list_find2(basequeue, index);
1100 vm_page_and_queue_spin_lock(m);
1101 if (m->queue == basequeue + index) {
1102 _vm_page_rem_queue_spinlocked(m);
1103 /* vm_page_t spin held, no queue spin */
1106 vm_page_and_queue_spin_unlock(m);
1112 _vm_page_list_find2(int basequeue, int index)
1116 struct vpgqueues *pq;
1118 pq = &vm_page_queues[basequeue];
1121 * Note that for the first loop, index+i and index-i wind up at the
1122 * same place. Even though this is not totally optimal, we've already
1123 * blown it by missing the cache case so we do not care.
1125 for (i = PQ_L2_SIZE / 2; i > 0; --i) {
1127 m = TAILQ_FIRST(&pq[(index + i) & PQ_L2_MASK].pl);
1129 _vm_page_and_queue_spin_lock(m);
1131 basequeue + ((index + i) & PQ_L2_MASK)) {
1132 _vm_page_rem_queue_spinlocked(m);
1135 _vm_page_and_queue_spin_unlock(m);
1138 m = TAILQ_FIRST(&pq[(index - i) & PQ_L2_MASK].pl);
1140 _vm_page_and_queue_spin_lock(m);
1142 basequeue + ((index - i) & PQ_L2_MASK)) {
1143 _vm_page_rem_queue_spinlocked(m);
1146 _vm_page_and_queue_spin_unlock(m);
1156 * Returns a vm_page candidate for allocation. The page is not busied so
1157 * it can move around. The caller must busy the page (and typically
1158 * deactivate it if it cannot be busied!)
1160 * Returns a spinlocked vm_page that has been removed from its queue.
1163 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
1165 return(_vm_page_list_find(basequeue, index, prefer_zero));
1169 * Find a page on the cache queue with color optimization, remove it
1170 * from the queue, and busy it. The returned page will not be spinlocked.
1172 * A candidate failure will be deactivated. Candidates can fail due to
1173 * being busied by someone else, in which case they will be deactivated.
1175 * This routine may not block.
1179 vm_page_select_cache(u_short pg_color)
1184 m = _vm_page_list_find(PQ_CACHE, pg_color & PQ_L2_MASK, FALSE);
1188 * (m) has been removed from its queue and spinlocked
1190 if (vm_page_busy_try(m, TRUE)) {
1191 _vm_page_deactivate_locked(m, 0);
1192 vm_page_spin_unlock(m);
1194 kprintf("Warning: busy page %p found in cache\n", m);
1198 * We successfully busied the page
1200 if ((m->flags & PG_UNMANAGED) == 0 &&
1201 m->hold_count == 0 &&
1202 m->wire_count == 0) {
1203 vm_page_spin_unlock(m);
1204 pagedaemon_wakeup();
1207 _vm_page_deactivate_locked(m, 0);
1208 if (_vm_page_wakeup(m)) {
1209 vm_page_spin_unlock(m);
1212 vm_page_spin_unlock(m);
1220 * Find a free or zero page, with specified preference. We attempt to
1221 * inline the nominal case and fall back to _vm_page_select_free()
1222 * otherwise. A busied page is removed from the queue and returned.
1224 * This routine may not block.
1226 static __inline vm_page_t
1227 vm_page_select_free(u_short pg_color, boolean_t prefer_zero)
1232 m = _vm_page_list_find(PQ_FREE, pg_color & PQ_L2_MASK,
1236 if (vm_page_busy_try(m, TRUE)) {
1238 * Various mechanisms such as a pmap_collect can
1239 * result in a busy page on the free queue. We
1240 * have to move the page out of the way so we can
1241 * retry the allocation. If the other thread is not
1242 * allocating the page then m->valid will remain 0 and
1243 * the pageout daemon will free the page later on.
1245 * Since we could not busy the page, however, we
1246 * cannot make assumptions as to whether the page
1247 * will be allocated by the other thread or not,
1248 * so all we can do is deactivate it to move it out
1249 * of the way. In particular, if the other thread
1250 * wires the page it may wind up on the inactive
1251 * queue and the pageout daemon will have to deal
1252 * with that case too.
1254 _vm_page_deactivate_locked(m, 0);
1255 vm_page_spin_unlock(m);
1257 kprintf("Warning: busy page %p found in cache\n", m);
1261 * Theoretically if we are able to busy the page
1262 * atomic with the queue removal (using the vm_page
1263 * lock) nobody else should be able to mess with the
1266 KKASSERT((m->flags & PG_UNMANAGED) == 0);
1267 KKASSERT(m->hold_count == 0);
1268 KKASSERT(m->wire_count == 0);
1269 vm_page_spin_unlock(m);
1270 pagedaemon_wakeup();
1272 /* return busied and removed page */
1280 * This implements a per-cpu cache of free, zero'd, ready-to-go pages.
1281 * The idea is to populate this cache prior to acquiring any locks so
1282 * we don't wind up potentially zeroing VM pages (under heavy loads) while
1283 * holding potentialy contending locks.
1285 * Note that we allocate the page uninserted into anything and use a pindex
1286 * of 0, the vm_page_alloc() will effectively add gd_cpuid so these
1287 * allocations should wind up being uncontended. However, we still want
1288 * to rove across PQ_L2_SIZE.
1291 vm_page_pcpu_cache(void)
1294 globaldata_t gd = mycpu;
1297 if (gd->gd_vmpg_count < GD_MINVMPG) {
1299 while (gd->gd_vmpg_count < GD_MAXVMPG) {
1300 m = vm_page_alloc(NULL, ticks & ~ncpus2_mask,
1301 VM_ALLOC_NULL_OK | VM_ALLOC_NORMAL |
1302 VM_ALLOC_NULL_OK | VM_ALLOC_ZERO);
1303 if (gd->gd_vmpg_count < GD_MAXVMPG) {
1304 if ((m->flags & PG_ZERO) == 0) {
1305 pmap_zero_page(VM_PAGE_TO_PHYS(m));
1306 vm_page_flag_set(m, PG_ZERO);
1308 gd->gd_vmpg_array[gd->gd_vmpg_count++] = m;
1321 * Allocate and return a memory cell associated with this VM object/offset
1322 * pair. If object is NULL an unassociated page will be allocated.
1324 * The returned page will be busied and removed from its queues. This
1325 * routine can block and may return NULL if a race occurs and the page
1326 * is found to already exist at the specified (object, pindex).
1328 * VM_ALLOC_NORMAL allow use of cache pages, nominal free drain
1329 * VM_ALLOC_QUICK like normal but cannot use cache
1330 * VM_ALLOC_SYSTEM greater free drain
1331 * VM_ALLOC_INTERRUPT allow free list to be completely drained
1332 * VM_ALLOC_ZERO advisory request for pre-zero'd page only
1333 * VM_ALLOC_FORCE_ZERO advisory request for pre-zero'd page only
1334 * VM_ALLOC_NULL_OK ok to return NULL on insertion collision
1335 * (see vm_page_grab())
1336 * VM_ALLOC_USE_GD ok to use per-gd cache
1338 * The object must be held if not NULL
1339 * This routine may not block
1341 * Additional special handling is required when called from an interrupt
1342 * (VM_ALLOC_INTERRUPT). We are not allowed to mess with the page cache
1346 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int page_req)
1349 globaldata_t gd = mycpu;
1356 * Special per-cpu free VM page cache. The pages are pre-busied
1357 * and pre-zerod for us.
1359 if (gd->gd_vmpg_count && (page_req & VM_ALLOC_USE_GD)) {
1361 if (gd->gd_vmpg_count) {
1362 m = gd->gd_vmpg_array[--gd->gd_vmpg_count];
1373 * Cpu twist - cpu localization algorithm
1376 pg_color = gd->gd_cpuid + (pindex & ~ncpus_fit_mask) +
1377 (object->pg_color & ~ncpus_fit_mask);
1379 pg_color = gd->gd_cpuid + (pindex & ~ncpus_fit_mask);
1383 * Normal page coloring algorithm
1386 pg_color = object->pg_color + pindex;
1392 (VM_ALLOC_NORMAL|VM_ALLOC_QUICK|
1393 VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
1396 * Certain system threads (pageout daemon, buf_daemon's) are
1397 * allowed to eat deeper into the free page list.
1399 if (curthread->td_flags & TDF_SYSTHREAD)
1400 page_req |= VM_ALLOC_SYSTEM;
1403 if (vmstats.v_free_count > vmstats.v_free_reserved ||
1404 ((page_req & VM_ALLOC_INTERRUPT) && vmstats.v_free_count > 0) ||
1405 ((page_req & VM_ALLOC_SYSTEM) && vmstats.v_cache_count == 0 &&
1406 vmstats.v_free_count > vmstats.v_interrupt_free_min)
1409 * The free queue has sufficient free pages to take one out.
1411 if (page_req & (VM_ALLOC_ZERO | VM_ALLOC_FORCE_ZERO))
1412 m = vm_page_select_free(pg_color, TRUE);
1414 m = vm_page_select_free(pg_color, FALSE);
1415 } else if (page_req & VM_ALLOC_NORMAL) {
1417 * Allocatable from the cache (non-interrupt only). On
1418 * success, we must free the page and try again, thus
1419 * ensuring that vmstats.v_*_free_min counters are replenished.
1422 if (curthread->td_preempted) {
1423 kprintf("vm_page_alloc(): warning, attempt to allocate"
1424 " cache page from preempting interrupt\n");
1427 m = vm_page_select_cache(pg_color);
1430 m = vm_page_select_cache(pg_color);
1433 * On success move the page into the free queue and loop.
1436 KASSERT(m->dirty == 0,
1437 ("Found dirty cache page %p", m));
1438 vm_page_protect(m, VM_PROT_NONE);
1444 * On failure return NULL
1446 #if defined(DIAGNOSTIC)
1447 if (vmstats.v_cache_count > 0)
1448 kprintf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", vmstats.v_cache_count);
1450 vm_pageout_deficit++;
1451 pagedaemon_wakeup();
1455 * No pages available, wakeup the pageout daemon and give up.
1457 vm_pageout_deficit++;
1458 pagedaemon_wakeup();
1463 * v_free_count can race so loop if we don't find the expected
1470 * Good page found. The page has already been busied for us and
1471 * removed from its queues.
1473 KASSERT(m->dirty == 0,
1474 ("vm_page_alloc: free/cache page %p was dirty", m));
1475 KKASSERT(m->queue == PQ_NONE);
1481 * Initialize the structure, inheriting some flags but clearing
1482 * all the rest. The page has already been busied for us.
1484 vm_page_flag_clear(m, ~(PG_ZERO | PG_BUSY | PG_SBUSY));
1485 KKASSERT(m->wire_count == 0);
1486 KKASSERT(m->busy == 0);
1491 * Caller must be holding the object lock (asserted by
1492 * vm_page_insert()).
1494 * NOTE: Inserting a page here does not insert it into any pmaps
1495 * (which could cause us to block allocating memory).
1497 * NOTE: If no object an unassociated page is allocated, m->pindex
1498 * can be used by the caller for any purpose.
1501 if (vm_page_insert(m, object, pindex) == FALSE) {
1502 kprintf("PAGE RACE (%p:%d,%"PRIu64")\n",
1503 object, object->type, pindex);
1506 if ((page_req & VM_ALLOC_NULL_OK) == 0)
1514 * Don't wakeup too often - wakeup the pageout daemon when
1515 * we would be nearly out of memory.
1517 pagedaemon_wakeup();
1520 * A PG_BUSY page is returned.
1526 * Wait for sufficient free memory for nominal heavy memory use kernel
1530 vm_wait_nominal(void)
1532 while (vm_page_count_min(0))
1537 * Test if vm_wait_nominal() would block.
1540 vm_test_nominal(void)
1542 if (vm_page_count_min(0))
1548 * Block until free pages are available for allocation, called in various
1549 * places before memory allocations.
1551 * The caller may loop if vm_page_count_min() == FALSE so we cannot be
1552 * more generous then that.
1558 * never wait forever
1562 lwkt_gettoken(&vm_token);
1564 if (curthread == pagethread) {
1566 * The pageout daemon itself needs pages, this is bad.
1568 if (vm_page_count_min(0)) {
1569 vm_pageout_pages_needed = 1;
1570 tsleep(&vm_pageout_pages_needed, 0, "VMWait", timo);
1574 * Wakeup the pageout daemon if necessary and wait.
1576 if (vm_page_count_target()) {
1577 if (vm_pages_needed == 0) {
1578 vm_pages_needed = 1;
1579 wakeup(&vm_pages_needed);
1581 ++vm_pages_waiting; /* SMP race ok */
1582 tsleep(&vmstats.v_free_count, 0, "vmwait", timo);
1585 lwkt_reltoken(&vm_token);
1589 * Block until free pages are available for allocation
1591 * Called only from vm_fault so that processes page faulting can be
1598 * Wakeup the pageout daemon if necessary and wait.
1600 if (vm_page_count_target()) {
1601 lwkt_gettoken(&vm_token);
1602 if (vm_page_count_target()) {
1603 if (vm_pages_needed == 0) {
1604 vm_pages_needed = 1;
1605 wakeup(&vm_pages_needed);
1607 ++vm_pages_waiting; /* SMP race ok */
1608 tsleep(&vmstats.v_free_count, 0, "pfault", hz);
1610 lwkt_reltoken(&vm_token);
1615 * Put the specified page on the active list (if appropriate). Ensure
1616 * that act_count is at least ACT_INIT but do not otherwise mess with it.
1618 * The caller should be holding the page busied ? XXX
1619 * This routine may not block.
1622 vm_page_activate(vm_page_t m)
1626 vm_page_spin_lock(m);
1627 if (m->queue - m->pc != PQ_ACTIVE) {
1628 _vm_page_queue_spin_lock(m);
1629 oqueue = _vm_page_rem_queue_spinlocked(m);
1630 /* page is left spinlocked, queue is unlocked */
1632 if (oqueue == PQ_CACHE)
1633 mycpu->gd_cnt.v_reactivated++;
1634 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1635 if (m->act_count < ACT_INIT)
1636 m->act_count = ACT_INIT;
1637 _vm_page_add_queue_spinlocked(m, PQ_ACTIVE + m->pc, 0);
1639 _vm_page_and_queue_spin_unlock(m);
1640 if (oqueue == PQ_CACHE || oqueue == PQ_FREE)
1641 pagedaemon_wakeup();
1643 if (m->act_count < ACT_INIT)
1644 m->act_count = ACT_INIT;
1645 vm_page_spin_unlock(m);
1650 * Helper routine for vm_page_free_toq() and vm_page_cache(). This
1651 * routine is called when a page has been added to the cache or free
1654 * This routine may not block.
1656 static __inline void
1657 vm_page_free_wakeup(void)
1660 * If the pageout daemon itself needs pages, then tell it that
1661 * there are some free.
1663 if (vm_pageout_pages_needed &&
1664 vmstats.v_cache_count + vmstats.v_free_count >=
1665 vmstats.v_pageout_free_min
1667 wakeup(&vm_pageout_pages_needed);
1668 vm_pageout_pages_needed = 0;
1672 * Wakeup processes that are waiting on memory.
1674 * NOTE: vm_paging_target() is the pageout daemon's target, while
1675 * vm_page_count_target() is somewhere inbetween. We want
1676 * to wake processes up prior to the pageout daemon reaching
1677 * its target to provide some hysteresis.
1679 if (vm_pages_waiting) {
1680 if (!vm_page_count_target()) {
1682 * Plenty of pages are free, wakeup everyone.
1684 vm_pages_waiting = 0;
1685 wakeup(&vmstats.v_free_count);
1686 ++mycpu->gd_cnt.v_ppwakeups;
1687 } else if (!vm_page_count_min(0)) {
1689 * Some pages are free, wakeup someone.
1691 int wcount = vm_pages_waiting;
1694 vm_pages_waiting = wcount;
1695 wakeup_one(&vmstats.v_free_count);
1696 ++mycpu->gd_cnt.v_ppwakeups;
1702 * Returns the given page to the PQ_FREE or PQ_HOLD list and disassociates
1703 * it from its VM object.
1705 * The vm_page must be PG_BUSY on entry. PG_BUSY will be released on
1706 * return (the page will have been freed).
1709 vm_page_free_toq(vm_page_t m)
1711 mycpu->gd_cnt.v_tfree++;
1712 KKASSERT((m->flags & PG_MAPPED) == 0);
1713 KKASSERT(m->flags & PG_BUSY);
1715 if (m->busy || ((m->queue - m->pc) == PQ_FREE)) {
1717 "vm_page_free: pindex(%lu), busy(%d), PG_BUSY(%d), hold(%d)\n",
1718 (u_long)m->pindex, m->busy, (m->flags & PG_BUSY) ? 1 : 0,
1720 if ((m->queue - m->pc) == PQ_FREE)
1721 panic("vm_page_free: freeing free page");
1723 panic("vm_page_free: freeing busy page");
1727 * Remove from object, spinlock the page and its queues and
1728 * remove from any queue. No queue spinlock will be held
1729 * after this section (because the page was removed from any
1733 vm_page_and_queue_spin_lock(m);
1734 _vm_page_rem_queue_spinlocked(m);
1737 * No further management of fictitious pages occurs beyond object
1738 * and queue removal.
1740 if ((m->flags & PG_FICTITIOUS) != 0) {
1741 vm_page_spin_unlock(m);
1749 if (m->wire_count != 0) {
1750 if (m->wire_count > 1) {
1752 "vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1753 m->wire_count, (long)m->pindex);
1755 panic("vm_page_free: freeing wired page");
1759 * Clear the UNMANAGED flag when freeing an unmanaged page.
1761 if (m->flags & PG_UNMANAGED) {
1762 vm_page_flag_clear(m, PG_UNMANAGED);
1765 if (m->hold_count != 0) {
1766 vm_page_flag_clear(m, PG_ZERO);
1767 _vm_page_add_queue_spinlocked(m, PQ_HOLD + m->pc, 0);
1769 _vm_page_add_queue_spinlocked(m, PQ_FREE + m->pc, 0);
1773 * This sequence allows us to clear PG_BUSY while still holding
1774 * its spin lock, which reduces contention vs allocators. We
1775 * must not leave the queue locked or _vm_page_wakeup() may
1778 _vm_page_queue_spin_unlock(m);
1779 if (_vm_page_wakeup(m)) {
1780 vm_page_spin_unlock(m);
1783 vm_page_spin_unlock(m);
1785 vm_page_free_wakeup();
1789 * vm_page_free_fromq_fast()
1791 * Remove a non-zero page from one of the free queues; the page is removed for
1792 * zeroing, so do not issue a wakeup.
1795 vm_page_free_fromq_fast(void)
1801 for (i = 0; i < PQ_L2_SIZE; ++i) {
1802 m = vm_page_list_find(PQ_FREE, qi, FALSE);
1803 /* page is returned spinlocked and removed from its queue */
1805 if (vm_page_busy_try(m, TRUE)) {
1807 * We were unable to busy the page, deactivate
1810 _vm_page_deactivate_locked(m, 0);
1811 vm_page_spin_unlock(m);
1812 } else if (m->flags & PG_ZERO) {
1814 * The page is PG_ZERO, requeue it and loop
1816 _vm_page_add_queue_spinlocked(m,
1819 vm_page_queue_spin_unlock(m);
1820 if (_vm_page_wakeup(m)) {
1821 vm_page_spin_unlock(m);
1824 vm_page_spin_unlock(m);
1828 * The page is not PG_ZERO'd so return it.
1830 vm_page_spin_unlock(m);
1831 KKASSERT((m->flags & PG_UNMANAGED) == 0);
1832 KKASSERT(m->hold_count == 0);
1833 KKASSERT(m->wire_count == 0);
1838 qi = (qi + PQ_PRIME2) & PQ_L2_MASK;
1844 * vm_page_unmanage()
1846 * Prevent PV management from being done on the page. The page is
1847 * removed from the paging queues as if it were wired, and as a
1848 * consequence of no longer being managed the pageout daemon will not
1849 * touch it (since there is no way to locate the pte mappings for the
1850 * page). madvise() calls that mess with the pmap will also no longer
1851 * operate on the page.
1853 * Beyond that the page is still reasonably 'normal'. Freeing the page
1854 * will clear the flag.
1856 * This routine is used by OBJT_PHYS objects - objects using unswappable
1857 * physical memory as backing store rather then swap-backed memory and
1858 * will eventually be extended to support 4MB unmanaged physical
1861 * Caller must be holding the page busy.
1864 vm_page_unmanage(vm_page_t m)
1866 KKASSERT(m->flags & PG_BUSY);
1867 if ((m->flags & PG_UNMANAGED) == 0) {
1868 if (m->wire_count == 0)
1871 vm_page_flag_set(m, PG_UNMANAGED);
1875 * Mark this page as wired down by yet another map, removing it from
1876 * paging queues as necessary.
1878 * Caller must be holding the page busy.
1881 vm_page_wire(vm_page_t m)
1884 * Only bump the wire statistics if the page is not already wired,
1885 * and only unqueue the page if it is on some queue (if it is unmanaged
1886 * it is already off the queues). Don't do anything with fictitious
1887 * pages because they are always wired.
1889 KKASSERT(m->flags & PG_BUSY);
1890 if ((m->flags & PG_FICTITIOUS) == 0) {
1891 if (atomic_fetchadd_int(&m->wire_count, 1) == 0) {
1892 if ((m->flags & PG_UNMANAGED) == 0)
1894 atomic_add_int(&vmstats.v_wire_count, 1);
1896 KASSERT(m->wire_count != 0,
1897 ("vm_page_wire: wire_count overflow m=%p", m));
1902 * Release one wiring of this page, potentially enabling it to be paged again.
1904 * Many pages placed on the inactive queue should actually go
1905 * into the cache, but it is difficult to figure out which. What
1906 * we do instead, if the inactive target is well met, is to put
1907 * clean pages at the head of the inactive queue instead of the tail.
1908 * This will cause them to be moved to the cache more quickly and
1909 * if not actively re-referenced, freed more quickly. If we just
1910 * stick these pages at the end of the inactive queue, heavy filesystem
1911 * meta-data accesses can cause an unnecessary paging load on memory bound
1912 * processes. This optimization causes one-time-use metadata to be
1913 * reused more quickly.
1915 * BUT, if we are in a low-memory situation we have no choice but to
1916 * put clean pages on the cache queue.
1918 * A number of routines use vm_page_unwire() to guarantee that the page
1919 * will go into either the inactive or active queues, and will NEVER
1920 * be placed in the cache - for example, just after dirtying a page.
1921 * dirty pages in the cache are not allowed.
1923 * The page queues must be locked.
1924 * This routine may not block.
1927 vm_page_unwire(vm_page_t m, int activate)
1929 KKASSERT(m->flags & PG_BUSY);
1930 if (m->flags & PG_FICTITIOUS) {
1932 } else if (m->wire_count <= 0) {
1933 panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
1935 if (atomic_fetchadd_int(&m->wire_count, -1) == 1) {
1936 atomic_add_int(&vmstats.v_wire_count, -1);
1937 if (m->flags & PG_UNMANAGED) {
1939 } else if (activate) {
1940 vm_page_spin_lock(m);
1941 _vm_page_add_queue_spinlocked(m,
1942 PQ_ACTIVE + m->pc, 0);
1943 _vm_page_and_queue_spin_unlock(m);
1945 vm_page_spin_lock(m);
1946 vm_page_flag_clear(m, PG_WINATCFLS);
1947 _vm_page_add_queue_spinlocked(m,
1948 PQ_INACTIVE + m->pc, 0);
1949 ++vm_swapcache_inactive_heuristic;
1950 _vm_page_and_queue_spin_unlock(m);
1957 * Move the specified page to the inactive queue. If the page has
1958 * any associated swap, the swap is deallocated.
1960 * Normally athead is 0 resulting in LRU operation. athead is set
1961 * to 1 if we want this page to be 'as if it were placed in the cache',
1962 * except without unmapping it from the process address space.
1964 * vm_page's spinlock must be held on entry and will remain held on return.
1965 * This routine may not block.
1968 _vm_page_deactivate_locked(vm_page_t m, int athead)
1973 * Ignore if already inactive.
1975 if (m->queue - m->pc == PQ_INACTIVE)
1977 _vm_page_queue_spin_lock(m);
1978 oqueue = _vm_page_rem_queue_spinlocked(m);
1980 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1981 if (oqueue == PQ_CACHE)
1982 mycpu->gd_cnt.v_reactivated++;
1983 vm_page_flag_clear(m, PG_WINATCFLS);
1984 _vm_page_add_queue_spinlocked(m, PQ_INACTIVE + m->pc, athead);
1986 ++vm_swapcache_inactive_heuristic;
1988 _vm_page_queue_spin_unlock(m);
1989 /* leaves vm_page spinlocked */
1993 * Attempt to deactivate a page.
1998 vm_page_deactivate(vm_page_t m)
2000 vm_page_spin_lock(m);
2001 _vm_page_deactivate_locked(m, 0);
2002 vm_page_spin_unlock(m);
2006 vm_page_deactivate_locked(vm_page_t m)
2008 _vm_page_deactivate_locked(m, 0);
2012 * Attempt to move a page to PQ_CACHE.
2014 * Returns 0 on failure, 1 on success
2016 * The page should NOT be busied by the caller. This function will validate
2017 * whether the page can be safely moved to the cache.
2020 vm_page_try_to_cache(vm_page_t m)
2022 vm_page_spin_lock(m);
2023 if (vm_page_busy_try(m, TRUE)) {
2024 vm_page_spin_unlock(m);
2027 if (m->dirty || m->hold_count || m->wire_count ||
2028 (m->flags & PG_UNMANAGED)) {
2029 if (_vm_page_wakeup(m)) {
2030 vm_page_spin_unlock(m);
2033 vm_page_spin_unlock(m);
2037 vm_page_spin_unlock(m);
2040 * Page busied by us and no longer spinlocked. Dirty pages cannot
2041 * be moved to the cache.
2043 vm_page_test_dirty(m);
2053 * Attempt to free the page. If we cannot free it, we do nothing.
2054 * 1 is returned on success, 0 on failure.
2059 vm_page_try_to_free(vm_page_t m)
2061 vm_page_spin_lock(m);
2062 if (vm_page_busy_try(m, TRUE)) {
2063 vm_page_spin_unlock(m);
2066 if (m->dirty || m->hold_count || m->wire_count ||
2067 (m->flags & PG_UNMANAGED)) {
2068 if (_vm_page_wakeup(m)) {
2069 vm_page_spin_unlock(m);
2072 vm_page_spin_unlock(m);
2076 vm_page_spin_unlock(m);
2079 * Page busied by us and no longer spinlocked. Dirty pages will
2080 * not be freed by this function. We have to re-test the
2081 * dirty bit after cleaning out the pmaps.
2083 vm_page_test_dirty(m);
2088 vm_page_protect(m, VM_PROT_NONE);
2100 * Put the specified page onto the page cache queue (if appropriate).
2102 * The page must be busy, and this routine will release the busy and
2103 * possibly even free the page.
2106 vm_page_cache(vm_page_t m)
2108 if ((m->flags & PG_UNMANAGED) || m->busy ||
2109 m->wire_count || m->hold_count) {
2110 kprintf("vm_page_cache: attempting to cache busy/held page\n");
2116 * Already in the cache (and thus not mapped)
2118 if ((m->queue - m->pc) == PQ_CACHE) {
2119 KKASSERT((m->flags & PG_MAPPED) == 0);
2125 * Caller is required to test m->dirty, but note that the act of
2126 * removing the page from its maps can cause it to become dirty
2127 * on an SMP system due to another cpu running in usermode.
2130 panic("vm_page_cache: caching a dirty page, pindex: %ld",
2135 * Remove all pmaps and indicate that the page is not
2136 * writeable or mapped. Our vm_page_protect() call may
2137 * have blocked (especially w/ VM_PROT_NONE), so recheck
2140 vm_page_protect(m, VM_PROT_NONE);
2141 if ((m->flags & (PG_UNMANAGED|PG_MAPPED)) || m->busy ||
2142 m->wire_count || m->hold_count) {
2144 } else if (m->dirty) {
2145 vm_page_deactivate(m);
2148 _vm_page_and_queue_spin_lock(m);
2149 _vm_page_rem_queue_spinlocked(m);
2150 _vm_page_add_queue_spinlocked(m, PQ_CACHE + m->pc, 0);
2151 _vm_page_queue_spin_unlock(m);
2152 if (_vm_page_wakeup(m)) {
2153 vm_page_spin_unlock(m);
2156 vm_page_spin_unlock(m);
2158 vm_page_free_wakeup();
2163 * vm_page_dontneed()
2165 * Cache, deactivate, or do nothing as appropriate. This routine
2166 * is typically used by madvise() MADV_DONTNEED.
2168 * Generally speaking we want to move the page into the cache so
2169 * it gets reused quickly. However, this can result in a silly syndrome
2170 * due to the page recycling too quickly. Small objects will not be
2171 * fully cached. On the otherhand, if we move the page to the inactive
2172 * queue we wind up with a problem whereby very large objects
2173 * unnecessarily blow away our inactive and cache queues.
2175 * The solution is to move the pages based on a fixed weighting. We
2176 * either leave them alone, deactivate them, or move them to the cache,
2177 * where moving them to the cache has the highest weighting.
2178 * By forcing some pages into other queues we eventually force the
2179 * system to balance the queues, potentially recovering other unrelated
2180 * space from active. The idea is to not force this to happen too
2183 * The page must be busied.
2186 vm_page_dontneed(vm_page_t m)
2188 static int dnweight;
2195 * occassionally leave the page alone
2197 if ((dnw & 0x01F0) == 0 ||
2198 m->queue - m->pc == PQ_INACTIVE ||
2199 m->queue - m->pc == PQ_CACHE
2201 if (m->act_count >= ACT_INIT)
2207 * If vm_page_dontneed() is inactivating a page, it must clear
2208 * the referenced flag; otherwise the pagedaemon will see references
2209 * on the page in the inactive queue and reactivate it. Until the
2210 * page can move to the cache queue, madvise's job is not done.
2212 vm_page_flag_clear(m, PG_REFERENCED);
2213 pmap_clear_reference(m);
2216 vm_page_test_dirty(m);
2218 if (m->dirty || (dnw & 0x0070) == 0) {
2220 * Deactivate the page 3 times out of 32.
2225 * Cache the page 28 times out of every 32. Note that
2226 * the page is deactivated instead of cached, but placed
2227 * at the head of the queue instead of the tail.
2231 vm_page_spin_lock(m);
2232 _vm_page_deactivate_locked(m, head);
2233 vm_page_spin_unlock(m);
2237 * These routines manipulate the 'soft busy' count for a page. A soft busy
2238 * is almost like PG_BUSY except that it allows certain compatible operations
2239 * to occur on the page while it is busy. For example, a page undergoing a
2240 * write can still be mapped read-only.
2242 * Because vm_pages can overlap buffers m->busy can be > 1. m->busy is only
2243 * adjusted while the vm_page is PG_BUSY so the flash will occur when the
2244 * busy bit is cleared.
2247 vm_page_io_start(vm_page_t m)
2249 KASSERT(m->flags & PG_BUSY, ("vm_page_io_start: page not busy!!!"));
2250 atomic_add_char(&m->busy, 1);
2251 vm_page_flag_set(m, PG_SBUSY);
2255 vm_page_io_finish(vm_page_t m)
2257 KASSERT(m->flags & PG_BUSY, ("vm_page_io_finish: page not busy!!!"));
2258 atomic_subtract_char(&m->busy, 1);
2260 vm_page_flag_clear(m, PG_SBUSY);
2264 * Grab a page, blocking if it is busy and allocating a page if necessary.
2265 * A busy page is returned or NULL. The page may or may not be valid and
2266 * might not be on a queue (the caller is responsible for the disposition of
2269 * If VM_ALLOC_ZERO is specified and the grab must allocate a new page, the
2270 * page will be zero'd and marked valid.
2272 * If VM_ALLOC_FORCE_ZERO is specified the page will be zero'd and marked
2273 * valid even if it already exists.
2275 * If VM_ALLOC_RETRY is specified this routine will never return NULL. Also
2276 * note that VM_ALLOC_NORMAL must be specified if VM_ALLOC_RETRY is specified.
2277 * VM_ALLOC_NULL_OK is implied when VM_ALLOC_RETRY is specified.
2279 * This routine may block, but if VM_ALLOC_RETRY is not set then NULL is
2280 * always returned if we had blocked.
2282 * This routine may not be called from an interrupt.
2284 * PG_ZERO is *ALWAYS* cleared by this routine.
2286 * No other requirements.
2289 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
2294 KKASSERT(allocflags &
2295 (VM_ALLOC_NORMAL|VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
2296 vm_object_hold(object);
2298 m = vm_page_lookup_busy_try(object, pindex, TRUE, &error);
2300 vm_page_sleep_busy(m, TRUE, "pgrbwt");
2301 if ((allocflags & VM_ALLOC_RETRY) == 0) {
2306 } else if (m == NULL) {
2307 if (allocflags & VM_ALLOC_RETRY)
2308 allocflags |= VM_ALLOC_NULL_OK;
2309 m = vm_page_alloc(object, pindex,
2310 allocflags & ~VM_ALLOC_RETRY);
2314 if ((allocflags & VM_ALLOC_RETRY) == 0)
2323 * If VM_ALLOC_ZERO an invalid page will be zero'd and set valid.
2325 * If VM_ALLOC_FORCE_ZERO the page is unconditionally zero'd and set
2326 * valid even if already valid.
2328 if (m->valid == 0) {
2329 if (allocflags & (VM_ALLOC_ZERO | VM_ALLOC_FORCE_ZERO)) {
2330 if ((m->flags & PG_ZERO) == 0)
2331 pmap_zero_page(VM_PAGE_TO_PHYS(m));
2332 m->valid = VM_PAGE_BITS_ALL;
2334 } else if (allocflags & VM_ALLOC_FORCE_ZERO) {
2335 pmap_zero_page(VM_PAGE_TO_PHYS(m));
2336 m->valid = VM_PAGE_BITS_ALL;
2338 vm_page_flag_clear(m, PG_ZERO);
2340 vm_object_drop(object);
2345 * Mapping function for valid bits or for dirty bits in
2346 * a page. May not block.
2348 * Inputs are required to range within a page.
2354 vm_page_bits(int base, int size)
2360 base + size <= PAGE_SIZE,
2361 ("vm_page_bits: illegal base/size %d/%d", base, size)
2364 if (size == 0) /* handle degenerate case */
2367 first_bit = base >> DEV_BSHIFT;
2368 last_bit = (base + size - 1) >> DEV_BSHIFT;
2370 return ((2 << last_bit) - (1 << first_bit));
2374 * Sets portions of a page valid and clean. The arguments are expected
2375 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
2376 * of any partial chunks touched by the range. The invalid portion of
2377 * such chunks will be zero'd.
2379 * NOTE: When truncating a buffer vnode_pager_setsize() will automatically
2380 * align base to DEV_BSIZE so as not to mark clean a partially
2381 * truncated device block. Otherwise the dirty page status might be
2384 * This routine may not block.
2386 * (base + size) must be less then or equal to PAGE_SIZE.
2389 _vm_page_zero_valid(vm_page_t m, int base, int size)
2394 if (size == 0) /* handle degenerate case */
2398 * If the base is not DEV_BSIZE aligned and the valid
2399 * bit is clear, we have to zero out a portion of the
2403 if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
2404 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0
2406 pmap_zero_page_area(
2414 * If the ending offset is not DEV_BSIZE aligned and the
2415 * valid bit is clear, we have to zero out a portion of
2419 endoff = base + size;
2421 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
2422 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0
2424 pmap_zero_page_area(
2427 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))
2433 * Set valid, clear dirty bits. If validating the entire
2434 * page we can safely clear the pmap modify bit. We also
2435 * use this opportunity to clear the PG_NOSYNC flag. If a process
2436 * takes a write fault on a MAP_NOSYNC memory area the flag will
2439 * We set valid bits inclusive of any overlap, but we can only
2440 * clear dirty bits for DEV_BSIZE chunks that are fully within
2443 * Page must be busied?
2444 * No other requirements.
2447 vm_page_set_valid(vm_page_t m, int base, int size)
2449 _vm_page_zero_valid(m, base, size);
2450 m->valid |= vm_page_bits(base, size);
2455 * Set valid bits and clear dirty bits.
2457 * NOTE: This function does not clear the pmap modified bit.
2458 * Also note that e.g. NFS may use a byte-granular base
2461 * WARNING: Page must be busied? But vfs_clean_one_page() will call
2462 * this without necessarily busying the page (via bdwrite()).
2463 * So for now vm_token must also be held.
2465 * No other requirements.
2468 vm_page_set_validclean(vm_page_t m, int base, int size)
2472 _vm_page_zero_valid(m, base, size);
2473 pagebits = vm_page_bits(base, size);
2474 m->valid |= pagebits;
2475 m->dirty &= ~pagebits;
2476 if (base == 0 && size == PAGE_SIZE) {
2477 /*pmap_clear_modify(m);*/
2478 vm_page_flag_clear(m, PG_NOSYNC);
2483 * Set valid & dirty. Used by buwrite()
2485 * WARNING: Page must be busied? But vfs_dirty_one_page() will
2486 * call this function in buwrite() so for now vm_token must
2489 * No other requirements.
2492 vm_page_set_validdirty(vm_page_t m, int base, int size)
2496 pagebits = vm_page_bits(base, size);
2497 m->valid |= pagebits;
2498 m->dirty |= pagebits;
2500 vm_object_set_writeable_dirty(m->object);
2506 * NOTE: This function does not clear the pmap modified bit.
2507 * Also note that e.g. NFS may use a byte-granular base
2510 * Page must be busied?
2511 * No other requirements.
2514 vm_page_clear_dirty(vm_page_t m, int base, int size)
2516 m->dirty &= ~vm_page_bits(base, size);
2517 if (base == 0 && size == PAGE_SIZE) {
2518 /*pmap_clear_modify(m);*/
2519 vm_page_flag_clear(m, PG_NOSYNC);
2524 * Make the page all-dirty.
2526 * Also make sure the related object and vnode reflect the fact that the
2527 * object may now contain a dirty page.
2529 * Page must be busied?
2530 * No other requirements.
2533 vm_page_dirty(vm_page_t m)
2536 int pqtype = m->queue - m->pc;
2538 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
2539 ("vm_page_dirty: page in free/cache queue!"));
2540 if (m->dirty != VM_PAGE_BITS_ALL) {
2541 m->dirty = VM_PAGE_BITS_ALL;
2543 vm_object_set_writeable_dirty(m->object);
2548 * Invalidates DEV_BSIZE'd chunks within a page. Both the
2549 * valid and dirty bits for the effected areas are cleared.
2551 * Page must be busied?
2553 * No other requirements.
2556 vm_page_set_invalid(vm_page_t m, int base, int size)
2560 bits = vm_page_bits(base, size);
2563 m->object->generation++;
2567 * The kernel assumes that the invalid portions of a page contain
2568 * garbage, but such pages can be mapped into memory by user code.
2569 * When this occurs, we must zero out the non-valid portions of the
2570 * page so user code sees what it expects.
2572 * Pages are most often semi-valid when the end of a file is mapped
2573 * into memory and the file's size is not page aligned.
2575 * Page must be busied?
2576 * No other requirements.
2579 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
2585 * Scan the valid bits looking for invalid sections that
2586 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
2587 * valid bit may be set ) have already been zerod by
2588 * vm_page_set_validclean().
2590 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
2591 if (i == (PAGE_SIZE / DEV_BSIZE) ||
2592 (m->valid & (1 << i))
2595 pmap_zero_page_area(
2598 (i - b) << DEV_BSHIFT
2606 * setvalid is TRUE when we can safely set the zero'd areas
2607 * as being valid. We can do this if there are no cache consistency
2608 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
2611 m->valid = VM_PAGE_BITS_ALL;
2615 * Is a (partial) page valid? Note that the case where size == 0
2616 * will return FALSE in the degenerate case where the page is entirely
2617 * invalid, and TRUE otherwise.
2620 * No other requirements.
2623 vm_page_is_valid(vm_page_t m, int base, int size)
2625 int bits = vm_page_bits(base, size);
2627 if (m->valid && ((m->valid & bits) == bits))
2634 * update dirty bits from pmap/mmu. May not block.
2636 * Caller must hold the page busy
2639 vm_page_test_dirty(vm_page_t m)
2641 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
2647 * Register an action, associating it with its vm_page
2650 vm_page_register_action(vm_page_action_t action, vm_page_event_t event)
2652 struct vm_page_action_list *list;
2655 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
2656 list = &action_list[hv];
2658 lwkt_gettoken(&vm_token);
2659 vm_page_flag_set(action->m, PG_ACTIONLIST);
2660 action->event = event;
2661 LIST_INSERT_HEAD(list, action, entry);
2662 lwkt_reltoken(&vm_token);
2666 * Unregister an action, disassociating it from its related vm_page
2669 vm_page_unregister_action(vm_page_action_t action)
2671 struct vm_page_action_list *list;
2674 lwkt_gettoken(&vm_token);
2675 if (action->event != VMEVENT_NONE) {
2676 action->event = VMEVENT_NONE;
2677 LIST_REMOVE(action, entry);
2679 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
2680 list = &action_list[hv];
2681 if (LIST_EMPTY(list))
2682 vm_page_flag_clear(action->m, PG_ACTIONLIST);
2684 lwkt_reltoken(&vm_token);
2688 * Issue an event on a VM page. Corresponding action structures are
2689 * removed from the page's list and called.
2691 * If the vm_page has no more pending action events we clear its
2692 * PG_ACTIONLIST flag.
2695 vm_page_event_internal(vm_page_t m, vm_page_event_t event)
2697 struct vm_page_action_list *list;
2698 struct vm_page_action *scan;
2699 struct vm_page_action *next;
2703 hv = (int)((intptr_t)m >> 8) & VMACTION_HMASK;
2704 list = &action_list[hv];
2707 lwkt_gettoken(&vm_token);
2708 LIST_FOREACH_MUTABLE(scan, list, entry, next) {
2710 if (scan->event == event) {
2711 scan->event = VMEVENT_NONE;
2712 LIST_REMOVE(scan, entry);
2713 scan->func(m, scan);
2721 vm_page_flag_clear(m, PG_ACTIONLIST);
2722 lwkt_reltoken(&vm_token);
2725 #include "opt_ddb.h"
2727 #include <sys/kernel.h>
2729 #include <ddb/ddb.h>
2731 DB_SHOW_COMMAND(page, vm_page_print_page_info)
2733 db_printf("vmstats.v_free_count: %d\n", vmstats.v_free_count);
2734 db_printf("vmstats.v_cache_count: %d\n", vmstats.v_cache_count);
2735 db_printf("vmstats.v_inactive_count: %d\n", vmstats.v_inactive_count);
2736 db_printf("vmstats.v_active_count: %d\n", vmstats.v_active_count);
2737 db_printf("vmstats.v_wire_count: %d\n", vmstats.v_wire_count);
2738 db_printf("vmstats.v_free_reserved: %d\n", vmstats.v_free_reserved);
2739 db_printf("vmstats.v_free_min: %d\n", vmstats.v_free_min);
2740 db_printf("vmstats.v_free_target: %d\n", vmstats.v_free_target);
2741 db_printf("vmstats.v_cache_min: %d\n", vmstats.v_cache_min);
2742 db_printf("vmstats.v_inactive_target: %d\n", vmstats.v_inactive_target);
2745 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
2748 db_printf("PQ_FREE:");
2749 for(i=0;i<PQ_L2_SIZE;i++) {
2750 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt);
2754 db_printf("PQ_CACHE:");
2755 for(i=0;i<PQ_L2_SIZE;i++) {
2756 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt);
2760 db_printf("PQ_ACTIVE:");
2761 for(i=0;i<PQ_L2_SIZE;i++) {
2762 db_printf(" %d", vm_page_queues[PQ_ACTIVE + i].lcnt);
2766 db_printf("PQ_INACTIVE:");
2767 for(i=0;i<PQ_L2_SIZE;i++) {
2768 db_printf(" %d", vm_page_queues[PQ_INACTIVE + i].lcnt);