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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
39 * $FreeBSD: src/sys/vm/vm_page.c,v 1.147.2.18 2002/03/10 05:03:19 alc Exp $
40 * $DragonFly: src/sys/vm/vm_page.c,v 1.40 2008/08/25 17:01:42 dillon Exp $
44 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
45 * All rights reserved.
47 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
49 * Permission to use, copy, modify and distribute this software and
50 * its documentation is hereby granted, provided that both the copyright
51 * notice and this permission notice appear in all copies of the
52 * software, derivative works or modified versions, and any portions
53 * thereof, and that both notices appear in supporting documentation.
55 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
59 * Carnegie Mellon requests users of this software to return to
61 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
62 * School of Computer Science
63 * Carnegie Mellon University
64 * Pittsburgh PA 15213-3890
66 * any improvements or extensions that they make and grant Carnegie the
67 * rights to redistribute these changes.
70 * Resident memory management module. The module manipulates 'VM pages'.
71 * A VM page is the core building block for memory management.
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/malloc.h>
78 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
82 #include <vm/vm_param.h>
84 #include <vm/vm_kern.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pageout.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_extern.h>
92 #include <vm/swap_pager.h>
94 #include <machine/md_var.h>
96 #include <vm/vm_page2.h>
97 #include <sys/mplock2.h>
99 #define VMACTION_HSIZE 256
100 #define VMACTION_HMASK (VMACTION_HSIZE - 1)
102 static void vm_page_queue_init(void);
103 static void vm_page_free_wakeup(void);
104 static vm_page_t vm_page_select_cache(vm_object_t, vm_pindex_t);
105 static vm_page_t _vm_page_list_find2(int basequeue, int index);
107 struct vpgqueues vm_page_queues[PQ_COUNT]; /* Array of tailq lists */
109 LIST_HEAD(vm_page_action_list, vm_page_action);
110 struct vm_page_action_list action_list[VMACTION_HSIZE];
113 #define ASSERT_IN_CRIT_SECTION() KKASSERT(crit_test(curthread));
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;
128 vm_page_queues[PQ_INACTIVE].cnt = &vmstats.v_inactive_count;
129 vm_page_queues[PQ_ACTIVE].cnt = &vmstats.v_active_count;
130 vm_page_queues[PQ_HOLD].cnt = &vmstats.v_active_count;
131 /* PQ_NONE has no queue */
133 for (i = 0; i < PQ_COUNT; i++)
134 TAILQ_INIT(&vm_page_queues[i].pl);
136 for (i = 0; i < VMACTION_HSIZE; i++)
137 LIST_INIT(&action_list[i]);
141 * note: place in initialized data section? Is this necessary?
144 int vm_page_array_size = 0;
145 int vm_page_zero_count = 0;
146 vm_page_t vm_page_array = 0;
151 * Sets the page size, perhaps based upon the memory size.
152 * Must be called before any use of page-size dependent functions.
155 vm_set_page_size(void)
157 if (vmstats.v_page_size == 0)
158 vmstats.v_page_size = PAGE_SIZE;
159 if (((vmstats.v_page_size - 1) & vmstats.v_page_size) != 0)
160 panic("vm_set_page_size: page size not a power of two");
166 * Add a new page to the freelist for use by the system. New pages
167 * are added to both the head and tail of the associated free page
168 * queue in a bottom-up fashion, so both zero'd and non-zero'd page
169 * requests pull 'recent' adds (higher physical addresses) first.
171 * Must be called in a critical section.
174 vm_add_new_page(vm_paddr_t pa)
176 struct vpgqueues *vpq;
179 ++vmstats.v_page_count;
180 ++vmstats.v_free_count;
181 m = PHYS_TO_VM_PAGE(pa);
184 m->pc = (pa >> PAGE_SHIFT) & PQ_L2_MASK;
185 m->queue = m->pc + PQ_FREE;
186 KKASSERT(m->dirty == 0);
188 vpq = &vm_page_queues[m->queue];
190 TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
192 TAILQ_INSERT_HEAD(&vpq->pl, m, pageq);
193 vpq->flipflop = 1 - vpq->flipflop;
195 vm_page_queues[m->queue].lcnt++;
202 * Initializes the resident memory module.
204 * Allocates memory for the page cells, and for the object/offset-to-page
205 * hash table headers. Each page cell is initialized and placed on the
208 * starta/enda represents the range of physical memory addresses available
209 * for use (skipping memory already used by the kernel), subject to
210 * phys_avail[]. Note that phys_avail[] has already mapped out memory
211 * already in use by the kernel.
214 vm_page_startup(vm_offset_t vaddr)
218 vm_paddr_t page_range;
225 vm_paddr_t biggestone, biggestsize;
232 vaddr = round_page(vaddr);
234 for (i = 0; phys_avail[i + 1]; i += 2) {
235 phys_avail[i] = round_page64(phys_avail[i]);
236 phys_avail[i + 1] = trunc_page64(phys_avail[i + 1]);
239 for (i = 0; phys_avail[i + 1]; i += 2) {
240 vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
242 if (size > biggestsize) {
250 end = phys_avail[biggestone+1];
251 end = trunc_page(end);
254 * Initialize the queue headers for the free queue, the active queue
255 * and the inactive queue.
258 vm_page_queue_init();
260 /* VKERNELs don't support minidumps and as such don't need vm_page_dump */
261 #if !defined(_KERNEL_VIRTUAL)
263 * Allocate a bitmap to indicate that a random physical page
264 * needs to be included in a minidump.
266 * The amd64 port needs this to indicate which direct map pages
267 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
269 * However, i386 still needs this workspace internally within the
270 * minidump code. In theory, they are not needed on i386, but are
271 * included should the sf_buf code decide to use them.
273 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
274 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
275 end -= vm_page_dump_size;
276 vm_page_dump = (void *)pmap_map(&vaddr, end, end + vm_page_dump_size,
277 VM_PROT_READ | VM_PROT_WRITE);
278 bzero((void *)vm_page_dump, vm_page_dump_size);
282 * Compute the number of pages of memory that will be available for
283 * use (taking into account the overhead of a page structure per
286 first_page = phys_avail[0] / PAGE_SIZE;
287 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE - first_page;
288 npages = (total - (page_range * sizeof(struct vm_page))) / PAGE_SIZE;
291 * Initialize the mem entry structures now, and put them in the free
294 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
295 mapped = pmap_map(&vaddr, new_end, end,
296 VM_PROT_READ | VM_PROT_WRITE);
297 vm_page_array = (vm_page_t)mapped;
299 #if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
301 * since pmap_map on amd64 returns stuff out of a direct-map region,
302 * we have to manually add these pages to the minidump tracking so
303 * that they can be dumped, including the vm_page_array.
305 for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
310 * Clear all of the page structures
312 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
313 vm_page_array_size = page_range;
316 * Construct the free queue(s) in ascending order (by physical
317 * address) so that the first 16MB of physical memory is allocated
318 * last rather than first. On large-memory machines, this avoids
319 * the exhaustion of low physical memory before isa_dmainit has run.
321 vmstats.v_page_count = 0;
322 vmstats.v_free_count = 0;
323 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) {
328 last_pa = phys_avail[i + 1];
329 while (pa < last_pa && npages-- > 0) {
338 * Scan comparison function for Red-Black tree scans. An inclusive
339 * (start,end) is expected. Other fields are not used.
342 rb_vm_page_scancmp(struct vm_page *p, void *data)
344 struct rb_vm_page_scan_info *info = data;
346 if (p->pindex < info->start_pindex)
348 if (p->pindex > info->end_pindex)
354 rb_vm_page_compare(struct vm_page *p1, struct vm_page *p2)
356 if (p1->pindex < p2->pindex)
358 if (p1->pindex > p2->pindex)
364 * Holding a page keeps it from being reused. Other parts of the system
365 * can still disassociate the page from its current object and free it, or
366 * perform read or write I/O on it and/or otherwise manipulate the page,
367 * but if the page is held the VM system will leave the page and its data
368 * intact and not reuse the page for other purposes until the last hold
369 * reference is released. (see vm_page_wire() if you want to prevent the
370 * page from being disassociated from its object too).
372 * The caller must hold vm_token.
374 * The caller must still validate the contents of the page and, if necessary,
375 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
376 * before manipulating the page.
379 vm_page_hold(vm_page_t m)
381 ASSERT_LWKT_TOKEN_HELD(&vm_token);
386 * The opposite of vm_page_hold(). A page can be freed while being held,
387 * which places it on the PQ_HOLD queue. We must call vm_page_free_toq()
388 * in this case to actually free it once the hold count drops to 0.
390 * The caller must hold vm_token if non-blocking operation is desired,
391 * but otherwise does not need to.
394 vm_page_unhold(vm_page_t m)
396 lwkt_gettoken(&vm_token);
398 KASSERT(m->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
399 if (m->hold_count == 0 && m->queue == PQ_HOLD) {
403 lwkt_reltoken(&vm_token);
407 * Inserts the given vm_page into the object and object list.
409 * The pagetables are not updated but will presumably fault the page
410 * in if necessary, or if a kernel page the caller will at some point
411 * enter the page into the kernel's pmap. We are not allowed to block
412 * here so we *can't* do this anyway.
414 * This routine may not block.
415 * This routine must be called with the vm_token held.
416 * This routine must be called with a critical section held.
419 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
421 ASSERT_IN_CRIT_SECTION();
422 ASSERT_LWKT_TOKEN_HELD(&vm_token);
423 if (m->object != NULL)
424 panic("vm_page_insert: already inserted");
427 * Record the object/offset pair in this page
433 * Insert it into the object.
435 vm_page_rb_tree_RB_INSERT(&object->rb_memq, m);
436 object->generation++;
439 * show that the object has one more resident page.
441 object->resident_page_count++;
444 * Since we are inserting a new and possibly dirty page,
445 * update the object's OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY flags.
447 if ((m->valid & m->dirty) || (m->flags & PG_WRITEABLE))
448 vm_object_set_writeable_dirty(object);
451 * Checks for a swap assignment and sets PG_SWAPPED if appropriate.
453 swap_pager_page_inserted(m);
457 * Removes the given vm_page_t from the global (object,index) hash table
458 * and from the object's memq.
460 * The underlying pmap entry (if any) is NOT removed here.
461 * This routine may not block.
463 * The page must be BUSY and will remain BUSY on return.
464 * No other requirements.
466 * NOTE: FreeBSD side effect was to unbusy the page on return. We leave
470 vm_page_remove(vm_page_t m)
475 lwkt_gettoken(&vm_token);
476 if (m->object == NULL) {
477 lwkt_reltoken(&vm_token);
482 if ((m->flags & PG_BUSY) == 0)
483 panic("vm_page_remove: page not busy");
488 * Remove the page from the object and update the object.
490 vm_page_rb_tree_RB_REMOVE(&object->rb_memq, m);
491 object->resident_page_count--;
492 object->generation++;
495 lwkt_reltoken(&vm_token);
500 * Locate and return the page at (object, pindex), or NULL if the
501 * page could not be found.
503 * The caller must hold vm_token.
506 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
511 * Search the hash table for this object/offset pair
513 ASSERT_LWKT_TOKEN_HELD(&vm_token);
515 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
517 KKASSERT(m == NULL || (m->object == object && m->pindex == pindex));
524 * Move the given memory entry from its current object to the specified
525 * target object/offset.
527 * The object must be locked.
528 * This routine may not block.
530 * Note: This routine will raise itself to splvm(), the caller need not.
532 * Note: Swap associated with the page must be invalidated by the move. We
533 * have to do this for several reasons: (1) we aren't freeing the
534 * page, (2) we are dirtying the page, (3) the VM system is probably
535 * moving the page from object A to B, and will then later move
536 * the backing store from A to B and we can't have a conflict.
538 * Note: We *always* dirty the page. It is necessary both for the
539 * fact that we moved it, and because we may be invalidating
540 * swap. If the page is on the cache, we have to deactivate it
541 * or vm_page_dirty() will panic. Dirty pages are not allowed
545 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
548 lwkt_gettoken(&vm_token);
550 vm_page_insert(m, new_object, new_pindex);
551 if (m->queue - m->pc == PQ_CACHE)
552 vm_page_deactivate(m);
555 lwkt_reltoken(&vm_token);
560 * vm_page_unqueue() without any wakeup. This routine is used when a page
561 * is being moved between queues or otherwise is to remain BUSYied by the
564 * The caller must hold vm_token
565 * This routine may not block.
568 vm_page_unqueue_nowakeup(vm_page_t m)
570 int queue = m->queue;
571 struct vpgqueues *pq;
573 ASSERT_LWKT_TOKEN_HELD(&vm_token);
574 if (queue != PQ_NONE) {
575 pq = &vm_page_queues[queue];
577 TAILQ_REMOVE(&pq->pl, m, pageq);
584 * vm_page_unqueue() - Remove a page from its queue, wakeup the pagedemon
587 * The caller must hold vm_token
588 * This routine may not block.
591 vm_page_unqueue(vm_page_t m)
593 int queue = m->queue;
594 struct vpgqueues *pq;
596 ASSERT_LWKT_TOKEN_HELD(&vm_token);
597 if (queue != PQ_NONE) {
599 pq = &vm_page_queues[queue];
600 TAILQ_REMOVE(&pq->pl, m, pageq);
603 if ((queue - m->pc) == PQ_CACHE || (queue - m->pc) == PQ_FREE)
609 * vm_page_list_find()
611 * Find a page on the specified queue with color optimization.
613 * The page coloring optimization attempts to locate a page that does
614 * not overload other nearby pages in the object in the cpu's L1 or L2
615 * caches. We need this optimization because cpu caches tend to be
616 * physical caches, while object spaces tend to be virtual.
618 * Must be called with vm_token held.
619 * This routine may not block.
621 * Note that this routine is carefully inlined. A non-inlined version
622 * is available for outside callers but the only critical path is
623 * from within this source file.
627 _vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
632 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
634 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
636 m = _vm_page_list_find2(basequeue, index);
641 _vm_page_list_find2(int basequeue, int index)
645 struct vpgqueues *pq;
647 pq = &vm_page_queues[basequeue];
650 * Note that for the first loop, index+i and index-i wind up at the
651 * same place. Even though this is not totally optimal, we've already
652 * blown it by missing the cache case so we do not care.
655 for(i = PQ_L2_SIZE / 2; i > 0; --i) {
656 if ((m = TAILQ_FIRST(&pq[(index + i) & PQ_L2_MASK].pl)) != NULL)
659 if ((m = TAILQ_FIRST(&pq[(index - i) & PQ_L2_MASK].pl)) != NULL)
666 * Must be called with vm_token held if the caller desired non-blocking
667 * operation and a stable result.
670 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
672 return(_vm_page_list_find(basequeue, index, prefer_zero));
676 * Find a page on the cache queue with color optimization. As pages
677 * might be found, but not applicable, they are deactivated. This
678 * keeps us from using potentially busy cached pages.
680 * This routine may not block.
681 * Must be called with vm_token held.
684 vm_page_select_cache(vm_object_t object, vm_pindex_t pindex)
688 ASSERT_LWKT_TOKEN_HELD(&vm_token);
690 m = _vm_page_list_find(
692 (pindex + object->pg_color) & PQ_L2_MASK,
695 if (m && ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
696 m->hold_count || m->wire_count)) {
697 vm_page_deactivate(m);
706 * Find a free or zero page, with specified preference. We attempt to
707 * inline the nominal case and fall back to _vm_page_select_free()
710 * This routine must be called with a critical section held.
711 * This routine may not block.
713 static __inline vm_page_t
714 vm_page_select_free(vm_object_t object, vm_pindex_t pindex, boolean_t prefer_zero)
718 m = _vm_page_list_find(
720 (pindex + object->pg_color) & PQ_L2_MASK,
729 * Allocate and return a memory cell associated with this VM object/offset
734 * VM_ALLOC_NORMAL allow use of cache pages, nominal free drain
735 * VM_ALLOC_QUICK like normal but cannot use cache
736 * VM_ALLOC_SYSTEM greater free drain
737 * VM_ALLOC_INTERRUPT allow free list to be completely drained
738 * VM_ALLOC_ZERO advisory request for pre-zero'd page
740 * The object must be locked.
741 * This routine may not block.
742 * The returned page will be marked PG_BUSY
744 * Additional special handling is required when called from an interrupt
745 * (VM_ALLOC_INTERRUPT). We are not allowed to mess with the page cache
749 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int page_req)
754 lwkt_gettoken(&vm_token);
756 KKASSERT(object != NULL);
757 KASSERT(!vm_page_lookup(object, pindex),
758 ("vm_page_alloc: page already allocated"));
760 (VM_ALLOC_NORMAL|VM_ALLOC_QUICK|
761 VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
764 * Certain system threads (pageout daemon, buf_daemon's) are
765 * allowed to eat deeper into the free page list.
767 if (curthread->td_flags & TDF_SYSTHREAD)
768 page_req |= VM_ALLOC_SYSTEM;
771 if (vmstats.v_free_count > vmstats.v_free_reserved ||
772 ((page_req & VM_ALLOC_INTERRUPT) && vmstats.v_free_count > 0) ||
773 ((page_req & VM_ALLOC_SYSTEM) && vmstats.v_cache_count == 0 &&
774 vmstats.v_free_count > vmstats.v_interrupt_free_min)
777 * The free queue has sufficient free pages to take one out.
779 if (page_req & VM_ALLOC_ZERO)
780 m = vm_page_select_free(object, pindex, TRUE);
782 m = vm_page_select_free(object, pindex, FALSE);
783 } else if (page_req & VM_ALLOC_NORMAL) {
785 * Allocatable from the cache (non-interrupt only). On
786 * success, we must free the page and try again, thus
787 * ensuring that vmstats.v_*_free_min counters are replenished.
790 if (curthread->td_preempted) {
791 kprintf("vm_page_alloc(): warning, attempt to allocate"
792 " cache page from preempting interrupt\n");
795 m = vm_page_select_cache(object, pindex);
798 m = vm_page_select_cache(object, pindex);
801 * On success move the page into the free queue and loop.
804 KASSERT(m->dirty == 0,
805 ("Found dirty cache page %p", m));
807 vm_page_protect(m, VM_PROT_NONE);
813 * On failure return NULL
815 lwkt_reltoken(&vm_token);
817 #if defined(DIAGNOSTIC)
818 if (vmstats.v_cache_count > 0)
819 kprintf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", vmstats.v_cache_count);
821 vm_pageout_deficit++;
826 * No pages available, wakeup the pageout daemon and give up.
828 lwkt_reltoken(&vm_token);
830 vm_pageout_deficit++;
836 * Good page found. The page has not yet been busied. We are in
837 * a critical section.
839 KASSERT(m != NULL, ("vm_page_alloc(): missing page on free queue\n"));
840 KASSERT(m->dirty == 0,
841 ("vm_page_alloc: free/cache page %p was dirty", m));
844 * Remove from free queue
846 vm_page_unqueue_nowakeup(m);
849 * Initialize structure. Only the PG_ZERO flag is inherited. Set
852 if (m->flags & PG_ZERO) {
853 vm_page_zero_count--;
854 m->flags = PG_ZERO | PG_BUSY;
865 * vm_page_insert() is safe prior to the crit_exit(). Note also that
866 * inserting a page here does not insert it into the pmap (which
867 * could cause us to block allocating memory). We cannot block
870 vm_page_insert(m, object, pindex);
873 * Don't wakeup too often - wakeup the pageout daemon when
874 * we would be nearly out of memory.
878 lwkt_reltoken(&vm_token);
882 * A PG_BUSY page is returned.
888 * Wait for sufficient free memory for nominal heavy memory use kernel
892 vm_wait_nominal(void)
894 while (vm_page_count_min(0))
899 * Test if vm_wait_nominal() would block.
902 vm_test_nominal(void)
904 if (vm_page_count_min(0))
910 * Block until free pages are available for allocation, called in various
911 * places before memory allocations.
917 lwkt_gettoken(&vm_token);
918 if (curthread == pagethread) {
919 vm_pageout_pages_needed = 1;
920 tsleep(&vm_pageout_pages_needed, 0, "VMWait", timo);
922 if (vm_pages_needed == 0) {
924 wakeup(&vm_pages_needed);
926 tsleep(&vmstats.v_free_count, 0, "vmwait", timo);
928 lwkt_reltoken(&vm_token);
933 * Block until free pages are available for allocation
935 * Called only in vm_fault so that processes page faulting can be
942 lwkt_gettoken(&vm_token);
943 if (vm_pages_needed == 0) {
945 wakeup(&vm_pages_needed);
947 tsleep(&vmstats.v_free_count, 0, "pfault", 0);
948 lwkt_reltoken(&vm_token);
953 * Put the specified page on the active list (if appropriate). Ensure
954 * that act_count is at least ACT_INIT but do not otherwise mess with it.
956 * The page queues must be locked.
957 * This routine may not block.
960 vm_page_activate(vm_page_t m)
963 lwkt_gettoken(&vm_token);
964 if (m->queue != PQ_ACTIVE) {
965 if ((m->queue - m->pc) == PQ_CACHE)
966 mycpu->gd_cnt.v_reactivated++;
970 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
971 m->queue = PQ_ACTIVE;
972 vm_page_queues[PQ_ACTIVE].lcnt++;
973 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_ACTIVE].pl,
975 if (m->act_count < ACT_INIT)
976 m->act_count = ACT_INIT;
977 vmstats.v_active_count++;
980 if (m->act_count < ACT_INIT)
981 m->act_count = ACT_INIT;
983 lwkt_reltoken(&vm_token);
988 * Helper routine for vm_page_free_toq() and vm_page_cache(). This
989 * routine is called when a page has been added to the cache or free
992 * This routine may not block.
993 * This routine must be called at splvm()
996 vm_page_free_wakeup(void)
999 * if pageout daemon needs pages, then tell it that there are
1002 if (vm_pageout_pages_needed &&
1003 vmstats.v_cache_count + vmstats.v_free_count >=
1004 vmstats.v_pageout_free_min
1006 wakeup(&vm_pageout_pages_needed);
1007 vm_pageout_pages_needed = 0;
1011 * wakeup processes that are waiting on memory if we hit a
1012 * high water mark. And wakeup scheduler process if we have
1013 * lots of memory. this process will swapin processes.
1015 if (vm_pages_needed && !vm_page_count_min(0)) {
1016 vm_pages_needed = 0;
1017 wakeup(&vmstats.v_free_count);
1024 * Returns the given page to the PQ_FREE list, disassociating it with
1027 * The vm_page must be PG_BUSY on entry. PG_BUSY will be released on
1028 * return (the page will have been freed). No particular spl is required
1031 * This routine may not block.
1034 vm_page_free_toq(vm_page_t m)
1036 struct vpgqueues *pq;
1039 lwkt_gettoken(&vm_token);
1040 mycpu->gd_cnt.v_tfree++;
1042 KKASSERT((m->flags & PG_MAPPED) == 0);
1044 if (m->busy || ((m->queue - m->pc) == PQ_FREE)) {
1046 "vm_page_free: pindex(%lu), busy(%d), PG_BUSY(%d), hold(%d)\n",
1047 (u_long)m->pindex, m->busy, (m->flags & PG_BUSY) ? 1 : 0,
1049 if ((m->queue - m->pc) == PQ_FREE)
1050 panic("vm_page_free: freeing free page");
1052 panic("vm_page_free: freeing busy page");
1056 * unqueue, then remove page. Note that we cannot destroy
1057 * the page here because we do not want to call the pager's
1058 * callback routine until after we've put the page on the
1059 * appropriate free queue.
1061 vm_page_unqueue_nowakeup(m);
1065 * No further management of fictitious pages occurs beyond object
1066 * and queue removal.
1068 if ((m->flags & PG_FICTITIOUS) != 0) {
1070 lwkt_reltoken(&vm_token);
1078 if (m->wire_count != 0) {
1079 if (m->wire_count > 1) {
1081 "vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1082 m->wire_count, (long)m->pindex);
1084 panic("vm_page_free: freeing wired page");
1088 * Clear the UNMANAGED flag when freeing an unmanaged page.
1090 if (m->flags & PG_UNMANAGED) {
1091 m->flags &= ~PG_UNMANAGED;
1094 if (m->hold_count != 0) {
1095 m->flags &= ~PG_ZERO;
1098 m->queue = PQ_FREE + m->pc;
1100 pq = &vm_page_queues[m->queue];
1105 * Put zero'd pages on the end ( where we look for zero'd pages
1106 * first ) and non-zerod pages at the head.
1108 if (m->flags & PG_ZERO) {
1109 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
1110 ++vm_page_zero_count;
1112 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
1115 vm_page_free_wakeup();
1116 lwkt_reltoken(&vm_token);
1121 * vm_page_free_fromq_fast()
1123 * Remove a non-zero page from one of the free queues; the page is removed for
1124 * zeroing, so do not issue a wakeup.
1129 vm_page_free_fromq_fast(void)
1136 lwkt_gettoken(&vm_token);
1137 for (i = 0; i < PQ_L2_SIZE; ++i) {
1138 m = vm_page_list_find(PQ_FREE, qi, FALSE);
1139 qi = (qi + PQ_PRIME2) & PQ_L2_MASK;
1140 if (m && (m->flags & PG_ZERO) == 0) {
1141 vm_page_unqueue_nowakeup(m);
1147 lwkt_reltoken(&vm_token);
1153 * vm_page_unmanage()
1155 * Prevent PV management from being done on the page. The page is
1156 * removed from the paging queues as if it were wired, and as a
1157 * consequence of no longer being managed the pageout daemon will not
1158 * touch it (since there is no way to locate the pte mappings for the
1159 * page). madvise() calls that mess with the pmap will also no longer
1160 * operate on the page.
1162 * Beyond that the page is still reasonably 'normal'. Freeing the page
1163 * will clear the flag.
1165 * This routine is used by OBJT_PHYS objects - objects using unswappable
1166 * physical memory as backing store rather then swap-backed memory and
1167 * will eventually be extended to support 4MB unmanaged physical
1170 * Must be called with a critical section held.
1171 * Must be called with vm_token held.
1174 vm_page_unmanage(vm_page_t m)
1176 ASSERT_IN_CRIT_SECTION();
1177 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1178 if ((m->flags & PG_UNMANAGED) == 0) {
1179 if (m->wire_count == 0)
1182 vm_page_flag_set(m, PG_UNMANAGED);
1186 * Mark this page as wired down by yet another map, removing it from
1187 * paging queues as necessary.
1189 * The page queues must be locked.
1190 * This routine may not block.
1193 vm_page_wire(vm_page_t m)
1196 * Only bump the wire statistics if the page is not already wired,
1197 * and only unqueue the page if it is on some queue (if it is unmanaged
1198 * it is already off the queues). Don't do anything with fictitious
1199 * pages because they are always wired.
1202 lwkt_gettoken(&vm_token);
1203 if ((m->flags & PG_FICTITIOUS) == 0) {
1204 if (m->wire_count == 0) {
1205 if ((m->flags & PG_UNMANAGED) == 0)
1207 vmstats.v_wire_count++;
1210 KASSERT(m->wire_count != 0,
1211 ("vm_page_wire: wire_count overflow m=%p", m));
1213 lwkt_reltoken(&vm_token);
1218 * Release one wiring of this page, potentially enabling it to be paged again.
1220 * Many pages placed on the inactive queue should actually go
1221 * into the cache, but it is difficult to figure out which. What
1222 * we do instead, if the inactive target is well met, is to put
1223 * clean pages at the head of the inactive queue instead of the tail.
1224 * This will cause them to be moved to the cache more quickly and
1225 * if not actively re-referenced, freed more quickly. If we just
1226 * stick these pages at the end of the inactive queue, heavy filesystem
1227 * meta-data accesses can cause an unnecessary paging load on memory bound
1228 * processes. This optimization causes one-time-use metadata to be
1229 * reused more quickly.
1231 * BUT, if we are in a low-memory situation we have no choice but to
1232 * put clean pages on the cache queue.
1234 * A number of routines use vm_page_unwire() to guarantee that the page
1235 * will go into either the inactive or active queues, and will NEVER
1236 * be placed in the cache - for example, just after dirtying a page.
1237 * dirty pages in the cache are not allowed.
1239 * The page queues must be locked.
1240 * This routine may not block.
1243 vm_page_unwire(vm_page_t m, int activate)
1246 lwkt_gettoken(&vm_token);
1247 if (m->flags & PG_FICTITIOUS) {
1249 } else if (m->wire_count <= 0) {
1250 panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
1252 if (--m->wire_count == 0) {
1253 --vmstats.v_wire_count;
1254 if (m->flags & PG_UNMANAGED) {
1256 } else if (activate) {
1258 &vm_page_queues[PQ_ACTIVE].pl, m, pageq);
1259 m->queue = PQ_ACTIVE;
1260 vm_page_queues[PQ_ACTIVE].lcnt++;
1261 vmstats.v_active_count++;
1263 vm_page_flag_clear(m, PG_WINATCFLS);
1265 &vm_page_queues[PQ_INACTIVE].pl, m, pageq);
1266 m->queue = PQ_INACTIVE;
1267 vm_page_queues[PQ_INACTIVE].lcnt++;
1268 vmstats.v_inactive_count++;
1269 ++vm_swapcache_inactive_heuristic;
1273 lwkt_reltoken(&vm_token);
1279 * Move the specified page to the inactive queue. If the page has
1280 * any associated swap, the swap is deallocated.
1282 * Normally athead is 0 resulting in LRU operation. athead is set
1283 * to 1 if we want this page to be 'as if it were placed in the cache',
1284 * except without unmapping it from the process address space.
1286 * This routine may not block.
1287 * The caller must hold vm_token.
1289 static __inline void
1290 _vm_page_deactivate(vm_page_t m, int athead)
1293 * Ignore if already inactive.
1295 if (m->queue == PQ_INACTIVE)
1298 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1299 if ((m->queue - m->pc) == PQ_CACHE)
1300 mycpu->gd_cnt.v_reactivated++;
1301 vm_page_flag_clear(m, PG_WINATCFLS);
1304 TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl,
1307 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl,
1309 ++vm_swapcache_inactive_heuristic;
1311 m->queue = PQ_INACTIVE;
1312 vm_page_queues[PQ_INACTIVE].lcnt++;
1313 vmstats.v_inactive_count++;
1318 * Attempt to deactivate a page.
1323 vm_page_deactivate(vm_page_t m)
1326 lwkt_gettoken(&vm_token);
1327 _vm_page_deactivate(m, 0);
1328 lwkt_reltoken(&vm_token);
1333 * Attempt to move a page to PQ_CACHE.
1334 * Returns 0 on failure, 1 on success
1339 vm_page_try_to_cache(vm_page_t m)
1342 lwkt_gettoken(&vm_token);
1343 if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1344 (m->flags & (PG_BUSY|PG_UNMANAGED))) {
1345 lwkt_reltoken(&vm_token);
1349 vm_page_test_dirty(m);
1351 lwkt_reltoken(&vm_token);
1356 lwkt_reltoken(&vm_token);
1362 * Attempt to free the page. If we cannot free it, we do nothing.
1363 * 1 is returned on success, 0 on failure.
1368 vm_page_try_to_free(vm_page_t m)
1371 lwkt_gettoken(&vm_token);
1372 if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1373 (m->flags & (PG_BUSY|PG_UNMANAGED))) {
1374 lwkt_reltoken(&vm_token);
1378 vm_page_test_dirty(m);
1380 lwkt_reltoken(&vm_token);
1385 vm_page_protect(m, VM_PROT_NONE);
1387 lwkt_reltoken(&vm_token);
1395 * Put the specified page onto the page cache queue (if appropriate).
1397 * The caller must hold vm_token.
1398 * This routine may not block.
1401 vm_page_cache(vm_page_t m)
1403 ASSERT_IN_CRIT_SECTION();
1404 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1406 if ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
1407 m->wire_count || m->hold_count) {
1408 kprintf("vm_page_cache: attempting to cache busy/held page\n");
1413 * Already in the cache (and thus not mapped)
1415 if ((m->queue - m->pc) == PQ_CACHE) {
1416 KKASSERT((m->flags & PG_MAPPED) == 0);
1421 * Caller is required to test m->dirty, but note that the act of
1422 * removing the page from its maps can cause it to become dirty
1423 * on an SMP system due to another cpu running in usermode.
1426 panic("vm_page_cache: caching a dirty page, pindex: %ld",
1431 * Remove all pmaps and indicate that the page is not
1432 * writeable or mapped. Our vm_page_protect() call may
1433 * have blocked (especially w/ VM_PROT_NONE), so recheck
1437 vm_page_protect(m, VM_PROT_NONE);
1439 if ((m->flags & (PG_BUSY|PG_UNMANAGED|PG_MAPPED)) || m->busy ||
1440 m->wire_count || m->hold_count) {
1442 } else if (m->dirty) {
1443 vm_page_deactivate(m);
1445 vm_page_unqueue_nowakeup(m);
1446 m->queue = PQ_CACHE + m->pc;
1447 vm_page_queues[m->queue].lcnt++;
1448 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, pageq);
1449 vmstats.v_cache_count++;
1450 vm_page_free_wakeup();
1455 * vm_page_dontneed()
1457 * Cache, deactivate, or do nothing as appropriate. This routine
1458 * is typically used by madvise() MADV_DONTNEED.
1460 * Generally speaking we want to move the page into the cache so
1461 * it gets reused quickly. However, this can result in a silly syndrome
1462 * due to the page recycling too quickly. Small objects will not be
1463 * fully cached. On the otherhand, if we move the page to the inactive
1464 * queue we wind up with a problem whereby very large objects
1465 * unnecessarily blow away our inactive and cache queues.
1467 * The solution is to move the pages based on a fixed weighting. We
1468 * either leave them alone, deactivate them, or move them to the cache,
1469 * where moving them to the cache has the highest weighting.
1470 * By forcing some pages into other queues we eventually force the
1471 * system to balance the queues, potentially recovering other unrelated
1472 * space from active. The idea is to not force this to happen too
1478 vm_page_dontneed(vm_page_t m)
1480 static int dnweight;
1487 * occassionally leave the page alone
1490 lwkt_gettoken(&vm_token);
1491 if ((dnw & 0x01F0) == 0 ||
1492 m->queue == PQ_INACTIVE ||
1493 m->queue - m->pc == PQ_CACHE
1495 if (m->act_count >= ACT_INIT)
1497 lwkt_reltoken(&vm_token);
1503 vm_page_test_dirty(m);
1505 if (m->dirty || (dnw & 0x0070) == 0) {
1507 * Deactivate the page 3 times out of 32.
1512 * Cache the page 28 times out of every 32. Note that
1513 * the page is deactivated instead of cached, but placed
1514 * at the head of the queue instead of the tail.
1518 _vm_page_deactivate(m, head);
1519 lwkt_reltoken(&vm_token);
1524 * Grab a page, blocking if it is busy and allocating a page if necessary.
1525 * A busy page is returned or NULL.
1527 * If VM_ALLOC_RETRY is specified VM_ALLOC_NORMAL must also be specified.
1528 * If VM_ALLOC_RETRY is not specified
1530 * This routine may block, but if VM_ALLOC_RETRY is not set then NULL is
1531 * always returned if we had blocked.
1532 * This routine will never return NULL if VM_ALLOC_RETRY is set.
1533 * This routine may not be called from an interrupt.
1534 * The returned page may not be entirely valid.
1536 * This routine may be called from mainline code without spl protection and
1537 * be guarenteed a busied page associated with the object at the specified
1543 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
1548 KKASSERT(allocflags &
1549 (VM_ALLOC_NORMAL|VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
1551 lwkt_gettoken(&vm_token);
1553 if ((m = vm_page_lookup(object, pindex)) != NULL) {
1554 if (m->busy || (m->flags & PG_BUSY)) {
1555 generation = object->generation;
1557 while ((object->generation == generation) &&
1558 (m->busy || (m->flags & PG_BUSY))) {
1559 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
1560 tsleep(m, 0, "pgrbwt", 0);
1561 if ((allocflags & VM_ALLOC_RETRY) == 0) {
1572 m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
1575 if ((allocflags & VM_ALLOC_RETRY) == 0)
1580 lwkt_reltoken(&vm_token);
1586 * Mapping function for valid bits or for dirty bits in
1587 * a page. May not block.
1589 * Inputs are required to range within a page.
1595 vm_page_bits(int base, int size)
1601 base + size <= PAGE_SIZE,
1602 ("vm_page_bits: illegal base/size %d/%d", base, size)
1605 if (size == 0) /* handle degenerate case */
1608 first_bit = base >> DEV_BSHIFT;
1609 last_bit = (base + size - 1) >> DEV_BSHIFT;
1611 return ((2 << last_bit) - (1 << first_bit));
1615 * Sets portions of a page valid and clean. The arguments are expected
1616 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1617 * of any partial chunks touched by the range. The invalid portion of
1618 * such chunks will be zero'd.
1620 * NOTE: When truncating a buffer vnode_pager_setsize() will automatically
1621 * align base to DEV_BSIZE so as not to mark clean a partially
1622 * truncated device block. Otherwise the dirty page status might be
1625 * This routine may not block.
1627 * (base + size) must be less then or equal to PAGE_SIZE.
1630 _vm_page_zero_valid(vm_page_t m, int base, int size)
1635 if (size == 0) /* handle degenerate case */
1639 * If the base is not DEV_BSIZE aligned and the valid
1640 * bit is clear, we have to zero out a portion of the
1644 if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1645 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0
1647 pmap_zero_page_area(
1655 * If the ending offset is not DEV_BSIZE aligned and the
1656 * valid bit is clear, we have to zero out a portion of
1660 endoff = base + size;
1662 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1663 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0
1665 pmap_zero_page_area(
1668 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))
1674 * Set valid, clear dirty bits. If validating the entire
1675 * page we can safely clear the pmap modify bit. We also
1676 * use this opportunity to clear the PG_NOSYNC flag. If a process
1677 * takes a write fault on a MAP_NOSYNC memory area the flag will
1680 * We set valid bits inclusive of any overlap, but we can only
1681 * clear dirty bits for DEV_BSIZE chunks that are fully within
1684 * Page must be busied?
1685 * No other requirements.
1688 vm_page_set_valid(vm_page_t m, int base, int size)
1690 _vm_page_zero_valid(m, base, size);
1691 m->valid |= vm_page_bits(base, size);
1696 * Set valid bits and clear dirty bits.
1698 * NOTE: This function does not clear the pmap modified bit.
1699 * Also note that e.g. NFS may use a byte-granular base
1702 * Page must be busied?
1703 * No other requirements.
1706 vm_page_set_validclean(vm_page_t m, int base, int size)
1710 _vm_page_zero_valid(m, base, size);
1711 pagebits = vm_page_bits(base, size);
1712 m->valid |= pagebits;
1713 m->dirty &= ~pagebits;
1714 if (base == 0 && size == PAGE_SIZE) {
1715 /*pmap_clear_modify(m);*/
1716 vm_page_flag_clear(m, PG_NOSYNC);
1721 * Set valid & dirty. Used by buwrite()
1723 * Page must be busied?
1724 * No other requirements.
1727 vm_page_set_validdirty(vm_page_t m, int base, int size)
1731 pagebits = vm_page_bits(base, size);
1732 m->valid |= pagebits;
1733 m->dirty |= pagebits;
1735 vm_object_set_writeable_dirty(m->object);
1741 * NOTE: This function does not clear the pmap modified bit.
1742 * Also note that e.g. NFS may use a byte-granular base
1745 * Page must be busied?
1746 * No other requirements.
1749 vm_page_clear_dirty(vm_page_t m, int base, int size)
1751 m->dirty &= ~vm_page_bits(base, size);
1752 if (base == 0 && size == PAGE_SIZE) {
1753 /*pmap_clear_modify(m);*/
1754 vm_page_flag_clear(m, PG_NOSYNC);
1759 * Make the page all-dirty.
1761 * Also make sure the related object and vnode reflect the fact that the
1762 * object may now contain a dirty page.
1764 * Page must be busied?
1765 * No other requirements.
1768 vm_page_dirty(vm_page_t m)
1771 int pqtype = m->queue - m->pc;
1773 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
1774 ("vm_page_dirty: page in free/cache queue!"));
1775 if (m->dirty != VM_PAGE_BITS_ALL) {
1776 m->dirty = VM_PAGE_BITS_ALL;
1778 vm_object_set_writeable_dirty(m->object);
1783 * Invalidates DEV_BSIZE'd chunks within a page. Both the
1784 * valid and dirty bits for the effected areas are cleared.
1786 * Page must be busied?
1788 * No other requirements.
1791 vm_page_set_invalid(vm_page_t m, int base, int size)
1795 bits = vm_page_bits(base, size);
1798 m->object->generation++;
1802 * The kernel assumes that the invalid portions of a page contain
1803 * garbage, but such pages can be mapped into memory by user code.
1804 * When this occurs, we must zero out the non-valid portions of the
1805 * page so user code sees what it expects.
1807 * Pages are most often semi-valid when the end of a file is mapped
1808 * into memory and the file's size is not page aligned.
1810 * Page must be busied?
1811 * No other requirements.
1814 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
1820 * Scan the valid bits looking for invalid sections that
1821 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
1822 * valid bit may be set ) have already been zerod by
1823 * vm_page_set_validclean().
1825 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
1826 if (i == (PAGE_SIZE / DEV_BSIZE) ||
1827 (m->valid & (1 << i))
1830 pmap_zero_page_area(
1833 (i - b) << DEV_BSHIFT
1841 * setvalid is TRUE when we can safely set the zero'd areas
1842 * as being valid. We can do this if there are no cache consistency
1843 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
1846 m->valid = VM_PAGE_BITS_ALL;
1850 * Is a (partial) page valid? Note that the case where size == 0
1851 * will return FALSE in the degenerate case where the page is entirely
1852 * invalid, and TRUE otherwise.
1855 * No other requirements.
1858 vm_page_is_valid(vm_page_t m, int base, int size)
1860 int bits = vm_page_bits(base, size);
1862 if (m->valid && ((m->valid & bits) == bits))
1869 * update dirty bits from pmap/mmu. May not block.
1871 * Caller must hold vm_token if non-blocking operation desired.
1872 * No other requirements.
1875 vm_page_test_dirty(vm_page_t m)
1877 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
1883 * Register an action, associating it with its vm_page
1886 vm_page_register_action(vm_page_action_t action, vm_page_event_t event)
1888 struct vm_page_action_list *list;
1891 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1892 list = &action_list[hv];
1894 lwkt_gettoken(&vm_token);
1895 vm_page_flag_set(action->m, PG_ACTIONLIST);
1896 action->event = event;
1897 LIST_INSERT_HEAD(list, action, entry);
1898 lwkt_reltoken(&vm_token);
1902 * Unregister an action, disassociating it from its related vm_page
1905 vm_page_unregister_action(vm_page_action_t action)
1907 struct vm_page_action_list *list;
1910 lwkt_gettoken(&vm_token);
1911 if (action->event != VMEVENT_NONE) {
1912 action->event = VMEVENT_NONE;
1913 LIST_REMOVE(action, entry);
1915 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1916 list = &action_list[hv];
1917 if (LIST_EMPTY(list))
1918 vm_page_flag_clear(action->m, PG_ACTIONLIST);
1920 lwkt_reltoken(&vm_token);
1924 * Issue an event on a VM page. Corresponding action structures are
1925 * removed from the page's list and called.
1927 * If the vm_page has no more pending action events we clear its
1928 * PG_ACTIONLIST flag.
1931 vm_page_event_internal(vm_page_t m, vm_page_event_t event)
1933 struct vm_page_action_list *list;
1934 struct vm_page_action *scan;
1935 struct vm_page_action *next;
1939 hv = (int)((intptr_t)m >> 8) & VMACTION_HMASK;
1940 list = &action_list[hv];
1943 lwkt_gettoken(&vm_token);
1944 LIST_FOREACH_MUTABLE(scan, list, entry, next) {
1946 if (scan->event == event) {
1947 scan->event = VMEVENT_NONE;
1948 LIST_REMOVE(scan, entry);
1949 scan->func(m, scan);
1957 vm_page_flag_clear(m, PG_ACTIONLIST);
1958 lwkt_reltoken(&vm_token);
1962 #include "opt_ddb.h"
1964 #include <sys/kernel.h>
1966 #include <ddb/ddb.h>
1968 DB_SHOW_COMMAND(page, vm_page_print_page_info)
1970 db_printf("vmstats.v_free_count: %d\n", vmstats.v_free_count);
1971 db_printf("vmstats.v_cache_count: %d\n", vmstats.v_cache_count);
1972 db_printf("vmstats.v_inactive_count: %d\n", vmstats.v_inactive_count);
1973 db_printf("vmstats.v_active_count: %d\n", vmstats.v_active_count);
1974 db_printf("vmstats.v_wire_count: %d\n", vmstats.v_wire_count);
1975 db_printf("vmstats.v_free_reserved: %d\n", vmstats.v_free_reserved);
1976 db_printf("vmstats.v_free_min: %d\n", vmstats.v_free_min);
1977 db_printf("vmstats.v_free_target: %d\n", vmstats.v_free_target);
1978 db_printf("vmstats.v_cache_min: %d\n", vmstats.v_cache_min);
1979 db_printf("vmstats.v_inactive_target: %d\n", vmstats.v_inactive_target);
1982 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
1985 db_printf("PQ_FREE:");
1986 for(i=0;i<PQ_L2_SIZE;i++) {
1987 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt);
1991 db_printf("PQ_CACHE:");
1992 for(i=0;i<PQ_L2_SIZE;i++) {
1993 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt);
1997 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n",
1998 vm_page_queues[PQ_ACTIVE].lcnt,
1999 vm_page_queues[PQ_INACTIVE].lcnt);