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>
80 #include <sys/kernel.h>
83 #include <vm/vm_param.h>
85 #include <vm/vm_kern.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pageout.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_extern.h>
93 #include <vm/swap_pager.h>
95 #include <machine/md_var.h>
97 #include <vm/vm_page2.h>
98 #include <sys/mplock2.h>
100 #define VMACTION_HSIZE 256
101 #define VMACTION_HMASK (VMACTION_HSIZE - 1)
103 static void vm_page_queue_init(void);
104 static void vm_page_free_wakeup(void);
105 static vm_page_t vm_page_select_cache(vm_object_t, vm_pindex_t);
106 static vm_page_t _vm_page_list_find2(int basequeue, int index);
108 struct vpgqueues vm_page_queues[PQ_COUNT]; /* Array of tailq lists */
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;
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 * Preallocates memory for critical VM structures and arrays prior to
205 * kernel_map becoming available.
207 * Memory is allocated from (virtual2_start, virtual2_end) if available,
208 * otherwise memory is allocated from (virtual_start, virtual_end).
210 * On x86-64 (virtual_start, virtual_end) is only 2GB and may not be
211 * large enough to hold vm_page_array & other structures for machines with
212 * large amounts of ram, so we want to use virtual2* when available.
215 vm_page_startup(void)
217 vm_offset_t vaddr = virtual2_start ? virtual2_start : virtual_start;
220 vm_paddr_t page_range;
227 vm_paddr_t biggestone, biggestsize;
234 vaddr = round_page(vaddr);
236 for (i = 0; phys_avail[i + 1]; i += 2) {
237 phys_avail[i] = round_page64(phys_avail[i]);
238 phys_avail[i + 1] = trunc_page64(phys_avail[i + 1]);
241 for (i = 0; phys_avail[i + 1]; i += 2) {
242 vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
244 if (size > biggestsize) {
252 end = phys_avail[biggestone+1];
253 end = trunc_page(end);
256 * Initialize the queue headers for the free queue, the active queue
257 * and the inactive queue.
260 vm_page_queue_init();
262 /* VKERNELs don't support minidumps and as such don't need vm_page_dump */
263 #if !defined(_KERNEL_VIRTUAL)
265 * Allocate a bitmap to indicate that a random physical page
266 * needs to be included in a minidump.
268 * The amd64 port needs this to indicate which direct map pages
269 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
271 * However, i386 still needs this workspace internally within the
272 * minidump code. In theory, they are not needed on i386, but are
273 * included should the sf_buf code decide to use them.
275 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
276 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
277 end -= vm_page_dump_size;
278 vm_page_dump = (void *)pmap_map(&vaddr, end, end + vm_page_dump_size,
279 VM_PROT_READ | VM_PROT_WRITE);
280 bzero((void *)vm_page_dump, vm_page_dump_size);
284 * Compute the number of pages of memory that will be available for
285 * use (taking into account the overhead of a page structure per
288 first_page = phys_avail[0] / PAGE_SIZE;
289 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE - first_page;
290 npages = (total - (page_range * sizeof(struct vm_page))) / PAGE_SIZE;
293 * Initialize the mem entry structures now, and put them in the free
296 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
297 mapped = pmap_map(&vaddr, new_end, end,
298 VM_PROT_READ | VM_PROT_WRITE);
299 vm_page_array = (vm_page_t)mapped;
301 #if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
303 * since pmap_map on amd64 returns stuff out of a direct-map region,
304 * we have to manually add these pages to the minidump tracking so
305 * that they can be dumped, including the vm_page_array.
307 for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
312 * Clear all of the page structures
314 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
315 vm_page_array_size = page_range;
318 * Construct the free queue(s) in ascending order (by physical
319 * address) so that the first 16MB of physical memory is allocated
320 * last rather than first. On large-memory machines, this avoids
321 * the exhaustion of low physical memory before isa_dmainit has run.
323 vmstats.v_page_count = 0;
324 vmstats.v_free_count = 0;
325 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) {
330 last_pa = phys_avail[i + 1];
331 while (pa < last_pa && npages-- > 0) {
337 virtual2_start = vaddr;
339 virtual_start = vaddr;
343 * Scan comparison function for Red-Black tree scans. An inclusive
344 * (start,end) is expected. Other fields are not used.
347 rb_vm_page_scancmp(struct vm_page *p, void *data)
349 struct rb_vm_page_scan_info *info = data;
351 if (p->pindex < info->start_pindex)
353 if (p->pindex > info->end_pindex)
359 rb_vm_page_compare(struct vm_page *p1, struct vm_page *p2)
361 if (p1->pindex < p2->pindex)
363 if (p1->pindex > p2->pindex)
369 * Holding a page keeps it from being reused. Other parts of the system
370 * can still disassociate the page from its current object and free it, or
371 * perform read or write I/O on it and/or otherwise manipulate the page,
372 * but if the page is held the VM system will leave the page and its data
373 * intact and not reuse the page for other purposes until the last hold
374 * reference is released. (see vm_page_wire() if you want to prevent the
375 * page from being disassociated from its object too).
377 * The caller must hold vm_token.
379 * The caller must still validate the contents of the page and, if necessary,
380 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
381 * before manipulating the page.
384 vm_page_hold(vm_page_t m)
386 ASSERT_LWKT_TOKEN_HELD(&vm_token);
391 * The opposite of vm_page_hold(). A page can be freed while being held,
392 * which places it on the PQ_HOLD queue. We must call vm_page_free_toq()
393 * in this case to actually free it once the hold count drops to 0.
395 * The caller must hold vm_token if non-blocking operation is desired,
396 * but otherwise does not need to.
399 vm_page_unhold(vm_page_t m)
401 lwkt_gettoken(&vm_token);
403 KASSERT(m->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
404 if (m->hold_count == 0 && m->queue == PQ_HOLD) {
408 lwkt_reltoken(&vm_token);
412 * Inserts the given vm_page into the object and object list.
414 * The pagetables are not updated but will presumably fault the page
415 * in if necessary, or if a kernel page the caller will at some point
416 * enter the page into the kernel's pmap. We are not allowed to block
417 * here so we *can't* do this anyway.
419 * This routine may not block.
420 * This routine must be called with the vm_token held.
421 * This routine must be called with a critical section held.
424 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
426 ASSERT_LWKT_TOKEN_HELD(&vm_token);
427 if (m->object != NULL)
428 panic("vm_page_insert: already inserted");
431 * Record the object/offset pair in this page
437 * Insert it into the object.
439 vm_page_rb_tree_RB_INSERT(&object->rb_memq, m);
440 object->generation++;
443 * show that the object has one more resident page.
445 object->resident_page_count++;
448 * Add the pv_list_cout of the page when its inserted in
451 object->agg_pv_list_count = object->agg_pv_list_count + m->md.pv_list_count;
454 * Since we are inserting a new and possibly dirty page,
455 * update the object's OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY flags.
457 if ((m->valid & m->dirty) || (m->flags & PG_WRITEABLE))
458 vm_object_set_writeable_dirty(object);
461 * Checks for a swap assignment and sets PG_SWAPPED if appropriate.
463 swap_pager_page_inserted(m);
467 * Removes the given vm_page_t from the global (object,index) hash table
468 * and from the object's memq.
470 * The underlying pmap entry (if any) is NOT removed here.
471 * This routine may not block.
473 * The page must be BUSY and will remain BUSY on return.
474 * No other requirements.
476 * NOTE: FreeBSD side effect was to unbusy the page on return. We leave
480 vm_page_remove(vm_page_t m)
484 lwkt_gettoken(&vm_token);
485 if (m->object == NULL) {
486 lwkt_reltoken(&vm_token);
490 if ((m->flags & PG_BUSY) == 0)
491 panic("vm_page_remove: page not busy");
496 * Remove the page from the object and update the object.
498 vm_page_rb_tree_RB_REMOVE(&object->rb_memq, m);
499 object->resident_page_count--;
500 object->agg_pv_list_count = object->agg_pv_list_count - m->md.pv_list_count;
501 object->generation++;
504 lwkt_reltoken(&vm_token);
508 * Locate and return the page at (object, pindex), or NULL if the
509 * page could not be found.
511 * The caller must hold vm_token.
514 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
519 * Search the hash table for this object/offset pair
521 ASSERT_LWKT_TOKEN_HELD(&vm_token);
522 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
523 KKASSERT(m == NULL || (m->object == object && m->pindex == pindex));
530 * Move the given memory entry from its current object to the specified
531 * target object/offset.
533 * The object must be locked.
534 * This routine may not block.
536 * Note: This routine will raise itself to splvm(), the caller need not.
538 * Note: Swap associated with the page must be invalidated by the move. We
539 * have to do this for several reasons: (1) we aren't freeing the
540 * page, (2) we are dirtying the page, (3) the VM system is probably
541 * moving the page from object A to B, and will then later move
542 * the backing store from A to B and we can't have a conflict.
544 * Note: We *always* dirty the page. It is necessary both for the
545 * fact that we moved it, and because we may be invalidating
546 * swap. If the page is on the cache, we have to deactivate it
547 * or vm_page_dirty() will panic. Dirty pages are not allowed
551 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
553 lwkt_gettoken(&vm_token);
555 vm_page_insert(m, new_object, new_pindex);
556 if (m->queue - m->pc == PQ_CACHE)
557 vm_page_deactivate(m);
560 lwkt_reltoken(&vm_token);
564 * vm_page_unqueue() without any wakeup. This routine is used when a page
565 * is being moved between queues or otherwise is to remain BUSYied by the
568 * The caller must hold vm_token
569 * This routine may not block.
572 vm_page_unqueue_nowakeup(vm_page_t m)
574 int queue = m->queue;
575 struct vpgqueues *pq;
577 ASSERT_LWKT_TOKEN_HELD(&vm_token);
578 if (queue != PQ_NONE) {
579 pq = &vm_page_queues[queue];
581 TAILQ_REMOVE(&pq->pl, m, pageq);
588 * vm_page_unqueue() - Remove a page from its queue, wakeup the pagedemon
591 * The caller must hold vm_token
592 * This routine may not block.
595 vm_page_unqueue(vm_page_t m)
597 int queue = m->queue;
598 struct vpgqueues *pq;
600 ASSERT_LWKT_TOKEN_HELD(&vm_token);
601 if (queue != PQ_NONE) {
603 pq = &vm_page_queues[queue];
604 TAILQ_REMOVE(&pq->pl, m, pageq);
607 if ((queue - m->pc) == PQ_CACHE || (queue - m->pc) == PQ_FREE)
613 * vm_page_list_find()
615 * Find a page on the specified queue with color optimization.
617 * The page coloring optimization attempts to locate a page that does
618 * not overload other nearby pages in the object in the cpu's L1 or L2
619 * caches. We need this optimization because cpu caches tend to be
620 * physical caches, while object spaces tend to be virtual.
622 * Must be called with vm_token held.
623 * This routine may not block.
625 * Note that this routine is carefully inlined. A non-inlined version
626 * is available for outside callers but the only critical path is
627 * from within this source file.
631 _vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
636 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
638 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
640 m = _vm_page_list_find2(basequeue, index);
645 _vm_page_list_find2(int basequeue, int index)
649 struct vpgqueues *pq;
651 pq = &vm_page_queues[basequeue];
654 * Note that for the first loop, index+i and index-i wind up at the
655 * same place. Even though this is not totally optimal, we've already
656 * blown it by missing the cache case so we do not care.
659 for(i = PQ_L2_SIZE / 2; i > 0; --i) {
660 if ((m = TAILQ_FIRST(&pq[(index + i) & PQ_L2_MASK].pl)) != NULL)
663 if ((m = TAILQ_FIRST(&pq[(index - i) & PQ_L2_MASK].pl)) != NULL)
670 * Must be called with vm_token held if the caller desired non-blocking
671 * operation and a stable result.
674 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
676 return(_vm_page_list_find(basequeue, index, prefer_zero));
680 * Find a page on the cache queue with color optimization. As pages
681 * might be found, but not applicable, they are deactivated. This
682 * keeps us from using potentially busy cached pages.
684 * This routine may not block.
685 * Must be called with vm_token held.
688 vm_page_select_cache(vm_object_t object, vm_pindex_t pindex)
692 ASSERT_LWKT_TOKEN_HELD(&vm_token);
694 m = _vm_page_list_find(
696 (pindex + object->pg_color) & PQ_L2_MASK,
699 if (m && ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
700 m->hold_count || m->wire_count)) {
701 vm_page_deactivate(m);
710 * Find a free or zero page, with specified preference. We attempt to
711 * inline the nominal case and fall back to _vm_page_select_free()
714 * This routine must be called with a critical section held.
715 * This routine may not block.
717 static __inline vm_page_t
718 vm_page_select_free(vm_object_t object, vm_pindex_t pindex, boolean_t prefer_zero)
722 m = _vm_page_list_find(
724 (pindex + object->pg_color) & PQ_L2_MASK,
733 * Allocate and return a memory cell associated with this VM object/offset
738 * VM_ALLOC_NORMAL allow use of cache pages, nominal free drain
739 * VM_ALLOC_QUICK like normal but cannot use cache
740 * VM_ALLOC_SYSTEM greater free drain
741 * VM_ALLOC_INTERRUPT allow free list to be completely drained
742 * VM_ALLOC_ZERO advisory request for pre-zero'd page
744 * The object must be locked.
745 * This routine may not block.
746 * The returned page will be marked PG_BUSY
748 * Additional special handling is required when called from an interrupt
749 * (VM_ALLOC_INTERRUPT). We are not allowed to mess with the page cache
753 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int page_req)
757 lwkt_gettoken(&vm_token);
759 KKASSERT(object != NULL);
760 KASSERT(!vm_page_lookup(object, pindex),
761 ("vm_page_alloc: page already allocated"));
763 (VM_ALLOC_NORMAL|VM_ALLOC_QUICK|
764 VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
767 * Certain system threads (pageout daemon, buf_daemon's) are
768 * allowed to eat deeper into the free page list.
770 if (curthread->td_flags & TDF_SYSTHREAD)
771 page_req |= VM_ALLOC_SYSTEM;
774 if (vmstats.v_free_count > vmstats.v_free_reserved ||
775 ((page_req & VM_ALLOC_INTERRUPT) && vmstats.v_free_count > 0) ||
776 ((page_req & VM_ALLOC_SYSTEM) && vmstats.v_cache_count == 0 &&
777 vmstats.v_free_count > vmstats.v_interrupt_free_min)
780 * The free queue has sufficient free pages to take one out.
782 if (page_req & VM_ALLOC_ZERO)
783 m = vm_page_select_free(object, pindex, TRUE);
785 m = vm_page_select_free(object, pindex, FALSE);
786 } else if (page_req & VM_ALLOC_NORMAL) {
788 * Allocatable from the cache (non-interrupt only). On
789 * success, we must free the page and try again, thus
790 * ensuring that vmstats.v_*_free_min counters are replenished.
793 if (curthread->td_preempted) {
794 kprintf("vm_page_alloc(): warning, attempt to allocate"
795 " cache page from preempting interrupt\n");
798 m = vm_page_select_cache(object, pindex);
801 m = vm_page_select_cache(object, pindex);
804 * On success move the page into the free queue and loop.
807 KASSERT(m->dirty == 0,
808 ("Found dirty cache page %p", m));
810 vm_page_protect(m, VM_PROT_NONE);
816 * On failure return NULL
818 lwkt_reltoken(&vm_token);
819 #if defined(DIAGNOSTIC)
820 if (vmstats.v_cache_count > 0)
821 kprintf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", vmstats.v_cache_count);
823 vm_pageout_deficit++;
828 * No pages available, wakeup the pageout daemon and give up.
830 lwkt_reltoken(&vm_token);
831 vm_pageout_deficit++;
837 * Good page found. The page has not yet been busied. We are in
838 * a critical section.
840 KASSERT(m != NULL, ("vm_page_alloc(): missing page on free queue\n"));
841 KASSERT(m->dirty == 0,
842 ("vm_page_alloc: free/cache page %p was dirty", m));
845 * Remove from free queue
847 vm_page_unqueue_nowakeup(m);
850 * Initialize structure. Only the PG_ZERO flag is inherited. Set
853 if (m->flags & PG_ZERO) {
854 vm_page_zero_count--;
855 m->flags = PG_ZERO | PG_BUSY;
866 * vm_page_insert() is safe while holding vm_token. Note also that
867 * inserting a page here does not insert it into the pmap (which
868 * could cause us to block allocating memory). We cannot block
871 vm_page_insert(m, object, pindex);
874 * Don't wakeup too often - wakeup the pageout daemon when
875 * we would be nearly out of memory.
879 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.
913 * The caller may loop if vm_page_count_min() == FALSE so we cannot be
914 * more generous then that.
924 lwkt_gettoken(&vm_token);
926 if (curthread == pagethread) {
928 * The pageout daemon itself needs pages, this is bad.
930 if (vm_page_count_min(0)) {
931 vm_pageout_pages_needed = 1;
932 tsleep(&vm_pageout_pages_needed, 0, "VMWait", timo);
936 * Wakeup the pageout daemon if necessary and wait.
938 if (vm_page_count_target()) {
939 if (vm_pages_needed == 0) {
941 wakeup(&vm_pages_needed);
943 ++vm_pages_waiting; /* SMP race ok */
944 tsleep(&vmstats.v_free_count, 0, "vmwait", timo);
947 lwkt_reltoken(&vm_token);
951 * Block until free pages are available for allocation
953 * Called only from vm_fault so that processes page faulting can be
960 * Wakeup the pageout daemon if necessary and wait.
962 if (vm_page_count_target()) {
963 lwkt_gettoken(&vm_token);
964 if (vm_page_count_target()) {
965 if (vm_pages_needed == 0) {
967 wakeup(&vm_pages_needed);
969 ++vm_pages_waiting; /* SMP race ok */
970 tsleep(&vmstats.v_free_count, 0, "pfault", hz);
972 lwkt_reltoken(&vm_token);
977 * Put the specified page on the active list (if appropriate). Ensure
978 * that act_count is at least ACT_INIT but do not otherwise mess with it.
980 * The page queues must be locked.
981 * This routine may not block.
984 vm_page_activate(vm_page_t m)
986 lwkt_gettoken(&vm_token);
987 if (m->queue != PQ_ACTIVE) {
988 if ((m->queue - m->pc) == PQ_CACHE)
989 mycpu->gd_cnt.v_reactivated++;
993 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
994 m->queue = PQ_ACTIVE;
995 vm_page_queues[PQ_ACTIVE].lcnt++;
996 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_ACTIVE].pl,
998 if (m->act_count < ACT_INIT)
999 m->act_count = ACT_INIT;
1000 vmstats.v_active_count++;
1003 if (m->act_count < ACT_INIT)
1004 m->act_count = ACT_INIT;
1006 lwkt_reltoken(&vm_token);
1010 * Helper routine for vm_page_free_toq() and vm_page_cache(). This
1011 * routine is called when a page has been added to the cache or free
1014 * This routine may not block.
1015 * This routine must be called at splvm()
1017 static __inline void
1018 vm_page_free_wakeup(void)
1021 * If the pageout daemon itself needs pages, then tell it that
1022 * there are some free.
1024 if (vm_pageout_pages_needed &&
1025 vmstats.v_cache_count + vmstats.v_free_count >=
1026 vmstats.v_pageout_free_min
1028 wakeup(&vm_pageout_pages_needed);
1029 vm_pageout_pages_needed = 0;
1033 * Wakeup processes that are waiting on memory.
1035 * NOTE: vm_paging_target() is the pageout daemon's target, while
1036 * vm_page_count_target() is somewhere inbetween. We want
1037 * to wake processes up prior to the pageout daemon reaching
1038 * its target to provide some hysteresis.
1040 if (vm_pages_waiting) {
1041 if (!vm_page_count_target()) {
1043 * Plenty of pages are free, wakeup everyone.
1045 vm_pages_waiting = 0;
1046 wakeup(&vmstats.v_free_count);
1047 ++mycpu->gd_cnt.v_ppwakeups;
1048 } else if (!vm_page_count_min(0)) {
1050 * Some pages are free, wakeup someone.
1052 int wcount = vm_pages_waiting;
1055 vm_pages_waiting = wcount;
1056 wakeup_one(&vmstats.v_free_count);
1057 ++mycpu->gd_cnt.v_ppwakeups;
1065 * Returns the given page to the PQ_FREE list, disassociating it with
1068 * The vm_page must be PG_BUSY on entry. PG_BUSY will be released on
1069 * return (the page will have been freed). No particular spl is required
1072 * This routine may not block.
1075 vm_page_free_toq(vm_page_t m)
1077 struct vpgqueues *pq;
1079 lwkt_gettoken(&vm_token);
1080 mycpu->gd_cnt.v_tfree++;
1082 KKASSERT((m->flags & PG_MAPPED) == 0);
1084 if (m->busy || ((m->queue - m->pc) == PQ_FREE)) {
1086 "vm_page_free: pindex(%lu), busy(%d), PG_BUSY(%d), hold(%d)\n",
1087 (u_long)m->pindex, m->busy, (m->flags & PG_BUSY) ? 1 : 0,
1089 if ((m->queue - m->pc) == PQ_FREE)
1090 panic("vm_page_free: freeing free page");
1092 panic("vm_page_free: freeing busy page");
1096 * unqueue, then remove page. Note that we cannot destroy
1097 * the page here because we do not want to call the pager's
1098 * callback routine until after we've put the page on the
1099 * appropriate free queue.
1101 vm_page_unqueue_nowakeup(m);
1105 * No further management of fictitious pages occurs beyond object
1106 * and queue removal.
1108 if ((m->flags & PG_FICTITIOUS) != 0) {
1110 lwkt_reltoken(&vm_token);
1117 if (m->wire_count != 0) {
1118 if (m->wire_count > 1) {
1120 "vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1121 m->wire_count, (long)m->pindex);
1123 panic("vm_page_free: freeing wired page");
1127 * Clear the UNMANAGED flag when freeing an unmanaged page.
1129 if (m->flags & PG_UNMANAGED) {
1130 m->flags &= ~PG_UNMANAGED;
1133 if (m->hold_count != 0) {
1134 m->flags &= ~PG_ZERO;
1137 m->queue = PQ_FREE + m->pc;
1139 pq = &vm_page_queues[m->queue];
1144 * Put zero'd pages on the end ( where we look for zero'd pages
1145 * first ) and non-zerod pages at the head.
1147 if (m->flags & PG_ZERO) {
1148 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
1149 ++vm_page_zero_count;
1151 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
1154 vm_page_free_wakeup();
1155 lwkt_reltoken(&vm_token);
1159 * vm_page_free_fromq_fast()
1161 * Remove a non-zero page from one of the free queues; the page is removed for
1162 * zeroing, so do not issue a wakeup.
1167 vm_page_free_fromq_fast(void)
1173 lwkt_gettoken(&vm_token);
1174 for (i = 0; i < PQ_L2_SIZE; ++i) {
1175 m = vm_page_list_find(PQ_FREE, qi, FALSE);
1176 qi = (qi + PQ_PRIME2) & PQ_L2_MASK;
1177 if (m && (m->flags & PG_ZERO) == 0) {
1178 KKASSERT(m->busy == 0 && (m->flags & PG_BUSY) == 0);
1179 vm_page_unqueue_nowakeup(m);
1185 lwkt_reltoken(&vm_token);
1190 * vm_page_unmanage()
1192 * Prevent PV management from being done on the page. The page is
1193 * removed from the paging queues as if it were wired, and as a
1194 * consequence of no longer being managed the pageout daemon will not
1195 * touch it (since there is no way to locate the pte mappings for the
1196 * page). madvise() calls that mess with the pmap will also no longer
1197 * operate on the page.
1199 * Beyond that the page is still reasonably 'normal'. Freeing the page
1200 * will clear the flag.
1202 * This routine is used by OBJT_PHYS objects - objects using unswappable
1203 * physical memory as backing store rather then swap-backed memory and
1204 * will eventually be extended to support 4MB unmanaged physical
1207 * Must be called with a critical section held.
1208 * Must be called with vm_token held.
1211 vm_page_unmanage(vm_page_t m)
1213 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1214 if ((m->flags & PG_UNMANAGED) == 0) {
1215 if (m->wire_count == 0)
1218 vm_page_flag_set(m, PG_UNMANAGED);
1222 * Mark this page as wired down by yet another map, removing it from
1223 * paging queues as necessary.
1225 * The page queues must be locked.
1226 * This routine may not block.
1229 vm_page_wire(vm_page_t m)
1232 * Only bump the wire statistics if the page is not already wired,
1233 * and only unqueue the page if it is on some queue (if it is unmanaged
1234 * it is already off the queues). Don't do anything with fictitious
1235 * pages because they are always wired.
1237 lwkt_gettoken(&vm_token);
1238 if ((m->flags & PG_FICTITIOUS) == 0) {
1239 if (m->wire_count == 0) {
1240 if ((m->flags & PG_UNMANAGED) == 0)
1242 vmstats.v_wire_count++;
1245 KASSERT(m->wire_count != 0,
1246 ("vm_page_wire: wire_count overflow m=%p", m));
1248 lwkt_reltoken(&vm_token);
1252 * Release one wiring of this page, potentially enabling it to be paged again.
1254 * Many pages placed on the inactive queue should actually go
1255 * into the cache, but it is difficult to figure out which. What
1256 * we do instead, if the inactive target is well met, is to put
1257 * clean pages at the head of the inactive queue instead of the tail.
1258 * This will cause them to be moved to the cache more quickly and
1259 * if not actively re-referenced, freed more quickly. If we just
1260 * stick these pages at the end of the inactive queue, heavy filesystem
1261 * meta-data accesses can cause an unnecessary paging load on memory bound
1262 * processes. This optimization causes one-time-use metadata to be
1263 * reused more quickly.
1265 * BUT, if we are in a low-memory situation we have no choice but to
1266 * put clean pages on the cache queue.
1268 * A number of routines use vm_page_unwire() to guarantee that the page
1269 * will go into either the inactive or active queues, and will NEVER
1270 * be placed in the cache - for example, just after dirtying a page.
1271 * dirty pages in the cache are not allowed.
1273 * The page queues must be locked.
1274 * This routine may not block.
1277 vm_page_unwire(vm_page_t m, int activate)
1279 lwkt_gettoken(&vm_token);
1280 if (m->flags & PG_FICTITIOUS) {
1282 } else if (m->wire_count <= 0) {
1283 panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
1285 if (--m->wire_count == 0) {
1286 --vmstats.v_wire_count;
1287 if (m->flags & PG_UNMANAGED) {
1289 } else if (activate) {
1291 &vm_page_queues[PQ_ACTIVE].pl, m, pageq);
1292 m->queue = PQ_ACTIVE;
1293 vm_page_queues[PQ_ACTIVE].lcnt++;
1294 vmstats.v_active_count++;
1296 vm_page_flag_clear(m, PG_WINATCFLS);
1298 &vm_page_queues[PQ_INACTIVE].pl, m, pageq);
1299 m->queue = PQ_INACTIVE;
1300 vm_page_queues[PQ_INACTIVE].lcnt++;
1301 vmstats.v_inactive_count++;
1302 ++vm_swapcache_inactive_heuristic;
1306 lwkt_reltoken(&vm_token);
1311 * Move the specified page to the inactive queue. If the page has
1312 * any associated swap, the swap is deallocated.
1314 * Normally athead is 0 resulting in LRU operation. athead is set
1315 * to 1 if we want this page to be 'as if it were placed in the cache',
1316 * except without unmapping it from the process address space.
1318 * This routine may not block.
1319 * The caller must hold vm_token.
1321 static __inline void
1322 _vm_page_deactivate(vm_page_t m, int athead)
1325 * Ignore if already inactive.
1327 if (m->queue == PQ_INACTIVE)
1330 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1331 if ((m->queue - m->pc) == PQ_CACHE)
1332 mycpu->gd_cnt.v_reactivated++;
1333 vm_page_flag_clear(m, PG_WINATCFLS);
1336 TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl,
1339 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl,
1341 ++vm_swapcache_inactive_heuristic;
1343 m->queue = PQ_INACTIVE;
1344 vm_page_queues[PQ_INACTIVE].lcnt++;
1345 vmstats.v_inactive_count++;
1350 * Attempt to deactivate a page.
1355 vm_page_deactivate(vm_page_t m)
1357 lwkt_gettoken(&vm_token);
1358 _vm_page_deactivate(m, 0);
1359 lwkt_reltoken(&vm_token);
1363 * Attempt to move a page to PQ_CACHE.
1364 * Returns 0 on failure, 1 on success
1369 vm_page_try_to_cache(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);
1377 vm_page_test_dirty(m);
1379 lwkt_reltoken(&vm_token);
1383 lwkt_reltoken(&vm_token);
1388 * Attempt to free the page. If we cannot free it, we do nothing.
1389 * 1 is returned on success, 0 on failure.
1394 vm_page_try_to_free(vm_page_t m)
1396 lwkt_gettoken(&vm_token);
1397 if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1398 (m->flags & (PG_BUSY|PG_UNMANAGED))) {
1399 lwkt_reltoken(&vm_token);
1402 vm_page_test_dirty(m);
1404 lwkt_reltoken(&vm_token);
1408 vm_page_protect(m, VM_PROT_NONE);
1410 lwkt_reltoken(&vm_token);
1417 * Put the specified page onto the page cache queue (if appropriate).
1419 * The caller must hold vm_token.
1420 * This routine may not block.
1423 vm_page_cache(vm_page_t m)
1425 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1427 if ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
1428 m->wire_count || m->hold_count) {
1429 kprintf("vm_page_cache: attempting to cache busy/held page\n");
1434 * Already in the cache (and thus not mapped)
1436 if ((m->queue - m->pc) == PQ_CACHE) {
1437 KKASSERT((m->flags & PG_MAPPED) == 0);
1442 * Caller is required to test m->dirty, but note that the act of
1443 * removing the page from its maps can cause it to become dirty
1444 * on an SMP system due to another cpu running in usermode.
1447 panic("vm_page_cache: caching a dirty page, pindex: %ld",
1452 * Remove all pmaps and indicate that the page is not
1453 * writeable or mapped. Our vm_page_protect() call may
1454 * have blocked (especially w/ VM_PROT_NONE), so recheck
1458 vm_page_protect(m, VM_PROT_NONE);
1460 if ((m->flags & (PG_BUSY|PG_UNMANAGED|PG_MAPPED)) || m->busy ||
1461 m->wire_count || m->hold_count) {
1463 } else if (m->dirty) {
1464 vm_page_deactivate(m);
1466 vm_page_unqueue_nowakeup(m);
1467 m->queue = PQ_CACHE + m->pc;
1468 vm_page_queues[m->queue].lcnt++;
1469 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, pageq);
1470 vmstats.v_cache_count++;
1471 vm_page_free_wakeup();
1476 * vm_page_dontneed()
1478 * Cache, deactivate, or do nothing as appropriate. This routine
1479 * is typically used by madvise() MADV_DONTNEED.
1481 * Generally speaking we want to move the page into the cache so
1482 * it gets reused quickly. However, this can result in a silly syndrome
1483 * due to the page recycling too quickly. Small objects will not be
1484 * fully cached. On the otherhand, if we move the page to the inactive
1485 * queue we wind up with a problem whereby very large objects
1486 * unnecessarily blow away our inactive and cache queues.
1488 * The solution is to move the pages based on a fixed weighting. We
1489 * either leave them alone, deactivate them, or move them to the cache,
1490 * where moving them to the cache has the highest weighting.
1491 * By forcing some pages into other queues we eventually force the
1492 * system to balance the queues, potentially recovering other unrelated
1493 * space from active. The idea is to not force this to happen too
1499 vm_page_dontneed(vm_page_t m)
1501 static int dnweight;
1508 * occassionally leave the page alone
1510 lwkt_gettoken(&vm_token);
1511 if ((dnw & 0x01F0) == 0 ||
1512 m->queue == PQ_INACTIVE ||
1513 m->queue - m->pc == PQ_CACHE
1515 if (m->act_count >= ACT_INIT)
1517 lwkt_reltoken(&vm_token);
1522 vm_page_test_dirty(m);
1524 if (m->dirty || (dnw & 0x0070) == 0) {
1526 * Deactivate the page 3 times out of 32.
1531 * Cache the page 28 times out of every 32. Note that
1532 * the page is deactivated instead of cached, but placed
1533 * at the head of the queue instead of the tail.
1537 _vm_page_deactivate(m, head);
1538 lwkt_reltoken(&vm_token);
1542 * Grab a page, blocking if it is busy and allocating a page if necessary.
1543 * A busy page is returned or NULL.
1545 * If VM_ALLOC_RETRY is specified VM_ALLOC_NORMAL must also be specified.
1546 * If VM_ALLOC_RETRY is not specified
1548 * This routine may block, but if VM_ALLOC_RETRY is not set then NULL is
1549 * always returned if we had blocked.
1550 * This routine will never return NULL if VM_ALLOC_RETRY is set.
1551 * This routine may not be called from an interrupt.
1552 * The returned page may not be entirely valid.
1554 * This routine may be called from mainline code without spl protection and
1555 * be guarenteed a busied page associated with the object at the specified
1561 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
1566 KKASSERT(allocflags &
1567 (VM_ALLOC_NORMAL|VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
1568 lwkt_gettoken(&vm_token);
1570 if ((m = vm_page_lookup(object, pindex)) != NULL) {
1571 if (m->busy || (m->flags & PG_BUSY)) {
1572 generation = object->generation;
1574 while ((object->generation == generation) &&
1575 (m->busy || (m->flags & PG_BUSY))) {
1576 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
1577 tsleep(m, 0, "pgrbwt", 0);
1578 if ((allocflags & VM_ALLOC_RETRY) == 0) {
1589 m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
1592 if ((allocflags & VM_ALLOC_RETRY) == 0)
1597 lwkt_reltoken(&vm_token);
1602 * Mapping function for valid bits or for dirty bits in
1603 * a page. May not block.
1605 * Inputs are required to range within a page.
1611 vm_page_bits(int base, int size)
1617 base + size <= PAGE_SIZE,
1618 ("vm_page_bits: illegal base/size %d/%d", base, size)
1621 if (size == 0) /* handle degenerate case */
1624 first_bit = base >> DEV_BSHIFT;
1625 last_bit = (base + size - 1) >> DEV_BSHIFT;
1627 return ((2 << last_bit) - (1 << first_bit));
1631 * Sets portions of a page valid and clean. The arguments are expected
1632 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1633 * of any partial chunks touched by the range. The invalid portion of
1634 * such chunks will be zero'd.
1636 * NOTE: When truncating a buffer vnode_pager_setsize() will automatically
1637 * align base to DEV_BSIZE so as not to mark clean a partially
1638 * truncated device block. Otherwise the dirty page status might be
1641 * This routine may not block.
1643 * (base + size) must be less then or equal to PAGE_SIZE.
1646 _vm_page_zero_valid(vm_page_t m, int base, int size)
1651 if (size == 0) /* handle degenerate case */
1655 * If the base is not DEV_BSIZE aligned and the valid
1656 * bit is clear, we have to zero out a portion of the
1660 if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1661 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0
1663 pmap_zero_page_area(
1671 * If the ending offset is not DEV_BSIZE aligned and the
1672 * valid bit is clear, we have to zero out a portion of
1676 endoff = base + size;
1678 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1679 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0
1681 pmap_zero_page_area(
1684 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))
1690 * Set valid, clear dirty bits. If validating the entire
1691 * page we can safely clear the pmap modify bit. We also
1692 * use this opportunity to clear the PG_NOSYNC flag. If a process
1693 * takes a write fault on a MAP_NOSYNC memory area the flag will
1696 * We set valid bits inclusive of any overlap, but we can only
1697 * clear dirty bits for DEV_BSIZE chunks that are fully within
1700 * Page must be busied?
1701 * No other requirements.
1704 vm_page_set_valid(vm_page_t m, int base, int size)
1706 _vm_page_zero_valid(m, base, size);
1707 m->valid |= vm_page_bits(base, size);
1712 * Set valid bits and clear dirty bits.
1714 * NOTE: This function does not clear the pmap modified bit.
1715 * Also note that e.g. NFS may use a byte-granular base
1718 * WARNING: Page must be busied? But vfs_clean_one_page() will call
1719 * this without necessarily busying the page (via bdwrite()).
1720 * So for now vm_token must also be held.
1722 * No other requirements.
1725 vm_page_set_validclean(vm_page_t m, int base, int size)
1729 _vm_page_zero_valid(m, base, size);
1730 pagebits = vm_page_bits(base, size);
1731 m->valid |= pagebits;
1732 m->dirty &= ~pagebits;
1733 if (base == 0 && size == PAGE_SIZE) {
1734 /*pmap_clear_modify(m);*/
1735 vm_page_flag_clear(m, PG_NOSYNC);
1740 * Set valid & dirty. Used by buwrite()
1742 * WARNING: Page must be busied? But vfs_dirty_one_page() will
1743 * call this function in buwrite() so for now vm_token must
1746 * No other requirements.
1749 vm_page_set_validdirty(vm_page_t m, int base, int size)
1753 pagebits = vm_page_bits(base, size);
1754 m->valid |= pagebits;
1755 m->dirty |= pagebits;
1757 vm_object_set_writeable_dirty(m->object);
1763 * NOTE: This function does not clear the pmap modified bit.
1764 * Also note that e.g. NFS may use a byte-granular base
1767 * Page must be busied?
1768 * No other requirements.
1771 vm_page_clear_dirty(vm_page_t m, int base, int size)
1773 m->dirty &= ~vm_page_bits(base, size);
1774 if (base == 0 && size == PAGE_SIZE) {
1775 /*pmap_clear_modify(m);*/
1776 vm_page_flag_clear(m, PG_NOSYNC);
1781 * Make the page all-dirty.
1783 * Also make sure the related object and vnode reflect the fact that the
1784 * object may now contain a dirty page.
1786 * Page must be busied?
1787 * No other requirements.
1790 vm_page_dirty(vm_page_t m)
1793 int pqtype = m->queue - m->pc;
1795 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
1796 ("vm_page_dirty: page in free/cache queue!"));
1797 if (m->dirty != VM_PAGE_BITS_ALL) {
1798 m->dirty = VM_PAGE_BITS_ALL;
1800 vm_object_set_writeable_dirty(m->object);
1805 * Invalidates DEV_BSIZE'd chunks within a page. Both the
1806 * valid and dirty bits for the effected areas are cleared.
1808 * Page must be busied?
1810 * No other requirements.
1813 vm_page_set_invalid(vm_page_t m, int base, int size)
1817 bits = vm_page_bits(base, size);
1820 m->object->generation++;
1824 * The kernel assumes that the invalid portions of a page contain
1825 * garbage, but such pages can be mapped into memory by user code.
1826 * When this occurs, we must zero out the non-valid portions of the
1827 * page so user code sees what it expects.
1829 * Pages are most often semi-valid when the end of a file is mapped
1830 * into memory and the file's size is not page aligned.
1832 * Page must be busied?
1833 * No other requirements.
1836 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
1842 * Scan the valid bits looking for invalid sections that
1843 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
1844 * valid bit may be set ) have already been zerod by
1845 * vm_page_set_validclean().
1847 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
1848 if (i == (PAGE_SIZE / DEV_BSIZE) ||
1849 (m->valid & (1 << i))
1852 pmap_zero_page_area(
1855 (i - b) << DEV_BSHIFT
1863 * setvalid is TRUE when we can safely set the zero'd areas
1864 * as being valid. We can do this if there are no cache consistency
1865 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
1868 m->valid = VM_PAGE_BITS_ALL;
1872 * Is a (partial) page valid? Note that the case where size == 0
1873 * will return FALSE in the degenerate case where the page is entirely
1874 * invalid, and TRUE otherwise.
1877 * No other requirements.
1880 vm_page_is_valid(vm_page_t m, int base, int size)
1882 int bits = vm_page_bits(base, size);
1884 if (m->valid && ((m->valid & bits) == bits))
1891 * update dirty bits from pmap/mmu. May not block.
1893 * Caller must hold vm_token if non-blocking operation desired.
1894 * No other requirements.
1897 vm_page_test_dirty(vm_page_t m)
1899 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
1905 * Register an action, associating it with its vm_page
1908 vm_page_register_action(vm_page_action_t action, vm_page_event_t event)
1910 struct vm_page_action_list *list;
1913 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1914 list = &action_list[hv];
1916 lwkt_gettoken(&vm_token);
1917 vm_page_flag_set(action->m, PG_ACTIONLIST);
1918 action->event = event;
1919 LIST_INSERT_HEAD(list, action, entry);
1920 lwkt_reltoken(&vm_token);
1924 * Unregister an action, disassociating it from its related vm_page
1927 vm_page_unregister_action(vm_page_action_t action)
1929 struct vm_page_action_list *list;
1932 lwkt_gettoken(&vm_token);
1933 if (action->event != VMEVENT_NONE) {
1934 action->event = VMEVENT_NONE;
1935 LIST_REMOVE(action, entry);
1937 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1938 list = &action_list[hv];
1939 if (LIST_EMPTY(list))
1940 vm_page_flag_clear(action->m, PG_ACTIONLIST);
1942 lwkt_reltoken(&vm_token);
1946 * Issue an event on a VM page. Corresponding action structures are
1947 * removed from the page's list and called.
1949 * If the vm_page has no more pending action events we clear its
1950 * PG_ACTIONLIST flag.
1953 vm_page_event_internal(vm_page_t m, vm_page_event_t event)
1955 struct vm_page_action_list *list;
1956 struct vm_page_action *scan;
1957 struct vm_page_action *next;
1961 hv = (int)((intptr_t)m >> 8) & VMACTION_HMASK;
1962 list = &action_list[hv];
1965 lwkt_gettoken(&vm_token);
1966 LIST_FOREACH_MUTABLE(scan, list, entry, next) {
1968 if (scan->event == event) {
1969 scan->event = VMEVENT_NONE;
1970 LIST_REMOVE(scan, entry);
1971 scan->func(m, scan);
1979 vm_page_flag_clear(m, PG_ACTIONLIST);
1980 lwkt_reltoken(&vm_token);
1984 #include "opt_ddb.h"
1986 #include <sys/kernel.h>
1988 #include <ddb/ddb.h>
1990 DB_SHOW_COMMAND(page, vm_page_print_page_info)
1992 db_printf("vmstats.v_free_count: %d\n", vmstats.v_free_count);
1993 db_printf("vmstats.v_cache_count: %d\n", vmstats.v_cache_count);
1994 db_printf("vmstats.v_inactive_count: %d\n", vmstats.v_inactive_count);
1995 db_printf("vmstats.v_active_count: %d\n", vmstats.v_active_count);
1996 db_printf("vmstats.v_wire_count: %d\n", vmstats.v_wire_count);
1997 db_printf("vmstats.v_free_reserved: %d\n", vmstats.v_free_reserved);
1998 db_printf("vmstats.v_free_min: %d\n", vmstats.v_free_min);
1999 db_printf("vmstats.v_free_target: %d\n", vmstats.v_free_target);
2000 db_printf("vmstats.v_cache_min: %d\n", vmstats.v_cache_min);
2001 db_printf("vmstats.v_inactive_target: %d\n", vmstats.v_inactive_target);
2004 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
2007 db_printf("PQ_FREE:");
2008 for(i=0;i<PQ_L2_SIZE;i++) {
2009 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt);
2013 db_printf("PQ_CACHE:");
2014 for(i=0;i<PQ_L2_SIZE;i++) {
2015 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt);
2019 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n",
2020 vm_page_queues[PQ_ACTIVE].lcnt,
2021 vm_page_queues[PQ_INACTIVE].lcnt);