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 * 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_LWKT_TOKEN_HELD(&vm_token);
422 if (m->object != NULL)
423 panic("vm_page_insert: already inserted");
426 * Record the object/offset pair in this page
432 * Insert it into the object.
434 vm_page_rb_tree_RB_INSERT(&object->rb_memq, m);
435 object->generation++;
438 * show that the object has one more resident page.
440 object->resident_page_count++;
443 * Add the pv_list_cout of the page when its inserted in
446 object->agg_pv_list_count = object->agg_pv_list_count + m->md.pv_list_count;
449 * Since we are inserting a new and possibly dirty page,
450 * update the object's OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY flags.
452 if ((m->valid & m->dirty) || (m->flags & PG_WRITEABLE))
453 vm_object_set_writeable_dirty(object);
456 * Checks for a swap assignment and sets PG_SWAPPED if appropriate.
458 swap_pager_page_inserted(m);
462 * Removes the given vm_page_t from the global (object,index) hash table
463 * and from the object's memq.
465 * The underlying pmap entry (if any) is NOT removed here.
466 * This routine may not block.
468 * The page must be BUSY and will remain BUSY on return.
469 * No other requirements.
471 * NOTE: FreeBSD side effect was to unbusy the page on return. We leave
475 vm_page_remove(vm_page_t m)
479 lwkt_gettoken(&vm_token);
480 if (m->object == NULL) {
481 lwkt_reltoken(&vm_token);
485 if ((m->flags & PG_BUSY) == 0)
486 panic("vm_page_remove: page not busy");
491 * Remove the page from the object and update the object.
493 vm_page_rb_tree_RB_REMOVE(&object->rb_memq, m);
494 object->resident_page_count--;
495 object->agg_pv_list_count = object->agg_pv_list_count - m->md.pv_list_count;
496 object->generation++;
499 lwkt_reltoken(&vm_token);
503 * Locate and return the page at (object, pindex), or NULL if the
504 * page could not be found.
506 * The caller must hold vm_token.
509 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
514 * Search the hash table for this object/offset pair
516 ASSERT_LWKT_TOKEN_HELD(&vm_token);
517 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
518 KKASSERT(m == NULL || (m->object == object && m->pindex == pindex));
525 * Move the given memory entry from its current object to the specified
526 * target object/offset.
528 * The object must be locked.
529 * This routine may not block.
531 * Note: This routine will raise itself to splvm(), the caller need not.
533 * Note: Swap associated with the page must be invalidated by the move. We
534 * have to do this for several reasons: (1) we aren't freeing the
535 * page, (2) we are dirtying the page, (3) the VM system is probably
536 * moving the page from object A to B, and will then later move
537 * the backing store from A to B and we can't have a conflict.
539 * Note: We *always* dirty the page. It is necessary both for the
540 * fact that we moved it, and because we may be invalidating
541 * swap. If the page is on the cache, we have to deactivate it
542 * or vm_page_dirty() will panic. Dirty pages are not allowed
546 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);
559 * vm_page_unqueue() without any wakeup. This routine is used when a page
560 * is being moved between queues or otherwise is to remain BUSYied by the
563 * The caller must hold vm_token
564 * This routine may not block.
567 vm_page_unqueue_nowakeup(vm_page_t m)
569 int queue = m->queue;
570 struct vpgqueues *pq;
572 ASSERT_LWKT_TOKEN_HELD(&vm_token);
573 if (queue != PQ_NONE) {
574 pq = &vm_page_queues[queue];
576 TAILQ_REMOVE(&pq->pl, m, pageq);
583 * vm_page_unqueue() - Remove a page from its queue, wakeup the pagedemon
586 * The caller must hold vm_token
587 * This routine may not block.
590 vm_page_unqueue(vm_page_t m)
592 int queue = m->queue;
593 struct vpgqueues *pq;
595 ASSERT_LWKT_TOKEN_HELD(&vm_token);
596 if (queue != PQ_NONE) {
598 pq = &vm_page_queues[queue];
599 TAILQ_REMOVE(&pq->pl, m, pageq);
602 if ((queue - m->pc) == PQ_CACHE || (queue - m->pc) == PQ_FREE)
608 * vm_page_list_find()
610 * Find a page on the specified queue with color optimization.
612 * The page coloring optimization attempts to locate a page that does
613 * not overload other nearby pages in the object in the cpu's L1 or L2
614 * caches. We need this optimization because cpu caches tend to be
615 * physical caches, while object spaces tend to be virtual.
617 * Must be called with vm_token held.
618 * This routine may not block.
620 * Note that this routine is carefully inlined. A non-inlined version
621 * is available for outside callers but the only critical path is
622 * from within this source file.
626 _vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
631 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
633 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
635 m = _vm_page_list_find2(basequeue, index);
640 _vm_page_list_find2(int basequeue, int index)
644 struct vpgqueues *pq;
646 pq = &vm_page_queues[basequeue];
649 * Note that for the first loop, index+i and index-i wind up at the
650 * same place. Even though this is not totally optimal, we've already
651 * blown it by missing the cache case so we do not care.
654 for(i = PQ_L2_SIZE / 2; i > 0; --i) {
655 if ((m = TAILQ_FIRST(&pq[(index + i) & PQ_L2_MASK].pl)) != NULL)
658 if ((m = TAILQ_FIRST(&pq[(index - i) & PQ_L2_MASK].pl)) != NULL)
665 * Must be called with vm_token held if the caller desired non-blocking
666 * operation and a stable result.
669 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
671 return(_vm_page_list_find(basequeue, index, prefer_zero));
675 * Find a page on the cache queue with color optimization. As pages
676 * might be found, but not applicable, they are deactivated. This
677 * keeps us from using potentially busy cached pages.
679 * This routine may not block.
680 * Must be called with vm_token held.
683 vm_page_select_cache(vm_object_t object, vm_pindex_t pindex)
687 ASSERT_LWKT_TOKEN_HELD(&vm_token);
689 m = _vm_page_list_find(
691 (pindex + object->pg_color) & PQ_L2_MASK,
694 if (m && ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
695 m->hold_count || m->wire_count)) {
696 vm_page_deactivate(m);
705 * Find a free or zero page, with specified preference. We attempt to
706 * inline the nominal case and fall back to _vm_page_select_free()
709 * This routine must be called with a critical section held.
710 * This routine may not block.
712 static __inline vm_page_t
713 vm_page_select_free(vm_object_t object, vm_pindex_t pindex, boolean_t prefer_zero)
717 m = _vm_page_list_find(
719 (pindex + object->pg_color) & PQ_L2_MASK,
728 * Allocate and return a memory cell associated with this VM object/offset
733 * VM_ALLOC_NORMAL allow use of cache pages, nominal free drain
734 * VM_ALLOC_QUICK like normal but cannot use cache
735 * VM_ALLOC_SYSTEM greater free drain
736 * VM_ALLOC_INTERRUPT allow free list to be completely drained
737 * VM_ALLOC_ZERO advisory request for pre-zero'd page
739 * The object must be locked.
740 * This routine may not block.
741 * The returned page will be marked PG_BUSY
743 * Additional special handling is required when called from an interrupt
744 * (VM_ALLOC_INTERRUPT). We are not allowed to mess with the page cache
748 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int page_req)
752 lwkt_gettoken(&vm_token);
754 KKASSERT(object != NULL);
755 KASSERT(!vm_page_lookup(object, pindex),
756 ("vm_page_alloc: page already allocated"));
758 (VM_ALLOC_NORMAL|VM_ALLOC_QUICK|
759 VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
762 * Certain system threads (pageout daemon, buf_daemon's) are
763 * allowed to eat deeper into the free page list.
765 if (curthread->td_flags & TDF_SYSTHREAD)
766 page_req |= VM_ALLOC_SYSTEM;
769 if (vmstats.v_free_count > vmstats.v_free_reserved ||
770 ((page_req & VM_ALLOC_INTERRUPT) && vmstats.v_free_count > 0) ||
771 ((page_req & VM_ALLOC_SYSTEM) && vmstats.v_cache_count == 0 &&
772 vmstats.v_free_count > vmstats.v_interrupt_free_min)
775 * The free queue has sufficient free pages to take one out.
777 if (page_req & VM_ALLOC_ZERO)
778 m = vm_page_select_free(object, pindex, TRUE);
780 m = vm_page_select_free(object, pindex, FALSE);
781 } else if (page_req & VM_ALLOC_NORMAL) {
783 * Allocatable from the cache (non-interrupt only). On
784 * success, we must free the page and try again, thus
785 * ensuring that vmstats.v_*_free_min counters are replenished.
788 if (curthread->td_preempted) {
789 kprintf("vm_page_alloc(): warning, attempt to allocate"
790 " cache page from preempting interrupt\n");
793 m = vm_page_select_cache(object, pindex);
796 m = vm_page_select_cache(object, pindex);
799 * On success move the page into the free queue and loop.
802 KASSERT(m->dirty == 0,
803 ("Found dirty cache page %p", m));
805 vm_page_protect(m, VM_PROT_NONE);
811 * On failure return NULL
813 lwkt_reltoken(&vm_token);
814 #if defined(DIAGNOSTIC)
815 if (vmstats.v_cache_count > 0)
816 kprintf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", vmstats.v_cache_count);
818 vm_pageout_deficit++;
823 * No pages available, wakeup the pageout daemon and give up.
825 lwkt_reltoken(&vm_token);
826 vm_pageout_deficit++;
832 * Good page found. The page has not yet been busied. We are in
833 * a critical section.
835 KASSERT(m != NULL, ("vm_page_alloc(): missing page on free queue\n"));
836 KASSERT(m->dirty == 0,
837 ("vm_page_alloc: free/cache page %p was dirty", m));
840 * Remove from free queue
842 vm_page_unqueue_nowakeup(m);
845 * Initialize structure. Only the PG_ZERO flag is inherited. Set
848 if (m->flags & PG_ZERO) {
849 vm_page_zero_count--;
850 m->flags = PG_ZERO | PG_BUSY;
861 * vm_page_insert() is safe while holding vm_token. Note also that
862 * inserting a page here does not insert it into the pmap (which
863 * could cause us to block allocating memory). We cannot block
866 vm_page_insert(m, object, pindex);
869 * Don't wakeup too often - wakeup the pageout daemon when
870 * we would be nearly out of memory.
874 lwkt_reltoken(&vm_token);
877 * A PG_BUSY page is returned.
883 * Wait for sufficient free memory for nominal heavy memory use kernel
887 vm_wait_nominal(void)
889 while (vm_page_count_min(0))
894 * Test if vm_wait_nominal() would block.
897 vm_test_nominal(void)
899 if (vm_page_count_min(0))
905 * Block until free pages are available for allocation, called in various
906 * places before memory allocations.
908 * The caller may loop if vm_page_count_min() == FALSE so we cannot be
909 * more generous then that.
919 lwkt_gettoken(&vm_token);
921 if (curthread == pagethread) {
923 * The pageout daemon itself needs pages, this is bad.
925 if (vm_page_count_min(0)) {
926 vm_pageout_pages_needed = 1;
927 tsleep(&vm_pageout_pages_needed, 0, "VMWait", timo);
931 * Wakeup the pageout daemon if necessary and wait.
933 if (vm_page_count_target()) {
934 if (vm_pages_needed == 0) {
936 wakeup(&vm_pages_needed);
938 ++vm_pages_waiting; /* SMP race ok */
939 tsleep(&vmstats.v_free_count, 0, "vmwait", timo);
942 lwkt_reltoken(&vm_token);
946 * Block until free pages are available for allocation
948 * Called only from vm_fault so that processes page faulting can be
955 * Wakeup the pageout daemon if necessary and wait.
957 if (vm_page_count_target()) {
958 lwkt_gettoken(&vm_token);
959 if (vm_page_count_target()) {
960 if (vm_pages_needed == 0) {
962 wakeup(&vm_pages_needed);
964 ++vm_pages_waiting; /* SMP race ok */
965 tsleep(&vmstats.v_free_count, 0, "pfault", hz);
967 lwkt_reltoken(&vm_token);
972 * Put the specified page on the active list (if appropriate). Ensure
973 * that act_count is at least ACT_INIT but do not otherwise mess with it.
975 * The page queues must be locked.
976 * This routine may not block.
979 vm_page_activate(vm_page_t m)
981 lwkt_gettoken(&vm_token);
982 if (m->queue != PQ_ACTIVE) {
983 if ((m->queue - m->pc) == PQ_CACHE)
984 mycpu->gd_cnt.v_reactivated++;
988 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
989 m->queue = PQ_ACTIVE;
990 vm_page_queues[PQ_ACTIVE].lcnt++;
991 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_ACTIVE].pl,
993 if (m->act_count < ACT_INIT)
994 m->act_count = ACT_INIT;
995 vmstats.v_active_count++;
998 if (m->act_count < ACT_INIT)
999 m->act_count = ACT_INIT;
1001 lwkt_reltoken(&vm_token);
1005 * Helper routine for vm_page_free_toq() and vm_page_cache(). This
1006 * routine is called when a page has been added to the cache or free
1009 * This routine may not block.
1010 * This routine must be called at splvm()
1012 static __inline void
1013 vm_page_free_wakeup(void)
1016 * If the pageout daemon itself needs pages, then tell it that
1017 * there are some free.
1019 if (vm_pageout_pages_needed &&
1020 vmstats.v_cache_count + vmstats.v_free_count >=
1021 vmstats.v_pageout_free_min
1023 wakeup(&vm_pageout_pages_needed);
1024 vm_pageout_pages_needed = 0;
1028 * Wakeup processes that are waiting on memory.
1030 * NOTE: vm_paging_target() is the pageout daemon's target, while
1031 * vm_page_count_target() is somewhere inbetween. We want
1032 * to wake processes up prior to the pageout daemon reaching
1033 * its target to provide some hysteresis.
1035 if (vm_pages_waiting) {
1036 if (!vm_page_count_target()) {
1038 * Plenty of pages are free, wakeup everyone.
1040 vm_pages_waiting = 0;
1041 wakeup(&vmstats.v_free_count);
1042 ++mycpu->gd_cnt.v_ppwakeups;
1043 } else if (!vm_page_count_min(0)) {
1045 * Some pages are free, wakeup someone.
1047 int wcount = vm_pages_waiting;
1050 vm_pages_waiting = wcount;
1051 wakeup_one(&vmstats.v_free_count);
1052 ++mycpu->gd_cnt.v_ppwakeups;
1060 * Returns the given page to the PQ_FREE list, disassociating it with
1063 * The vm_page must be PG_BUSY on entry. PG_BUSY will be released on
1064 * return (the page will have been freed). No particular spl is required
1067 * This routine may not block.
1070 vm_page_free_toq(vm_page_t m)
1072 struct vpgqueues *pq;
1074 lwkt_gettoken(&vm_token);
1075 mycpu->gd_cnt.v_tfree++;
1077 KKASSERT((m->flags & PG_MAPPED) == 0);
1079 if (m->busy || ((m->queue - m->pc) == PQ_FREE)) {
1081 "vm_page_free: pindex(%lu), busy(%d), PG_BUSY(%d), hold(%d)\n",
1082 (u_long)m->pindex, m->busy, (m->flags & PG_BUSY) ? 1 : 0,
1084 if ((m->queue - m->pc) == PQ_FREE)
1085 panic("vm_page_free: freeing free page");
1087 panic("vm_page_free: freeing busy page");
1091 * unqueue, then remove page. Note that we cannot destroy
1092 * the page here because we do not want to call the pager's
1093 * callback routine until after we've put the page on the
1094 * appropriate free queue.
1096 vm_page_unqueue_nowakeup(m);
1100 * No further management of fictitious pages occurs beyond object
1101 * and queue removal.
1103 if ((m->flags & PG_FICTITIOUS) != 0) {
1105 lwkt_reltoken(&vm_token);
1112 if (m->wire_count != 0) {
1113 if (m->wire_count > 1) {
1115 "vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1116 m->wire_count, (long)m->pindex);
1118 panic("vm_page_free: freeing wired page");
1122 * Clear the UNMANAGED flag when freeing an unmanaged page.
1124 if (m->flags & PG_UNMANAGED) {
1125 m->flags &= ~PG_UNMANAGED;
1128 if (m->hold_count != 0) {
1129 m->flags &= ~PG_ZERO;
1132 m->queue = PQ_FREE + m->pc;
1134 pq = &vm_page_queues[m->queue];
1139 * Put zero'd pages on the end ( where we look for zero'd pages
1140 * first ) and non-zerod pages at the head.
1142 if (m->flags & PG_ZERO) {
1143 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
1144 ++vm_page_zero_count;
1146 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
1149 vm_page_free_wakeup();
1150 lwkt_reltoken(&vm_token);
1154 * vm_page_free_fromq_fast()
1156 * Remove a non-zero page from one of the free queues; the page is removed for
1157 * zeroing, so do not issue a wakeup.
1162 vm_page_free_fromq_fast(void)
1168 lwkt_gettoken(&vm_token);
1169 for (i = 0; i < PQ_L2_SIZE; ++i) {
1170 m = vm_page_list_find(PQ_FREE, qi, FALSE);
1171 qi = (qi + PQ_PRIME2) & PQ_L2_MASK;
1172 if (m && (m->flags & PG_ZERO) == 0) {
1173 KKASSERT(m->busy == 0 && (m->flags & PG_BUSY) == 0);
1174 vm_page_unqueue_nowakeup(m);
1180 lwkt_reltoken(&vm_token);
1185 * vm_page_unmanage()
1187 * Prevent PV management from being done on the page. The page is
1188 * removed from the paging queues as if it were wired, and as a
1189 * consequence of no longer being managed the pageout daemon will not
1190 * touch it (since there is no way to locate the pte mappings for the
1191 * page). madvise() calls that mess with the pmap will also no longer
1192 * operate on the page.
1194 * Beyond that the page is still reasonably 'normal'. Freeing the page
1195 * will clear the flag.
1197 * This routine is used by OBJT_PHYS objects - objects using unswappable
1198 * physical memory as backing store rather then swap-backed memory and
1199 * will eventually be extended to support 4MB unmanaged physical
1202 * Must be called with a critical section held.
1203 * Must be called with vm_token held.
1206 vm_page_unmanage(vm_page_t m)
1208 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1209 if ((m->flags & PG_UNMANAGED) == 0) {
1210 if (m->wire_count == 0)
1213 vm_page_flag_set(m, PG_UNMANAGED);
1217 * Mark this page as wired down by yet another map, removing it from
1218 * paging queues as necessary.
1220 * The page queues must be locked.
1221 * This routine may not block.
1224 vm_page_wire(vm_page_t m)
1227 * Only bump the wire statistics if the page is not already wired,
1228 * and only unqueue the page if it is on some queue (if it is unmanaged
1229 * it is already off the queues). Don't do anything with fictitious
1230 * pages because they are always wired.
1232 lwkt_gettoken(&vm_token);
1233 if ((m->flags & PG_FICTITIOUS) == 0) {
1234 if (m->wire_count == 0) {
1235 if ((m->flags & PG_UNMANAGED) == 0)
1237 vmstats.v_wire_count++;
1240 KASSERT(m->wire_count != 0,
1241 ("vm_page_wire: wire_count overflow m=%p", m));
1243 lwkt_reltoken(&vm_token);
1247 * Release one wiring of this page, potentially enabling it to be paged again.
1249 * Many pages placed on the inactive queue should actually go
1250 * into the cache, but it is difficult to figure out which. What
1251 * we do instead, if the inactive target is well met, is to put
1252 * clean pages at the head of the inactive queue instead of the tail.
1253 * This will cause them to be moved to the cache more quickly and
1254 * if not actively re-referenced, freed more quickly. If we just
1255 * stick these pages at the end of the inactive queue, heavy filesystem
1256 * meta-data accesses can cause an unnecessary paging load on memory bound
1257 * processes. This optimization causes one-time-use metadata to be
1258 * reused more quickly.
1260 * BUT, if we are in a low-memory situation we have no choice but to
1261 * put clean pages on the cache queue.
1263 * A number of routines use vm_page_unwire() to guarantee that the page
1264 * will go into either the inactive or active queues, and will NEVER
1265 * be placed in the cache - for example, just after dirtying a page.
1266 * dirty pages in the cache are not allowed.
1268 * The page queues must be locked.
1269 * This routine may not block.
1272 vm_page_unwire(vm_page_t m, int activate)
1274 lwkt_gettoken(&vm_token);
1275 if (m->flags & PG_FICTITIOUS) {
1277 } else if (m->wire_count <= 0) {
1278 panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
1280 if (--m->wire_count == 0) {
1281 --vmstats.v_wire_count;
1282 if (m->flags & PG_UNMANAGED) {
1284 } else if (activate) {
1286 &vm_page_queues[PQ_ACTIVE].pl, m, pageq);
1287 m->queue = PQ_ACTIVE;
1288 vm_page_queues[PQ_ACTIVE].lcnt++;
1289 vmstats.v_active_count++;
1291 vm_page_flag_clear(m, PG_WINATCFLS);
1293 &vm_page_queues[PQ_INACTIVE].pl, m, pageq);
1294 m->queue = PQ_INACTIVE;
1295 vm_page_queues[PQ_INACTIVE].lcnt++;
1296 vmstats.v_inactive_count++;
1297 ++vm_swapcache_inactive_heuristic;
1301 lwkt_reltoken(&vm_token);
1306 * Move the specified page to the inactive queue. If the page has
1307 * any associated swap, the swap is deallocated.
1309 * Normally athead is 0 resulting in LRU operation. athead is set
1310 * to 1 if we want this page to be 'as if it were placed in the cache',
1311 * except without unmapping it from the process address space.
1313 * This routine may not block.
1314 * The caller must hold vm_token.
1316 static __inline void
1317 _vm_page_deactivate(vm_page_t m, int athead)
1320 * Ignore if already inactive.
1322 if (m->queue == PQ_INACTIVE)
1325 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1326 if ((m->queue - m->pc) == PQ_CACHE)
1327 mycpu->gd_cnt.v_reactivated++;
1328 vm_page_flag_clear(m, PG_WINATCFLS);
1331 TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl,
1334 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl,
1336 ++vm_swapcache_inactive_heuristic;
1338 m->queue = PQ_INACTIVE;
1339 vm_page_queues[PQ_INACTIVE].lcnt++;
1340 vmstats.v_inactive_count++;
1345 * Attempt to deactivate a page.
1350 vm_page_deactivate(vm_page_t m)
1352 lwkt_gettoken(&vm_token);
1353 _vm_page_deactivate(m, 0);
1354 lwkt_reltoken(&vm_token);
1358 * Attempt to move a page to PQ_CACHE.
1359 * Returns 0 on failure, 1 on success
1364 vm_page_try_to_cache(vm_page_t m)
1366 lwkt_gettoken(&vm_token);
1367 if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1368 (m->flags & (PG_BUSY|PG_UNMANAGED))) {
1369 lwkt_reltoken(&vm_token);
1372 vm_page_test_dirty(m);
1374 lwkt_reltoken(&vm_token);
1378 lwkt_reltoken(&vm_token);
1383 * Attempt to free the page. If we cannot free it, we do nothing.
1384 * 1 is returned on success, 0 on failure.
1389 vm_page_try_to_free(vm_page_t m)
1391 lwkt_gettoken(&vm_token);
1392 if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1393 (m->flags & (PG_BUSY|PG_UNMANAGED))) {
1394 lwkt_reltoken(&vm_token);
1397 vm_page_test_dirty(m);
1399 lwkt_reltoken(&vm_token);
1403 vm_page_protect(m, VM_PROT_NONE);
1405 lwkt_reltoken(&vm_token);
1412 * Put the specified page onto the page cache queue (if appropriate).
1414 * The caller must hold vm_token.
1415 * This routine may not block.
1418 vm_page_cache(vm_page_t m)
1420 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1422 if ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
1423 m->wire_count || m->hold_count) {
1424 kprintf("vm_page_cache: attempting to cache busy/held page\n");
1429 * Already in the cache (and thus not mapped)
1431 if ((m->queue - m->pc) == PQ_CACHE) {
1432 KKASSERT((m->flags & PG_MAPPED) == 0);
1437 * Caller is required to test m->dirty, but note that the act of
1438 * removing the page from its maps can cause it to become dirty
1439 * on an SMP system due to another cpu running in usermode.
1442 panic("vm_page_cache: caching a dirty page, pindex: %ld",
1447 * Remove all pmaps and indicate that the page is not
1448 * writeable or mapped. Our vm_page_protect() call may
1449 * have blocked (especially w/ VM_PROT_NONE), so recheck
1453 vm_page_protect(m, VM_PROT_NONE);
1455 if ((m->flags & (PG_BUSY|PG_UNMANAGED|PG_MAPPED)) || m->busy ||
1456 m->wire_count || m->hold_count) {
1458 } else if (m->dirty) {
1459 vm_page_deactivate(m);
1461 vm_page_unqueue_nowakeup(m);
1462 m->queue = PQ_CACHE + m->pc;
1463 vm_page_queues[m->queue].lcnt++;
1464 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, pageq);
1465 vmstats.v_cache_count++;
1466 vm_page_free_wakeup();
1471 * vm_page_dontneed()
1473 * Cache, deactivate, or do nothing as appropriate. This routine
1474 * is typically used by madvise() MADV_DONTNEED.
1476 * Generally speaking we want to move the page into the cache so
1477 * it gets reused quickly. However, this can result in a silly syndrome
1478 * due to the page recycling too quickly. Small objects will not be
1479 * fully cached. On the otherhand, if we move the page to the inactive
1480 * queue we wind up with a problem whereby very large objects
1481 * unnecessarily blow away our inactive and cache queues.
1483 * The solution is to move the pages based on a fixed weighting. We
1484 * either leave them alone, deactivate them, or move them to the cache,
1485 * where moving them to the cache has the highest weighting.
1486 * By forcing some pages into other queues we eventually force the
1487 * system to balance the queues, potentially recovering other unrelated
1488 * space from active. The idea is to not force this to happen too
1494 vm_page_dontneed(vm_page_t m)
1496 static int dnweight;
1503 * occassionally leave the page alone
1505 lwkt_gettoken(&vm_token);
1506 if ((dnw & 0x01F0) == 0 ||
1507 m->queue == PQ_INACTIVE ||
1508 m->queue - m->pc == PQ_CACHE
1510 if (m->act_count >= ACT_INIT)
1512 lwkt_reltoken(&vm_token);
1517 vm_page_test_dirty(m);
1519 if (m->dirty || (dnw & 0x0070) == 0) {
1521 * Deactivate the page 3 times out of 32.
1526 * Cache the page 28 times out of every 32. Note that
1527 * the page is deactivated instead of cached, but placed
1528 * at the head of the queue instead of the tail.
1532 _vm_page_deactivate(m, head);
1533 lwkt_reltoken(&vm_token);
1537 * Grab a page, blocking if it is busy and allocating a page if necessary.
1538 * A busy page is returned or NULL.
1540 * If VM_ALLOC_RETRY is specified VM_ALLOC_NORMAL must also be specified.
1541 * If VM_ALLOC_RETRY is not specified
1543 * This routine may block, but if VM_ALLOC_RETRY is not set then NULL is
1544 * always returned if we had blocked.
1545 * This routine will never return NULL if VM_ALLOC_RETRY is set.
1546 * This routine may not be called from an interrupt.
1547 * The returned page may not be entirely valid.
1549 * This routine may be called from mainline code without spl protection and
1550 * be guarenteed a busied page associated with the object at the specified
1556 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
1561 KKASSERT(allocflags &
1562 (VM_ALLOC_NORMAL|VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
1563 lwkt_gettoken(&vm_token);
1565 if ((m = vm_page_lookup(object, pindex)) != NULL) {
1566 if (m->busy || (m->flags & PG_BUSY)) {
1567 generation = object->generation;
1569 while ((object->generation == generation) &&
1570 (m->busy || (m->flags & PG_BUSY))) {
1571 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
1572 tsleep(m, 0, "pgrbwt", 0);
1573 if ((allocflags & VM_ALLOC_RETRY) == 0) {
1584 m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
1587 if ((allocflags & VM_ALLOC_RETRY) == 0)
1592 lwkt_reltoken(&vm_token);
1597 * Mapping function for valid bits or for dirty bits in
1598 * a page. May not block.
1600 * Inputs are required to range within a page.
1606 vm_page_bits(int base, int size)
1612 base + size <= PAGE_SIZE,
1613 ("vm_page_bits: illegal base/size %d/%d", base, size)
1616 if (size == 0) /* handle degenerate case */
1619 first_bit = base >> DEV_BSHIFT;
1620 last_bit = (base + size - 1) >> DEV_BSHIFT;
1622 return ((2 << last_bit) - (1 << first_bit));
1626 * Sets portions of a page valid and clean. The arguments are expected
1627 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1628 * of any partial chunks touched by the range. The invalid portion of
1629 * such chunks will be zero'd.
1631 * NOTE: When truncating a buffer vnode_pager_setsize() will automatically
1632 * align base to DEV_BSIZE so as not to mark clean a partially
1633 * truncated device block. Otherwise the dirty page status might be
1636 * This routine may not block.
1638 * (base + size) must be less then or equal to PAGE_SIZE.
1641 _vm_page_zero_valid(vm_page_t m, int base, int size)
1646 if (size == 0) /* handle degenerate case */
1650 * If the base is not DEV_BSIZE aligned and the valid
1651 * bit is clear, we have to zero out a portion of the
1655 if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1656 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0
1658 pmap_zero_page_area(
1666 * If the ending offset is not DEV_BSIZE aligned and the
1667 * valid bit is clear, we have to zero out a portion of
1671 endoff = base + size;
1673 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1674 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0
1676 pmap_zero_page_area(
1679 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))
1685 * Set valid, clear dirty bits. If validating the entire
1686 * page we can safely clear the pmap modify bit. We also
1687 * use this opportunity to clear the PG_NOSYNC flag. If a process
1688 * takes a write fault on a MAP_NOSYNC memory area the flag will
1691 * We set valid bits inclusive of any overlap, but we can only
1692 * clear dirty bits for DEV_BSIZE chunks that are fully within
1695 * Page must be busied?
1696 * No other requirements.
1699 vm_page_set_valid(vm_page_t m, int base, int size)
1701 _vm_page_zero_valid(m, base, size);
1702 m->valid |= vm_page_bits(base, size);
1707 * Set valid bits and clear dirty bits.
1709 * NOTE: This function does not clear the pmap modified bit.
1710 * Also note that e.g. NFS may use a byte-granular base
1713 * WARNING: Page must be busied? But vfs_clean_one_page() will call
1714 * this without necessarily busying the page (via bdwrite()).
1715 * So for now vm_token must also be held.
1717 * No other requirements.
1720 vm_page_set_validclean(vm_page_t m, int base, int size)
1724 _vm_page_zero_valid(m, base, size);
1725 pagebits = vm_page_bits(base, size);
1726 m->valid |= pagebits;
1727 m->dirty &= ~pagebits;
1728 if (base == 0 && size == PAGE_SIZE) {
1729 /*pmap_clear_modify(m);*/
1730 vm_page_flag_clear(m, PG_NOSYNC);
1735 * Set valid & dirty. Used by buwrite()
1737 * WARNING: Page must be busied? But vfs_dirty_one_page() will
1738 * call this function in buwrite() so for now vm_token must
1741 * No other requirements.
1744 vm_page_set_validdirty(vm_page_t m, int base, int size)
1748 pagebits = vm_page_bits(base, size);
1749 m->valid |= pagebits;
1750 m->dirty |= pagebits;
1752 vm_object_set_writeable_dirty(m->object);
1758 * NOTE: This function does not clear the pmap modified bit.
1759 * Also note that e.g. NFS may use a byte-granular base
1762 * Page must be busied?
1763 * No other requirements.
1766 vm_page_clear_dirty(vm_page_t m, int base, int size)
1768 m->dirty &= ~vm_page_bits(base, size);
1769 if (base == 0 && size == PAGE_SIZE) {
1770 /*pmap_clear_modify(m);*/
1771 vm_page_flag_clear(m, PG_NOSYNC);
1776 * Make the page all-dirty.
1778 * Also make sure the related object and vnode reflect the fact that the
1779 * object may now contain a dirty page.
1781 * Page must be busied?
1782 * No other requirements.
1785 vm_page_dirty(vm_page_t m)
1788 int pqtype = m->queue - m->pc;
1790 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
1791 ("vm_page_dirty: page in free/cache queue!"));
1792 if (m->dirty != VM_PAGE_BITS_ALL) {
1793 m->dirty = VM_PAGE_BITS_ALL;
1795 vm_object_set_writeable_dirty(m->object);
1800 * Invalidates DEV_BSIZE'd chunks within a page. Both the
1801 * valid and dirty bits for the effected areas are cleared.
1803 * Page must be busied?
1805 * No other requirements.
1808 vm_page_set_invalid(vm_page_t m, int base, int size)
1812 bits = vm_page_bits(base, size);
1815 m->object->generation++;
1819 * The kernel assumes that the invalid portions of a page contain
1820 * garbage, but such pages can be mapped into memory by user code.
1821 * When this occurs, we must zero out the non-valid portions of the
1822 * page so user code sees what it expects.
1824 * Pages are most often semi-valid when the end of a file is mapped
1825 * into memory and the file's size is not page aligned.
1827 * Page must be busied?
1828 * No other requirements.
1831 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
1837 * Scan the valid bits looking for invalid sections that
1838 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
1839 * valid bit may be set ) have already been zerod by
1840 * vm_page_set_validclean().
1842 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
1843 if (i == (PAGE_SIZE / DEV_BSIZE) ||
1844 (m->valid & (1 << i))
1847 pmap_zero_page_area(
1850 (i - b) << DEV_BSHIFT
1858 * setvalid is TRUE when we can safely set the zero'd areas
1859 * as being valid. We can do this if there are no cache consistency
1860 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
1863 m->valid = VM_PAGE_BITS_ALL;
1867 * Is a (partial) page valid? Note that the case where size == 0
1868 * will return FALSE in the degenerate case where the page is entirely
1869 * invalid, and TRUE otherwise.
1872 * No other requirements.
1875 vm_page_is_valid(vm_page_t m, int base, int size)
1877 int bits = vm_page_bits(base, size);
1879 if (m->valid && ((m->valid & bits) == bits))
1886 * update dirty bits from pmap/mmu. May not block.
1888 * Caller must hold vm_token if non-blocking operation desired.
1889 * No other requirements.
1892 vm_page_test_dirty(vm_page_t m)
1894 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
1900 * Register an action, associating it with its vm_page
1903 vm_page_register_action(vm_page_action_t action, vm_page_event_t event)
1905 struct vm_page_action_list *list;
1908 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1909 list = &action_list[hv];
1911 lwkt_gettoken(&vm_token);
1912 vm_page_flag_set(action->m, PG_ACTIONLIST);
1913 action->event = event;
1914 LIST_INSERT_HEAD(list, action, entry);
1915 lwkt_reltoken(&vm_token);
1919 * Unregister an action, disassociating it from its related vm_page
1922 vm_page_unregister_action(vm_page_action_t action)
1924 struct vm_page_action_list *list;
1927 lwkt_gettoken(&vm_token);
1928 if (action->event != VMEVENT_NONE) {
1929 action->event = VMEVENT_NONE;
1930 LIST_REMOVE(action, entry);
1932 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1933 list = &action_list[hv];
1934 if (LIST_EMPTY(list))
1935 vm_page_flag_clear(action->m, PG_ACTIONLIST);
1937 lwkt_reltoken(&vm_token);
1941 * Issue an event on a VM page. Corresponding action structures are
1942 * removed from the page's list and called.
1944 * If the vm_page has no more pending action events we clear its
1945 * PG_ACTIONLIST flag.
1948 vm_page_event_internal(vm_page_t m, vm_page_event_t event)
1950 struct vm_page_action_list *list;
1951 struct vm_page_action *scan;
1952 struct vm_page_action *next;
1956 hv = (int)((intptr_t)m >> 8) & VMACTION_HMASK;
1957 list = &action_list[hv];
1960 lwkt_gettoken(&vm_token);
1961 LIST_FOREACH_MUTABLE(scan, list, entry, next) {
1963 if (scan->event == event) {
1964 scan->event = VMEVENT_NONE;
1965 LIST_REMOVE(scan, entry);
1966 scan->func(m, scan);
1974 vm_page_flag_clear(m, PG_ACTIONLIST);
1975 lwkt_reltoken(&vm_token);
1979 #include "opt_ddb.h"
1981 #include <sys/kernel.h>
1983 #include <ddb/ddb.h>
1985 DB_SHOW_COMMAND(page, vm_page_print_page_info)
1987 db_printf("vmstats.v_free_count: %d\n", vmstats.v_free_count);
1988 db_printf("vmstats.v_cache_count: %d\n", vmstats.v_cache_count);
1989 db_printf("vmstats.v_inactive_count: %d\n", vmstats.v_inactive_count);
1990 db_printf("vmstats.v_active_count: %d\n", vmstats.v_active_count);
1991 db_printf("vmstats.v_wire_count: %d\n", vmstats.v_wire_count);
1992 db_printf("vmstats.v_free_reserved: %d\n", vmstats.v_free_reserved);
1993 db_printf("vmstats.v_free_min: %d\n", vmstats.v_free_min);
1994 db_printf("vmstats.v_free_target: %d\n", vmstats.v_free_target);
1995 db_printf("vmstats.v_cache_min: %d\n", vmstats.v_cache_min);
1996 db_printf("vmstats.v_inactive_target: %d\n", vmstats.v_inactive_target);
1999 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
2002 db_printf("PQ_FREE:");
2003 for(i=0;i<PQ_L2_SIZE;i++) {
2004 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt);
2008 db_printf("PQ_CACHE:");
2009 for(i=0;i<PQ_L2_SIZE;i++) {
2010 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt);
2014 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n",
2015 vm_page_queues[PQ_ACTIVE].lcnt,
2016 vm_page_queues[PQ_INACTIVE].lcnt);