2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991 Regents of the University of California.
6 * Copyright (c) 1998 Matthew Dillon. All Rights Reserved.
8 * This code is derived from software contributed to Berkeley by
9 * The Mach Operating System project at Carnegie-Mellon University.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
54 * Carnegie Mellon requests users of this software to return to
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
66 * GENERAL RULES ON VM_PAGE MANIPULATION
68 * - A page queue lock is required when adding or removing a page from a
69 * page queue regardless of other locks or the busy state of a page.
71 * * In general, no thread besides the page daemon can acquire or
72 * hold more than one page queue lock at a time.
74 * * The page daemon can acquire and hold any pair of page queue
77 * - The object lock is required when inserting or removing
78 * pages from an object (vm_page_insert() or vm_page_remove()).
83 * Resident memory management module.
86 #include <sys/cdefs.h>
87 __FBSDID("$FreeBSD$");
91 #include <sys/param.h>
92 #include <sys/systm.h>
94 #include <sys/domainset.h>
95 #include <sys/kernel.h>
96 #include <sys/limits.h>
97 #include <sys/linker.h>
98 #include <sys/malloc.h>
100 #include <sys/msgbuf.h>
101 #include <sys/mutex.h>
102 #include <sys/proc.h>
103 #include <sys/rwlock.h>
104 #include <sys/sbuf.h>
106 #include <sys/sysctl.h>
107 #include <sys/vmmeter.h>
108 #include <sys/vnode.h>
112 #include <vm/vm_param.h>
113 #include <vm/vm_domainset.h>
114 #include <vm/vm_kern.h>
115 #include <vm/vm_map.h>
116 #include <vm/vm_object.h>
117 #include <vm/vm_page.h>
118 #include <vm/vm_pageout.h>
119 #include <vm/vm_phys.h>
120 #include <vm/vm_pagequeue.h>
121 #include <vm/vm_pager.h>
122 #include <vm/vm_radix.h>
123 #include <vm/vm_reserv.h>
124 #include <vm/vm_extern.h>
126 #include <vm/uma_int.h>
128 #include <machine/md_var.h>
130 extern int uma_startup_count(int);
131 extern void uma_startup(void *, int);
132 extern int vmem_startup_count(void);
135 * Associated with page of user-allocatable memory is a
139 struct vm_domain vm_dom[MAXMEMDOM];
141 struct mtx_padalign __exclusive_cache_line pa_lock[PA_LOCK_COUNT];
142 struct mtx_padalign __exclusive_cache_line vm_domainset_lock;
143 domainset_t __exclusive_cache_line vm_min_domains;
144 domainset_t __exclusive_cache_line vm_severe_domains;
145 static int vm_min_waiters;
146 static int vm_severe_waiters;
147 static int vm_pageproc_waiters;
151 * bogus page -- for I/O to/from partially complete buffers,
152 * or for paging into sparsely invalid regions.
154 vm_page_t bogus_page;
156 vm_page_t vm_page_array;
157 long vm_page_array_size;
160 static int boot_pages;
161 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
163 "number of pages allocated for bootstrapping the VM system");
165 static int pa_tryrelock_restart;
166 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
167 &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
169 static TAILQ_HEAD(, vm_page) blacklist_head;
170 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
171 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
172 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
174 static uma_zone_t fakepg_zone;
176 static void vm_page_alloc_check(vm_page_t m);
177 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
178 static void vm_page_enqueue(uint8_t queue, vm_page_t m);
179 static void vm_page_free_phys(struct vm_domain *vmd, vm_page_t m);
180 static void vm_page_init(void *dummy);
181 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
182 vm_pindex_t pindex, vm_page_t mpred);
183 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
185 static int vm_page_reclaim_run(int req_class, int domain, u_long npages,
186 vm_page_t m_run, vm_paddr_t high);
187 static void vm_domain_free_wakeup(struct vm_domain *);
188 static int vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object,
191 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init, NULL);
194 vm_page_init(void *dummy)
197 fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
198 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
199 bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ |
200 VM_ALLOC_NORMAL | VM_ALLOC_WIRED);
203 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
204 #if PAGE_SIZE == 32768
206 CTASSERT(sizeof(u_long) >= 8);
211 * Try to acquire a physical address lock while a pmap is locked. If we
212 * fail to trylock we unlock and lock the pmap directly and cache the
213 * locked pa in *locked. The caller should then restart their loop in case
214 * the virtual to physical mapping has changed.
217 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
224 PA_LOCK_ASSERT(lockpa, MA_OWNED);
225 if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
232 atomic_add_int(&pa_tryrelock_restart, 1);
241 * Sets the page size, perhaps based upon the memory
242 * size. Must be called before any use of page-size
243 * dependent functions.
246 vm_set_page_size(void)
248 if (vm_cnt.v_page_size == 0)
249 vm_cnt.v_page_size = PAGE_SIZE;
250 if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
251 panic("vm_set_page_size: page size not a power of two");
255 * vm_page_blacklist_next:
257 * Find the next entry in the provided string of blacklist
258 * addresses. Entries are separated by space, comma, or newline.
259 * If an invalid integer is encountered then the rest of the
260 * string is skipped. Updates the list pointer to the next
261 * character, or NULL if the string is exhausted or invalid.
264 vm_page_blacklist_next(char **list, char *end)
269 if (list == NULL || *list == NULL)
277 * If there's no end pointer then the buffer is coming from
278 * the kenv and we know it's null-terminated.
281 end = *list + strlen(*list);
283 /* Ensure that strtoq() won't walk off the end */
285 if (*end == '\n' || *end == ' ' || *end == ',')
288 printf("Blacklist not terminated, skipping\n");
294 for (pos = *list; *pos != '\0'; pos = cp) {
295 bad = strtoq(pos, &cp, 0);
296 if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
305 if (*cp == '\0' || ++cp >= end)
309 return (trunc_page(bad));
311 printf("Garbage in RAM blacklist, skipping\n");
317 * vm_page_blacklist_check:
319 * Iterate through the provided string of blacklist addresses, pulling
320 * each entry out of the physical allocator free list and putting it
321 * onto a list for reporting via the vm.page_blacklist sysctl.
324 vm_page_blacklist_check(char *list, char *end)
326 struct vm_domain *vmd;
333 while (next != NULL) {
334 if ((pa = vm_page_blacklist_next(&next, end)) == 0)
336 m = vm_phys_paddr_to_vm_page(pa);
339 vmd = vm_pagequeue_domain(m);
340 vm_domain_free_lock(vmd);
341 ret = vm_phys_unfree_page(m);
342 vm_domain_free_unlock(vmd);
344 TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
346 printf("Skipping page with pa 0x%jx\n",
353 * vm_page_blacklist_load:
355 * Search for a special module named "ram_blacklist". It'll be a
356 * plain text file provided by the user via the loader directive
360 vm_page_blacklist_load(char **list, char **end)
369 mod = preload_search_by_type("ram_blacklist");
371 ptr = preload_fetch_addr(mod);
372 len = preload_fetch_size(mod);
383 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
390 error = sysctl_wire_old_buffer(req, 0);
393 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
394 TAILQ_FOREACH(m, &blacklist_head, listq) {
395 sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
396 (uintmax_t)m->phys_addr);
399 error = sbuf_finish(&sbuf);
405 vm_page_domain_init(int domain)
407 struct vm_domain *vmd;
408 struct vm_pagequeue *pq;
411 vmd = VM_DOMAIN(domain);
412 bzero(vmd, sizeof(*vmd));
413 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
414 "vm inactive pagequeue";
415 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
416 "vm active pagequeue";
417 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_name) =
418 "vm laundry pagequeue";
419 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_name) =
420 "vm unswappable pagequeue";
421 vmd->vmd_domain = domain;
422 vmd->vmd_page_count = 0;
423 vmd->vmd_free_count = 0;
425 vmd->vmd_oom = FALSE;
426 for (i = 0; i < PQ_COUNT; i++) {
427 pq = &vmd->vmd_pagequeues[i];
428 TAILQ_INIT(&pq->pq_pl);
429 mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
430 MTX_DEF | MTX_DUPOK);
432 mtx_init(&vmd->vmd_free_mtx, "vm page free queue", NULL, MTX_DEF);
433 snprintf(vmd->vmd_name, sizeof(vmd->vmd_name), "%d", domain);
437 * Initialize a physical page in preparation for adding it to the free
441 vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind)
446 m->busy_lock = VPB_UNBUSIED;
453 m->order = VM_NFREEORDER;
454 m->pool = VM_FREEPOOL_DEFAULT;
455 m->valid = m->dirty = 0;
462 * Initializes the resident memory module. Allocates physical memory for
463 * bootstrapping UMA and some data structures that are used to manage
464 * physical pages. Initializes these structures, and populates the free
468 vm_page_startup(vm_offset_t vaddr)
470 struct vm_phys_seg *seg;
472 char *list, *listend;
474 vm_paddr_t end, high_avail, low_avail, new_end, page_range, size;
475 vm_paddr_t biggestsize, last_pa, pa;
477 int biggestone, i, segind;
481 vaddr = round_page(vaddr);
483 for (i = 0; phys_avail[i + 1]; i += 2) {
484 phys_avail[i] = round_page(phys_avail[i]);
485 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
487 for (i = 0; phys_avail[i + 1]; i += 2) {
488 size = phys_avail[i + 1] - phys_avail[i];
489 if (size > biggestsize) {
495 end = phys_avail[biggestone+1];
498 * Initialize the page and queue locks.
500 mtx_init(&vm_domainset_lock, "vm domainset lock", NULL, MTX_DEF);
501 for (i = 0; i < PA_LOCK_COUNT; i++)
502 mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
503 for (i = 0; i < vm_ndomains; i++)
504 vm_page_domain_init(i);
507 * Allocate memory for use when boot strapping the kernel memory
508 * allocator. Tell UMA how many zones we are going to create
509 * before going fully functional. UMA will add its zones.
511 * VM startup zones: vmem, vmem_btag, VM OBJECT, RADIX NODE, MAP,
512 * KMAP ENTRY, MAP ENTRY, VMSPACE.
514 boot_pages = uma_startup_count(8);
516 #ifndef UMA_MD_SMALL_ALLOC
517 /* vmem_startup() calls uma_prealloc(). */
518 boot_pages += vmem_startup_count();
519 /* vm_map_startup() calls uma_prealloc(). */
520 boot_pages += howmany(MAX_KMAP,
521 UMA_SLAB_SPACE / sizeof(struct vm_map));
524 * Before going fully functional kmem_init() does allocation
525 * from "KMAP ENTRY" and vmem_create() does allocation from "vmem".
530 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
531 * manually fetch the value.
533 TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
534 new_end = end - (boot_pages * UMA_SLAB_SIZE);
535 new_end = trunc_page(new_end);
536 mapped = pmap_map(&vaddr, new_end, end,
537 VM_PROT_READ | VM_PROT_WRITE);
538 bzero((void *)mapped, end - new_end);
539 uma_startup((void *)mapped, boot_pages);
541 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
542 defined(__i386__) || defined(__mips__)
544 * Allocate a bitmap to indicate that a random physical page
545 * needs to be included in a minidump.
547 * The amd64 port needs this to indicate which direct map pages
548 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
550 * However, i386 still needs this workspace internally within the
551 * minidump code. In theory, they are not needed on i386, but are
552 * included should the sf_buf code decide to use them.
555 for (i = 0; dump_avail[i + 1] != 0; i += 2)
556 if (dump_avail[i + 1] > last_pa)
557 last_pa = dump_avail[i + 1];
558 page_range = last_pa / PAGE_SIZE;
559 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
560 new_end -= vm_page_dump_size;
561 vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
562 new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
563 bzero((void *)vm_page_dump, vm_page_dump_size);
567 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
569 * Include the UMA bootstrap pages and vm_page_dump in a crash dump.
570 * When pmap_map() uses the direct map, they are not automatically
573 for (pa = new_end; pa < end; pa += PAGE_SIZE)
576 phys_avail[biggestone + 1] = new_end;
579 * Request that the physical pages underlying the message buffer be
580 * included in a crash dump. Since the message buffer is accessed
581 * through the direct map, they are not automatically included.
583 pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
584 last_pa = pa + round_page(msgbufsize);
585 while (pa < last_pa) {
591 * Compute the number of pages of memory that will be available for
592 * use, taking into account the overhead of a page structure per page.
593 * In other words, solve
594 * "available physical memory" - round_page(page_range *
595 * sizeof(struct vm_page)) = page_range * PAGE_SIZE
598 low_avail = phys_avail[0];
599 high_avail = phys_avail[1];
600 for (i = 0; i < vm_phys_nsegs; i++) {
601 if (vm_phys_segs[i].start < low_avail)
602 low_avail = vm_phys_segs[i].start;
603 if (vm_phys_segs[i].end > high_avail)
604 high_avail = vm_phys_segs[i].end;
606 /* Skip the first chunk. It is already accounted for. */
607 for (i = 2; phys_avail[i + 1] != 0; i += 2) {
608 if (phys_avail[i] < low_avail)
609 low_avail = phys_avail[i];
610 if (phys_avail[i + 1] > high_avail)
611 high_avail = phys_avail[i + 1];
613 first_page = low_avail / PAGE_SIZE;
614 #ifdef VM_PHYSSEG_SPARSE
616 for (i = 0; i < vm_phys_nsegs; i++)
617 size += vm_phys_segs[i].end - vm_phys_segs[i].start;
618 for (i = 0; phys_avail[i + 1] != 0; i += 2)
619 size += phys_avail[i + 1] - phys_avail[i];
620 #elif defined(VM_PHYSSEG_DENSE)
621 size = high_avail - low_avail;
623 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
626 #ifdef VM_PHYSSEG_DENSE
628 * In the VM_PHYSSEG_DENSE case, the number of pages can account for
629 * the overhead of a page structure per page only if vm_page_array is
630 * allocated from the last physical memory chunk. Otherwise, we must
631 * allocate page structures representing the physical memory
632 * underlying vm_page_array, even though they will not be used.
634 if (new_end != high_avail)
635 page_range = size / PAGE_SIZE;
639 page_range = size / (PAGE_SIZE + sizeof(struct vm_page));
642 * If the partial bytes remaining are large enough for
643 * a page (PAGE_SIZE) without a corresponding
644 * 'struct vm_page', then new_end will contain an
645 * extra page after subtracting the length of the VM
646 * page array. Compensate by subtracting an extra
649 if (size % (PAGE_SIZE + sizeof(struct vm_page)) >= PAGE_SIZE) {
650 if (new_end == high_avail)
651 high_avail -= PAGE_SIZE;
652 new_end -= PAGE_SIZE;
658 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
659 * However, because this page is allocated from KVM, out-of-bounds
660 * accesses using the direct map will not be trapped.
665 * Allocate physical memory for the page structures, and map it.
667 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
668 mapped = pmap_map(&vaddr, new_end, end,
669 VM_PROT_READ | VM_PROT_WRITE);
670 vm_page_array = (vm_page_t)mapped;
671 vm_page_array_size = page_range;
673 #if VM_NRESERVLEVEL > 0
675 * Allocate physical memory for the reservation management system's
676 * data structures, and map it.
678 if (high_avail == end)
679 high_avail = new_end;
680 new_end = vm_reserv_startup(&vaddr, new_end, high_avail);
682 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
684 * Include vm_page_array and vm_reserv_array in a crash dump.
686 for (pa = new_end; pa < end; pa += PAGE_SIZE)
689 phys_avail[biggestone + 1] = new_end;
692 * Add physical memory segments corresponding to the available
695 for (i = 0; phys_avail[i + 1] != 0; i += 2)
696 vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
699 * Initialize the physical memory allocator.
704 * Initialize the page structures and add every available page to the
705 * physical memory allocator's free lists.
707 vm_cnt.v_page_count = 0;
708 for (segind = 0; segind < vm_phys_nsegs; segind++) {
709 seg = &vm_phys_segs[segind];
710 for (m = seg->first_page, pa = seg->start; pa < seg->end;
711 m++, pa += PAGE_SIZE)
712 vm_page_init_page(m, pa, segind);
715 * Add the segment to the free lists only if it is covered by
716 * one of the ranges in phys_avail. Because we've added the
717 * ranges to the vm_phys_segs array, we can assume that each
718 * segment is either entirely contained in one of the ranges,
719 * or doesn't overlap any of them.
721 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
722 struct vm_domain *vmd;
724 if (seg->start < phys_avail[i] ||
725 seg->end > phys_avail[i + 1])
729 pagecount = (u_long)atop(seg->end - seg->start);
731 vmd = VM_DOMAIN(seg->domain);
732 vm_domain_free_lock(vmd);
733 vm_phys_free_contig(m, pagecount);
734 vm_domain_freecnt_adj(vmd, (int)pagecount);
735 vm_domain_free_unlock(vmd);
736 vm_cnt.v_page_count += (u_int)pagecount;
738 vmd = VM_DOMAIN(seg->domain);
739 vmd->vmd_page_count += (u_int)pagecount;
740 vmd->vmd_segs |= 1UL << m->segind;
746 * Remove blacklisted pages from the physical memory allocator.
748 TAILQ_INIT(&blacklist_head);
749 vm_page_blacklist_load(&list, &listend);
750 vm_page_blacklist_check(list, listend);
752 list = kern_getenv("vm.blacklist");
753 vm_page_blacklist_check(list, NULL);
756 #if VM_NRESERVLEVEL > 0
758 * Initialize the reservation management system.
763 * Set an initial domain policy for thread0 so that allocations
772 vm_page_reference(vm_page_t m)
775 vm_page_aflag_set(m, PGA_REFERENCED);
779 * vm_page_busy_downgrade:
781 * Downgrade an exclusive busy page into a single shared busy page.
784 vm_page_busy_downgrade(vm_page_t m)
789 vm_page_assert_xbusied(m);
790 locked = mtx_owned(vm_page_lockptr(m));
794 x &= VPB_BIT_WAITERS;
795 if (x != 0 && !locked)
797 if (atomic_cmpset_rel_int(&m->busy_lock,
798 VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1)))
800 if (x != 0 && !locked)
813 * Return a positive value if the page is shared busied, 0 otherwise.
816 vm_page_sbusied(vm_page_t m)
821 return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
827 * Shared unbusy a page.
830 vm_page_sunbusy(vm_page_t m)
834 vm_page_lock_assert(m, MA_NOTOWNED);
835 vm_page_assert_sbusied(m);
839 if (VPB_SHARERS(x) > 1) {
840 if (atomic_cmpset_int(&m->busy_lock, x,
845 if ((x & VPB_BIT_WAITERS) == 0) {
846 KASSERT(x == VPB_SHARERS_WORD(1),
847 ("vm_page_sunbusy: invalid lock state"));
848 if (atomic_cmpset_int(&m->busy_lock,
849 VPB_SHARERS_WORD(1), VPB_UNBUSIED))
853 KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
854 ("vm_page_sunbusy: invalid lock state for waiters"));
857 if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
868 * vm_page_busy_sleep:
870 * Sleep and release the page lock, using the page pointer as wchan.
871 * This is used to implement the hard-path of busying mechanism.
873 * The given page must be locked.
875 * If nonshared is true, sleep only if the page is xbusy.
878 vm_page_busy_sleep(vm_page_t m, const char *wmesg, bool nonshared)
882 vm_page_assert_locked(m);
885 if (x == VPB_UNBUSIED || (nonshared && (x & VPB_BIT_SHARED) != 0) ||
886 ((x & VPB_BIT_WAITERS) == 0 &&
887 !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS))) {
891 msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
897 * Try to shared busy a page.
898 * If the operation succeeds 1 is returned otherwise 0.
899 * The operation never sleeps.
902 vm_page_trysbusy(vm_page_t m)
908 if ((x & VPB_BIT_SHARED) == 0)
910 if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
916 vm_page_xunbusy_locked(vm_page_t m)
919 vm_page_assert_xbusied(m);
920 vm_page_assert_locked(m);
922 atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
923 /* There is a waiter, do wakeup() instead of vm_page_flash(). */
928 vm_page_xunbusy_maybelocked(vm_page_t m)
932 vm_page_assert_xbusied(m);
935 * Fast path for unbusy. If it succeeds, we know that there
936 * are no waiters, so we do not need a wakeup.
938 if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER,
942 lockacq = !mtx_owned(vm_page_lockptr(m));
945 vm_page_xunbusy_locked(m);
951 * vm_page_xunbusy_hard:
953 * Called after the first try the exclusive unbusy of a page failed.
954 * It is assumed that the waiters bit is on.
957 vm_page_xunbusy_hard(vm_page_t m)
960 vm_page_assert_xbusied(m);
963 vm_page_xunbusy_locked(m);
970 * Wakeup anyone waiting for the page.
971 * The ownership bits do not change.
973 * The given page must be locked.
976 vm_page_flash(vm_page_t m)
980 vm_page_lock_assert(m, MA_OWNED);
984 if ((x & VPB_BIT_WAITERS) == 0)
986 if (atomic_cmpset_int(&m->busy_lock, x,
987 x & (~VPB_BIT_WAITERS)))
994 * Avoid releasing and reacquiring the same page lock.
997 vm_page_change_lock(vm_page_t m, struct mtx **mtx)
1001 mtx1 = vm_page_lockptr(m);
1011 * Keep page from being freed by the page daemon
1012 * much of the same effect as wiring, except much lower
1013 * overhead and should be used only for *very* temporary
1014 * holding ("wiring").
1017 vm_page_hold(vm_page_t mem)
1020 vm_page_lock_assert(mem, MA_OWNED);
1025 vm_page_unhold(vm_page_t mem)
1028 vm_page_lock_assert(mem, MA_OWNED);
1029 KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
1031 if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
1032 vm_page_free_toq(mem);
1036 * vm_page_unhold_pages:
1038 * Unhold each of the pages that is referenced by the given array.
1041 vm_page_unhold_pages(vm_page_t *ma, int count)
1046 for (; count != 0; count--) {
1047 vm_page_change_lock(*ma, &mtx);
1048 vm_page_unhold(*ma);
1056 PHYS_TO_VM_PAGE(vm_paddr_t pa)
1060 #ifdef VM_PHYSSEG_SPARSE
1061 m = vm_phys_paddr_to_vm_page(pa);
1063 m = vm_phys_fictitious_to_vm_page(pa);
1065 #elif defined(VM_PHYSSEG_DENSE)
1069 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
1070 m = &vm_page_array[pi - first_page];
1073 return (vm_phys_fictitious_to_vm_page(pa));
1075 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
1082 * Create a fictitious page with the specified physical address and
1083 * memory attribute. The memory attribute is the only the machine-
1084 * dependent aspect of a fictitious page that must be initialized.
1087 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
1091 m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
1092 vm_page_initfake(m, paddr, memattr);
1097 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1100 if ((m->flags & PG_FICTITIOUS) != 0) {
1102 * The page's memattr might have changed since the
1103 * previous initialization. Update the pmap to the
1108 m->phys_addr = paddr;
1110 /* Fictitious pages don't use "segind". */
1111 m->flags = PG_FICTITIOUS;
1112 /* Fictitious pages don't use "order" or "pool". */
1113 m->oflags = VPO_UNMANAGED;
1114 m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1118 pmap_page_set_memattr(m, memattr);
1124 * Release a fictitious page.
1127 vm_page_putfake(vm_page_t m)
1130 KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
1131 KASSERT((m->flags & PG_FICTITIOUS) != 0,
1132 ("vm_page_putfake: bad page %p", m));
1133 uma_zfree(fakepg_zone, m);
1137 * vm_page_updatefake:
1139 * Update the given fictitious page to the specified physical address and
1143 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1146 KASSERT((m->flags & PG_FICTITIOUS) != 0,
1147 ("vm_page_updatefake: bad page %p", m));
1148 m->phys_addr = paddr;
1149 pmap_page_set_memattr(m, memattr);
1158 vm_page_free(vm_page_t m)
1161 m->flags &= ~PG_ZERO;
1162 vm_page_free_toq(m);
1166 * vm_page_free_zero:
1168 * Free a page to the zerod-pages queue
1171 vm_page_free_zero(vm_page_t m)
1174 m->flags |= PG_ZERO;
1175 vm_page_free_toq(m);
1179 * Unbusy and handle the page queueing for a page from a getpages request that
1180 * was optionally read ahead or behind.
1183 vm_page_readahead_finish(vm_page_t m)
1186 /* We shouldn't put invalid pages on queues. */
1187 KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m));
1190 * Since the page is not the actually needed one, whether it should
1191 * be activated or deactivated is not obvious. Empirical results
1192 * have shown that deactivating the page is usually the best choice,
1193 * unless the page is wanted by another thread.
1196 if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1197 vm_page_activate(m);
1199 vm_page_deactivate(m);
1205 * vm_page_sleep_if_busy:
1207 * Sleep and release the page queues lock if the page is busied.
1208 * Returns TRUE if the thread slept.
1210 * The given page must be unlocked and object containing it must
1214 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1218 vm_page_lock_assert(m, MA_NOTOWNED);
1219 VM_OBJECT_ASSERT_WLOCKED(m->object);
1221 if (vm_page_busied(m)) {
1223 * The page-specific object must be cached because page
1224 * identity can change during the sleep, causing the
1225 * re-lock of a different object.
1226 * It is assumed that a reference to the object is already
1227 * held by the callers.
1231 VM_OBJECT_WUNLOCK(obj);
1232 vm_page_busy_sleep(m, msg, false);
1233 VM_OBJECT_WLOCK(obj);
1240 * vm_page_dirty_KBI: [ internal use only ]
1242 * Set all bits in the page's dirty field.
1244 * The object containing the specified page must be locked if the
1245 * call is made from the machine-independent layer.
1247 * See vm_page_clear_dirty_mask().
1249 * This function should only be called by vm_page_dirty().
1252 vm_page_dirty_KBI(vm_page_t m)
1255 /* Refer to this operation by its public name. */
1256 KASSERT(m->valid == VM_PAGE_BITS_ALL,
1257 ("vm_page_dirty: page is invalid!"));
1258 m->dirty = VM_PAGE_BITS_ALL;
1262 * vm_page_insert: [ internal use only ]
1264 * Inserts the given mem entry into the object and object list.
1266 * The object must be locked.
1269 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1273 VM_OBJECT_ASSERT_WLOCKED(object);
1274 mpred = vm_radix_lookup_le(&object->rtree, pindex);
1275 return (vm_page_insert_after(m, object, pindex, mpred));
1279 * vm_page_insert_after:
1281 * Inserts the page "m" into the specified object at offset "pindex".
1283 * The page "mpred" must immediately precede the offset "pindex" within
1284 * the specified object.
1286 * The object must be locked.
1289 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1294 VM_OBJECT_ASSERT_WLOCKED(object);
1295 KASSERT(m->object == NULL,
1296 ("vm_page_insert_after: page already inserted"));
1297 if (mpred != NULL) {
1298 KASSERT(mpred->object == object,
1299 ("vm_page_insert_after: object doesn't contain mpred"));
1300 KASSERT(mpred->pindex < pindex,
1301 ("vm_page_insert_after: mpred doesn't precede pindex"));
1302 msucc = TAILQ_NEXT(mpred, listq);
1304 msucc = TAILQ_FIRST(&object->memq);
1306 KASSERT(msucc->pindex > pindex,
1307 ("vm_page_insert_after: msucc doesn't succeed pindex"));
1310 * Record the object/offset pair in this page
1316 * Now link into the object's ordered list of backed pages.
1318 if (vm_radix_insert(&object->rtree, m)) {
1323 vm_page_insert_radixdone(m, object, mpred);
1328 * vm_page_insert_radixdone:
1330 * Complete page "m" insertion into the specified object after the
1331 * radix trie hooking.
1333 * The page "mpred" must precede the offset "m->pindex" within the
1336 * The object must be locked.
1339 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1342 VM_OBJECT_ASSERT_WLOCKED(object);
1343 KASSERT(object != NULL && m->object == object,
1344 ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1345 if (mpred != NULL) {
1346 KASSERT(mpred->object == object,
1347 ("vm_page_insert_after: object doesn't contain mpred"));
1348 KASSERT(mpred->pindex < m->pindex,
1349 ("vm_page_insert_after: mpred doesn't precede pindex"));
1353 TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1355 TAILQ_INSERT_HEAD(&object->memq, m, listq);
1358 * Show that the object has one more resident page.
1360 object->resident_page_count++;
1363 * Hold the vnode until the last page is released.
1365 if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1366 vhold(object->handle);
1369 * Since we are inserting a new and possibly dirty page,
1370 * update the object's OBJ_MIGHTBEDIRTY flag.
1372 if (pmap_page_is_write_mapped(m))
1373 vm_object_set_writeable_dirty(object);
1379 * Removes the specified page from its containing object, but does not
1380 * invalidate any backing storage.
1382 * The object must be locked. The page must be locked if it is managed.
1385 vm_page_remove(vm_page_t m)
1390 if ((m->oflags & VPO_UNMANAGED) == 0)
1391 vm_page_assert_locked(m);
1392 if ((object = m->object) == NULL)
1394 VM_OBJECT_ASSERT_WLOCKED(object);
1395 if (vm_page_xbusied(m))
1396 vm_page_xunbusy_maybelocked(m);
1397 mrem = vm_radix_remove(&object->rtree, m->pindex);
1398 KASSERT(mrem == m, ("removed page %p, expected page %p", mrem, m));
1401 * Now remove from the object's list of backed pages.
1403 TAILQ_REMOVE(&object->memq, m, listq);
1406 * And show that the object has one fewer resident page.
1408 object->resident_page_count--;
1411 * The vnode may now be recycled.
1413 if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1414 vdrop(object->handle);
1422 * Returns the page associated with the object/offset
1423 * pair specified; if none is found, NULL is returned.
1425 * The object must be locked.
1428 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1431 VM_OBJECT_ASSERT_LOCKED(object);
1432 return (vm_radix_lookup(&object->rtree, pindex));
1436 * vm_page_find_least:
1438 * Returns the page associated with the object with least pindex
1439 * greater than or equal to the parameter pindex, or NULL.
1441 * The object must be locked.
1444 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1448 VM_OBJECT_ASSERT_LOCKED(object);
1449 if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1450 m = vm_radix_lookup_ge(&object->rtree, pindex);
1455 * Returns the given page's successor (by pindex) within the object if it is
1456 * resident; if none is found, NULL is returned.
1458 * The object must be locked.
1461 vm_page_next(vm_page_t m)
1465 VM_OBJECT_ASSERT_LOCKED(m->object);
1466 if ((next = TAILQ_NEXT(m, listq)) != NULL) {
1467 MPASS(next->object == m->object);
1468 if (next->pindex != m->pindex + 1)
1475 * Returns the given page's predecessor (by pindex) within the object if it is
1476 * resident; if none is found, NULL is returned.
1478 * The object must be locked.
1481 vm_page_prev(vm_page_t m)
1485 VM_OBJECT_ASSERT_LOCKED(m->object);
1486 if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL) {
1487 MPASS(prev->object == m->object);
1488 if (prev->pindex != m->pindex - 1)
1495 * Uses the page mnew as a replacement for an existing page at index
1496 * pindex which must be already present in the object.
1498 * The existing page must not be on a paging queue.
1501 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1505 VM_OBJECT_ASSERT_WLOCKED(object);
1506 KASSERT(mnew->object == NULL,
1507 ("vm_page_replace: page already in object"));
1510 * This function mostly follows vm_page_insert() and
1511 * vm_page_remove() without the radix, object count and vnode
1512 * dance. Double check such functions for more comments.
1515 mnew->object = object;
1516 mnew->pindex = pindex;
1517 mold = vm_radix_replace(&object->rtree, mnew);
1518 KASSERT(mold->queue == PQ_NONE,
1519 ("vm_page_replace: mold is on a paging queue"));
1521 /* Keep the resident page list in sorted order. */
1522 TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1523 TAILQ_REMOVE(&object->memq, mold, listq);
1525 mold->object = NULL;
1526 vm_page_xunbusy_maybelocked(mold);
1529 * The object's resident_page_count does not change because we have
1530 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1532 if (pmap_page_is_write_mapped(mnew))
1533 vm_object_set_writeable_dirty(object);
1540 * Move the given memory entry from its
1541 * current object to the specified target object/offset.
1543 * Note: swap associated with the page must be invalidated by the move. We
1544 * have to do this for several reasons: (1) we aren't freeing the
1545 * page, (2) we are dirtying the page, (3) the VM system is probably
1546 * moving the page from object A to B, and will then later move
1547 * the backing store from A to B and we can't have a conflict.
1549 * Note: we *always* dirty the page. It is necessary both for the
1550 * fact that we moved it, and because we may be invalidating
1553 * The objects must be locked.
1556 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1561 VM_OBJECT_ASSERT_WLOCKED(new_object);
1563 mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1564 KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1565 ("vm_page_rename: pindex already renamed"));
1568 * Create a custom version of vm_page_insert() which does not depend
1569 * by m_prev and can cheat on the implementation aspects of the
1573 m->pindex = new_pindex;
1574 if (vm_radix_insert(&new_object->rtree, m)) {
1580 * The operation cannot fail anymore. The removal must happen before
1581 * the listq iterator is tainted.
1587 /* Return back to the new pindex to complete vm_page_insert(). */
1588 m->pindex = new_pindex;
1589 m->object = new_object;
1591 vm_page_insert_radixdone(m, new_object, mpred);
1599 * Allocate and return a page that is associated with the specified
1600 * object and offset pair. By default, this page is exclusive busied.
1602 * The caller must always specify an allocation class.
1604 * allocation classes:
1605 * VM_ALLOC_NORMAL normal process request
1606 * VM_ALLOC_SYSTEM system *really* needs a page
1607 * VM_ALLOC_INTERRUPT interrupt time request
1609 * optional allocation flags:
1610 * VM_ALLOC_COUNT(number) the number of additional pages that the caller
1611 * intends to allocate
1612 * VM_ALLOC_NOBUSY do not exclusive busy the page
1613 * VM_ALLOC_NODUMP do not include the page in a kernel core dump
1614 * VM_ALLOC_NOOBJ page is not associated with an object and
1615 * should not be exclusive busy
1616 * VM_ALLOC_SBUSY shared busy the allocated page
1617 * VM_ALLOC_WIRED wire the allocated page
1618 * VM_ALLOC_ZERO prefer a zeroed page
1621 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1624 return (vm_page_alloc_after(object, pindex, req, object != NULL ?
1625 vm_radix_lookup_le(&object->rtree, pindex) : NULL));
1629 vm_page_alloc_domain(vm_object_t object, vm_pindex_t pindex, int domain,
1633 return (vm_page_alloc_domain_after(object, pindex, domain, req,
1634 object != NULL ? vm_radix_lookup_le(&object->rtree, pindex) :
1639 * Allocate a page in the specified object with the given page index. To
1640 * optimize insertion of the page into the object, the caller must also specifiy
1641 * the resident page in the object with largest index smaller than the given
1642 * page index, or NULL if no such page exists.
1645 vm_page_alloc_after(vm_object_t object, vm_pindex_t pindex,
1646 int req, vm_page_t mpred)
1648 struct vm_domainset_iter di;
1652 vm_domainset_iter_page_init(&di, object, &domain, &req);
1654 m = vm_page_alloc_domain_after(object, pindex, domain, req,
1658 } while (vm_domainset_iter_page(&di, &domain, &req) == 0);
1664 * Returns true if the number of free pages exceeds the minimum
1665 * for the request class and false otherwise.
1668 vm_domain_available(struct vm_domain *vmd, int req, int npages)
1671 vm_domain_free_assert_locked(vmd);
1672 req = req & VM_ALLOC_CLASS_MASK;
1675 * The page daemon is allowed to dig deeper into the free page list.
1677 if (curproc == pageproc && req != VM_ALLOC_INTERRUPT)
1678 req = VM_ALLOC_SYSTEM;
1680 if (vmd->vmd_free_count >= npages + vmd->vmd_free_reserved ||
1681 (req == VM_ALLOC_SYSTEM &&
1682 vmd->vmd_free_count >= npages + vmd->vmd_interrupt_free_min) ||
1683 (req == VM_ALLOC_INTERRUPT &&
1684 vmd->vmd_free_count >= npages))
1691 vm_page_alloc_domain_after(vm_object_t object, vm_pindex_t pindex, int domain,
1692 int req, vm_page_t mpred)
1694 struct vm_domain *vmd;
1699 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1700 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1701 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1702 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1703 ("inconsistent object(%p)/req(%x)", object, req));
1704 KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0,
1705 ("Can't sleep and retry object insertion."));
1706 KASSERT(mpred == NULL || mpred->pindex < pindex,
1707 ("mpred %p doesn't precede pindex 0x%jx", mpred,
1708 (uintmax_t)pindex));
1710 VM_OBJECT_ASSERT_WLOCKED(object);
1714 #if VM_NRESERVLEVEL > 0
1715 if (vm_object_reserv(object) &&
1716 (m = vm_reserv_extend(req, object, pindex, domain, mpred))
1718 domain = vm_phys_domain(m);
1719 vmd = VM_DOMAIN(domain);
1723 vmd = VM_DOMAIN(domain);
1724 vm_domain_free_lock(vmd);
1725 if (vm_domain_available(vmd, req, 1)) {
1727 * Can we allocate the page from a reservation?
1729 #if VM_NRESERVLEVEL > 0
1730 if (!vm_object_reserv(object) ||
1731 (m = vm_reserv_alloc_page(object, pindex,
1732 domain, mpred)) == NULL)
1736 * If not, allocate it from the free page queues.
1738 m = vm_phys_alloc_pages(domain, object != NULL ?
1739 VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1740 #if VM_NRESERVLEVEL > 0
1741 if (m == NULL && vm_reserv_reclaim_inactive(domain)) {
1742 m = vm_phys_alloc_pages(domain,
1744 VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1752 * Not allocatable, give up.
1754 if (vm_domain_alloc_fail(vmd, object, req))
1760 * At this point we had better have found a good page.
1762 KASSERT(m != NULL, ("missing page"));
1763 free_count = vm_domain_freecnt_adj(vmd, -1);
1764 vm_domain_free_unlock(vmd);
1767 * Don't wakeup too often - wakeup the pageout daemon when
1768 * we would be nearly out of memory.
1770 if (vm_paging_needed(vmd, free_count))
1771 pagedaemon_wakeup(vmd->vmd_domain);
1772 #if VM_NRESERVLEVEL > 0
1775 vm_page_alloc_check(m);
1778 * Initialize the page. Only the PG_ZERO flag is inherited.
1781 if ((req & VM_ALLOC_ZERO) != 0)
1784 if ((req & VM_ALLOC_NODUMP) != 0)
1788 m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1790 m->busy_lock = VPB_UNBUSIED;
1791 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1792 m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1793 if ((req & VM_ALLOC_SBUSY) != 0)
1794 m->busy_lock = VPB_SHARERS_WORD(1);
1795 if (req & VM_ALLOC_WIRED) {
1797 * The page lock is not required for wiring a page until that
1798 * page is inserted into the object.
1805 if (object != NULL) {
1806 if (vm_page_insert_after(m, object, pindex, mpred)) {
1807 pagedaemon_wakeup(domain);
1808 if (req & VM_ALLOC_WIRED) {
1812 KASSERT(m->object == NULL, ("page %p has object", m));
1813 m->oflags = VPO_UNMANAGED;
1814 m->busy_lock = VPB_UNBUSIED;
1815 /* Don't change PG_ZERO. */
1816 vm_page_free_toq(m);
1817 if (req & VM_ALLOC_WAITFAIL) {
1818 VM_OBJECT_WUNLOCK(object);
1820 VM_OBJECT_WLOCK(object);
1825 /* Ignore device objects; the pager sets "memattr" for them. */
1826 if (object->memattr != VM_MEMATTR_DEFAULT &&
1827 (object->flags & OBJ_FICTITIOUS) == 0)
1828 pmap_page_set_memattr(m, object->memattr);
1836 * vm_page_alloc_contig:
1838 * Allocate a contiguous set of physical pages of the given size "npages"
1839 * from the free lists. All of the physical pages must be at or above
1840 * the given physical address "low" and below the given physical address
1841 * "high". The given value "alignment" determines the alignment of the
1842 * first physical page in the set. If the given value "boundary" is
1843 * non-zero, then the set of physical pages cannot cross any physical
1844 * address boundary that is a multiple of that value. Both "alignment"
1845 * and "boundary" must be a power of two.
1847 * If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1848 * then the memory attribute setting for the physical pages is configured
1849 * to the object's memory attribute setting. Otherwise, the memory
1850 * attribute setting for the physical pages is configured to "memattr",
1851 * overriding the object's memory attribute setting. However, if the
1852 * object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1853 * memory attribute setting for the physical pages cannot be configured
1854 * to VM_MEMATTR_DEFAULT.
1856 * The specified object may not contain fictitious pages.
1858 * The caller must always specify an allocation class.
1860 * allocation classes:
1861 * VM_ALLOC_NORMAL normal process request
1862 * VM_ALLOC_SYSTEM system *really* needs a page
1863 * VM_ALLOC_INTERRUPT interrupt time request
1865 * optional allocation flags:
1866 * VM_ALLOC_NOBUSY do not exclusive busy the page
1867 * VM_ALLOC_NODUMP do not include the page in a kernel core dump
1868 * VM_ALLOC_NOOBJ page is not associated with an object and
1869 * should not be exclusive busy
1870 * VM_ALLOC_SBUSY shared busy the allocated page
1871 * VM_ALLOC_WIRED wire the allocated page
1872 * VM_ALLOC_ZERO prefer a zeroed page
1875 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1876 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1877 vm_paddr_t boundary, vm_memattr_t memattr)
1879 struct vm_domainset_iter di;
1883 vm_domainset_iter_page_init(&di, object, &domain, &req);
1885 m = vm_page_alloc_contig_domain(object, pindex, domain, req,
1886 npages, low, high, alignment, boundary, memattr);
1889 } while (vm_domainset_iter_page(&di, &domain, &req) == 0);
1895 vm_page_alloc_contig_domain(vm_object_t object, vm_pindex_t pindex, int domain,
1896 int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1897 vm_paddr_t boundary, vm_memattr_t memattr)
1899 struct vm_domain *vmd;
1900 vm_page_t m, m_ret, mpred;
1901 u_int busy_lock, flags, oflags;
1903 mpred = NULL; /* XXX: pacify gcc */
1904 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1905 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1906 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1907 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1908 ("vm_page_alloc_contig: inconsistent object(%p)/req(%x)", object,
1910 KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0,
1911 ("Can't sleep and retry object insertion."));
1912 if (object != NULL) {
1913 VM_OBJECT_ASSERT_WLOCKED(object);
1914 KASSERT((object->flags & OBJ_FICTITIOUS) == 0,
1915 ("vm_page_alloc_contig: object %p has fictitious pages",
1918 KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1920 if (object != NULL) {
1921 mpred = vm_radix_lookup_le(&object->rtree, pindex);
1922 KASSERT(mpred == NULL || mpred->pindex != pindex,
1923 ("vm_page_alloc_contig: pindex already allocated"));
1927 * Can we allocate the pages without the number of free pages falling
1928 * below the lower bound for the allocation class?
1931 #if VM_NRESERVLEVEL > 0
1932 if (vm_object_reserv(object) &&
1933 (m_ret = vm_reserv_extend_contig(req, object, pindex, domain,
1934 npages, low, high, alignment, boundary, mpred)) != NULL) {
1935 domain = vm_phys_domain(m_ret);
1936 vmd = VM_DOMAIN(domain);
1941 vmd = VM_DOMAIN(domain);
1942 vm_domain_free_lock(vmd);
1943 if (vm_domain_available(vmd, req, npages)) {
1945 * Can we allocate the pages from a reservation?
1947 #if VM_NRESERVLEVEL > 0
1949 if (!vm_object_reserv(object) ||
1950 (m_ret = vm_reserv_alloc_contig(object, pindex, domain,
1951 npages, low, high, alignment, boundary, mpred)) == NULL)
1954 * If not, allocate them from the free page queues.
1956 m_ret = vm_phys_alloc_contig(domain, npages, low, high,
1957 alignment, boundary);
1958 #if VM_NRESERVLEVEL > 0
1959 if (m_ret == NULL && vm_reserv_reclaim_contig(
1960 domain, npages, low, high, alignment, boundary))
1964 if (m_ret == NULL) {
1965 if (vm_domain_alloc_fail(vmd, object, req))
1969 vm_domain_freecnt_adj(vmd, -npages);
1970 vm_domain_free_unlock(vmd);
1971 #if VM_NRESERVLEVEL > 0
1974 for (m = m_ret; m < &m_ret[npages]; m++)
1975 vm_page_alloc_check(m);
1978 * Initialize the pages. Only the PG_ZERO flag is inherited.
1981 if ((req & VM_ALLOC_ZERO) != 0)
1983 if ((req & VM_ALLOC_NODUMP) != 0)
1985 oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1987 busy_lock = VPB_UNBUSIED;
1988 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1989 busy_lock = VPB_SINGLE_EXCLUSIVER;
1990 if ((req & VM_ALLOC_SBUSY) != 0)
1991 busy_lock = VPB_SHARERS_WORD(1);
1992 if ((req & VM_ALLOC_WIRED) != 0)
1993 vm_wire_add(npages);
1994 if (object != NULL) {
1995 if (object->memattr != VM_MEMATTR_DEFAULT &&
1996 memattr == VM_MEMATTR_DEFAULT)
1997 memattr = object->memattr;
1999 for (m = m_ret; m < &m_ret[npages]; m++) {
2001 m->flags = (m->flags | PG_NODUMP) & flags;
2002 m->busy_lock = busy_lock;
2003 if ((req & VM_ALLOC_WIRED) != 0)
2007 if (object != NULL) {
2008 if (vm_page_insert_after(m, object, pindex, mpred)) {
2009 pagedaemon_wakeup(domain);
2010 if ((req & VM_ALLOC_WIRED) != 0)
2011 vm_wire_sub(npages);
2012 KASSERT(m->object == NULL,
2013 ("page %p has object", m));
2015 for (m = m_ret; m < &m_ret[npages]; m++) {
2017 (req & VM_ALLOC_WIRED) != 0)
2019 m->oflags = VPO_UNMANAGED;
2020 m->busy_lock = VPB_UNBUSIED;
2021 /* Don't change PG_ZERO. */
2022 vm_page_free_toq(m);
2024 if (req & VM_ALLOC_WAITFAIL) {
2025 VM_OBJECT_WUNLOCK(object);
2027 VM_OBJECT_WLOCK(object);
2034 if (memattr != VM_MEMATTR_DEFAULT)
2035 pmap_page_set_memattr(m, memattr);
2038 vmd = VM_DOMAIN(domain);
2039 if (vm_paging_needed(vmd, vmd->vmd_free_count))
2040 pagedaemon_wakeup(domain);
2045 * Check a page that has been freshly dequeued from a freelist.
2048 vm_page_alloc_check(vm_page_t m)
2051 KASSERT(m->object == NULL, ("page %p has object", m));
2052 KASSERT(m->queue == PQ_NONE,
2053 ("page %p has unexpected queue %d", m, m->queue));
2054 KASSERT(!vm_page_held(m), ("page %p is held", m));
2055 KASSERT(!vm_page_busied(m), ("page %p is busy", m));
2056 KASSERT(m->dirty == 0, ("page %p is dirty", m));
2057 KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
2058 ("page %p has unexpected memattr %d",
2059 m, pmap_page_get_memattr(m)));
2060 KASSERT(m->valid == 0, ("free page %p is valid", m));
2064 * vm_page_alloc_freelist:
2066 * Allocate a physical page from the specified free page list.
2068 * The caller must always specify an allocation class.
2070 * allocation classes:
2071 * VM_ALLOC_NORMAL normal process request
2072 * VM_ALLOC_SYSTEM system *really* needs a page
2073 * VM_ALLOC_INTERRUPT interrupt time request
2075 * optional allocation flags:
2076 * VM_ALLOC_COUNT(number) the number of additional pages that the caller
2077 * intends to allocate
2078 * VM_ALLOC_WIRED wire the allocated page
2079 * VM_ALLOC_ZERO prefer a zeroed page
2082 vm_page_alloc_freelist(int freelist, int req)
2084 struct vm_domainset_iter di;
2088 vm_domainset_iter_page_init(&di, kernel_object, &domain, &req);
2090 m = vm_page_alloc_freelist_domain(domain, freelist, req);
2093 } while (vm_domainset_iter_page(&di, &domain, &req) == 0);
2099 vm_page_alloc_freelist_domain(int domain, int freelist, int req)
2101 struct vm_domain *vmd;
2103 u_int flags, free_count;
2106 * Do not allocate reserved pages unless the req has asked for it.
2108 vmd = VM_DOMAIN(domain);
2110 vm_domain_free_lock(vmd);
2111 if (vm_domain_available(vmd, req, 1))
2112 m = vm_phys_alloc_freelist_pages(domain, freelist,
2113 VM_FREEPOOL_DIRECT, 0);
2115 if (vm_domain_alloc_fail(vmd, NULL, req))
2119 free_count = vm_domain_freecnt_adj(vmd, -1);
2120 vm_domain_free_unlock(vmd);
2121 vm_page_alloc_check(m);
2124 * Initialize the page. Only the PG_ZERO flag is inherited.
2128 if ((req & VM_ALLOC_ZERO) != 0)
2131 if ((req & VM_ALLOC_WIRED) != 0) {
2133 * The page lock is not required for wiring a page that does
2134 * not belong to an object.
2139 /* Unmanaged pages don't use "act_count". */
2140 m->oflags = VPO_UNMANAGED;
2141 if (vm_paging_needed(vmd, free_count))
2142 pagedaemon_wakeup(domain);
2146 #define VPSC_ANY 0 /* No restrictions. */
2147 #define VPSC_NORESERV 1 /* Skip reservations; implies VPSC_NOSUPER. */
2148 #define VPSC_NOSUPER 2 /* Skip superpages. */
2151 * vm_page_scan_contig:
2153 * Scan vm_page_array[] between the specified entries "m_start" and
2154 * "m_end" for a run of contiguous physical pages that satisfy the
2155 * specified conditions, and return the lowest page in the run. The
2156 * specified "alignment" determines the alignment of the lowest physical
2157 * page in the run. If the specified "boundary" is non-zero, then the
2158 * run of physical pages cannot span a physical address that is a
2159 * multiple of "boundary".
2161 * "m_end" is never dereferenced, so it need not point to a vm_page
2162 * structure within vm_page_array[].
2164 * "npages" must be greater than zero. "m_start" and "m_end" must not
2165 * span a hole (or discontiguity) in the physical address space. Both
2166 * "alignment" and "boundary" must be a power of two.
2169 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
2170 u_long alignment, vm_paddr_t boundary, int options)
2176 #if VM_NRESERVLEVEL > 0
2179 int m_inc, order, run_ext, run_len;
2181 KASSERT(npages > 0, ("npages is 0"));
2182 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2183 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2187 for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
2188 KASSERT((m->flags & PG_MARKER) == 0,
2189 ("page %p is PG_MARKER", m));
2190 KASSERT((m->flags & PG_FICTITIOUS) == 0 || m->wire_count == 1,
2191 ("fictitious page %p has invalid wire count", m));
2194 * If the current page would be the start of a run, check its
2195 * physical address against the end, alignment, and boundary
2196 * conditions. If it doesn't satisfy these conditions, either
2197 * terminate the scan or advance to the next page that
2198 * satisfies the failed condition.
2201 KASSERT(m_run == NULL, ("m_run != NULL"));
2202 if (m + npages > m_end)
2204 pa = VM_PAGE_TO_PHYS(m);
2205 if ((pa & (alignment - 1)) != 0) {
2206 m_inc = atop(roundup2(pa, alignment) - pa);
2209 if (rounddown2(pa ^ (pa + ptoa(npages) - 1),
2211 m_inc = atop(roundup2(pa, boundary) - pa);
2215 KASSERT(m_run != NULL, ("m_run == NULL"));
2217 vm_page_change_lock(m, &m_mtx);
2220 if (vm_page_held(m))
2222 #if VM_NRESERVLEVEL > 0
2223 else if ((level = vm_reserv_level(m)) >= 0 &&
2224 (options & VPSC_NORESERV) != 0) {
2226 /* Advance to the end of the reservation. */
2227 pa = VM_PAGE_TO_PHYS(m);
2228 m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
2232 else if ((object = m->object) != NULL) {
2234 * The page is considered eligible for relocation if
2235 * and only if it could be laundered or reclaimed by
2238 if (!VM_OBJECT_TRYRLOCK(object)) {
2240 VM_OBJECT_RLOCK(object);
2242 if (m->object != object) {
2244 * The page may have been freed.
2246 VM_OBJECT_RUNLOCK(object);
2248 } else if (vm_page_held(m)) {
2253 KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2254 ("page %p is PG_UNHOLDFREE", m));
2255 /* Don't care: PG_NODUMP, PG_ZERO. */
2256 if (object->type != OBJT_DEFAULT &&
2257 object->type != OBJT_SWAP &&
2258 object->type != OBJT_VNODE) {
2260 #if VM_NRESERVLEVEL > 0
2261 } else if ((options & VPSC_NOSUPER) != 0 &&
2262 (level = vm_reserv_level_iffullpop(m)) >= 0) {
2264 /* Advance to the end of the superpage. */
2265 pa = VM_PAGE_TO_PHYS(m);
2266 m_inc = atop(roundup2(pa + 1,
2267 vm_reserv_size(level)) - pa);
2269 } else if (object->memattr == VM_MEMATTR_DEFAULT &&
2270 m->queue != PQ_NONE && !vm_page_busied(m)) {
2272 * The page is allocated but eligible for
2273 * relocation. Extend the current run by one
2276 KASSERT(pmap_page_get_memattr(m) ==
2278 ("page %p has an unexpected memattr", m));
2279 KASSERT((m->oflags & (VPO_SWAPINPROG |
2280 VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2281 ("page %p has unexpected oflags", m));
2282 /* Don't care: VPO_NOSYNC. */
2287 VM_OBJECT_RUNLOCK(object);
2288 #if VM_NRESERVLEVEL > 0
2289 } else if (level >= 0) {
2291 * The page is reserved but not yet allocated. In
2292 * other words, it is still free. Extend the current
2297 } else if ((order = m->order) < VM_NFREEORDER) {
2299 * The page is enqueued in the physical memory
2300 * allocator's free page queues. Moreover, it is the
2301 * first page in a power-of-two-sized run of
2302 * contiguous free pages. Add these pages to the end
2303 * of the current run, and jump ahead.
2305 run_ext = 1 << order;
2309 * Skip the page for one of the following reasons: (1)
2310 * It is enqueued in the physical memory allocator's
2311 * free page queues. However, it is not the first
2312 * page in a run of contiguous free pages. (This case
2313 * rarely occurs because the scan is performed in
2314 * ascending order.) (2) It is not reserved, and it is
2315 * transitioning from free to allocated. (Conversely,
2316 * the transition from allocated to free for managed
2317 * pages is blocked by the page lock.) (3) It is
2318 * allocated but not contained by an object and not
2319 * wired, e.g., allocated by Xen's balloon driver.
2325 * Extend or reset the current run of pages.
2340 if (run_len >= npages)
2346 * vm_page_reclaim_run:
2348 * Try to relocate each of the allocated virtual pages within the
2349 * specified run of physical pages to a new physical address. Free the
2350 * physical pages underlying the relocated virtual pages. A virtual page
2351 * is relocatable if and only if it could be laundered or reclaimed by
2352 * the page daemon. Whenever possible, a virtual page is relocated to a
2353 * physical address above "high".
2355 * Returns 0 if every physical page within the run was already free or
2356 * just freed by a successful relocation. Otherwise, returns a non-zero
2357 * value indicating why the last attempt to relocate a virtual page was
2360 * "req_class" must be an allocation class.
2363 vm_page_reclaim_run(int req_class, int domain, u_long npages, vm_page_t m_run,
2366 struct vm_domain *vmd;
2368 struct spglist free;
2371 vm_page_t m, m_end, m_new;
2372 int error, order, req;
2374 KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
2375 ("req_class is not an allocation class"));
2379 m_end = m_run + npages;
2381 for (; error == 0 && m < m_end; m++) {
2382 KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2383 ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2386 * Avoid releasing and reacquiring the same page lock.
2388 vm_page_change_lock(m, &m_mtx);
2390 if (vm_page_held(m))
2392 else if ((object = m->object) != NULL) {
2394 * The page is relocated if and only if it could be
2395 * laundered or reclaimed by the page daemon.
2397 if (!VM_OBJECT_TRYWLOCK(object)) {
2399 VM_OBJECT_WLOCK(object);
2401 if (m->object != object) {
2403 * The page may have been freed.
2405 VM_OBJECT_WUNLOCK(object);
2407 } else if (vm_page_held(m)) {
2412 KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2413 ("page %p is PG_UNHOLDFREE", m));
2414 /* Don't care: PG_NODUMP, PG_ZERO. */
2415 if (object->type != OBJT_DEFAULT &&
2416 object->type != OBJT_SWAP &&
2417 object->type != OBJT_VNODE)
2419 else if (object->memattr != VM_MEMATTR_DEFAULT)
2421 else if (m->queue != PQ_NONE && !vm_page_busied(m)) {
2422 KASSERT(pmap_page_get_memattr(m) ==
2424 ("page %p has an unexpected memattr", m));
2425 KASSERT((m->oflags & (VPO_SWAPINPROG |
2426 VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2427 ("page %p has unexpected oflags", m));
2428 /* Don't care: VPO_NOSYNC. */
2429 if (m->valid != 0) {
2431 * First, try to allocate a new page
2432 * that is above "high". Failing
2433 * that, try to allocate a new page
2434 * that is below "m_run". Allocate
2435 * the new page between the end of
2436 * "m_run" and "high" only as a last
2439 req = req_class | VM_ALLOC_NOOBJ;
2440 if ((m->flags & PG_NODUMP) != 0)
2441 req |= VM_ALLOC_NODUMP;
2442 if (trunc_page(high) !=
2443 ~(vm_paddr_t)PAGE_MASK) {
2444 m_new = vm_page_alloc_contig(
2449 VM_MEMATTR_DEFAULT);
2452 if (m_new == NULL) {
2453 pa = VM_PAGE_TO_PHYS(m_run);
2454 m_new = vm_page_alloc_contig(
2456 0, pa - 1, PAGE_SIZE, 0,
2457 VM_MEMATTR_DEFAULT);
2459 if (m_new == NULL) {
2461 m_new = vm_page_alloc_contig(
2463 pa, high, PAGE_SIZE, 0,
2464 VM_MEMATTR_DEFAULT);
2466 if (m_new == NULL) {
2470 KASSERT(m_new->wire_count == 0,
2471 ("page %p is wired", m));
2474 * Replace "m" with the new page. For
2475 * vm_page_replace(), "m" must be busy
2476 * and dequeued. Finally, change "m"
2477 * as if vm_page_free() was called.
2479 if (object->ref_count != 0)
2481 m_new->aflags = m->aflags;
2482 KASSERT(m_new->oflags == VPO_UNMANAGED,
2483 ("page %p is managed", m));
2484 m_new->oflags = m->oflags & VPO_NOSYNC;
2485 pmap_copy_page(m, m_new);
2486 m_new->valid = m->valid;
2487 m_new->dirty = m->dirty;
2488 m->flags &= ~PG_ZERO;
2491 vm_page_replace_checked(m_new, object,
2497 * The new page must be deactivated
2498 * before the object is unlocked.
2500 vm_page_change_lock(m_new, &m_mtx);
2501 vm_page_deactivate(m_new);
2503 m->flags &= ~PG_ZERO;
2506 KASSERT(m->dirty == 0,
2507 ("page %p is dirty", m));
2509 SLIST_INSERT_HEAD(&free, m, plinks.s.ss);
2513 VM_OBJECT_WUNLOCK(object);
2515 MPASS(vm_phys_domain(m) == domain);
2516 vmd = VM_DOMAIN(domain);
2517 vm_domain_free_lock(vmd);
2519 if (order < VM_NFREEORDER) {
2521 * The page is enqueued in the physical memory
2522 * allocator's free page queues. Moreover, it
2523 * is the first page in a power-of-two-sized
2524 * run of contiguous free pages. Jump ahead
2525 * to the last page within that run, and
2526 * continue from there.
2528 m += (1 << order) - 1;
2530 #if VM_NRESERVLEVEL > 0
2531 else if (vm_reserv_is_page_free(m))
2534 vm_domain_free_unlock(vmd);
2535 if (order == VM_NFREEORDER)
2541 if ((m = SLIST_FIRST(&free)) != NULL) {
2542 vmd = VM_DOMAIN(domain);
2543 vm_domain_free_lock(vmd);
2545 MPASS(vm_phys_domain(m) == domain);
2546 SLIST_REMOVE_HEAD(&free, plinks.s.ss);
2547 vm_page_free_phys(vmd, m);
2548 } while ((m = SLIST_FIRST(&free)) != NULL);
2549 vm_domain_free_wakeup(vmd);
2550 vm_domain_free_unlock(vmd);
2557 CTASSERT(powerof2(NRUNS));
2559 #define RUN_INDEX(count) ((count) & (NRUNS - 1))
2561 #define MIN_RECLAIM 8
2564 * vm_page_reclaim_contig:
2566 * Reclaim allocated, contiguous physical memory satisfying the specified
2567 * conditions by relocating the virtual pages using that physical memory.
2568 * Returns true if reclamation is successful and false otherwise. Since
2569 * relocation requires the allocation of physical pages, reclamation may
2570 * fail due to a shortage of free pages. When reclamation fails, callers
2571 * are expected to perform vm_wait() before retrying a failed allocation
2572 * operation, e.g., vm_page_alloc_contig().
2574 * The caller must always specify an allocation class through "req".
2576 * allocation classes:
2577 * VM_ALLOC_NORMAL normal process request
2578 * VM_ALLOC_SYSTEM system *really* needs a page
2579 * VM_ALLOC_INTERRUPT interrupt time request
2581 * The optional allocation flags are ignored.
2583 * "npages" must be greater than zero. Both "alignment" and "boundary"
2584 * must be a power of two.
2587 vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
2588 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
2590 struct vm_domain *vmd;
2591 vm_paddr_t curr_low;
2592 vm_page_t m_run, m_runs[NRUNS];
2593 u_long count, reclaimed;
2594 int error, i, options, req_class;
2596 KASSERT(npages > 0, ("npages is 0"));
2597 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2598 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2599 req_class = req & VM_ALLOC_CLASS_MASK;
2602 * The page daemon is allowed to dig deeper into the free page list.
2604 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2605 req_class = VM_ALLOC_SYSTEM;
2608 * Return if the number of free pages cannot satisfy the requested
2611 vmd = VM_DOMAIN(domain);
2612 count = vmd->vmd_free_count;
2613 if (count < npages + vmd->vmd_free_reserved || (count < npages +
2614 vmd->vmd_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
2615 (count < npages && req_class == VM_ALLOC_INTERRUPT))
2619 * Scan up to three times, relaxing the restrictions ("options") on
2620 * the reclamation of reservations and superpages each time.
2622 for (options = VPSC_NORESERV;;) {
2624 * Find the highest runs that satisfy the given constraints
2625 * and restrictions, and record them in "m_runs".
2630 m_run = vm_phys_scan_contig(domain, npages, curr_low,
2631 high, alignment, boundary, options);
2634 curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages);
2635 m_runs[RUN_INDEX(count)] = m_run;
2640 * Reclaim the highest runs in LIFO (descending) order until
2641 * the number of reclaimed pages, "reclaimed", is at least
2642 * MIN_RECLAIM. Reset "reclaimed" each time because each
2643 * reclamation is idempotent, and runs will (likely) recur
2644 * from one scan to the next as restrictions are relaxed.
2647 for (i = 0; count > 0 && i < NRUNS; i++) {
2649 m_run = m_runs[RUN_INDEX(count)];
2650 error = vm_page_reclaim_run(req_class, domain, npages,
2653 reclaimed += npages;
2654 if (reclaimed >= MIN_RECLAIM)
2660 * Either relax the restrictions on the next scan or return if
2661 * the last scan had no restrictions.
2663 if (options == VPSC_NORESERV)
2664 options = VPSC_NOSUPER;
2665 else if (options == VPSC_NOSUPER)
2667 else if (options == VPSC_ANY)
2668 return (reclaimed != 0);
2673 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
2674 u_long alignment, vm_paddr_t boundary)
2676 struct vm_domainset_iter di;
2680 vm_domainset_iter_page_init(&di, kernel_object, &domain, &req);
2682 ret = vm_page_reclaim_contig_domain(domain, req, npages, low,
2683 high, alignment, boundary);
2686 } while (vm_domainset_iter_page(&di, &domain, &req) == 0);
2692 * Set the domain in the appropriate page level domainset.
2695 vm_domain_set(struct vm_domain *vmd)
2698 mtx_lock(&vm_domainset_lock);
2699 if (!vmd->vmd_minset && vm_paging_min(vmd)) {
2700 vmd->vmd_minset = 1;
2701 DOMAINSET_SET(vmd->vmd_domain, &vm_min_domains);
2703 if (!vmd->vmd_severeset && vm_paging_severe(vmd)) {
2704 vmd->vmd_severeset = 1;
2705 DOMAINSET_CLR(vmd->vmd_domain, &vm_severe_domains);
2707 mtx_unlock(&vm_domainset_lock);
2711 * Clear the domain from the appropriate page level domainset.
2714 vm_domain_clear(struct vm_domain *vmd)
2717 mtx_lock(&vm_domainset_lock);
2718 if (vmd->vmd_minset && !vm_paging_min(vmd)) {
2719 vmd->vmd_minset = 0;
2720 DOMAINSET_CLR(vmd->vmd_domain, &vm_min_domains);
2721 if (vm_min_waiters != 0) {
2723 wakeup(&vm_min_domains);
2726 if (vmd->vmd_severeset && !vm_paging_severe(vmd)) {
2727 vmd->vmd_severeset = 0;
2728 DOMAINSET_CLR(vmd->vmd_domain, &vm_severe_domains);
2729 if (vm_severe_waiters != 0) {
2730 vm_severe_waiters = 0;
2731 wakeup(&vm_severe_domains);
2734 mtx_unlock(&vm_domainset_lock);
2738 * Wait for free pages to exceed the min threshold globally.
2744 mtx_lock(&vm_domainset_lock);
2745 while (vm_page_count_min()) {
2747 msleep(&vm_min_domains, &vm_domainset_lock, PVM, "vmwait", 0);
2749 mtx_unlock(&vm_domainset_lock);
2753 * Wait for free pages to exceed the severe threshold globally.
2756 vm_wait_severe(void)
2759 mtx_lock(&vm_domainset_lock);
2760 while (vm_page_count_severe()) {
2761 vm_severe_waiters++;
2762 msleep(&vm_severe_domains, &vm_domainset_lock, PVM,
2765 mtx_unlock(&vm_domainset_lock);
2772 return (vm_severe_waiters + vm_min_waiters);
2776 vm_wait_doms(const domainset_t *wdoms)
2780 * We use racey wakeup synchronization to avoid expensive global
2781 * locking for the pageproc when sleeping with a non-specific vm_wait.
2782 * To handle this, we only sleep for one tick in this instance. It
2783 * is expected that most allocations for the pageproc will come from
2784 * kmem or vm_page_grab* which will use the more specific and
2785 * race-free vm_wait_domain().
2787 if (curproc == pageproc) {
2788 mtx_lock(&vm_domainset_lock);
2789 vm_pageproc_waiters++;
2790 msleep(&vm_pageproc_waiters, &vm_domainset_lock, PVM,
2792 mtx_unlock(&vm_domainset_lock);
2795 * XXX Ideally we would wait only until the allocation could
2796 * be satisfied. This condition can cause new allocators to
2797 * consume all freed pages while old allocators wait.
2799 mtx_lock(&vm_domainset_lock);
2800 if (DOMAINSET_SUBSET(&vm_min_domains, wdoms)) {
2802 msleep(&vm_min_domains, &vm_domainset_lock, PVM,
2805 mtx_unlock(&vm_domainset_lock);
2812 * Sleep until free pages are available for allocation.
2813 * - Called in various places after failed memory allocations.
2816 vm_wait_domain(int domain)
2818 struct vm_domain *vmd;
2821 vmd = VM_DOMAIN(domain);
2822 vm_domain_free_assert_locked(vmd);
2824 if (curproc == pageproc) {
2825 vmd->vmd_pageout_pages_needed = 1;
2826 msleep(&vmd->vmd_pageout_pages_needed,
2827 vm_domain_free_lockptr(vmd), PDROP | PSWP, "VMWait", 0);
2829 vm_domain_free_unlock(vmd);
2830 if (pageproc == NULL)
2831 panic("vm_wait in early boot");
2832 DOMAINSET_ZERO(&wdom);
2833 DOMAINSET_SET(vmd->vmd_domain, &wdom);
2834 vm_wait_doms(&wdom);
2841 * Sleep until free pages are available for allocation in the
2842 * affinity domains of the obj. If obj is NULL, the domain set
2843 * for the calling thread is used.
2844 * Called in various places after failed memory allocations.
2847 vm_wait(vm_object_t obj)
2849 struct domainset *d;
2854 * Carefully fetch pointers only once: the struct domainset
2855 * itself is ummutable but the pointer might change.
2858 d = obj->domain.dr_policy;
2860 d = curthread->td_domain.dr_policy;
2862 vm_wait_doms(&d->ds_mask);
2866 * vm_domain_alloc_fail:
2868 * Called when a page allocation function fails. Informs the
2869 * pagedaemon and performs the requested wait. Requires the
2870 * domain_free and object lock on entry. Returns with the
2871 * object lock held and free lock released. Returns an error when
2872 * retry is necessary.
2876 vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object, int req)
2879 vm_domain_free_assert_locked(vmd);
2881 atomic_add_int(&vmd->vmd_pageout_deficit,
2882 max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
2883 if (req & (VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL)) {
2885 VM_OBJECT_WUNLOCK(object);
2886 vm_wait_domain(vmd->vmd_domain);
2888 VM_OBJECT_WLOCK(object);
2889 if (req & VM_ALLOC_WAITOK)
2892 vm_domain_free_unlock(vmd);
2893 pagedaemon_wakeup(vmd->vmd_domain);
2901 * Sleep until free pages are available for allocation.
2902 * - Called only in vm_fault so that processes page faulting
2903 * can be easily tracked.
2904 * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2905 * processes will be able to grab memory first. Do not change
2906 * this balance without careful testing first.
2912 mtx_lock(&vm_domainset_lock);
2913 if (vm_page_count_min()) {
2915 msleep(&vm_min_domains, &vm_domainset_lock, PUSER, "pfault", 0);
2917 mtx_unlock(&vm_domainset_lock);
2920 struct vm_pagequeue *
2921 vm_page_pagequeue(vm_page_t m)
2924 return (&vm_pagequeue_domain(m)->vmd_pagequeues[m->queue]);
2930 * Remove the given page from its current page queue.
2932 * The page must be locked.
2935 vm_page_dequeue(vm_page_t m)
2937 struct vm_pagequeue *pq;
2939 vm_page_assert_locked(m);
2940 KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2942 pq = vm_page_pagequeue(m);
2943 vm_pagequeue_lock(pq);
2945 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2946 vm_pagequeue_cnt_dec(pq);
2947 vm_pagequeue_unlock(pq);
2951 * vm_page_dequeue_locked:
2953 * Remove the given page from its current page queue.
2955 * The page and page queue must be locked.
2958 vm_page_dequeue_locked(vm_page_t m)
2960 struct vm_pagequeue *pq;
2962 vm_page_lock_assert(m, MA_OWNED);
2963 pq = vm_page_pagequeue(m);
2964 vm_pagequeue_assert_locked(pq);
2966 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2967 vm_pagequeue_cnt_dec(pq);
2973 * Add the given page to the specified page queue.
2975 * The page must be locked.
2978 vm_page_enqueue(uint8_t queue, vm_page_t m)
2980 struct vm_pagequeue *pq;
2982 vm_page_lock_assert(m, MA_OWNED);
2983 KASSERT(queue < PQ_COUNT,
2984 ("vm_page_enqueue: invalid queue %u request for page %p",
2986 pq = &vm_pagequeue_domain(m)->vmd_pagequeues[queue];
2987 vm_pagequeue_lock(pq);
2989 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2990 vm_pagequeue_cnt_inc(pq);
2991 vm_pagequeue_unlock(pq);
2997 * Move the given page to the tail of its current page queue.
2999 * The page must be locked.
3002 vm_page_requeue(vm_page_t m)
3004 struct vm_pagequeue *pq;
3006 vm_page_lock_assert(m, MA_OWNED);
3007 KASSERT(m->queue != PQ_NONE,
3008 ("vm_page_requeue: page %p is not queued", m));
3009 pq = vm_page_pagequeue(m);
3010 vm_pagequeue_lock(pq);
3011 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
3012 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3013 vm_pagequeue_unlock(pq);
3017 * vm_page_requeue_locked:
3019 * Move the given page to the tail of its current page queue.
3021 * The page queue must be locked.
3024 vm_page_requeue_locked(vm_page_t m)
3026 struct vm_pagequeue *pq;
3028 KASSERT(m->queue != PQ_NONE,
3029 ("vm_page_requeue_locked: page %p is not queued", m));
3030 pq = vm_page_pagequeue(m);
3031 vm_pagequeue_assert_locked(pq);
3032 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
3033 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3039 * Put the specified page on the active list (if appropriate).
3040 * Ensure that act_count is at least ACT_INIT but do not otherwise
3043 * The page must be locked.
3046 vm_page_activate(vm_page_t m)
3050 vm_page_lock_assert(m, MA_OWNED);
3051 if ((queue = m->queue) != PQ_ACTIVE) {
3052 if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
3053 if (m->act_count < ACT_INIT)
3054 m->act_count = ACT_INIT;
3055 if (queue != PQ_NONE)
3057 vm_page_enqueue(PQ_ACTIVE, m);
3060 if (m->act_count < ACT_INIT)
3061 m->act_count = ACT_INIT;
3066 * vm_domain_free_wakeup:
3068 * Helper routine for vm_page_free_toq(). This routine is called
3069 * when a page is added to the free queues.
3071 * The page queues must be locked.
3074 vm_domain_free_wakeup(struct vm_domain *vmd)
3077 vm_domain_free_assert_locked(vmd);
3080 * if pageout daemon needs pages, then tell it that there are
3083 if (vmd->vmd_pageout_pages_needed &&
3084 vmd->vmd_free_count >= vmd->vmd_pageout_free_min) {
3085 wakeup(&vmd->vmd_pageout_pages_needed);
3086 vmd->vmd_pageout_pages_needed = 0;
3089 * wakeup processes that are waiting on memory if we hit a
3090 * high water mark. And wakeup scheduler process if we have
3091 * lots of memory. this process will swapin processes.
3093 if ((vmd->vmd_minset && !vm_paging_min(vmd)) ||
3094 (vmd->vmd_severeset && !vm_paging_severe(vmd)))
3095 vm_domain_clear(vmd);
3097 /* See comments in vm_wait(); */
3098 if (vm_pageproc_waiters) {
3099 vm_pageproc_waiters = 0;
3100 wakeup(&vm_pageproc_waiters);
3106 * vm_page_free_prep:
3108 * Prepares the given page to be put on the free list,
3109 * disassociating it from any VM object. The caller may return
3110 * the page to the free list only if this function returns true.
3112 * The object must be locked. The page must be locked if it is
3113 * managed. For a queued managed page, the pagequeue_locked
3114 * argument specifies whether the page queue is already locked.
3117 vm_page_free_prep(vm_page_t m, bool pagequeue_locked)
3120 #if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP)
3121 if (PMAP_HAS_DMAP && (m->flags & PG_ZERO) != 0) {
3124 p = (uint64_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
3125 for (i = 0; i < PAGE_SIZE / sizeof(uint64_t); i++, p++)
3126 KASSERT(*p == 0, ("vm_page_free_prep %p PG_ZERO %d %jx",
3127 m, i, (uintmax_t)*p));
3130 if ((m->oflags & VPO_UNMANAGED) == 0) {
3131 vm_page_lock_assert(m, MA_OWNED);
3132 KASSERT(!pmap_page_is_mapped(m),
3133 ("vm_page_free_toq: freeing mapped page %p", m));
3135 KASSERT(m->queue == PQ_NONE,
3136 ("vm_page_free_toq: unmanaged page %p is queued", m));
3137 VM_CNT_INC(v_tfree);
3139 if (vm_page_sbusied(m))
3140 panic("vm_page_free: freeing busy page %p", m);
3145 * If fictitious remove object association and
3148 if ((m->flags & PG_FICTITIOUS) != 0) {
3149 KASSERT(m->wire_count == 1,
3150 ("fictitious page %p is not wired", m));
3151 KASSERT(m->queue == PQ_NONE,
3152 ("fictitious page %p is queued", m));
3156 if (m->queue != PQ_NONE) {
3157 if (pagequeue_locked)
3158 vm_page_dequeue_locked(m);
3165 if (m->wire_count != 0)
3166 panic("vm_page_free: freeing wired page %p", m);
3167 if (m->hold_count != 0) {
3168 m->flags &= ~PG_ZERO;
3169 KASSERT((m->flags & PG_UNHOLDFREE) == 0,
3170 ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
3171 m->flags |= PG_UNHOLDFREE;
3176 * Restore the default memory attribute to the page.
3178 if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
3179 pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
3185 * Insert the page into the physical memory allocator's free page
3186 * queues. This is the last step to free a page.
3189 vm_page_free_phys(struct vm_domain *vmd, vm_page_t m)
3192 vm_domain_free_assert_locked(vmd);
3194 vm_domain_freecnt_adj(vmd, 1);
3195 #if VM_NRESERVLEVEL > 0
3196 if (!vm_reserv_free_page(m))
3198 vm_phys_free_pages(m, 0);
3202 vm_page_free_phys_pglist(struct pglist *tq)
3204 struct vm_domain *vmd;
3207 if (TAILQ_EMPTY(tq))
3210 TAILQ_FOREACH(m, tq, listq) {
3211 if (vmd != vm_pagequeue_domain(m)) {
3213 vm_domain_free_wakeup(vmd);
3214 vm_domain_free_unlock(vmd);
3216 vmd = vm_pagequeue_domain(m);
3217 vm_domain_free_lock(vmd);
3219 vm_page_free_phys(vmd, m);
3222 vm_domain_free_wakeup(vmd);
3223 vm_domain_free_unlock(vmd);
3230 * Returns the given page to the free list, disassociating it
3231 * from any VM object.
3233 * The object must be locked. The page must be locked if it is
3237 vm_page_free_toq(vm_page_t m)
3239 struct vm_domain *vmd;
3241 if (!vm_page_free_prep(m, false))
3243 vmd = vm_pagequeue_domain(m);
3244 vm_domain_free_lock(vmd);
3245 vm_page_free_phys(vmd, m);
3246 vm_domain_free_wakeup(vmd);
3247 vm_domain_free_unlock(vmd);
3251 * vm_page_free_pages_toq:
3253 * Returns a list of pages to the free list, disassociating it
3254 * from any VM object. In other words, this is equivalent to
3255 * calling vm_page_free_toq() for each page of a list of VM objects.
3257 * The objects must be locked. The pages must be locked if it is
3261 vm_page_free_pages_toq(struct spglist *free, bool update_wire_count)
3267 if (SLIST_EMPTY(free))
3272 while ((m = SLIST_FIRST(free)) != NULL) {
3274 SLIST_REMOVE_HEAD(free, plinks.s.ss);
3275 if (vm_page_free_prep(m, false))
3276 TAILQ_INSERT_TAIL(&pgl, m, listq);
3279 vm_page_free_phys_pglist(&pgl);
3281 if (update_wire_count)
3288 * Mark this page as wired down. If the page is fictitious, then
3289 * its wire count must remain one.
3291 * The page must be locked.
3294 vm_page_wire(vm_page_t m)
3297 vm_page_assert_locked(m);
3298 if ((m->flags & PG_FICTITIOUS) != 0) {
3299 KASSERT(m->wire_count == 1,
3300 ("vm_page_wire: fictitious page %p's wire count isn't one",
3304 if (m->wire_count == 0) {
3305 KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
3306 m->queue == PQ_NONE,
3307 ("vm_page_wire: unmanaged page %p is queued", m));
3311 KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
3317 * Release one wiring of the specified page, potentially allowing it to be
3318 * paged out. Returns TRUE if the number of wirings transitions to zero and
3321 * Only managed pages belonging to an object can be paged out. If the number
3322 * of wirings transitions to zero and the page is eligible for page out, then
3323 * the page is added to the specified paging queue (unless PQ_NONE is
3324 * specified, in which case the page is dequeued if it belongs to a paging
3327 * If a page is fictitious, then its wire count must always be one.
3329 * A managed page must be locked.
3332 vm_page_unwire(vm_page_t m, uint8_t queue)
3336 KASSERT(queue < PQ_COUNT || queue == PQ_NONE,
3337 ("vm_page_unwire: invalid queue %u request for page %p",
3340 unwired = vm_page_unwire_noq(m);
3341 if (unwired && (m->oflags & VPO_UNMANAGED) == 0 && m->object != NULL) {
3342 if (m->queue == queue) {
3343 if (queue == PQ_ACTIVE)
3344 vm_page_reference(m);
3345 else if (queue != PQ_NONE)
3349 if (queue != PQ_NONE) {
3350 vm_page_enqueue(queue, m);
3351 if (queue == PQ_ACTIVE)
3352 /* Initialize act_count. */
3353 vm_page_activate(m);
3362 * vm_page_unwire_noq:
3364 * Unwire a page without (re-)inserting it into a page queue. It is up
3365 * to the caller to enqueue, requeue, or free the page as appropriate.
3366 * In most cases, vm_page_unwire() should be used instead.
3369 vm_page_unwire_noq(vm_page_t m)
3372 if ((m->oflags & VPO_UNMANAGED) == 0)
3373 vm_page_assert_locked(m);
3374 if ((m->flags & PG_FICTITIOUS) != 0) {
3375 KASSERT(m->wire_count == 1,
3376 ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
3379 if (m->wire_count == 0)
3380 panic("vm_page_unwire: page %p's wire count is zero", m);
3382 if (m->wire_count == 0) {
3390 * Move the specified page to the inactive queue.
3392 * Normally, "noreuse" is FALSE, resulting in LRU ordering of the inactive
3393 * queue. However, setting "noreuse" to TRUE will accelerate the specified
3394 * page's reclamation, but it will not unmap the page from any address space.
3395 * This is implemented by inserting the page near the head of the inactive
3396 * queue, using a marker page to guide FIFO insertion ordering.
3398 * The page must be locked.
3401 _vm_page_deactivate(vm_page_t m, boolean_t noreuse)
3403 struct vm_pagequeue *pq;
3406 vm_page_assert_locked(m);
3409 * Ignore if the page is already inactive, unless it is unlikely to be
3412 if ((queue = m->queue) == PQ_INACTIVE && !noreuse)
3414 if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
3415 pq = &vm_pagequeue_domain(m)->vmd_pagequeues[PQ_INACTIVE];
3416 /* Avoid multiple acquisitions of the inactive queue lock. */
3417 if (queue == PQ_INACTIVE) {
3418 vm_pagequeue_lock(pq);
3419 vm_page_dequeue_locked(m);
3421 if (queue != PQ_NONE)
3423 vm_pagequeue_lock(pq);
3425 m->queue = PQ_INACTIVE;
3427 TAILQ_INSERT_BEFORE(
3428 &vm_pagequeue_domain(m)->vmd_inacthead, m,
3431 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3432 vm_pagequeue_cnt_inc(pq);
3433 vm_pagequeue_unlock(pq);
3438 * Move the specified page to the inactive queue.
3440 * The page must be locked.
3443 vm_page_deactivate(vm_page_t m)
3446 _vm_page_deactivate(m, FALSE);
3450 * Move the specified page to the inactive queue with the expectation
3451 * that it is unlikely to be reused.
3453 * The page must be locked.
3456 vm_page_deactivate_noreuse(vm_page_t m)
3459 _vm_page_deactivate(m, TRUE);
3465 * Put a page in the laundry.
3468 vm_page_launder(vm_page_t m)
3472 vm_page_assert_locked(m);
3473 if ((queue = m->queue) != PQ_LAUNDRY && m->wire_count == 0 &&
3474 (m->oflags & VPO_UNMANAGED) == 0) {
3475 if (queue != PQ_NONE)
3477 vm_page_enqueue(PQ_LAUNDRY, m);
3482 * vm_page_unswappable
3484 * Put a page in the PQ_UNSWAPPABLE holding queue.
3487 vm_page_unswappable(vm_page_t m)
3490 vm_page_assert_locked(m);
3491 KASSERT(m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0,
3492 ("page %p already unswappable", m));
3493 if (m->queue != PQ_NONE)
3495 vm_page_enqueue(PQ_UNSWAPPABLE, m);
3499 * Attempt to free the page. If it cannot be freed, do nothing. Returns true
3500 * if the page is freed and false otherwise.
3502 * The page must be managed. The page and its containing object must be
3506 vm_page_try_to_free(vm_page_t m)
3509 vm_page_assert_locked(m);
3510 VM_OBJECT_ASSERT_WLOCKED(m->object);
3511 KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("page %p is unmanaged", m));
3512 if (m->dirty != 0 || vm_page_held(m) || vm_page_busied(m))
3514 if (m->object->ref_count != 0) {
3526 * Apply the specified advice to the given page.
3528 * The object and page must be locked.
3531 vm_page_advise(vm_page_t m, int advice)
3534 vm_page_assert_locked(m);
3535 VM_OBJECT_ASSERT_WLOCKED(m->object);
3536 if (advice == MADV_FREE)
3538 * Mark the page clean. This will allow the page to be freed
3539 * without first paging it out. MADV_FREE pages are often
3540 * quickly reused by malloc(3), so we do not do anything that
3541 * would result in a page fault on a later access.
3544 else if (advice != MADV_DONTNEED) {
3545 if (advice == MADV_WILLNEED)
3546 vm_page_activate(m);
3551 * Clear any references to the page. Otherwise, the page daemon will
3552 * immediately reactivate the page.
3554 vm_page_aflag_clear(m, PGA_REFERENCED);
3556 if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
3560 * Place clean pages near the head of the inactive queue rather than
3561 * the tail, thus defeating the queue's LRU operation and ensuring that
3562 * the page will be reused quickly. Dirty pages not already in the
3563 * laundry are moved there.
3566 vm_page_deactivate_noreuse(m);
3572 * Grab a page, waiting until we are waken up due to the page
3573 * changing state. We keep on waiting, if the page continues
3574 * to be in the object. If the page doesn't exist, first allocate it
3575 * and then conditionally zero it.
3577 * This routine may sleep.
3579 * The object must be locked on entry. The lock will, however, be released
3580 * and reacquired if the routine sleeps.
3583 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
3589 VM_OBJECT_ASSERT_WLOCKED(object);
3590 KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3591 (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3592 ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
3593 pflags = allocflags &
3594 ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
3595 if ((allocflags & VM_ALLOC_NOWAIT) == 0)
3596 pflags |= VM_ALLOC_WAITFAIL;
3598 if ((m = vm_page_lookup(object, pindex)) != NULL) {
3599 sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3600 vm_page_xbusied(m) : vm_page_busied(m);
3602 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3605 * Reference the page before unlocking and
3606 * sleeping so that the page daemon is less
3607 * likely to reclaim it.
3609 vm_page_aflag_set(m, PGA_REFERENCED);
3611 VM_OBJECT_WUNLOCK(object);
3612 vm_page_busy_sleep(m, "pgrbwt", (allocflags &
3613 VM_ALLOC_IGN_SBUSY) != 0);
3614 VM_OBJECT_WLOCK(object);
3617 if ((allocflags & VM_ALLOC_WIRED) != 0) {
3623 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
3625 if ((allocflags & VM_ALLOC_SBUSY) != 0)
3630 m = vm_page_alloc(object, pindex, pflags);
3632 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3636 if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
3642 * Return the specified range of pages from the given object. For each
3643 * page offset within the range, if a page already exists within the object
3644 * at that offset and it is busy, then wait for it to change state. If,
3645 * instead, the page doesn't exist, then allocate it.
3647 * The caller must always specify an allocation class.
3649 * allocation classes:
3650 * VM_ALLOC_NORMAL normal process request
3651 * VM_ALLOC_SYSTEM system *really* needs the pages
3653 * The caller must always specify that the pages are to be busied and/or
3656 * optional allocation flags:
3657 * VM_ALLOC_IGN_SBUSY do not sleep on soft busy pages
3658 * VM_ALLOC_NOBUSY do not exclusive busy the page
3659 * VM_ALLOC_NOWAIT do not sleep
3660 * VM_ALLOC_SBUSY set page to sbusy state
3661 * VM_ALLOC_WIRED wire the pages
3662 * VM_ALLOC_ZERO zero and validate any invalid pages
3664 * If VM_ALLOC_NOWAIT is not specified, this routine may sleep. Otherwise, it
3665 * may return a partial prefix of the requested range.
3668 vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
3669 vm_page_t *ma, int count)
3676 VM_OBJECT_ASSERT_WLOCKED(object);
3677 KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0,
3678 ("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed"));
3679 KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 ||
3680 (allocflags & VM_ALLOC_WIRED) != 0,
3681 ("vm_page_grab_pages: the pages must be busied or wired"));
3682 KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3683 (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3684 ("vm_page_grab_pages: VM_ALLOC_SBUSY/IGN_SBUSY mismatch"));
3687 pflags = allocflags & ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK |
3688 VM_ALLOC_WAITFAIL | VM_ALLOC_IGN_SBUSY);
3689 if ((allocflags & VM_ALLOC_NOWAIT) == 0)
3690 pflags |= VM_ALLOC_WAITFAIL;
3693 m = vm_radix_lookup_le(&object->rtree, pindex + i);
3694 if (m == NULL || m->pindex != pindex + i) {
3698 mpred = TAILQ_PREV(m, pglist, listq);
3699 for (; i < count; i++) {
3701 sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3702 vm_page_xbusied(m) : vm_page_busied(m);
3704 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3707 * Reference the page before unlocking and
3708 * sleeping so that the page daemon is less
3709 * likely to reclaim it.
3711 vm_page_aflag_set(m, PGA_REFERENCED);
3713 VM_OBJECT_WUNLOCK(object);
3714 vm_page_busy_sleep(m, "grbmaw", (allocflags &
3715 VM_ALLOC_IGN_SBUSY) != 0);
3716 VM_OBJECT_WLOCK(object);
3719 if ((allocflags & VM_ALLOC_WIRED) != 0) {
3724 if ((allocflags & (VM_ALLOC_NOBUSY |
3725 VM_ALLOC_SBUSY)) == 0)
3727 if ((allocflags & VM_ALLOC_SBUSY) != 0)
3730 m = vm_page_alloc_after(object, pindex + i,
3731 pflags | VM_ALLOC_COUNT(count - i), mpred);
3733 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3738 if (m->valid == 0 && (allocflags & VM_ALLOC_ZERO) != 0) {
3739 if ((m->flags & PG_ZERO) == 0)
3741 m->valid = VM_PAGE_BITS_ALL;
3744 m = vm_page_next(m);
3750 * Mapping function for valid or dirty bits in a page.
3752 * Inputs are required to range within a page.
3755 vm_page_bits(int base, int size)
3761 base + size <= PAGE_SIZE,
3762 ("vm_page_bits: illegal base/size %d/%d", base, size)
3765 if (size == 0) /* handle degenerate case */
3768 first_bit = base >> DEV_BSHIFT;
3769 last_bit = (base + size - 1) >> DEV_BSHIFT;
3771 return (((vm_page_bits_t)2 << last_bit) -
3772 ((vm_page_bits_t)1 << first_bit));
3776 * vm_page_set_valid_range:
3778 * Sets portions of a page valid. The arguments are expected
3779 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3780 * of any partial chunks touched by the range. The invalid portion of
3781 * such chunks will be zeroed.
3783 * (base + size) must be less then or equal to PAGE_SIZE.
3786 vm_page_set_valid_range(vm_page_t m, int base, int size)
3790 VM_OBJECT_ASSERT_WLOCKED(m->object);
3791 if (size == 0) /* handle degenerate case */
3795 * If the base is not DEV_BSIZE aligned and the valid
3796 * bit is clear, we have to zero out a portion of the
3799 if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3800 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
3801 pmap_zero_page_area(m, frag, base - frag);
3804 * If the ending offset is not DEV_BSIZE aligned and the
3805 * valid bit is clear, we have to zero out a portion of
3808 endoff = base + size;
3809 if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3810 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
3811 pmap_zero_page_area(m, endoff,
3812 DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3815 * Assert that no previously invalid block that is now being validated
3818 KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
3819 ("vm_page_set_valid_range: page %p is dirty", m));
3822 * Set valid bits inclusive of any overlap.
3824 m->valid |= vm_page_bits(base, size);
3828 * Clear the given bits from the specified page's dirty field.
3830 static __inline void
3831 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
3834 #if PAGE_SIZE < 16384
3839 * If the object is locked and the page is neither exclusive busy nor
3840 * write mapped, then the page's dirty field cannot possibly be
3841 * set by a concurrent pmap operation.
3843 VM_OBJECT_ASSERT_WLOCKED(m->object);
3844 if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
3845 m->dirty &= ~pagebits;
3848 * The pmap layer can call vm_page_dirty() without
3849 * holding a distinguished lock. The combination of
3850 * the object's lock and an atomic operation suffice
3851 * to guarantee consistency of the page dirty field.
3853 * For PAGE_SIZE == 32768 case, compiler already
3854 * properly aligns the dirty field, so no forcible
3855 * alignment is needed. Only require existence of
3856 * atomic_clear_64 when page size is 32768.
3858 addr = (uintptr_t)&m->dirty;
3859 #if PAGE_SIZE == 32768
3860 atomic_clear_64((uint64_t *)addr, pagebits);
3861 #elif PAGE_SIZE == 16384
3862 atomic_clear_32((uint32_t *)addr, pagebits);
3863 #else /* PAGE_SIZE <= 8192 */
3865 * Use a trick to perform a 32-bit atomic on the
3866 * containing aligned word, to not depend on the existence
3867 * of atomic_clear_{8, 16}.
3869 shift = addr & (sizeof(uint32_t) - 1);
3870 #if BYTE_ORDER == BIG_ENDIAN
3871 shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
3875 addr &= ~(sizeof(uint32_t) - 1);
3876 atomic_clear_32((uint32_t *)addr, pagebits << shift);
3877 #endif /* PAGE_SIZE */
3882 * vm_page_set_validclean:
3884 * Sets portions of a page valid and clean. The arguments are expected
3885 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3886 * of any partial chunks touched by the range. The invalid portion of
3887 * such chunks will be zero'd.
3889 * (base + size) must be less then or equal to PAGE_SIZE.
3892 vm_page_set_validclean(vm_page_t m, int base, int size)
3894 vm_page_bits_t oldvalid, pagebits;
3897 VM_OBJECT_ASSERT_WLOCKED(m->object);
3898 if (size == 0) /* handle degenerate case */
3902 * If the base is not DEV_BSIZE aligned and the valid
3903 * bit is clear, we have to zero out a portion of the
3906 if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3907 (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3908 pmap_zero_page_area(m, frag, base - frag);
3911 * If the ending offset is not DEV_BSIZE aligned and the
3912 * valid bit is clear, we have to zero out a portion of
3915 endoff = base + size;
3916 if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3917 (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3918 pmap_zero_page_area(m, endoff,
3919 DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3922 * Set valid, clear dirty bits. If validating the entire
3923 * page we can safely clear the pmap modify bit. We also
3924 * use this opportunity to clear the VPO_NOSYNC flag. If a process
3925 * takes a write fault on a MAP_NOSYNC memory area the flag will
3928 * We set valid bits inclusive of any overlap, but we can only
3929 * clear dirty bits for DEV_BSIZE chunks that are fully within
3932 oldvalid = m->valid;
3933 pagebits = vm_page_bits(base, size);
3934 m->valid |= pagebits;
3936 if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3937 frag = DEV_BSIZE - frag;
3943 pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3945 if (base == 0 && size == PAGE_SIZE) {
3947 * The page can only be modified within the pmap if it is
3948 * mapped, and it can only be mapped if it was previously
3951 if (oldvalid == VM_PAGE_BITS_ALL)
3953 * Perform the pmap_clear_modify() first. Otherwise,
3954 * a concurrent pmap operation, such as
3955 * pmap_protect(), could clear a modification in the
3956 * pmap and set the dirty field on the page before
3957 * pmap_clear_modify() had begun and after the dirty
3958 * field was cleared here.
3960 pmap_clear_modify(m);
3962 m->oflags &= ~VPO_NOSYNC;
3963 } else if (oldvalid != VM_PAGE_BITS_ALL)
3964 m->dirty &= ~pagebits;
3966 vm_page_clear_dirty_mask(m, pagebits);
3970 vm_page_clear_dirty(vm_page_t m, int base, int size)
3973 vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3977 * vm_page_set_invalid:
3979 * Invalidates DEV_BSIZE'd chunks within a page. Both the
3980 * valid and dirty bits for the effected areas are cleared.
3983 vm_page_set_invalid(vm_page_t m, int base, int size)
3985 vm_page_bits_t bits;
3989 VM_OBJECT_ASSERT_WLOCKED(object);
3990 if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3991 size >= object->un_pager.vnp.vnp_size)
3992 bits = VM_PAGE_BITS_ALL;
3994 bits = vm_page_bits(base, size);
3995 if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
3998 KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3999 !pmap_page_is_mapped(m),
4000 ("vm_page_set_invalid: page %p is mapped", m));
4006 * vm_page_zero_invalid()
4008 * The kernel assumes that the invalid portions of a page contain
4009 * garbage, but such pages can be mapped into memory by user code.
4010 * When this occurs, we must zero out the non-valid portions of the
4011 * page so user code sees what it expects.
4013 * Pages are most often semi-valid when the end of a file is mapped
4014 * into memory and the file's size is not page aligned.
4017 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
4022 VM_OBJECT_ASSERT_WLOCKED(m->object);
4024 * Scan the valid bits looking for invalid sections that
4025 * must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the
4026 * valid bit may be set ) have already been zeroed by
4027 * vm_page_set_validclean().
4029 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
4030 if (i == (PAGE_SIZE / DEV_BSIZE) ||
4031 (m->valid & ((vm_page_bits_t)1 << i))) {
4033 pmap_zero_page_area(m,
4034 b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
4041 * setvalid is TRUE when we can safely set the zero'd areas
4042 * as being valid. We can do this if there are no cache consistancy
4043 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
4046 m->valid = VM_PAGE_BITS_ALL;
4052 * Is (partial) page valid? Note that the case where size == 0
4053 * will return FALSE in the degenerate case where the page is
4054 * entirely invalid, and TRUE otherwise.
4057 vm_page_is_valid(vm_page_t m, int base, int size)
4059 vm_page_bits_t bits;
4061 VM_OBJECT_ASSERT_LOCKED(m->object);
4062 bits = vm_page_bits(base, size);
4063 return (m->valid != 0 && (m->valid & bits) == bits);
4067 * Returns true if all of the specified predicates are true for the entire
4068 * (super)page and false otherwise.
4071 vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m)
4077 VM_OBJECT_ASSERT_LOCKED(object);
4078 npages = atop(pagesizes[m->psind]);
4081 * The physically contiguous pages that make up a superpage, i.e., a
4082 * page with a page size index ("psind") greater than zero, will
4083 * occupy adjacent entries in vm_page_array[].
4085 for (i = 0; i < npages; i++) {
4086 /* Always test object consistency, including "skip_m". */
4087 if (m[i].object != object)
4089 if (&m[i] == skip_m)
4091 if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i]))
4093 if ((flags & PS_ALL_DIRTY) != 0) {
4095 * Calling vm_page_test_dirty() or pmap_is_modified()
4096 * might stop this case from spuriously returning
4097 * "false". However, that would require a write lock
4098 * on the object containing "m[i]".
4100 if (m[i].dirty != VM_PAGE_BITS_ALL)
4103 if ((flags & PS_ALL_VALID) != 0 &&
4104 m[i].valid != VM_PAGE_BITS_ALL)
4111 * Set the page's dirty bits if the page is modified.
4114 vm_page_test_dirty(vm_page_t m)
4117 VM_OBJECT_ASSERT_WLOCKED(m->object);
4118 if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
4123 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
4126 mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
4130 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
4133 mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
4137 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
4140 return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
4143 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
4145 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
4148 vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
4152 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
4155 mtx_assert_(vm_page_lockptr(m), a, file, line);
4161 vm_page_object_lock_assert(vm_page_t m)
4165 * Certain of the page's fields may only be modified by the
4166 * holder of the containing object's lock or the exclusive busy.
4167 * holder. Unfortunately, the holder of the write busy is
4168 * not recorded, and thus cannot be checked here.
4170 if (m->object != NULL && !vm_page_xbusied(m))
4171 VM_OBJECT_ASSERT_WLOCKED(m->object);
4175 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
4178 if ((bits & PGA_WRITEABLE) == 0)
4182 * The PGA_WRITEABLE flag can only be set if the page is
4183 * managed, is exclusively busied or the object is locked.
4184 * Currently, this flag is only set by pmap_enter().
4186 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
4187 ("PGA_WRITEABLE on unmanaged page"));
4188 if (!vm_page_xbusied(m))
4189 VM_OBJECT_ASSERT_LOCKED(m->object);
4193 #include "opt_ddb.h"
4195 #include <sys/kernel.h>
4197 #include <ddb/ddb.h>
4199 DB_SHOW_COMMAND(page, vm_page_print_page_info)
4202 db_printf("vm_cnt.v_free_count: %d\n", vm_free_count());
4203 db_printf("vm_cnt.v_inactive_count: %d\n", vm_inactive_count());
4204 db_printf("vm_cnt.v_active_count: %d\n", vm_active_count());
4205 db_printf("vm_cnt.v_laundry_count: %d\n", vm_laundry_count());
4206 db_printf("vm_cnt.v_wire_count: %d\n", vm_wire_count());
4207 db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
4208 db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
4209 db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
4210 db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
4213 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
4217 db_printf("pq_free %d\n", vm_free_count());
4218 for (dom = 0; dom < vm_ndomains; dom++) {
4220 "dom %d page_cnt %d free %d pq_act %d pq_inact %d pq_laund %d pq_unsw %d\n",
4222 vm_dom[dom].vmd_page_count,
4223 vm_dom[dom].vmd_free_count,
4224 vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
4225 vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
4226 vm_dom[dom].vmd_pagequeues[PQ_LAUNDRY].pq_cnt,
4227 vm_dom[dom].vmd_pagequeues[PQ_UNSWAPPABLE].pq_cnt);
4231 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
4237 db_printf("show pginfo addr\n");
4241 phys = strchr(modif, 'p') != NULL;
4243 m = PHYS_TO_VM_PAGE(addr);
4245 m = (vm_page_t)addr;
4247 "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
4248 " af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
4249 m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
4250 m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
4251 m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);