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 * Resident memory management module.
69 #include <sys/cdefs.h>
70 __FBSDID("$FreeBSD$");
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/counter.h>
77 #include <sys/domainset.h>
78 #include <sys/kernel.h>
79 #include <sys/limits.h>
80 #include <sys/linker.h>
82 #include <sys/malloc.h>
84 #include <sys/msgbuf.h>
85 #include <sys/mutex.h>
87 #include <sys/rwlock.h>
88 #include <sys/sleepqueue.h>
90 #include <sys/sched.h>
92 #include <sys/sysctl.h>
93 #include <sys/vmmeter.h>
94 #include <sys/vnode.h>
98 #include <vm/vm_param.h>
99 #include <vm/vm_domainset.h>
100 #include <vm/vm_kern.h>
101 #include <vm/vm_map.h>
102 #include <vm/vm_object.h>
103 #include <vm/vm_page.h>
104 #include <vm/vm_pageout.h>
105 #include <vm/vm_phys.h>
106 #include <vm/vm_pagequeue.h>
107 #include <vm/vm_pager.h>
108 #include <vm/vm_radix.h>
109 #include <vm/vm_reserv.h>
110 #include <vm/vm_extern.h>
112 #include <vm/uma_int.h>
114 #include <machine/md_var.h>
116 extern int uma_startup_count(int);
117 extern void uma_startup(void *, int);
118 extern int vmem_startup_count(void);
120 struct vm_domain vm_dom[MAXMEMDOM];
122 DPCPU_DEFINE_STATIC(struct vm_batchqueue, pqbatch[MAXMEMDOM][PQ_COUNT]);
124 struct mtx_padalign __exclusive_cache_line pa_lock[PA_LOCK_COUNT];
126 struct mtx_padalign __exclusive_cache_line vm_domainset_lock;
127 /* The following fields are protected by the domainset lock. */
128 domainset_t __exclusive_cache_line vm_min_domains;
129 domainset_t __exclusive_cache_line vm_severe_domains;
130 static int vm_min_waiters;
131 static int vm_severe_waiters;
132 static int vm_pageproc_waiters;
134 static SYSCTL_NODE(_vm_stats, OID_AUTO, page, CTLFLAG_RD, 0,
135 "VM page statistics");
137 static counter_u64_t queue_ops = EARLY_COUNTER;
138 SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, queue_ops,
139 CTLFLAG_RD, &queue_ops,
140 "Number of batched queue operations");
142 static counter_u64_t queue_nops = EARLY_COUNTER;
143 SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, queue_nops,
144 CTLFLAG_RD, &queue_nops,
145 "Number of batched queue operations with no effects");
148 counter_startup(void)
151 queue_ops = counter_u64_alloc(M_WAITOK);
152 queue_nops = counter_u64_alloc(M_WAITOK);
154 SYSINIT(page_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL);
157 * bogus page -- for I/O to/from partially complete buffers,
158 * or for paging into sparsely invalid regions.
160 vm_page_t bogus_page;
162 vm_page_t vm_page_array;
163 long vm_page_array_size;
166 static int boot_pages;
167 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
169 "number of pages allocated for bootstrapping the VM system");
171 static int pa_tryrelock_restart;
172 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
173 &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
175 static TAILQ_HEAD(, vm_page) blacklist_head;
176 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
177 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
178 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
180 static uma_zone_t fakepg_zone;
182 static void vm_page_alloc_check(vm_page_t m);
183 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
184 static void vm_page_dequeue_complete(vm_page_t m);
185 static void vm_page_enqueue(vm_page_t m, uint8_t queue);
186 static void vm_page_init(void *dummy);
187 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
188 vm_pindex_t pindex, vm_page_t mpred);
189 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
191 static void vm_page_mvqueue(vm_page_t m, uint8_t queue);
192 static int vm_page_reclaim_run(int req_class, int domain, u_long npages,
193 vm_page_t m_run, vm_paddr_t high);
194 static int vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object,
196 static int vm_page_zone_import(void *arg, void **store, int cnt, int domain,
198 static void vm_page_zone_release(void *arg, void **store, int cnt);
200 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init, NULL);
203 vm_page_init(void *dummy)
206 fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
207 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
208 bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ |
209 VM_ALLOC_NORMAL | VM_ALLOC_WIRED);
213 * The cache page zone is initialized later since we need to be able to allocate
214 * pages before UMA is fully initialized.
217 vm_page_init_cache_zones(void *dummy __unused)
219 struct vm_domain *vmd;
220 struct vm_pgcache *pgcache;
223 for (domain = 0; domain < vm_ndomains; domain++) {
224 vmd = VM_DOMAIN(domain);
227 * Don't allow the page caches to take up more than .25% of
230 if (vmd->vmd_page_count / 400 < 256 * mp_ncpus * VM_NFREEPOOL)
232 for (pool = 0; pool < VM_NFREEPOOL; pool++) {
233 pgcache = &vmd->vmd_pgcache[pool];
234 pgcache->domain = domain;
235 pgcache->pool = pool;
236 pgcache->zone = uma_zcache_create("vm pgcache",
237 sizeof(struct vm_page), NULL, NULL, NULL, NULL,
238 vm_page_zone_import, vm_page_zone_release, pgcache,
239 UMA_ZONE_MAXBUCKET | UMA_ZONE_VM);
240 (void)uma_zone_set_maxcache(pgcache->zone, 0);
244 SYSINIT(vm_page2, SI_SUB_VM_CONF, SI_ORDER_ANY, vm_page_init_cache_zones, NULL);
246 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
247 #if PAGE_SIZE == 32768
249 CTASSERT(sizeof(u_long) >= 8);
254 * Try to acquire a physical address lock while a pmap is locked. If we
255 * fail to trylock we unlock and lock the pmap directly and cache the
256 * locked pa in *locked. The caller should then restart their loop in case
257 * the virtual to physical mapping has changed.
260 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
267 PA_LOCK_ASSERT(lockpa, MA_OWNED);
268 if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
275 atomic_add_int(&pa_tryrelock_restart, 1);
284 * Sets the page size, perhaps based upon the memory
285 * size. Must be called before any use of page-size
286 * dependent functions.
289 vm_set_page_size(void)
291 if (vm_cnt.v_page_size == 0)
292 vm_cnt.v_page_size = PAGE_SIZE;
293 if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
294 panic("vm_set_page_size: page size not a power of two");
298 * vm_page_blacklist_next:
300 * Find the next entry in the provided string of blacklist
301 * addresses. Entries are separated by space, comma, or newline.
302 * If an invalid integer is encountered then the rest of the
303 * string is skipped. Updates the list pointer to the next
304 * character, or NULL if the string is exhausted or invalid.
307 vm_page_blacklist_next(char **list, char *end)
312 if (list == NULL || *list == NULL)
320 * If there's no end pointer then the buffer is coming from
321 * the kenv and we know it's null-terminated.
324 end = *list + strlen(*list);
326 /* Ensure that strtoq() won't walk off the end */
328 if (*end == '\n' || *end == ' ' || *end == ',')
331 printf("Blacklist not terminated, skipping\n");
337 for (pos = *list; *pos != '\0'; pos = cp) {
338 bad = strtoq(pos, &cp, 0);
339 if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
348 if (*cp == '\0' || ++cp >= end)
352 return (trunc_page(bad));
354 printf("Garbage in RAM blacklist, skipping\n");
360 vm_page_blacklist_add(vm_paddr_t pa, bool verbose)
362 struct vm_domain *vmd;
366 m = vm_phys_paddr_to_vm_page(pa);
368 return (true); /* page does not exist, no failure */
370 vmd = vm_pagequeue_domain(m);
371 vm_domain_free_lock(vmd);
372 ret = vm_phys_unfree_page(m);
373 vm_domain_free_unlock(vmd);
375 vm_domain_freecnt_inc(vmd, -1);
376 TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
378 printf("Skipping page with pa 0x%jx\n", (uintmax_t)pa);
384 * vm_page_blacklist_check:
386 * Iterate through the provided string of blacklist addresses, pulling
387 * each entry out of the physical allocator free list and putting it
388 * onto a list for reporting via the vm.page_blacklist sysctl.
391 vm_page_blacklist_check(char *list, char *end)
397 while (next != NULL) {
398 if ((pa = vm_page_blacklist_next(&next, end)) == 0)
400 vm_page_blacklist_add(pa, bootverbose);
405 * vm_page_blacklist_load:
407 * Search for a special module named "ram_blacklist". It'll be a
408 * plain text file provided by the user via the loader directive
412 vm_page_blacklist_load(char **list, char **end)
421 mod = preload_search_by_type("ram_blacklist");
423 ptr = preload_fetch_addr(mod);
424 len = preload_fetch_size(mod);
435 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
442 error = sysctl_wire_old_buffer(req, 0);
445 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
446 TAILQ_FOREACH(m, &blacklist_head, listq) {
447 sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
448 (uintmax_t)m->phys_addr);
451 error = sbuf_finish(&sbuf);
457 * Initialize a dummy page for use in scans of the specified paging queue.
458 * In principle, this function only needs to set the flag PG_MARKER.
459 * Nonetheless, it write busies the page as a safety precaution.
462 vm_page_init_marker(vm_page_t marker, int queue, uint8_t aflags)
465 bzero(marker, sizeof(*marker));
466 marker->flags = PG_MARKER;
467 marker->aflags = aflags;
468 marker->busy_lock = VPB_SINGLE_EXCLUSIVER;
469 marker->queue = queue;
473 vm_page_domain_init(int domain)
475 struct vm_domain *vmd;
476 struct vm_pagequeue *pq;
479 vmd = VM_DOMAIN(domain);
480 bzero(vmd, sizeof(*vmd));
481 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
482 "vm inactive pagequeue";
483 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
484 "vm active pagequeue";
485 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_name) =
486 "vm laundry pagequeue";
487 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_name) =
488 "vm unswappable pagequeue";
489 vmd->vmd_domain = domain;
490 vmd->vmd_page_count = 0;
491 vmd->vmd_free_count = 0;
493 vmd->vmd_oom = FALSE;
494 for (i = 0; i < PQ_COUNT; i++) {
495 pq = &vmd->vmd_pagequeues[i];
496 TAILQ_INIT(&pq->pq_pl);
497 mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
498 MTX_DEF | MTX_DUPOK);
500 vm_page_init_marker(&vmd->vmd_markers[i], i, 0);
502 mtx_init(&vmd->vmd_free_mtx, "vm page free queue", NULL, MTX_DEF);
503 mtx_init(&vmd->vmd_pageout_mtx, "vm pageout lock", NULL, MTX_DEF);
504 snprintf(vmd->vmd_name, sizeof(vmd->vmd_name), "%d", domain);
507 * inacthead is used to provide FIFO ordering for LRU-bypassing
510 vm_page_init_marker(&vmd->vmd_inacthead, PQ_INACTIVE, PGA_ENQUEUED);
511 TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_INACTIVE].pq_pl,
512 &vmd->vmd_inacthead, plinks.q);
515 * The clock pages are used to implement active queue scanning without
516 * requeues. Scans start at clock[0], which is advanced after the scan
517 * ends. When the two clock hands meet, they are reset and scanning
518 * resumes from the head of the queue.
520 vm_page_init_marker(&vmd->vmd_clock[0], PQ_ACTIVE, PGA_ENQUEUED);
521 vm_page_init_marker(&vmd->vmd_clock[1], PQ_ACTIVE, PGA_ENQUEUED);
522 TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl,
523 &vmd->vmd_clock[0], plinks.q);
524 TAILQ_INSERT_TAIL(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl,
525 &vmd->vmd_clock[1], plinks.q);
529 * Initialize a physical page in preparation for adding it to the free
533 vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind)
538 m->busy_lock = VPB_UNBUSIED;
539 m->flags = m->aflags = 0;
544 m->order = VM_NFREEORDER;
545 m->pool = VM_FREEPOOL_DEFAULT;
546 m->valid = m->dirty = 0;
550 #ifndef PMAP_HAS_PAGE_ARRAY
552 vm_page_array_alloc(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t page_range)
557 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
558 * However, because this page is allocated from KVM, out-of-bounds
559 * accesses using the direct map will not be trapped.
564 * Allocate physical memory for the page structures, and map it.
566 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
567 vm_page_array = (vm_page_t)pmap_map(vaddr, new_end, end,
568 VM_PROT_READ | VM_PROT_WRITE);
569 vm_page_array_size = page_range;
578 * Initializes the resident memory module. Allocates physical memory for
579 * bootstrapping UMA and some data structures that are used to manage
580 * physical pages. Initializes these structures, and populates the free
584 vm_page_startup(vm_offset_t vaddr)
586 struct vm_phys_seg *seg;
588 char *list, *listend;
590 vm_paddr_t end, high_avail, low_avail, new_end, page_range, size;
591 vm_paddr_t last_pa, pa;
593 int biggestone, i, segind;
597 #if defined(__i386__) && defined(VM_PHYSSEG_DENSE)
601 vaddr = round_page(vaddr);
603 vm_phys_early_startup();
604 biggestone = vm_phys_avail_largest();
605 end = phys_avail[biggestone+1];
608 * Initialize the page and queue locks.
610 mtx_init(&vm_domainset_lock, "vm domainset lock", NULL, MTX_DEF);
611 for (i = 0; i < PA_LOCK_COUNT; i++)
612 mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
613 for (i = 0; i < vm_ndomains; i++)
614 vm_page_domain_init(i);
617 * Allocate memory for use when boot strapping the kernel memory
618 * allocator. Tell UMA how many zones we are going to create
619 * before going fully functional. UMA will add its zones.
621 * VM startup zones: vmem, vmem_btag, VM OBJECT, RADIX NODE, MAP,
622 * KMAP ENTRY, MAP ENTRY, VMSPACE.
624 boot_pages = uma_startup_count(8);
626 #ifndef UMA_MD_SMALL_ALLOC
627 /* vmem_startup() calls uma_prealloc(). */
628 boot_pages += vmem_startup_count();
629 /* vm_map_startup() calls uma_prealloc(). */
630 boot_pages += howmany(MAX_KMAP,
631 UMA_SLAB_SPACE / sizeof(struct vm_map));
634 * Before going fully functional kmem_init() does allocation
635 * from "KMAP ENTRY" and vmem_create() does allocation from "vmem".
640 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
641 * manually fetch the value.
643 TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
644 new_end = end - (boot_pages * UMA_SLAB_SIZE);
645 new_end = trunc_page(new_end);
646 mapped = pmap_map(&vaddr, new_end, end,
647 VM_PROT_READ | VM_PROT_WRITE);
648 bzero((void *)mapped, end - new_end);
649 uma_startup((void *)mapped, boot_pages);
652 witness_size = round_page(witness_startup_count());
653 new_end -= witness_size;
654 mapped = pmap_map(&vaddr, new_end, new_end + witness_size,
655 VM_PROT_READ | VM_PROT_WRITE);
656 bzero((void *)mapped, witness_size);
657 witness_startup((void *)mapped);
660 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
661 defined(__i386__) || defined(__mips__) || defined(__riscv) || \
662 defined(__powerpc64__)
664 * Allocate a bitmap to indicate that a random physical page
665 * needs to be included in a minidump.
667 * The amd64 port needs this to indicate which direct map pages
668 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
670 * However, i386 still needs this workspace internally within the
671 * minidump code. In theory, they are not needed on i386, but are
672 * included should the sf_buf code decide to use them.
675 for (i = 0; dump_avail[i + 1] != 0; i += 2)
676 if (dump_avail[i + 1] > last_pa)
677 last_pa = dump_avail[i + 1];
678 page_range = last_pa / PAGE_SIZE;
679 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
680 new_end -= vm_page_dump_size;
681 vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
682 new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
683 bzero((void *)vm_page_dump, vm_page_dump_size);
687 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \
688 defined(__riscv) || defined(__powerpc64__)
690 * Include the UMA bootstrap pages, witness pages and vm_page_dump
691 * in a crash dump. When pmap_map() uses the direct map, they are
692 * not automatically included.
694 for (pa = new_end; pa < end; pa += PAGE_SIZE)
697 phys_avail[biggestone + 1] = new_end;
700 * Request that the physical pages underlying the message buffer be
701 * included in a crash dump. Since the message buffer is accessed
702 * through the direct map, they are not automatically included.
704 pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
705 last_pa = pa + round_page(msgbufsize);
706 while (pa < last_pa) {
712 * Compute the number of pages of memory that will be available for
713 * use, taking into account the overhead of a page structure per page.
714 * In other words, solve
715 * "available physical memory" - round_page(page_range *
716 * sizeof(struct vm_page)) = page_range * PAGE_SIZE
719 low_avail = phys_avail[0];
720 high_avail = phys_avail[1];
721 for (i = 0; i < vm_phys_nsegs; i++) {
722 if (vm_phys_segs[i].start < low_avail)
723 low_avail = vm_phys_segs[i].start;
724 if (vm_phys_segs[i].end > high_avail)
725 high_avail = vm_phys_segs[i].end;
727 /* Skip the first chunk. It is already accounted for. */
728 for (i = 2; phys_avail[i + 1] != 0; i += 2) {
729 if (phys_avail[i] < low_avail)
730 low_avail = phys_avail[i];
731 if (phys_avail[i + 1] > high_avail)
732 high_avail = phys_avail[i + 1];
734 first_page = low_avail / PAGE_SIZE;
735 #ifdef VM_PHYSSEG_SPARSE
737 for (i = 0; i < vm_phys_nsegs; i++)
738 size += vm_phys_segs[i].end - vm_phys_segs[i].start;
739 for (i = 0; phys_avail[i + 1] != 0; i += 2)
740 size += phys_avail[i + 1] - phys_avail[i];
741 #elif defined(VM_PHYSSEG_DENSE)
742 size = high_avail - low_avail;
744 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
747 #ifdef PMAP_HAS_PAGE_ARRAY
748 pmap_page_array_startup(size / PAGE_SIZE);
749 biggestone = vm_phys_avail_largest();
750 end = new_end = phys_avail[biggestone + 1];
752 #ifdef VM_PHYSSEG_DENSE
754 * In the VM_PHYSSEG_DENSE case, the number of pages can account for
755 * the overhead of a page structure per page only if vm_page_array is
756 * allocated from the last physical memory chunk. Otherwise, we must
757 * allocate page structures representing the physical memory
758 * underlying vm_page_array, even though they will not be used.
760 if (new_end != high_avail)
761 page_range = size / PAGE_SIZE;
765 page_range = size / (PAGE_SIZE + sizeof(struct vm_page));
768 * If the partial bytes remaining are large enough for
769 * a page (PAGE_SIZE) without a corresponding
770 * 'struct vm_page', then new_end will contain an
771 * extra page after subtracting the length of the VM
772 * page array. Compensate by subtracting an extra
775 if (size % (PAGE_SIZE + sizeof(struct vm_page)) >= PAGE_SIZE) {
776 if (new_end == high_avail)
777 high_avail -= PAGE_SIZE;
778 new_end -= PAGE_SIZE;
782 new_end = vm_page_array_alloc(&vaddr, end, page_range);
785 #if VM_NRESERVLEVEL > 0
787 * Allocate physical memory for the reservation management system's
788 * data structures, and map it.
790 new_end = vm_reserv_startup(&vaddr, new_end);
792 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \
793 defined(__riscv) || defined(__powerpc64__)
795 * Include vm_page_array and vm_reserv_array in a crash dump.
797 for (pa = new_end; pa < end; pa += PAGE_SIZE)
800 phys_avail[biggestone + 1] = new_end;
803 * Add physical memory segments corresponding to the available
806 for (i = 0; phys_avail[i + 1] != 0; i += 2)
807 if (vm_phys_avail_size(i) != 0)
808 vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
811 * Initialize the physical memory allocator.
816 * Initialize the page structures and add every available page to the
817 * physical memory allocator's free lists.
819 #if defined(__i386__) && defined(VM_PHYSSEG_DENSE)
820 for (ii = 0; ii < vm_page_array_size; ii++) {
821 m = &vm_page_array[ii];
822 vm_page_init_page(m, (first_page + ii) << PAGE_SHIFT, 0);
823 m->flags = PG_FICTITIOUS;
826 vm_cnt.v_page_count = 0;
827 for (segind = 0; segind < vm_phys_nsegs; segind++) {
828 seg = &vm_phys_segs[segind];
829 for (m = seg->first_page, pa = seg->start; pa < seg->end;
830 m++, pa += PAGE_SIZE)
831 vm_page_init_page(m, pa, segind);
834 * Add the segment to the free lists only if it is covered by
835 * one of the ranges in phys_avail. Because we've added the
836 * ranges to the vm_phys_segs array, we can assume that each
837 * segment is either entirely contained in one of the ranges,
838 * or doesn't overlap any of them.
840 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
841 struct vm_domain *vmd;
843 if (seg->start < phys_avail[i] ||
844 seg->end > phys_avail[i + 1])
848 pagecount = (u_long)atop(seg->end - seg->start);
850 vmd = VM_DOMAIN(seg->domain);
851 vm_domain_free_lock(vmd);
852 vm_phys_enqueue_contig(m, pagecount);
853 vm_domain_free_unlock(vmd);
854 vm_domain_freecnt_inc(vmd, pagecount);
855 vm_cnt.v_page_count += (u_int)pagecount;
857 vmd = VM_DOMAIN(seg->domain);
858 vmd->vmd_page_count += (u_int)pagecount;
859 vmd->vmd_segs |= 1UL << m->segind;
865 * Remove blacklisted pages from the physical memory allocator.
867 TAILQ_INIT(&blacklist_head);
868 vm_page_blacklist_load(&list, &listend);
869 vm_page_blacklist_check(list, listend);
871 list = kern_getenv("vm.blacklist");
872 vm_page_blacklist_check(list, NULL);
875 #if VM_NRESERVLEVEL > 0
877 * Initialize the reservation management system.
886 vm_page_reference(vm_page_t m)
889 vm_page_aflag_set(m, PGA_REFERENCED);
893 * vm_page_busy_acquire:
895 * Acquire the busy lock as described by VM_ALLOC_* flags. Will loop
896 * and drop the object lock if necessary.
899 vm_page_busy_acquire(vm_page_t m, int allocflags)
906 * The page-specific object must be cached because page
907 * identity can change during the sleep, causing the
908 * re-lock of a different object.
909 * It is assumed that a reference to the object is already
910 * held by the callers.
914 if ((allocflags & VM_ALLOC_SBUSY) == 0) {
915 if (vm_page_tryxbusy(m))
918 if (vm_page_trysbusy(m))
921 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
924 locked = VM_OBJECT_WOWNED(obj);
926 MPASS(vm_page_wired(m));
931 if (x == VPB_UNBUSIED ||
932 ((allocflags & VM_ALLOC_SBUSY) != 0 &&
933 (x & VPB_BIT_SHARED) != 0) ||
934 ((x & VPB_BIT_WAITERS) == 0 &&
935 !atomic_cmpset_int(&m->busy_lock, x,
936 x | VPB_BIT_WAITERS))) {
941 VM_OBJECT_WUNLOCK(obj);
942 sleepq_add(m, NULL, "vmpba", 0, 0);
945 VM_OBJECT_WLOCK(obj);
946 MPASS(m->object == obj || m->object == NULL);
947 if ((allocflags & VM_ALLOC_WAITFAIL) != 0)
953 * vm_page_busy_downgrade:
955 * Downgrade an exclusive busy page into a single shared busy page.
958 vm_page_busy_downgrade(vm_page_t m)
962 vm_page_assert_xbusied(m);
966 if (atomic_fcmpset_rel_int(&m->busy_lock,
967 &x, VPB_SHARERS_WORD(1)))
970 if ((x & VPB_BIT_WAITERS) != 0)
976 * vm_page_busy_tryupgrade:
978 * Attempt to upgrade a single shared busy into an exclusive busy.
981 vm_page_busy_tryupgrade(vm_page_t m)
985 vm_page_assert_sbusied(m);
989 if (VPB_SHARERS(x) > 1)
991 KASSERT((x & ~VPB_BIT_WAITERS) == VPB_SHARERS_WORD(1),
992 ("vm_page_busy_tryupgrade: invalid lock state"));
993 if (!atomic_fcmpset_acq_int(&m->busy_lock, &x,
994 VPB_SINGLE_EXCLUSIVER | (x & VPB_BIT_WAITERS)))
1003 * Return a positive value if the page is shared busied, 0 otherwise.
1006 vm_page_sbusied(vm_page_t m)
1011 return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
1017 * Shared unbusy a page.
1020 vm_page_sunbusy(vm_page_t m)
1024 vm_page_assert_sbusied(m);
1028 if (VPB_SHARERS(x) > 1) {
1029 if (atomic_fcmpset_int(&m->busy_lock, &x,
1030 x - VPB_ONE_SHARER))
1034 KASSERT((x & ~VPB_BIT_WAITERS) == VPB_SHARERS_WORD(1),
1035 ("vm_page_sunbusy: invalid lock state"));
1036 if (!atomic_fcmpset_rel_int(&m->busy_lock, &x, VPB_UNBUSIED))
1038 if ((x & VPB_BIT_WAITERS) == 0)
1046 * vm_page_busy_sleep:
1048 * Sleep if the page is busy, using the page pointer as wchan.
1049 * This is used to implement the hard-path of busying mechanism.
1051 * If nonshared is true, sleep only if the page is xbusy.
1053 * The object lock must be held on entry and will be released on exit.
1056 vm_page_busy_sleep(vm_page_t m, const char *wmesg, bool nonshared)
1062 vm_page_lock_assert(m, MA_NOTOWNED);
1063 VM_OBJECT_ASSERT_LOCKED(obj);
1067 if (x == VPB_UNBUSIED || (nonshared && (x & VPB_BIT_SHARED) != 0) ||
1068 ((x & VPB_BIT_WAITERS) == 0 &&
1069 !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS))) {
1070 VM_OBJECT_DROP(obj);
1074 VM_OBJECT_DROP(obj);
1075 sleepq_add(m, NULL, wmesg, 0, 0);
1076 sleepq_wait(m, PVM);
1082 * Try to shared busy a page.
1083 * If the operation succeeds 1 is returned otherwise 0.
1084 * The operation never sleeps.
1087 vm_page_trysbusy(vm_page_t m)
1093 if ((x & VPB_BIT_SHARED) == 0)
1095 if (atomic_fcmpset_acq_int(&m->busy_lock, &x,
1096 x + VPB_ONE_SHARER))
1102 * vm_page_xunbusy_hard:
1104 * Called when unbusy has failed because there is a waiter.
1107 vm_page_xunbusy_hard(vm_page_t m)
1110 vm_page_assert_xbusied(m);
1115 atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
1120 * Avoid releasing and reacquiring the same page lock.
1123 vm_page_change_lock(vm_page_t m, struct mtx **mtx)
1127 mtx1 = vm_page_lockptr(m);
1137 * vm_page_unhold_pages:
1139 * Unhold each of the pages that is referenced by the given array.
1142 vm_page_unhold_pages(vm_page_t *ma, int count)
1145 for (; count != 0; count--) {
1146 vm_page_unwire(*ma, PQ_ACTIVE);
1152 PHYS_TO_VM_PAGE(vm_paddr_t pa)
1156 #ifdef VM_PHYSSEG_SPARSE
1157 m = vm_phys_paddr_to_vm_page(pa);
1159 m = vm_phys_fictitious_to_vm_page(pa);
1161 #elif defined(VM_PHYSSEG_DENSE)
1165 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
1166 m = &vm_page_array[pi - first_page];
1169 return (vm_phys_fictitious_to_vm_page(pa));
1171 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
1178 * Create a fictitious page with the specified physical address and
1179 * memory attribute. The memory attribute is the only the machine-
1180 * dependent aspect of a fictitious page that must be initialized.
1183 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
1187 m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
1188 vm_page_initfake(m, paddr, memattr);
1193 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1196 if ((m->flags & PG_FICTITIOUS) != 0) {
1198 * The page's memattr might have changed since the
1199 * previous initialization. Update the pmap to the
1204 m->phys_addr = paddr;
1206 /* Fictitious pages don't use "segind". */
1207 m->flags = PG_FICTITIOUS;
1208 /* Fictitious pages don't use "order" or "pool". */
1209 m->oflags = VPO_UNMANAGED;
1210 m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1211 /* Fictitious pages are unevictable. */
1215 pmap_page_set_memattr(m, memattr);
1221 * Release a fictitious page.
1224 vm_page_putfake(vm_page_t m)
1227 KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
1228 KASSERT((m->flags & PG_FICTITIOUS) != 0,
1229 ("vm_page_putfake: bad page %p", m));
1230 if (vm_page_xbusied(m))
1232 uma_zfree(fakepg_zone, m);
1236 * vm_page_updatefake:
1238 * Update the given fictitious page to the specified physical address and
1242 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1245 KASSERT((m->flags & PG_FICTITIOUS) != 0,
1246 ("vm_page_updatefake: bad page %p", m));
1247 m->phys_addr = paddr;
1248 pmap_page_set_memattr(m, memattr);
1257 vm_page_free(vm_page_t m)
1260 m->flags &= ~PG_ZERO;
1261 vm_page_free_toq(m);
1265 * vm_page_free_zero:
1267 * Free a page to the zerod-pages queue
1270 vm_page_free_zero(vm_page_t m)
1273 m->flags |= PG_ZERO;
1274 vm_page_free_toq(m);
1278 * Unbusy and handle the page queueing for a page from a getpages request that
1279 * was optionally read ahead or behind.
1282 vm_page_readahead_finish(vm_page_t m)
1285 /* We shouldn't put invalid pages on queues. */
1286 KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m));
1289 * Since the page is not the actually needed one, whether it should
1290 * be activated or deactivated is not obvious. Empirical results
1291 * have shown that deactivating the page is usually the best choice,
1292 * unless the page is wanted by another thread.
1295 if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1296 vm_page_activate(m);
1298 vm_page_deactivate(m);
1304 * vm_page_sleep_if_busy:
1306 * Sleep and release the object lock if the page is busied.
1307 * Returns TRUE if the thread slept.
1309 * The given page must be unlocked and object containing it must
1313 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1317 vm_page_lock_assert(m, MA_NOTOWNED);
1318 VM_OBJECT_ASSERT_WLOCKED(m->object);
1320 if (vm_page_busied(m)) {
1322 * The page-specific object must be cached because page
1323 * identity can change during the sleep, causing the
1324 * re-lock of a different object.
1325 * It is assumed that a reference to the object is already
1326 * held by the callers.
1329 vm_page_busy_sleep(m, msg, false);
1330 VM_OBJECT_WLOCK(obj);
1337 * vm_page_sleep_if_xbusy:
1339 * Sleep and release the object lock if the page is xbusied.
1340 * Returns TRUE if the thread slept.
1342 * The given page must be unlocked and object containing it must
1346 vm_page_sleep_if_xbusy(vm_page_t m, const char *msg)
1350 vm_page_lock_assert(m, MA_NOTOWNED);
1351 VM_OBJECT_ASSERT_WLOCKED(m->object);
1353 if (vm_page_xbusied(m)) {
1355 * The page-specific object must be cached because page
1356 * identity can change during the sleep, causing the
1357 * re-lock of a different object.
1358 * It is assumed that a reference to the object is already
1359 * held by the callers.
1362 vm_page_busy_sleep(m, msg, true);
1363 VM_OBJECT_WLOCK(obj);
1370 * vm_page_dirty_KBI: [ internal use only ]
1372 * Set all bits in the page's dirty field.
1374 * The object containing the specified page must be locked if the
1375 * call is made from the machine-independent layer.
1377 * See vm_page_clear_dirty_mask().
1379 * This function should only be called by vm_page_dirty().
1382 vm_page_dirty_KBI(vm_page_t m)
1385 /* Refer to this operation by its public name. */
1386 KASSERT(m->valid == VM_PAGE_BITS_ALL,
1387 ("vm_page_dirty: page is invalid!"));
1388 m->dirty = VM_PAGE_BITS_ALL;
1392 * vm_page_insert: [ internal use only ]
1394 * Inserts the given mem entry into the object and object list.
1396 * The object must be locked.
1399 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1403 VM_OBJECT_ASSERT_WLOCKED(object);
1404 mpred = vm_radix_lookup_le(&object->rtree, pindex);
1405 return (vm_page_insert_after(m, object, pindex, mpred));
1409 * vm_page_insert_after:
1411 * Inserts the page "m" into the specified object at offset "pindex".
1413 * The page "mpred" must immediately precede the offset "pindex" within
1414 * the specified object.
1416 * The object must be locked.
1419 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1424 VM_OBJECT_ASSERT_WLOCKED(object);
1425 KASSERT(m->object == NULL,
1426 ("vm_page_insert_after: page already inserted"));
1427 if (mpred != NULL) {
1428 KASSERT(mpred->object == object,
1429 ("vm_page_insert_after: object doesn't contain mpred"));
1430 KASSERT(mpred->pindex < pindex,
1431 ("vm_page_insert_after: mpred doesn't precede pindex"));
1432 msucc = TAILQ_NEXT(mpred, listq);
1434 msucc = TAILQ_FIRST(&object->memq);
1436 KASSERT(msucc->pindex > pindex,
1437 ("vm_page_insert_after: msucc doesn't succeed pindex"));
1440 * Record the object/offset pair in this page.
1444 m->ref_count |= VPRC_OBJREF;
1447 * Now link into the object's ordered list of backed pages.
1449 if (vm_radix_insert(&object->rtree, m)) {
1452 m->ref_count &= ~VPRC_OBJREF;
1455 vm_page_insert_radixdone(m, object, mpred);
1460 * vm_page_insert_radixdone:
1462 * Complete page "m" insertion into the specified object after the
1463 * radix trie hooking.
1465 * The page "mpred" must precede the offset "m->pindex" within the
1468 * The object must be locked.
1471 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1474 VM_OBJECT_ASSERT_WLOCKED(object);
1475 KASSERT(object != NULL && m->object == object,
1476 ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1477 KASSERT((m->ref_count & VPRC_OBJREF) != 0,
1478 ("vm_page_insert_radixdone: page %p is missing object ref", m));
1479 if (mpred != NULL) {
1480 KASSERT(mpred->object == object,
1481 ("vm_page_insert_radixdone: object doesn't contain mpred"));
1482 KASSERT(mpred->pindex < m->pindex,
1483 ("vm_page_insert_radixdone: mpred doesn't precede pindex"));
1487 TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1489 TAILQ_INSERT_HEAD(&object->memq, m, listq);
1492 * Show that the object has one more resident page.
1494 object->resident_page_count++;
1497 * Hold the vnode until the last page is released.
1499 if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1500 vhold(object->handle);
1503 * Since we are inserting a new and possibly dirty page,
1504 * update the object's OBJ_MIGHTBEDIRTY flag.
1506 if (pmap_page_is_write_mapped(m))
1507 vm_object_set_writeable_dirty(object);
1511 * Do the work to remove a page from its object. The caller is responsible for
1512 * updating the page's fields to reflect this removal.
1515 vm_page_object_remove(vm_page_t m)
1521 VM_OBJECT_ASSERT_WLOCKED(object);
1522 KASSERT((m->ref_count & VPRC_OBJREF) != 0,
1523 ("page %p is missing its object ref", m));
1525 mrem = vm_radix_remove(&object->rtree, m->pindex);
1526 KASSERT(mrem == m, ("removed page %p, expected page %p", mrem, m));
1529 * Now remove from the object's list of backed pages.
1531 TAILQ_REMOVE(&object->memq, m, listq);
1534 * And show that the object has one fewer resident page.
1536 object->resident_page_count--;
1539 * The vnode may now be recycled.
1541 if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1542 vdrop(object->handle);
1548 * Removes the specified page from its containing object, but does not
1549 * invalidate any backing storage. Returns true if the object's reference
1550 * was the last reference to the page, and false otherwise.
1552 * The object must be locked.
1555 vm_page_remove(vm_page_t m)
1558 vm_page_object_remove(m);
1560 return (vm_page_drop(m, VPRC_OBJREF) == VPRC_OBJREF);
1566 * Returns the page associated with the object/offset
1567 * pair specified; if none is found, NULL is returned.
1569 * The object must be locked.
1572 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1575 VM_OBJECT_ASSERT_LOCKED(object);
1576 return (vm_radix_lookup(&object->rtree, pindex));
1580 * vm_page_find_least:
1582 * Returns the page associated with the object with least pindex
1583 * greater than or equal to the parameter pindex, or NULL.
1585 * The object must be locked.
1588 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1592 VM_OBJECT_ASSERT_LOCKED(object);
1593 if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1594 m = vm_radix_lookup_ge(&object->rtree, pindex);
1599 * Returns the given page's successor (by pindex) within the object if it is
1600 * resident; if none is found, NULL is returned.
1602 * The object must be locked.
1605 vm_page_next(vm_page_t m)
1609 VM_OBJECT_ASSERT_LOCKED(m->object);
1610 if ((next = TAILQ_NEXT(m, listq)) != NULL) {
1611 MPASS(next->object == m->object);
1612 if (next->pindex != m->pindex + 1)
1619 * Returns the given page's predecessor (by pindex) within the object if it is
1620 * resident; if none is found, NULL is returned.
1622 * The object must be locked.
1625 vm_page_prev(vm_page_t m)
1629 VM_OBJECT_ASSERT_LOCKED(m->object);
1630 if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL) {
1631 MPASS(prev->object == m->object);
1632 if (prev->pindex != m->pindex - 1)
1639 * Uses the page mnew as a replacement for an existing page at index
1640 * pindex which must be already present in the object.
1643 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1647 VM_OBJECT_ASSERT_WLOCKED(object);
1648 KASSERT(mnew->object == NULL && (mnew->ref_count & VPRC_OBJREF) == 0,
1649 ("vm_page_replace: page %p already in object", mnew));
1652 * This function mostly follows vm_page_insert() and
1653 * vm_page_remove() without the radix, object count and vnode
1654 * dance. Double check such functions for more comments.
1657 mnew->object = object;
1658 mnew->pindex = pindex;
1659 atomic_set_int(&mnew->ref_count, VPRC_OBJREF);
1660 mold = vm_radix_replace(&object->rtree, mnew);
1661 KASSERT(mold->queue == PQ_NONE,
1662 ("vm_page_replace: old page %p is on a paging queue", mold));
1664 /* Keep the resident page list in sorted order. */
1665 TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1666 TAILQ_REMOVE(&object->memq, mold, listq);
1668 mold->object = NULL;
1669 atomic_clear_int(&mold->ref_count, VPRC_OBJREF);
1670 vm_page_xunbusy(mold);
1673 * The object's resident_page_count does not change because we have
1674 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1676 if (pmap_page_is_write_mapped(mnew))
1677 vm_object_set_writeable_dirty(object);
1684 * Move the given memory entry from its
1685 * current object to the specified target object/offset.
1687 * Note: swap associated with the page must be invalidated by the move. We
1688 * have to do this for several reasons: (1) we aren't freeing the
1689 * page, (2) we are dirtying the page, (3) the VM system is probably
1690 * moving the page from object A to B, and will then later move
1691 * the backing store from A to B and we can't have a conflict.
1693 * Note: we *always* dirty the page. It is necessary both for the
1694 * fact that we moved it, and because we may be invalidating
1697 * The objects must be locked.
1700 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1705 VM_OBJECT_ASSERT_WLOCKED(new_object);
1707 KASSERT(m->ref_count != 0, ("vm_page_rename: page %p has no refs", m));
1708 mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1709 KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1710 ("vm_page_rename: pindex already renamed"));
1713 * Create a custom version of vm_page_insert() which does not depend
1714 * by m_prev and can cheat on the implementation aspects of the
1718 m->pindex = new_pindex;
1719 if (vm_radix_insert(&new_object->rtree, m)) {
1725 * The operation cannot fail anymore. The removal must happen before
1726 * the listq iterator is tainted.
1729 vm_page_object_remove(m);
1731 /* Return back to the new pindex to complete vm_page_insert(). */
1732 m->pindex = new_pindex;
1733 m->object = new_object;
1735 vm_page_insert_radixdone(m, new_object, mpred);
1743 * Allocate and return a page that is associated with the specified
1744 * object and offset pair. By default, this page is exclusive busied.
1746 * The caller must always specify an allocation class.
1748 * allocation classes:
1749 * VM_ALLOC_NORMAL normal process request
1750 * VM_ALLOC_SYSTEM system *really* needs a page
1751 * VM_ALLOC_INTERRUPT interrupt time request
1753 * optional allocation flags:
1754 * VM_ALLOC_COUNT(number) the number of additional pages that the caller
1755 * intends to allocate
1756 * VM_ALLOC_NOBUSY do not exclusive busy the page
1757 * VM_ALLOC_NODUMP do not include the page in a kernel core dump
1758 * VM_ALLOC_NOOBJ page is not associated with an object and
1759 * should not be exclusive busy
1760 * VM_ALLOC_SBUSY shared busy the allocated page
1761 * VM_ALLOC_WIRED wire the allocated page
1762 * VM_ALLOC_ZERO prefer a zeroed page
1765 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1768 return (vm_page_alloc_after(object, pindex, req, object != NULL ?
1769 vm_radix_lookup_le(&object->rtree, pindex) : NULL));
1773 vm_page_alloc_domain(vm_object_t object, vm_pindex_t pindex, int domain,
1777 return (vm_page_alloc_domain_after(object, pindex, domain, req,
1778 object != NULL ? vm_radix_lookup_le(&object->rtree, pindex) :
1783 * Allocate a page in the specified object with the given page index. To
1784 * optimize insertion of the page into the object, the caller must also specifiy
1785 * the resident page in the object with largest index smaller than the given
1786 * page index, or NULL if no such page exists.
1789 vm_page_alloc_after(vm_object_t object, vm_pindex_t pindex,
1790 int req, vm_page_t mpred)
1792 struct vm_domainset_iter di;
1796 vm_domainset_iter_page_init(&di, object, pindex, &domain, &req);
1798 m = vm_page_alloc_domain_after(object, pindex, domain, req,
1802 } while (vm_domainset_iter_page(&di, object, &domain) == 0);
1808 * Returns true if the number of free pages exceeds the minimum
1809 * for the request class and false otherwise.
1812 vm_domain_allocate(struct vm_domain *vmd, int req, int npages)
1814 u_int limit, old, new;
1816 req = req & VM_ALLOC_CLASS_MASK;
1819 * The page daemon is allowed to dig deeper into the free page list.
1821 if (curproc == pageproc && req != VM_ALLOC_INTERRUPT)
1822 req = VM_ALLOC_SYSTEM;
1823 if (req == VM_ALLOC_INTERRUPT)
1825 else if (req == VM_ALLOC_SYSTEM)
1826 limit = vmd->vmd_interrupt_free_min;
1828 limit = vmd->vmd_free_reserved;
1831 * Attempt to reserve the pages. Fail if we're below the limit.
1834 old = vmd->vmd_free_count;
1839 } while (atomic_fcmpset_int(&vmd->vmd_free_count, &old, new) == 0);
1841 /* Wake the page daemon if we've crossed the threshold. */
1842 if (vm_paging_needed(vmd, new) && !vm_paging_needed(vmd, old))
1843 pagedaemon_wakeup(vmd->vmd_domain);
1845 /* Only update bitsets on transitions. */
1846 if ((old >= vmd->vmd_free_min && new < vmd->vmd_free_min) ||
1847 (old >= vmd->vmd_free_severe && new < vmd->vmd_free_severe))
1854 vm_page_alloc_domain_after(vm_object_t object, vm_pindex_t pindex, int domain,
1855 int req, vm_page_t mpred)
1857 struct vm_domain *vmd;
1861 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1862 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1863 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1864 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1865 ("inconsistent object(%p)/req(%x)", object, req));
1866 KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0,
1867 ("Can't sleep and retry object insertion."));
1868 KASSERT(mpred == NULL || mpred->pindex < pindex,
1869 ("mpred %p doesn't precede pindex 0x%jx", mpred,
1870 (uintmax_t)pindex));
1872 VM_OBJECT_ASSERT_WLOCKED(object);
1876 pool = object != NULL ? VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT;
1878 #if VM_NRESERVLEVEL > 0
1880 * Can we allocate the page from a reservation?
1882 if (vm_object_reserv(object) &&
1883 (m = vm_reserv_alloc_page(object, pindex, domain, req, mpred)) !=
1885 domain = vm_phys_domain(m);
1886 vmd = VM_DOMAIN(domain);
1890 vmd = VM_DOMAIN(domain);
1891 if (vmd->vmd_pgcache[pool].zone != NULL) {
1892 m = uma_zalloc(vmd->vmd_pgcache[pool].zone, M_NOWAIT);
1894 flags |= PG_PCPU_CACHE;
1898 if (vm_domain_allocate(vmd, req, 1)) {
1900 * If not, allocate it from the free page queues.
1902 vm_domain_free_lock(vmd);
1903 m = vm_phys_alloc_pages(domain, pool, 0);
1904 vm_domain_free_unlock(vmd);
1906 vm_domain_freecnt_inc(vmd, 1);
1907 #if VM_NRESERVLEVEL > 0
1908 if (vm_reserv_reclaim_inactive(domain))
1915 * Not allocatable, give up.
1917 if (vm_domain_alloc_fail(vmd, object, req))
1923 * At this point we had better have found a good page.
1927 vm_page_alloc_check(m);
1930 * Initialize the page. Only the PG_ZERO flag is inherited.
1932 if ((req & VM_ALLOC_ZERO) != 0)
1933 flags |= (m->flags & PG_ZERO);
1934 if ((req & VM_ALLOC_NODUMP) != 0)
1938 m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1940 m->busy_lock = VPB_UNBUSIED;
1941 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1942 m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1943 if ((req & VM_ALLOC_SBUSY) != 0)
1944 m->busy_lock = VPB_SHARERS_WORD(1);
1945 if (req & VM_ALLOC_WIRED) {
1947 * The page lock is not required for wiring a page until that
1948 * page is inserted into the object.
1955 if (object != NULL) {
1956 if (vm_page_insert_after(m, object, pindex, mpred)) {
1957 if (req & VM_ALLOC_WIRED) {
1961 KASSERT(m->object == NULL, ("page %p has object", m));
1962 m->oflags = VPO_UNMANAGED;
1963 m->busy_lock = VPB_UNBUSIED;
1964 /* Don't change PG_ZERO. */
1965 vm_page_free_toq(m);
1966 if (req & VM_ALLOC_WAITFAIL) {
1967 VM_OBJECT_WUNLOCK(object);
1969 VM_OBJECT_WLOCK(object);
1974 /* Ignore device objects; the pager sets "memattr" for them. */
1975 if (object->memattr != VM_MEMATTR_DEFAULT &&
1976 (object->flags & OBJ_FICTITIOUS) == 0)
1977 pmap_page_set_memattr(m, object->memattr);
1985 * vm_page_alloc_contig:
1987 * Allocate a contiguous set of physical pages of the given size "npages"
1988 * from the free lists. All of the physical pages must be at or above
1989 * the given physical address "low" and below the given physical address
1990 * "high". The given value "alignment" determines the alignment of the
1991 * first physical page in the set. If the given value "boundary" is
1992 * non-zero, then the set of physical pages cannot cross any physical
1993 * address boundary that is a multiple of that value. Both "alignment"
1994 * and "boundary" must be a power of two.
1996 * If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1997 * then the memory attribute setting for the physical pages is configured
1998 * to the object's memory attribute setting. Otherwise, the memory
1999 * attribute setting for the physical pages is configured to "memattr",
2000 * overriding the object's memory attribute setting. However, if the
2001 * object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
2002 * memory attribute setting for the physical pages cannot be configured
2003 * to VM_MEMATTR_DEFAULT.
2005 * The specified object may not contain fictitious pages.
2007 * The caller must always specify an allocation class.
2009 * allocation classes:
2010 * VM_ALLOC_NORMAL normal process request
2011 * VM_ALLOC_SYSTEM system *really* needs a page
2012 * VM_ALLOC_INTERRUPT interrupt time request
2014 * optional allocation flags:
2015 * VM_ALLOC_NOBUSY do not exclusive busy the page
2016 * VM_ALLOC_NODUMP do not include the page in a kernel core dump
2017 * VM_ALLOC_NOOBJ page is not associated with an object and
2018 * should not be exclusive busy
2019 * VM_ALLOC_SBUSY shared busy the allocated page
2020 * VM_ALLOC_WIRED wire the allocated page
2021 * VM_ALLOC_ZERO prefer a zeroed page
2024 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
2025 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
2026 vm_paddr_t boundary, vm_memattr_t memattr)
2028 struct vm_domainset_iter di;
2032 vm_domainset_iter_page_init(&di, object, pindex, &domain, &req);
2034 m = vm_page_alloc_contig_domain(object, pindex, domain, req,
2035 npages, low, high, alignment, boundary, memattr);
2038 } while (vm_domainset_iter_page(&di, object, &domain) == 0);
2044 vm_page_alloc_contig_domain(vm_object_t object, vm_pindex_t pindex, int domain,
2045 int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
2046 vm_paddr_t boundary, vm_memattr_t memattr)
2048 struct vm_domain *vmd;
2049 vm_page_t m, m_ret, mpred;
2050 u_int busy_lock, flags, oflags;
2052 mpred = NULL; /* XXX: pacify gcc */
2053 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
2054 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
2055 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
2056 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
2057 ("vm_page_alloc_contig: inconsistent object(%p)/req(%x)", object,
2059 KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0,
2060 ("Can't sleep and retry object insertion."));
2061 if (object != NULL) {
2062 VM_OBJECT_ASSERT_WLOCKED(object);
2063 KASSERT((object->flags & OBJ_FICTITIOUS) == 0,
2064 ("vm_page_alloc_contig: object %p has fictitious pages",
2067 KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
2069 if (object != NULL) {
2070 mpred = vm_radix_lookup_le(&object->rtree, pindex);
2071 KASSERT(mpred == NULL || mpred->pindex != pindex,
2072 ("vm_page_alloc_contig: pindex already allocated"));
2076 * Can we allocate the pages without the number of free pages falling
2077 * below the lower bound for the allocation class?
2081 #if VM_NRESERVLEVEL > 0
2083 * Can we allocate the pages from a reservation?
2085 if (vm_object_reserv(object) &&
2086 (m_ret = vm_reserv_alloc_contig(object, pindex, domain, req,
2087 mpred, npages, low, high, alignment, boundary)) != NULL) {
2088 domain = vm_phys_domain(m_ret);
2089 vmd = VM_DOMAIN(domain);
2093 vmd = VM_DOMAIN(domain);
2094 if (vm_domain_allocate(vmd, req, npages)) {
2096 * allocate them from the free page queues.
2098 vm_domain_free_lock(vmd);
2099 m_ret = vm_phys_alloc_contig(domain, npages, low, high,
2100 alignment, boundary);
2101 vm_domain_free_unlock(vmd);
2102 if (m_ret == NULL) {
2103 vm_domain_freecnt_inc(vmd, npages);
2104 #if VM_NRESERVLEVEL > 0
2105 if (vm_reserv_reclaim_contig(domain, npages, low,
2106 high, alignment, boundary))
2111 if (m_ret == NULL) {
2112 if (vm_domain_alloc_fail(vmd, object, req))
2116 #if VM_NRESERVLEVEL > 0
2119 for (m = m_ret; m < &m_ret[npages]; m++) {
2121 vm_page_alloc_check(m);
2125 * Initialize the pages. Only the PG_ZERO flag is inherited.
2128 if ((req & VM_ALLOC_ZERO) != 0)
2130 if ((req & VM_ALLOC_NODUMP) != 0)
2132 oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
2134 busy_lock = VPB_UNBUSIED;
2135 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
2136 busy_lock = VPB_SINGLE_EXCLUSIVER;
2137 if ((req & VM_ALLOC_SBUSY) != 0)
2138 busy_lock = VPB_SHARERS_WORD(1);
2139 if ((req & VM_ALLOC_WIRED) != 0)
2140 vm_wire_add(npages);
2141 if (object != NULL) {
2142 if (object->memattr != VM_MEMATTR_DEFAULT &&
2143 memattr == VM_MEMATTR_DEFAULT)
2144 memattr = object->memattr;
2146 for (m = m_ret; m < &m_ret[npages]; m++) {
2148 m->flags = (m->flags | PG_NODUMP) & flags;
2149 m->busy_lock = busy_lock;
2150 if ((req & VM_ALLOC_WIRED) != 0)
2154 if (object != NULL) {
2155 if (vm_page_insert_after(m, object, pindex, mpred)) {
2156 if ((req & VM_ALLOC_WIRED) != 0)
2157 vm_wire_sub(npages);
2158 KASSERT(m->object == NULL,
2159 ("page %p has object", m));
2161 for (m = m_ret; m < &m_ret[npages]; m++) {
2163 (req & VM_ALLOC_WIRED) != 0)
2165 m->oflags = VPO_UNMANAGED;
2166 m->busy_lock = VPB_UNBUSIED;
2167 /* Don't change PG_ZERO. */
2168 vm_page_free_toq(m);
2170 if (req & VM_ALLOC_WAITFAIL) {
2171 VM_OBJECT_WUNLOCK(object);
2173 VM_OBJECT_WLOCK(object);
2180 if (memattr != VM_MEMATTR_DEFAULT)
2181 pmap_page_set_memattr(m, memattr);
2188 * Check a page that has been freshly dequeued from a freelist.
2191 vm_page_alloc_check(vm_page_t m)
2194 KASSERT(m->object == NULL, ("page %p has object", m));
2195 KASSERT(m->queue == PQ_NONE && (m->aflags & PGA_QUEUE_STATE_MASK) == 0,
2196 ("page %p has unexpected queue %d, flags %#x",
2197 m, m->queue, (m->aflags & PGA_QUEUE_STATE_MASK)));
2198 KASSERT(m->ref_count == 0, ("page %p has references", m));
2199 KASSERT(!vm_page_busied(m), ("page %p is busy", m));
2200 KASSERT(m->dirty == 0, ("page %p is dirty", m));
2201 KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
2202 ("page %p has unexpected memattr %d",
2203 m, pmap_page_get_memattr(m)));
2204 KASSERT(m->valid == 0, ("free page %p is valid", m));
2208 * vm_page_alloc_freelist:
2210 * Allocate a physical page from the specified free page list.
2212 * The caller must always specify an allocation class.
2214 * allocation classes:
2215 * VM_ALLOC_NORMAL normal process request
2216 * VM_ALLOC_SYSTEM system *really* needs a page
2217 * VM_ALLOC_INTERRUPT interrupt time request
2219 * optional allocation flags:
2220 * VM_ALLOC_COUNT(number) the number of additional pages that the caller
2221 * intends to allocate
2222 * VM_ALLOC_WIRED wire the allocated page
2223 * VM_ALLOC_ZERO prefer a zeroed page
2226 vm_page_alloc_freelist(int freelist, int req)
2228 struct vm_domainset_iter di;
2232 vm_domainset_iter_page_init(&di, NULL, 0, &domain, &req);
2234 m = vm_page_alloc_freelist_domain(domain, freelist, req);
2237 } while (vm_domainset_iter_page(&di, NULL, &domain) == 0);
2243 vm_page_alloc_freelist_domain(int domain, int freelist, int req)
2245 struct vm_domain *vmd;
2250 vmd = VM_DOMAIN(domain);
2252 if (vm_domain_allocate(vmd, req, 1)) {
2253 vm_domain_free_lock(vmd);
2254 m = vm_phys_alloc_freelist_pages(domain, freelist,
2255 VM_FREEPOOL_DIRECT, 0);
2256 vm_domain_free_unlock(vmd);
2258 vm_domain_freecnt_inc(vmd, 1);
2261 if (vm_domain_alloc_fail(vmd, NULL, req))
2266 vm_page_alloc_check(m);
2269 * Initialize the page. Only the PG_ZERO flag is inherited.
2273 if ((req & VM_ALLOC_ZERO) != 0)
2276 if ((req & VM_ALLOC_WIRED) != 0) {
2278 * The page lock is not required for wiring a page that does
2279 * not belong to an object.
2284 /* Unmanaged pages don't use "act_count". */
2285 m->oflags = VPO_UNMANAGED;
2290 vm_page_zone_import(void *arg, void **store, int cnt, int domain, int flags)
2292 struct vm_domain *vmd;
2293 struct vm_pgcache *pgcache;
2297 vmd = VM_DOMAIN(pgcache->domain);
2298 /* Only import if we can bring in a full bucket. */
2299 if (cnt == 1 || !vm_domain_allocate(vmd, VM_ALLOC_NORMAL, cnt))
2301 domain = vmd->vmd_domain;
2302 vm_domain_free_lock(vmd);
2303 i = vm_phys_alloc_npages(domain, pgcache->pool, cnt,
2304 (vm_page_t *)store);
2305 vm_domain_free_unlock(vmd);
2307 vm_domain_freecnt_inc(vmd, cnt - i);
2313 vm_page_zone_release(void *arg, void **store, int cnt)
2315 struct vm_domain *vmd;
2316 struct vm_pgcache *pgcache;
2321 vmd = VM_DOMAIN(pgcache->domain);
2322 vm_domain_free_lock(vmd);
2323 for (i = 0; i < cnt; i++) {
2324 m = (vm_page_t)store[i];
2325 vm_phys_free_pages(m, 0);
2327 vm_domain_free_unlock(vmd);
2328 vm_domain_freecnt_inc(vmd, cnt);
2331 #define VPSC_ANY 0 /* No restrictions. */
2332 #define VPSC_NORESERV 1 /* Skip reservations; implies VPSC_NOSUPER. */
2333 #define VPSC_NOSUPER 2 /* Skip superpages. */
2336 * vm_page_scan_contig:
2338 * Scan vm_page_array[] between the specified entries "m_start" and
2339 * "m_end" for a run of contiguous physical pages that satisfy the
2340 * specified conditions, and return the lowest page in the run. The
2341 * specified "alignment" determines the alignment of the lowest physical
2342 * page in the run. If the specified "boundary" is non-zero, then the
2343 * run of physical pages cannot span a physical address that is a
2344 * multiple of "boundary".
2346 * "m_end" is never dereferenced, so it need not point to a vm_page
2347 * structure within vm_page_array[].
2349 * "npages" must be greater than zero. "m_start" and "m_end" must not
2350 * span a hole (or discontiguity) in the physical address space. Both
2351 * "alignment" and "boundary" must be a power of two.
2354 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
2355 u_long alignment, vm_paddr_t boundary, int options)
2361 #if VM_NRESERVLEVEL > 0
2364 int m_inc, order, run_ext, run_len;
2366 KASSERT(npages > 0, ("npages is 0"));
2367 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2368 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2372 for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
2373 KASSERT((m->flags & PG_MARKER) == 0,
2374 ("page %p is PG_MARKER", m));
2375 KASSERT((m->flags & PG_FICTITIOUS) == 0 || m->ref_count >= 1,
2376 ("fictitious page %p has invalid ref count", m));
2379 * If the current page would be the start of a run, check its
2380 * physical address against the end, alignment, and boundary
2381 * conditions. If it doesn't satisfy these conditions, either
2382 * terminate the scan or advance to the next page that
2383 * satisfies the failed condition.
2386 KASSERT(m_run == NULL, ("m_run != NULL"));
2387 if (m + npages > m_end)
2389 pa = VM_PAGE_TO_PHYS(m);
2390 if ((pa & (alignment - 1)) != 0) {
2391 m_inc = atop(roundup2(pa, alignment) - pa);
2394 if (rounddown2(pa ^ (pa + ptoa(npages) - 1),
2396 m_inc = atop(roundup2(pa, boundary) - pa);
2400 KASSERT(m_run != NULL, ("m_run == NULL"));
2402 vm_page_change_lock(m, &m_mtx);
2405 if (vm_page_wired(m))
2407 #if VM_NRESERVLEVEL > 0
2408 else if ((level = vm_reserv_level(m)) >= 0 &&
2409 (options & VPSC_NORESERV) != 0) {
2411 /* Advance to the end of the reservation. */
2412 pa = VM_PAGE_TO_PHYS(m);
2413 m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
2417 else if ((object = m->object) != NULL) {
2419 * The page is considered eligible for relocation if
2420 * and only if it could be laundered or reclaimed by
2423 if (!VM_OBJECT_TRYRLOCK(object)) {
2425 VM_OBJECT_RLOCK(object);
2427 if (m->object != object) {
2429 * The page may have been freed.
2431 VM_OBJECT_RUNLOCK(object);
2435 /* Don't care: PG_NODUMP, PG_ZERO. */
2436 if (object->type != OBJT_DEFAULT &&
2437 object->type != OBJT_SWAP &&
2438 object->type != OBJT_VNODE) {
2440 #if VM_NRESERVLEVEL > 0
2441 } else if ((options & VPSC_NOSUPER) != 0 &&
2442 (level = vm_reserv_level_iffullpop(m)) >= 0) {
2444 /* Advance to the end of the superpage. */
2445 pa = VM_PAGE_TO_PHYS(m);
2446 m_inc = atop(roundup2(pa + 1,
2447 vm_reserv_size(level)) - pa);
2449 } else if (object->memattr == VM_MEMATTR_DEFAULT &&
2450 vm_page_queue(m) != PQ_NONE && !vm_page_busied(m) &&
2451 !vm_page_wired(m)) {
2453 * The page is allocated but eligible for
2454 * relocation. Extend the current run by one
2457 KASSERT(pmap_page_get_memattr(m) ==
2459 ("page %p has an unexpected memattr", m));
2460 KASSERT((m->oflags & (VPO_SWAPINPROG |
2461 VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2462 ("page %p has unexpected oflags", m));
2463 /* Don't care: VPO_NOSYNC. */
2467 VM_OBJECT_RUNLOCK(object);
2468 #if VM_NRESERVLEVEL > 0
2469 } else if (level >= 0) {
2471 * The page is reserved but not yet allocated. In
2472 * other words, it is still free. Extend the current
2477 } else if ((order = m->order) < VM_NFREEORDER) {
2479 * The page is enqueued in the physical memory
2480 * allocator's free page queues. Moreover, it is the
2481 * first page in a power-of-two-sized run of
2482 * contiguous free pages. Add these pages to the end
2483 * of the current run, and jump ahead.
2485 run_ext = 1 << order;
2489 * Skip the page for one of the following reasons: (1)
2490 * It is enqueued in the physical memory allocator's
2491 * free page queues. However, it is not the first
2492 * page in a run of contiguous free pages. (This case
2493 * rarely occurs because the scan is performed in
2494 * ascending order.) (2) It is not reserved, and it is
2495 * transitioning from free to allocated. (Conversely,
2496 * the transition from allocated to free for managed
2497 * pages is blocked by the page lock.) (3) It is
2498 * allocated but not contained by an object and not
2499 * wired, e.g., allocated by Xen's balloon driver.
2505 * Extend or reset the current run of pages.
2520 if (run_len >= npages)
2526 * vm_page_reclaim_run:
2528 * Try to relocate each of the allocated virtual pages within the
2529 * specified run of physical pages to a new physical address. Free the
2530 * physical pages underlying the relocated virtual pages. A virtual page
2531 * is relocatable if and only if it could be laundered or reclaimed by
2532 * the page daemon. Whenever possible, a virtual page is relocated to a
2533 * physical address above "high".
2535 * Returns 0 if every physical page within the run was already free or
2536 * just freed by a successful relocation. Otherwise, returns a non-zero
2537 * value indicating why the last attempt to relocate a virtual page was
2540 * "req_class" must be an allocation class.
2543 vm_page_reclaim_run(int req_class, int domain, u_long npages, vm_page_t m_run,
2546 struct vm_domain *vmd;
2548 struct spglist free;
2551 vm_page_t m, m_end, m_new;
2552 int error, order, req;
2554 KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
2555 ("req_class is not an allocation class"));
2559 m_end = m_run + npages;
2561 for (; error == 0 && m < m_end; m++) {
2562 KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2563 ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2566 * Avoid releasing and reacquiring the same page lock.
2568 vm_page_change_lock(m, &m_mtx);
2571 * Racily check for wirings. Races are handled below.
2573 if (vm_page_wired(m))
2575 else if ((object = m->object) != NULL) {
2577 * The page is relocated if and only if it could be
2578 * laundered or reclaimed by the page daemon.
2580 if (!VM_OBJECT_TRYWLOCK(object)) {
2582 VM_OBJECT_WLOCK(object);
2584 if (m->object != object) {
2586 * The page may have been freed.
2588 VM_OBJECT_WUNLOCK(object);
2592 /* Don't care: PG_NODUMP, PG_ZERO. */
2593 if (object->type != OBJT_DEFAULT &&
2594 object->type != OBJT_SWAP &&
2595 object->type != OBJT_VNODE)
2597 else if (object->memattr != VM_MEMATTR_DEFAULT)
2599 else if (vm_page_queue(m) != PQ_NONE &&
2600 vm_page_tryxbusy(m) != 0) {
2601 if (vm_page_wired(m)) {
2606 KASSERT(pmap_page_get_memattr(m) ==
2608 ("page %p has an unexpected memattr", m));
2609 KASSERT((m->oflags & (VPO_SWAPINPROG |
2610 VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2611 ("page %p has unexpected oflags", m));
2612 /* Don't care: VPO_NOSYNC. */
2613 if (m->valid != 0) {
2615 * First, try to allocate a new page
2616 * that is above "high". Failing
2617 * that, try to allocate a new page
2618 * that is below "m_run". Allocate
2619 * the new page between the end of
2620 * "m_run" and "high" only as a last
2623 req = req_class | VM_ALLOC_NOOBJ;
2624 if ((m->flags & PG_NODUMP) != 0)
2625 req |= VM_ALLOC_NODUMP;
2626 if (trunc_page(high) !=
2627 ~(vm_paddr_t)PAGE_MASK) {
2628 m_new = vm_page_alloc_contig(
2633 VM_MEMATTR_DEFAULT);
2636 if (m_new == NULL) {
2637 pa = VM_PAGE_TO_PHYS(m_run);
2638 m_new = vm_page_alloc_contig(
2640 0, pa - 1, PAGE_SIZE, 0,
2641 VM_MEMATTR_DEFAULT);
2643 if (m_new == NULL) {
2645 m_new = vm_page_alloc_contig(
2647 pa, high, PAGE_SIZE, 0,
2648 VM_MEMATTR_DEFAULT);
2650 if (m_new == NULL) {
2657 * Unmap the page and check for new
2658 * wirings that may have been acquired
2659 * through a pmap lookup.
2661 if (object->ref_count != 0 &&
2662 !vm_page_try_remove_all(m)) {
2663 vm_page_free(m_new);
2669 * Replace "m" with the new page. For
2670 * vm_page_replace(), "m" must be busy
2671 * and dequeued. Finally, change "m"
2672 * as if vm_page_free() was called.
2674 m_new->aflags = m->aflags &
2675 ~PGA_QUEUE_STATE_MASK;
2676 KASSERT(m_new->oflags == VPO_UNMANAGED,
2677 ("page %p is managed", m_new));
2678 m_new->oflags = m->oflags & VPO_NOSYNC;
2679 pmap_copy_page(m, m_new);
2680 m_new->valid = m->valid;
2681 m_new->dirty = m->dirty;
2682 m->flags &= ~PG_ZERO;
2684 vm_page_replace_checked(m_new, object,
2686 if (vm_page_free_prep(m))
2687 SLIST_INSERT_HEAD(&free, m,
2691 * The new page must be deactivated
2692 * before the object is unlocked.
2694 vm_page_change_lock(m_new, &m_mtx);
2695 vm_page_deactivate(m_new);
2697 m->flags &= ~PG_ZERO;
2699 if (vm_page_free_prep(m))
2700 SLIST_INSERT_HEAD(&free, m,
2702 KASSERT(m->dirty == 0,
2703 ("page %p is dirty", m));
2708 VM_OBJECT_WUNLOCK(object);
2710 MPASS(vm_phys_domain(m) == domain);
2711 vmd = VM_DOMAIN(domain);
2712 vm_domain_free_lock(vmd);
2714 if (order < VM_NFREEORDER) {
2716 * The page is enqueued in the physical memory
2717 * allocator's free page queues. Moreover, it
2718 * is the first page in a power-of-two-sized
2719 * run of contiguous free pages. Jump ahead
2720 * to the last page within that run, and
2721 * continue from there.
2723 m += (1 << order) - 1;
2725 #if VM_NRESERVLEVEL > 0
2726 else if (vm_reserv_is_page_free(m))
2729 vm_domain_free_unlock(vmd);
2730 if (order == VM_NFREEORDER)
2736 if ((m = SLIST_FIRST(&free)) != NULL) {
2739 vmd = VM_DOMAIN(domain);
2741 vm_domain_free_lock(vmd);
2743 MPASS(vm_phys_domain(m) == domain);
2744 SLIST_REMOVE_HEAD(&free, plinks.s.ss);
2745 vm_phys_free_pages(m, 0);
2747 } while ((m = SLIST_FIRST(&free)) != NULL);
2748 vm_domain_free_unlock(vmd);
2749 vm_domain_freecnt_inc(vmd, cnt);
2756 CTASSERT(powerof2(NRUNS));
2758 #define RUN_INDEX(count) ((count) & (NRUNS - 1))
2760 #define MIN_RECLAIM 8
2763 * vm_page_reclaim_contig:
2765 * Reclaim allocated, contiguous physical memory satisfying the specified
2766 * conditions by relocating the virtual pages using that physical memory.
2767 * Returns true if reclamation is successful and false otherwise. Since
2768 * relocation requires the allocation of physical pages, reclamation may
2769 * fail due to a shortage of free pages. When reclamation fails, callers
2770 * are expected to perform vm_wait() before retrying a failed allocation
2771 * operation, e.g., vm_page_alloc_contig().
2773 * The caller must always specify an allocation class through "req".
2775 * allocation classes:
2776 * VM_ALLOC_NORMAL normal process request
2777 * VM_ALLOC_SYSTEM system *really* needs a page
2778 * VM_ALLOC_INTERRUPT interrupt time request
2780 * The optional allocation flags are ignored.
2782 * "npages" must be greater than zero. Both "alignment" and "boundary"
2783 * must be a power of two.
2786 vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
2787 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
2789 struct vm_domain *vmd;
2790 vm_paddr_t curr_low;
2791 vm_page_t m_run, m_runs[NRUNS];
2792 u_long count, reclaimed;
2793 int error, i, options, req_class;
2795 KASSERT(npages > 0, ("npages is 0"));
2796 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2797 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2798 req_class = req & VM_ALLOC_CLASS_MASK;
2801 * The page daemon is allowed to dig deeper into the free page list.
2803 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2804 req_class = VM_ALLOC_SYSTEM;
2807 * Return if the number of free pages cannot satisfy the requested
2810 vmd = VM_DOMAIN(domain);
2811 count = vmd->vmd_free_count;
2812 if (count < npages + vmd->vmd_free_reserved || (count < npages +
2813 vmd->vmd_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
2814 (count < npages && req_class == VM_ALLOC_INTERRUPT))
2818 * Scan up to three times, relaxing the restrictions ("options") on
2819 * the reclamation of reservations and superpages each time.
2821 for (options = VPSC_NORESERV;;) {
2823 * Find the highest runs that satisfy the given constraints
2824 * and restrictions, and record them in "m_runs".
2829 m_run = vm_phys_scan_contig(domain, npages, curr_low,
2830 high, alignment, boundary, options);
2833 curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages);
2834 m_runs[RUN_INDEX(count)] = m_run;
2839 * Reclaim the highest runs in LIFO (descending) order until
2840 * the number of reclaimed pages, "reclaimed", is at least
2841 * MIN_RECLAIM. Reset "reclaimed" each time because each
2842 * reclamation is idempotent, and runs will (likely) recur
2843 * from one scan to the next as restrictions are relaxed.
2846 for (i = 0; count > 0 && i < NRUNS; i++) {
2848 m_run = m_runs[RUN_INDEX(count)];
2849 error = vm_page_reclaim_run(req_class, domain, npages,
2852 reclaimed += npages;
2853 if (reclaimed >= MIN_RECLAIM)
2859 * Either relax the restrictions on the next scan or return if
2860 * the last scan had no restrictions.
2862 if (options == VPSC_NORESERV)
2863 options = VPSC_NOSUPER;
2864 else if (options == VPSC_NOSUPER)
2866 else if (options == VPSC_ANY)
2867 return (reclaimed != 0);
2872 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
2873 u_long alignment, vm_paddr_t boundary)
2875 struct vm_domainset_iter di;
2879 vm_domainset_iter_page_init(&di, NULL, 0, &domain, &req);
2881 ret = vm_page_reclaim_contig_domain(domain, req, npages, low,
2882 high, alignment, boundary);
2885 } while (vm_domainset_iter_page(&di, NULL, &domain) == 0);
2891 * Set the domain in the appropriate page level domainset.
2894 vm_domain_set(struct vm_domain *vmd)
2897 mtx_lock(&vm_domainset_lock);
2898 if (!vmd->vmd_minset && vm_paging_min(vmd)) {
2899 vmd->vmd_minset = 1;
2900 DOMAINSET_SET(vmd->vmd_domain, &vm_min_domains);
2902 if (!vmd->vmd_severeset && vm_paging_severe(vmd)) {
2903 vmd->vmd_severeset = 1;
2904 DOMAINSET_SET(vmd->vmd_domain, &vm_severe_domains);
2906 mtx_unlock(&vm_domainset_lock);
2910 * Clear the domain from the appropriate page level domainset.
2913 vm_domain_clear(struct vm_domain *vmd)
2916 mtx_lock(&vm_domainset_lock);
2917 if (vmd->vmd_minset && !vm_paging_min(vmd)) {
2918 vmd->vmd_minset = 0;
2919 DOMAINSET_CLR(vmd->vmd_domain, &vm_min_domains);
2920 if (vm_min_waiters != 0) {
2922 wakeup(&vm_min_domains);
2925 if (vmd->vmd_severeset && !vm_paging_severe(vmd)) {
2926 vmd->vmd_severeset = 0;
2927 DOMAINSET_CLR(vmd->vmd_domain, &vm_severe_domains);
2928 if (vm_severe_waiters != 0) {
2929 vm_severe_waiters = 0;
2930 wakeup(&vm_severe_domains);
2935 * If pageout daemon needs pages, then tell it that there are
2938 if (vmd->vmd_pageout_pages_needed &&
2939 vmd->vmd_free_count >= vmd->vmd_pageout_free_min) {
2940 wakeup(&vmd->vmd_pageout_pages_needed);
2941 vmd->vmd_pageout_pages_needed = 0;
2944 /* See comments in vm_wait_doms(). */
2945 if (vm_pageproc_waiters) {
2946 vm_pageproc_waiters = 0;
2947 wakeup(&vm_pageproc_waiters);
2949 mtx_unlock(&vm_domainset_lock);
2953 * Wait for free pages to exceed the min threshold globally.
2959 mtx_lock(&vm_domainset_lock);
2960 while (vm_page_count_min()) {
2962 msleep(&vm_min_domains, &vm_domainset_lock, PVM, "vmwait", 0);
2964 mtx_unlock(&vm_domainset_lock);
2968 * Wait for free pages to exceed the severe threshold globally.
2971 vm_wait_severe(void)
2974 mtx_lock(&vm_domainset_lock);
2975 while (vm_page_count_severe()) {
2976 vm_severe_waiters++;
2977 msleep(&vm_severe_domains, &vm_domainset_lock, PVM,
2980 mtx_unlock(&vm_domainset_lock);
2987 return (vm_severe_waiters + vm_min_waiters + vm_pageproc_waiters);
2991 vm_wait_doms(const domainset_t *wdoms)
2995 * We use racey wakeup synchronization to avoid expensive global
2996 * locking for the pageproc when sleeping with a non-specific vm_wait.
2997 * To handle this, we only sleep for one tick in this instance. It
2998 * is expected that most allocations for the pageproc will come from
2999 * kmem or vm_page_grab* which will use the more specific and
3000 * race-free vm_wait_domain().
3002 if (curproc == pageproc) {
3003 mtx_lock(&vm_domainset_lock);
3004 vm_pageproc_waiters++;
3005 msleep(&vm_pageproc_waiters, &vm_domainset_lock, PVM | PDROP,
3009 * XXX Ideally we would wait only until the allocation could
3010 * be satisfied. This condition can cause new allocators to
3011 * consume all freed pages while old allocators wait.
3013 mtx_lock(&vm_domainset_lock);
3014 if (vm_page_count_min_set(wdoms)) {
3016 msleep(&vm_min_domains, &vm_domainset_lock,
3017 PVM | PDROP, "vmwait", 0);
3019 mtx_unlock(&vm_domainset_lock);
3026 * Sleep until free pages are available for allocation.
3027 * - Called in various places after failed memory allocations.
3030 vm_wait_domain(int domain)
3032 struct vm_domain *vmd;
3035 vmd = VM_DOMAIN(domain);
3036 vm_domain_free_assert_unlocked(vmd);
3038 if (curproc == pageproc) {
3039 mtx_lock(&vm_domainset_lock);
3040 if (vmd->vmd_free_count < vmd->vmd_pageout_free_min) {
3041 vmd->vmd_pageout_pages_needed = 1;
3042 msleep(&vmd->vmd_pageout_pages_needed,
3043 &vm_domainset_lock, PDROP | PSWP, "VMWait", 0);
3045 mtx_unlock(&vm_domainset_lock);
3047 if (pageproc == NULL)
3048 panic("vm_wait in early boot");
3049 DOMAINSET_ZERO(&wdom);
3050 DOMAINSET_SET(vmd->vmd_domain, &wdom);
3051 vm_wait_doms(&wdom);
3058 * Sleep until free pages are available for allocation in the
3059 * affinity domains of the obj. If obj is NULL, the domain set
3060 * for the calling thread is used.
3061 * Called in various places after failed memory allocations.
3064 vm_wait(vm_object_t obj)
3066 struct domainset *d;
3071 * Carefully fetch pointers only once: the struct domainset
3072 * itself is ummutable but the pointer might change.
3075 d = obj->domain.dr_policy;
3077 d = curthread->td_domain.dr_policy;
3079 vm_wait_doms(&d->ds_mask);
3083 * vm_domain_alloc_fail:
3085 * Called when a page allocation function fails. Informs the
3086 * pagedaemon and performs the requested wait. Requires the
3087 * domain_free and object lock on entry. Returns with the
3088 * object lock held and free lock released. Returns an error when
3089 * retry is necessary.
3093 vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object, int req)
3096 vm_domain_free_assert_unlocked(vmd);
3098 atomic_add_int(&vmd->vmd_pageout_deficit,
3099 max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
3100 if (req & (VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL)) {
3102 VM_OBJECT_WUNLOCK(object);
3103 vm_wait_domain(vmd->vmd_domain);
3105 VM_OBJECT_WLOCK(object);
3106 if (req & VM_ALLOC_WAITOK)
3116 * Sleep until free pages are available for allocation.
3117 * - Called only in vm_fault so that processes page faulting
3118 * can be easily tracked.
3119 * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing
3120 * processes will be able to grab memory first. Do not change
3121 * this balance without careful testing first.
3124 vm_waitpfault(struct domainset *dset, int timo)
3128 * XXX Ideally we would wait only until the allocation could
3129 * be satisfied. This condition can cause new allocators to
3130 * consume all freed pages while old allocators wait.
3132 mtx_lock(&vm_domainset_lock);
3133 if (vm_page_count_min_set(&dset->ds_mask)) {
3135 msleep(&vm_min_domains, &vm_domainset_lock, PUSER | PDROP,
3138 mtx_unlock(&vm_domainset_lock);
3141 static struct vm_pagequeue *
3142 vm_page_pagequeue(vm_page_t m)
3147 if ((queue = atomic_load_8(&m->queue)) == PQ_NONE)
3149 return (&vm_pagequeue_domain(m)->vmd_pagequeues[queue]);
3153 vm_pqbatch_process_page(struct vm_pagequeue *pq, vm_page_t m)
3155 struct vm_domain *vmd;
3158 CRITICAL_ASSERT(curthread);
3159 vm_pagequeue_assert_locked(pq);
3162 * The page daemon is allowed to set m->queue = PQ_NONE without
3163 * the page queue lock held. In this case it is about to free the page,
3164 * which must not have any queue state.
3166 qflags = atomic_load_8(&m->aflags);
3167 KASSERT(pq == vm_page_pagequeue(m) ||
3168 (qflags & PGA_QUEUE_STATE_MASK) == 0,
3169 ("page %p doesn't belong to queue %p but has aflags %#x",
3172 if ((qflags & PGA_DEQUEUE) != 0) {
3173 if (__predict_true((qflags & PGA_ENQUEUED) != 0))
3174 vm_pagequeue_remove(pq, m);
3175 vm_page_dequeue_complete(m);
3176 counter_u64_add(queue_ops, 1);
3177 } else if ((qflags & (PGA_REQUEUE | PGA_REQUEUE_HEAD)) != 0) {
3178 if ((qflags & PGA_ENQUEUED) != 0)
3179 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
3181 vm_pagequeue_cnt_inc(pq);
3182 vm_page_aflag_set(m, PGA_ENQUEUED);
3186 * Give PGA_REQUEUE_HEAD precedence over PGA_REQUEUE.
3187 * In particular, if both flags are set in close succession,
3188 * only PGA_REQUEUE_HEAD will be applied, even if it was set
3191 if ((qflags & PGA_REQUEUE_HEAD) != 0) {
3192 KASSERT(m->queue == PQ_INACTIVE,
3193 ("head enqueue not supported for page %p", m));
3194 vmd = vm_pagequeue_domain(m);
3195 TAILQ_INSERT_BEFORE(&vmd->vmd_inacthead, m, plinks.q);
3197 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3199 vm_page_aflag_clear(m, qflags & (PGA_REQUEUE |
3201 counter_u64_add(queue_ops, 1);
3203 counter_u64_add(queue_nops, 1);
3208 vm_pqbatch_process(struct vm_pagequeue *pq, struct vm_batchqueue *bq,
3214 for (i = 0; i < bq->bq_cnt; i++) {
3216 if (__predict_false(m->queue != queue))
3218 vm_pqbatch_process_page(pq, m);
3220 vm_batchqueue_init(bq);
3224 * vm_page_pqbatch_submit: [ internal use only ]
3226 * Enqueue a page in the specified page queue's batched work queue.
3227 * The caller must have encoded the requested operation in the page
3228 * structure's aflags field.
3231 vm_page_pqbatch_submit(vm_page_t m, uint8_t queue)
3233 struct vm_batchqueue *bq;
3234 struct vm_pagequeue *pq;
3237 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3238 ("page %p is unmanaged", m));
3239 KASSERT(mtx_owned(vm_page_lockptr(m)) || m->object == NULL,
3240 ("missing synchronization for page %p", m));
3241 KASSERT(queue < PQ_COUNT, ("invalid queue %d", queue));
3243 domain = vm_phys_domain(m);
3244 pq = &vm_pagequeue_domain(m)->vmd_pagequeues[queue];
3247 bq = DPCPU_PTR(pqbatch[domain][queue]);
3248 if (vm_batchqueue_insert(bq, m)) {
3253 vm_pagequeue_lock(pq);
3255 bq = DPCPU_PTR(pqbatch[domain][queue]);
3256 vm_pqbatch_process(pq, bq, queue);
3259 * The page may have been logically dequeued before we acquired the
3260 * page queue lock. In this case, since we either hold the page lock
3261 * or the page is being freed, a different thread cannot be concurrently
3262 * enqueuing the page.
3264 if (__predict_true(m->queue == queue))
3265 vm_pqbatch_process_page(pq, m);
3267 KASSERT(m->queue == PQ_NONE,
3268 ("invalid queue transition for page %p", m));
3269 KASSERT((m->aflags & PGA_ENQUEUED) == 0,
3270 ("page %p is enqueued with invalid queue index", m));
3272 vm_pagequeue_unlock(pq);
3277 * vm_page_pqbatch_drain: [ internal use only ]
3279 * Force all per-CPU page queue batch queues to be drained. This is
3280 * intended for use in severe memory shortages, to ensure that pages
3281 * do not remain stuck in the batch queues.
3284 vm_page_pqbatch_drain(void)
3287 struct vm_domain *vmd;
3288 struct vm_pagequeue *pq;
3289 int cpu, domain, queue;
3294 sched_bind(td, cpu);
3297 for (domain = 0; domain < vm_ndomains; domain++) {
3298 vmd = VM_DOMAIN(domain);
3299 for (queue = 0; queue < PQ_COUNT; queue++) {
3300 pq = &vmd->vmd_pagequeues[queue];
3301 vm_pagequeue_lock(pq);
3303 vm_pqbatch_process(pq,
3304 DPCPU_PTR(pqbatch[domain][queue]), queue);
3306 vm_pagequeue_unlock(pq);
3316 * Complete the logical removal of a page from a page queue. We must be
3317 * careful to synchronize with the page daemon, which may be concurrently
3318 * examining the page with only the page lock held. The page must not be
3319 * in a state where it appears to be logically enqueued.
3322 vm_page_dequeue_complete(vm_page_t m)
3326 atomic_thread_fence_rel();
3327 vm_page_aflag_clear(m, PGA_QUEUE_STATE_MASK);
3331 * vm_page_dequeue_deferred: [ internal use only ]
3333 * Request removal of the given page from its current page
3334 * queue. Physical removal from the queue may be deferred
3337 * The page must be locked.
3340 vm_page_dequeue_deferred(vm_page_t m)
3344 vm_page_assert_locked(m);
3346 if ((queue = vm_page_queue(m)) == PQ_NONE)
3350 * Set PGA_DEQUEUE if it is not already set to handle a concurrent call
3351 * to vm_page_dequeue_deferred_free(). In particular, avoid modifying
3352 * the page's queue state once vm_page_dequeue_deferred_free() has been
3353 * called. In the event of a race, two batch queue entries for the page
3354 * will be created, but the second will have no effect.
3356 if (vm_page_pqstate_cmpset(m, queue, queue, PGA_DEQUEUE, PGA_DEQUEUE))
3357 vm_page_pqbatch_submit(m, queue);
3361 * A variant of vm_page_dequeue_deferred() that does not assert the page
3362 * lock and is only to be called from vm_page_free_prep(). Because the
3363 * page is being freed, we can assume that nothing other than the page
3364 * daemon is scheduling queue operations on this page, so we get for
3365 * free the mutual exclusion that is otherwise provided by the page lock.
3366 * To handle races, the page daemon must take care to atomically check
3367 * for PGA_DEQUEUE when updating queue state.
3370 vm_page_dequeue_deferred_free(vm_page_t m)
3374 KASSERT(m->ref_count == 0, ("page %p has references", m));
3377 if ((m->aflags & PGA_DEQUEUE) != 0)
3379 atomic_thread_fence_acq();
3380 if ((queue = atomic_load_8(&m->queue)) == PQ_NONE)
3382 if (vm_page_pqstate_cmpset(m, queue, queue, PGA_DEQUEUE,
3384 vm_page_pqbatch_submit(m, queue);
3393 * Remove the page from whichever page queue it's in, if any.
3394 * The page must either be locked or unallocated. This constraint
3395 * ensures that the queue state of the page will remain consistent
3396 * after this function returns.
3399 vm_page_dequeue(vm_page_t m)
3401 struct vm_pagequeue *pq, *pq1;
3404 KASSERT(mtx_owned(vm_page_lockptr(m)) || m->object == NULL,
3405 ("page %p is allocated and unlocked", m));
3407 for (pq = vm_page_pagequeue(m);; pq = pq1) {
3410 * A thread may be concurrently executing
3411 * vm_page_dequeue_complete(). Ensure that all queue
3412 * state is cleared before we return.
3414 aflags = atomic_load_8(&m->aflags);
3415 if ((aflags & PGA_QUEUE_STATE_MASK) == 0)
3417 KASSERT((aflags & PGA_DEQUEUE) != 0,
3418 ("page %p has unexpected queue state flags %#x",
3422 * Busy wait until the thread updating queue state is
3423 * finished. Such a thread must be executing in a
3427 pq1 = vm_page_pagequeue(m);
3430 vm_pagequeue_lock(pq);
3431 if ((pq1 = vm_page_pagequeue(m)) == pq)
3433 vm_pagequeue_unlock(pq);
3435 KASSERT(pq == vm_page_pagequeue(m),
3436 ("%s: page %p migrated directly between queues", __func__, m));
3437 KASSERT((m->aflags & PGA_DEQUEUE) != 0 ||
3438 mtx_owned(vm_page_lockptr(m)),
3439 ("%s: queued unlocked page %p", __func__, m));
3441 if ((m->aflags & PGA_ENQUEUED) != 0)
3442 vm_pagequeue_remove(pq, m);
3443 vm_page_dequeue_complete(m);
3444 vm_pagequeue_unlock(pq);
3448 * Schedule the given page for insertion into the specified page queue.
3449 * Physical insertion of the page may be deferred indefinitely.
3452 vm_page_enqueue(vm_page_t m, uint8_t queue)
3455 vm_page_assert_locked(m);
3456 KASSERT(m->queue == PQ_NONE && (m->aflags & PGA_QUEUE_STATE_MASK) == 0,
3457 ("%s: page %p is already enqueued", __func__, m));
3460 if ((m->aflags & PGA_REQUEUE) == 0)
3461 vm_page_aflag_set(m, PGA_REQUEUE);
3462 vm_page_pqbatch_submit(m, queue);
3466 * vm_page_requeue: [ internal use only ]
3468 * Schedule a requeue of the given page.
3470 * The page must be locked.
3473 vm_page_requeue(vm_page_t m)
3476 vm_page_assert_locked(m);
3477 KASSERT(vm_page_queue(m) != PQ_NONE,
3478 ("%s: page %p is not logically enqueued", __func__, m));
3480 if ((m->aflags & PGA_REQUEUE) == 0)
3481 vm_page_aflag_set(m, PGA_REQUEUE);
3482 vm_page_pqbatch_submit(m, atomic_load_8(&m->queue));
3486 * vm_page_swapqueue: [ internal use only ]
3488 * Move the page from one queue to another, or to the tail of its
3489 * current queue, in the face of a possible concurrent call to
3490 * vm_page_dequeue_deferred_free().
3493 vm_page_swapqueue(vm_page_t m, uint8_t oldq, uint8_t newq)
3495 struct vm_pagequeue *pq;
3499 KASSERT(oldq < PQ_COUNT && newq < PQ_COUNT && oldq != newq,
3500 ("vm_page_swapqueue: invalid queues (%d, %d)", oldq, newq));
3501 vm_page_assert_locked(m);
3503 pq = &vm_pagequeue_domain(m)->vmd_pagequeues[oldq];
3504 vm_pagequeue_lock(pq);
3507 * The physical queue state might change at any point before the page
3508 * queue lock is acquired, so we must verify that we hold the correct
3509 * lock before proceeding.
3511 if (__predict_false(m->queue != oldq)) {
3512 vm_pagequeue_unlock(pq);
3517 * Once the queue index of the page changes, there is nothing
3518 * synchronizing with further updates to the physical queue state.
3519 * Therefore we must remove the page from the queue now in anticipation
3520 * of a successful commit, and be prepared to roll back.
3522 if (__predict_true((m->aflags & PGA_ENQUEUED) != 0)) {
3523 next = TAILQ_NEXT(m, plinks.q);
3524 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
3525 vm_page_aflag_clear(m, PGA_ENQUEUED);
3532 * Atomically update the queue field and set PGA_REQUEUE while
3533 * ensuring that PGA_DEQUEUE has not been set.
3535 if (__predict_false(!vm_page_pqstate_cmpset(m, oldq, newq, PGA_DEQUEUE,
3538 vm_page_aflag_set(m, PGA_ENQUEUED);
3540 TAILQ_INSERT_BEFORE(next, m, plinks.q);
3542 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3544 vm_pagequeue_unlock(pq);
3547 vm_pagequeue_cnt_dec(pq);
3548 vm_pagequeue_unlock(pq);
3549 vm_page_pqbatch_submit(m, newq);
3553 * vm_page_free_prep:
3555 * Prepares the given page to be put on the free list,
3556 * disassociating it from any VM object. The caller may return
3557 * the page to the free list only if this function returns true.
3559 * The object must be locked. The page must be locked if it is
3563 vm_page_free_prep(vm_page_t m)
3567 * Synchronize with threads that have dropped a reference to this
3570 atomic_thread_fence_acq();
3572 #if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP)
3573 if (PMAP_HAS_DMAP && (m->flags & PG_ZERO) != 0) {
3576 p = (uint64_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
3577 for (i = 0; i < PAGE_SIZE / sizeof(uint64_t); i++, p++)
3578 KASSERT(*p == 0, ("vm_page_free_prep %p PG_ZERO %d %jx",
3579 m, i, (uintmax_t)*p));
3582 if ((m->oflags & VPO_UNMANAGED) == 0) {
3583 KASSERT(!pmap_page_is_mapped(m),
3584 ("vm_page_free_prep: freeing mapped page %p", m));
3585 KASSERT((m->aflags & (PGA_EXECUTABLE | PGA_WRITEABLE)) == 0,
3586 ("vm_page_free_prep: mapping flags set in page %p", m));
3588 KASSERT(m->queue == PQ_NONE,
3589 ("vm_page_free_prep: unmanaged page %p is queued", m));
3591 VM_CNT_INC(v_tfree);
3593 if (vm_page_sbusied(m))
3594 panic("vm_page_free_prep: freeing shared busy page %p", m);
3596 if (m->object != NULL) {
3597 vm_page_object_remove(m);
3600 * The object reference can be released without an atomic
3603 KASSERT((m->flags & PG_FICTITIOUS) != 0 ||
3604 m->ref_count == VPRC_OBJREF,
3605 ("vm_page_free_prep: page %p has unexpected ref_count %u",
3608 m->ref_count -= VPRC_OBJREF;
3611 if (vm_page_xbusied(m))
3615 * If fictitious remove object association and
3618 if ((m->flags & PG_FICTITIOUS) != 0) {
3619 KASSERT(m->ref_count == 1,
3620 ("fictitious page %p is referenced", m));
3621 KASSERT(m->queue == PQ_NONE,
3622 ("fictitious page %p is queued", m));
3627 * Pages need not be dequeued before they are returned to the physical
3628 * memory allocator, but they must at least be marked for a deferred
3631 if ((m->oflags & VPO_UNMANAGED) == 0)
3632 vm_page_dequeue_deferred_free(m);
3637 if (m->ref_count != 0)
3638 panic("vm_page_free_prep: page %p has references", m);
3641 * Restore the default memory attribute to the page.
3643 if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
3644 pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
3646 #if VM_NRESERVLEVEL > 0
3648 * Determine whether the page belongs to a reservation. If the page was
3649 * allocated from a per-CPU cache, it cannot belong to a reservation, so
3650 * as an optimization, we avoid the check in that case.
3652 if ((m->flags & PG_PCPU_CACHE) == 0 && vm_reserv_free_page(m))
3662 * Returns the given page to the free list, disassociating it
3663 * from any VM object.
3665 * The object must be locked. The page must be locked if it is
3669 vm_page_free_toq(vm_page_t m)
3671 struct vm_domain *vmd;
3674 if (!vm_page_free_prep(m))
3677 vmd = vm_pagequeue_domain(m);
3678 zone = vmd->vmd_pgcache[m->pool].zone;
3679 if ((m->flags & PG_PCPU_CACHE) != 0 && zone != NULL) {
3683 vm_domain_free_lock(vmd);
3684 vm_phys_free_pages(m, 0);
3685 vm_domain_free_unlock(vmd);
3686 vm_domain_freecnt_inc(vmd, 1);
3690 * vm_page_free_pages_toq:
3692 * Returns a list of pages to the free list, disassociating it
3693 * from any VM object. In other words, this is equivalent to
3694 * calling vm_page_free_toq() for each page of a list of VM objects.
3696 * The objects must be locked. The pages must be locked if it is
3700 vm_page_free_pages_toq(struct spglist *free, bool update_wire_count)
3705 if (SLIST_EMPTY(free))
3709 while ((m = SLIST_FIRST(free)) != NULL) {
3711 SLIST_REMOVE_HEAD(free, plinks.s.ss);
3712 vm_page_free_toq(m);
3715 if (update_wire_count)
3720 * Mark this page as wired down, preventing reclamation by the page daemon
3721 * or when the containing object is destroyed.
3724 vm_page_wire(vm_page_t m)
3728 KASSERT(m->object != NULL,
3729 ("vm_page_wire: page %p does not belong to an object", m));
3730 if (!vm_page_busied(m))
3731 VM_OBJECT_ASSERT_LOCKED(m->object);
3732 KASSERT((m->flags & PG_FICTITIOUS) == 0 ||
3733 VPRC_WIRE_COUNT(m->ref_count) >= 1,
3734 ("vm_page_wire: fictitious page %p has zero wirings", m));
3736 old = atomic_fetchadd_int(&m->ref_count, 1);
3737 KASSERT(VPRC_WIRE_COUNT(old) != VPRC_WIRE_COUNT_MAX,
3738 ("vm_page_wire: counter overflow for page %p", m));
3739 if (VPRC_WIRE_COUNT(old) == 0)
3744 * Attempt to wire a mapped page following a pmap lookup of that page.
3745 * This may fail if a thread is concurrently tearing down mappings of the page.
3748 vm_page_wire_mapped(vm_page_t m)
3755 ("vm_page_wire_mapped: wiring unreferenced page %p", m));
3756 if ((old & VPRC_BLOCKED) != 0)
3758 } while (!atomic_fcmpset_int(&m->ref_count, &old, old + 1));
3760 if (VPRC_WIRE_COUNT(old) == 0)
3766 * Release one wiring of the specified page, potentially allowing it to be
3769 * Only managed pages belonging to an object can be paged out. If the number
3770 * of wirings transitions to zero and the page is eligible for page out, then
3771 * the page is added to the specified paging queue. If the released wiring
3772 * represented the last reference to the page, the page is freed.
3774 * A managed page must be locked.
3777 vm_page_unwire(vm_page_t m, uint8_t queue)
3782 KASSERT(queue < PQ_COUNT,
3783 ("vm_page_unwire: invalid queue %u request for page %p", queue, m));
3785 if ((m->oflags & VPO_UNMANAGED) != 0) {
3786 if (vm_page_unwire_noq(m) && m->ref_count == 0)
3792 * Update LRU state before releasing the wiring reference.
3793 * We only need to do this once since we hold the page lock.
3794 * Use a release store when updating the reference count to
3795 * synchronize with vm_page_free_prep().
3800 KASSERT(VPRC_WIRE_COUNT(old) > 0,
3801 ("vm_page_unwire: wire count underflow for page %p", m));
3802 if (!locked && VPRC_WIRE_COUNT(old) == 1) {
3805 if (queue == PQ_ACTIVE && vm_page_queue(m) == PQ_ACTIVE)
3806 vm_page_reference(m);
3808 vm_page_mvqueue(m, queue);
3810 } while (!atomic_fcmpset_rel_int(&m->ref_count, &old, old - 1));
3813 * Release the lock only after the wiring is released, to ensure that
3814 * the page daemon does not encounter and dequeue the page while it is
3820 if (VPRC_WIRE_COUNT(old) == 1) {
3828 * Unwire a page without (re-)inserting it into a page queue. It is up
3829 * to the caller to enqueue, requeue, or free the page as appropriate.
3830 * In most cases involving managed pages, vm_page_unwire() should be used
3834 vm_page_unwire_noq(vm_page_t m)
3838 old = vm_page_drop(m, 1);
3839 KASSERT(VPRC_WIRE_COUNT(old) != 0,
3840 ("vm_page_unref: counter underflow for page %p", m));
3841 KASSERT((m->flags & PG_FICTITIOUS) == 0 || VPRC_WIRE_COUNT(old) > 1,
3842 ("vm_page_unref: missing ref on fictitious page %p", m));
3844 if (VPRC_WIRE_COUNT(old) > 1)
3851 * Ensure that the page is in the specified page queue. If the page is
3852 * active or being moved to the active queue, ensure that its act_count is
3853 * at least ACT_INIT but do not otherwise mess with it. Otherwise, ensure that
3854 * the page is at the tail of its page queue.
3856 * The page may be wired. The caller should release its wiring reference
3857 * before releasing the page lock, otherwise the page daemon may immediately
3860 * A managed page must be locked.
3862 static __always_inline void
3863 vm_page_mvqueue(vm_page_t m, const uint8_t nqueue)
3866 vm_page_assert_locked(m);
3867 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3868 ("vm_page_mvqueue: page %p is unmanaged", m));
3870 if (vm_page_queue(m) != nqueue) {
3872 vm_page_enqueue(m, nqueue);
3873 } else if (nqueue != PQ_ACTIVE) {
3877 if (nqueue == PQ_ACTIVE && m->act_count < ACT_INIT)
3878 m->act_count = ACT_INIT;
3882 * Put the specified page on the active list (if appropriate).
3885 vm_page_activate(vm_page_t m)
3888 if ((m->oflags & VPO_UNMANAGED) != 0 || vm_page_wired(m))
3890 vm_page_mvqueue(m, PQ_ACTIVE);
3894 * Move the specified page to the tail of the inactive queue, or requeue
3895 * the page if it is already in the inactive queue.
3898 vm_page_deactivate(vm_page_t m)
3901 if ((m->oflags & VPO_UNMANAGED) != 0 || vm_page_wired(m))
3903 vm_page_mvqueue(m, PQ_INACTIVE);
3907 * Move the specified page close to the head of the inactive queue,
3908 * bypassing LRU. A marker page is used to maintain FIFO ordering.
3909 * As with regular enqueues, we use a per-CPU batch queue to reduce
3910 * contention on the page queue lock.
3913 _vm_page_deactivate_noreuse(vm_page_t m)
3916 vm_page_assert_locked(m);
3918 if (!vm_page_inactive(m)) {
3920 m->queue = PQ_INACTIVE;
3922 if ((m->aflags & PGA_REQUEUE_HEAD) == 0)
3923 vm_page_aflag_set(m, PGA_REQUEUE_HEAD);
3924 vm_page_pqbatch_submit(m, PQ_INACTIVE);
3928 vm_page_deactivate_noreuse(vm_page_t m)
3931 KASSERT(m->object != NULL,
3932 ("vm_page_deactivate_noreuse: page %p has no object", m));
3934 if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_wired(m))
3935 _vm_page_deactivate_noreuse(m);
3939 * Put a page in the laundry, or requeue it if it is already there.
3942 vm_page_launder(vm_page_t m)
3945 if ((m->oflags & VPO_UNMANAGED) != 0 || vm_page_wired(m))
3947 vm_page_mvqueue(m, PQ_LAUNDRY);
3951 * Put a page in the PQ_UNSWAPPABLE holding queue.
3954 vm_page_unswappable(vm_page_t m)
3957 vm_page_assert_locked(m);
3958 KASSERT(!vm_page_wired(m) && (m->oflags & VPO_UNMANAGED) == 0,
3959 ("page %p already unswappable", m));
3962 vm_page_enqueue(m, PQ_UNSWAPPABLE);
3966 vm_page_release_toq(vm_page_t m, int flags)
3969 vm_page_assert_locked(m);
3972 * Use a check of the valid bits to determine whether we should
3973 * accelerate reclamation of the page. The object lock might not be
3974 * held here, in which case the check is racy. At worst we will either
3975 * accelerate reclamation of a valid page and violate LRU, or
3976 * unnecessarily defer reclamation of an invalid page.
3978 * If we were asked to not cache the page, place it near the head of the
3979 * inactive queue so that is reclaimed sooner.
3981 if ((flags & (VPR_TRYFREE | VPR_NOREUSE)) != 0 || m->valid == 0)
3982 _vm_page_deactivate_noreuse(m);
3983 else if (vm_page_active(m))
3984 vm_page_reference(m);
3986 vm_page_mvqueue(m, PQ_INACTIVE);
3990 * Unwire a page and either attempt to free it or re-add it to the page queues.
3993 vm_page_release(vm_page_t m, int flags)
3999 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
4000 ("vm_page_release: page %p is unmanaged", m));
4002 if ((flags & VPR_TRYFREE) != 0) {
4004 object = (vm_object_t)atomic_load_ptr(&m->object);
4007 /* Depends on type-stability. */
4008 if (vm_page_busied(m) || !VM_OBJECT_TRYWLOCK(object)) {
4012 if (object == m->object)
4014 VM_OBJECT_WUNLOCK(object);
4016 if (__predict_true(object != NULL)) {
4017 vm_page_release_locked(m, flags);
4018 VM_OBJECT_WUNLOCK(object);
4024 * Update LRU state before releasing the wiring reference.
4025 * Use a release store when updating the reference count to
4026 * synchronize with vm_page_free_prep().
4031 KASSERT(VPRC_WIRE_COUNT(old) > 0,
4032 ("vm_page_unwire: wire count underflow for page %p", m));
4033 if (!locked && VPRC_WIRE_COUNT(old) == 1) {
4036 vm_page_release_toq(m, flags);
4038 } while (!atomic_fcmpset_rel_int(&m->ref_count, &old, old - 1));
4041 * Release the lock only after the wiring is released, to ensure that
4042 * the page daemon does not encounter and dequeue the page while it is
4048 if (VPRC_WIRE_COUNT(old) == 1) {
4055 /* See vm_page_release(). */
4057 vm_page_release_locked(vm_page_t m, int flags)
4060 VM_OBJECT_ASSERT_WLOCKED(m->object);
4061 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
4062 ("vm_page_release_locked: page %p is unmanaged", m));
4064 if (vm_page_unwire_noq(m)) {
4065 if ((flags & VPR_TRYFREE) != 0 &&
4066 (m->object->ref_count == 0 || !pmap_page_is_mapped(m)) &&
4067 m->dirty == 0 && !vm_page_busied(m)) {
4071 vm_page_release_toq(m, flags);
4078 vm_page_try_blocked_op(vm_page_t m, void (*op)(vm_page_t))
4082 KASSERT(m->object != NULL && (m->oflags & VPO_UNMANAGED) == 0,
4083 ("vm_page_try_blocked_op: page %p has no object", m));
4084 KASSERT(vm_page_busied(m),
4085 ("vm_page_try_blocked_op: page %p is not busy", m));
4086 VM_OBJECT_ASSERT_LOCKED(m->object);
4091 ("vm_page_try_blocked_op: page %p has no references", m));
4092 if (VPRC_WIRE_COUNT(old) != 0)
4094 } while (!atomic_fcmpset_int(&m->ref_count, &old, old | VPRC_BLOCKED));
4099 * If the object is read-locked, new wirings may be created via an
4102 old = vm_page_drop(m, VPRC_BLOCKED);
4103 KASSERT(!VM_OBJECT_WOWNED(m->object) ||
4104 old == (VPRC_BLOCKED | VPRC_OBJREF),
4105 ("vm_page_try_blocked_op: unexpected refcount value %u for %p",
4111 * Atomically check for wirings and remove all mappings of the page.
4114 vm_page_try_remove_all(vm_page_t m)
4117 return (vm_page_try_blocked_op(m, pmap_remove_all));
4121 * Atomically check for wirings and remove all writeable mappings of the page.
4124 vm_page_try_remove_write(vm_page_t m)
4127 return (vm_page_try_blocked_op(m, pmap_remove_write));
4133 * Apply the specified advice to the given page.
4135 * The object and page must be locked.
4138 vm_page_advise(vm_page_t m, int advice)
4141 vm_page_assert_locked(m);
4142 VM_OBJECT_ASSERT_WLOCKED(m->object);
4143 if (advice == MADV_FREE)
4145 * Mark the page clean. This will allow the page to be freed
4146 * without first paging it out. MADV_FREE pages are often
4147 * quickly reused by malloc(3), so we do not do anything that
4148 * would result in a page fault on a later access.
4151 else if (advice != MADV_DONTNEED) {
4152 if (advice == MADV_WILLNEED)
4153 vm_page_activate(m);
4158 * Clear any references to the page. Otherwise, the page daemon will
4159 * immediately reactivate the page.
4161 vm_page_aflag_clear(m, PGA_REFERENCED);
4163 if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
4167 * Place clean pages near the head of the inactive queue rather than
4168 * the tail, thus defeating the queue's LRU operation and ensuring that
4169 * the page will be reused quickly. Dirty pages not already in the
4170 * laundry are moved there.
4173 vm_page_deactivate_noreuse(m);
4174 else if (!vm_page_in_laundry(m))
4179 * Grab a page, waiting until we are waken up due to the page
4180 * changing state. We keep on waiting, if the page continues
4181 * to be in the object. If the page doesn't exist, first allocate it
4182 * and then conditionally zero it.
4184 * This routine may sleep.
4186 * The object must be locked on entry. The lock will, however, be released
4187 * and reacquired if the routine sleeps.
4190 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
4196 VM_OBJECT_ASSERT_WLOCKED(object);
4197 KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
4198 (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
4199 ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
4200 pflags = allocflags &
4201 ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL |
4203 if ((allocflags & VM_ALLOC_NOWAIT) == 0)
4204 pflags |= VM_ALLOC_WAITFAIL;
4205 if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0)
4206 pflags |= VM_ALLOC_SBUSY;
4208 if ((m = vm_page_lookup(object, pindex)) != NULL) {
4209 if ((allocflags & (VM_ALLOC_IGN_SBUSY | VM_ALLOC_SBUSY)) != 0)
4210 sleep = !vm_page_trysbusy(m);
4212 sleep = !vm_page_tryxbusy(m);
4214 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
4217 * Reference the page before unlocking and
4218 * sleeping so that the page daemon is less
4219 * likely to reclaim it.
4221 if ((allocflags & VM_ALLOC_NOCREAT) == 0)
4222 vm_page_aflag_set(m, PGA_REFERENCED);
4223 vm_page_busy_sleep(m, "pgrbwt", (allocflags &
4224 VM_ALLOC_IGN_SBUSY) != 0);
4225 VM_OBJECT_WLOCK(object);
4226 if ((allocflags & VM_ALLOC_WAITFAIL) != 0)
4230 if ((allocflags & VM_ALLOC_WIRED) != 0)
4235 if ((allocflags & VM_ALLOC_NOCREAT) != 0)
4237 m = vm_page_alloc(object, pindex, pflags);
4239 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
4243 if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
4247 if ((allocflags & VM_ALLOC_NOBUSY) != 0) {
4248 if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0)
4257 * Grab a page and make it valid, paging in if necessary. Pages missing from
4258 * their pager are zero filled and validated.
4261 vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex, int allocflags)
4268 KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
4269 (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
4270 ("vm_page_grab_valid: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
4271 KASSERT((allocflags &
4272 (VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL | VM_ALLOC_ZERO)) == 0,
4273 ("vm_page_grab_valid: Invalid flags 0x%X", allocflags));
4274 VM_OBJECT_ASSERT_WLOCKED(object);
4275 pflags = allocflags & ~(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY);
4276 pflags |= VM_ALLOC_WAITFAIL;
4280 if ((m = vm_page_lookup(object, pindex)) != NULL) {
4282 * If the page is fully valid it can only become invalid
4283 * with the object lock held. If it is not valid it can
4284 * become valid with the busy lock held. Therefore, we
4285 * may unnecessarily lock the exclusive busy here if we
4286 * race with I/O completion not using the object lock.
4287 * However, we will not end up with an invalid page and a
4290 if (m->valid != VM_PAGE_BITS_ALL ||
4291 (allocflags & (VM_ALLOC_IGN_SBUSY | VM_ALLOC_SBUSY)) == 0) {
4292 sleep = !vm_page_tryxbusy(m);
4295 sleep = !vm_page_trysbusy(m);
4298 * Reference the page before unlocking and
4299 * sleeping so that the page daemon is less
4300 * likely to reclaim it.
4302 if ((allocflags & VM_ALLOC_NOCREAT) == 0)
4303 vm_page_aflag_set(m, PGA_REFERENCED);
4304 vm_page_busy_sleep(m, "pgrbwt", (allocflags &
4305 VM_ALLOC_IGN_SBUSY) != 0);
4306 VM_OBJECT_WLOCK(object);
4309 if ((allocflags & VM_ALLOC_NOCREAT) != 0 &&
4310 m->valid != VM_PAGE_BITS_ALL) {
4316 return (VM_PAGER_FAIL);
4318 if ((allocflags & VM_ALLOC_WIRED) != 0)
4320 if (m->valid == VM_PAGE_BITS_ALL)
4322 } else if ((allocflags & VM_ALLOC_NOCREAT) != 0) {
4324 return (VM_PAGER_FAIL);
4325 } else if ((m = vm_page_alloc(object, pindex, pflags)) != NULL) {
4331 vm_page_assert_xbusied(m);
4333 if (vm_pager_has_page(object, pindex, NULL, NULL)) {
4334 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
4335 if (rv != VM_PAGER_OK) {
4336 if (allocflags & VM_ALLOC_WIRED)
4337 vm_page_unwire_noq(m);
4342 MPASS(m->valid == VM_PAGE_BITS_ALL);
4344 vm_page_zero_invalid(m, TRUE);
4347 if ((allocflags & VM_ALLOC_NOBUSY) != 0) {
4353 if ((allocflags & VM_ALLOC_SBUSY) != 0 && xbusy)
4354 vm_page_busy_downgrade(m);
4356 return (VM_PAGER_OK);
4360 * Return the specified range of pages from the given object. For each
4361 * page offset within the range, if a page already exists within the object
4362 * at that offset and it is busy, then wait for it to change state. If,
4363 * instead, the page doesn't exist, then allocate it.
4365 * The caller must always specify an allocation class.
4367 * allocation classes:
4368 * VM_ALLOC_NORMAL normal process request
4369 * VM_ALLOC_SYSTEM system *really* needs the pages
4371 * The caller must always specify that the pages are to be busied and/or
4374 * optional allocation flags:
4375 * VM_ALLOC_IGN_SBUSY do not sleep on soft busy pages
4376 * VM_ALLOC_NOBUSY do not exclusive busy the page
4377 * VM_ALLOC_NOWAIT do not sleep
4378 * VM_ALLOC_SBUSY set page to sbusy state
4379 * VM_ALLOC_WIRED wire the pages
4380 * VM_ALLOC_ZERO zero and validate any invalid pages
4382 * If VM_ALLOC_NOWAIT is not specified, this routine may sleep. Otherwise, it
4383 * may return a partial prefix of the requested range.
4386 vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
4387 vm_page_t *ma, int count)
4394 VM_OBJECT_ASSERT_WLOCKED(object);
4395 KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0,
4396 ("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed"));
4397 KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 ||
4398 (allocflags & VM_ALLOC_WIRED) != 0,
4399 ("vm_page_grab_pages: the pages must be busied or wired"));
4400 KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
4401 (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
4402 ("vm_page_grab_pages: VM_ALLOC_SBUSY/IGN_SBUSY mismatch"));
4405 pflags = allocflags &
4406 ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL |
4408 if ((allocflags & VM_ALLOC_NOWAIT) == 0)
4409 pflags |= VM_ALLOC_WAITFAIL;
4410 if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0)
4411 pflags |= VM_ALLOC_SBUSY;
4414 m = vm_radix_lookup_le(&object->rtree, pindex + i);
4415 if (m == NULL || m->pindex != pindex + i) {
4419 mpred = TAILQ_PREV(m, pglist, listq);
4420 for (; i < count; i++) {
4423 (VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY)) != 0)
4424 sleep = !vm_page_trysbusy(m);
4426 sleep = !vm_page_tryxbusy(m);
4428 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
4431 * Reference the page before unlocking and
4432 * sleeping so that the page daemon is less
4433 * likely to reclaim it.
4435 if ((allocflags & VM_ALLOC_NOCREAT) == 0)
4436 vm_page_aflag_set(m, PGA_REFERENCED);
4437 vm_page_busy_sleep(m, "grbmaw", (allocflags &
4438 VM_ALLOC_IGN_SBUSY) != 0);
4439 VM_OBJECT_WLOCK(object);
4442 if ((allocflags & VM_ALLOC_WIRED) != 0)
4445 if ((allocflags & VM_ALLOC_NOCREAT) != 0)
4447 m = vm_page_alloc_after(object, pindex + i,
4448 pflags | VM_ALLOC_COUNT(count - i), mpred);
4450 if ((allocflags & VM_ALLOC_NOWAIT) != 0)
4455 if (m->valid == 0 && (allocflags & VM_ALLOC_ZERO) != 0) {
4456 if ((m->flags & PG_ZERO) == 0)
4458 m->valid = VM_PAGE_BITS_ALL;
4460 if ((allocflags & VM_ALLOC_NOBUSY) != 0) {
4461 if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0)
4467 m = vm_page_next(m);
4473 * Mapping function for valid or dirty bits in a page.
4475 * Inputs are required to range within a page.
4478 vm_page_bits(int base, int size)
4484 base + size <= PAGE_SIZE,
4485 ("vm_page_bits: illegal base/size %d/%d", base, size)
4488 if (size == 0) /* handle degenerate case */
4491 first_bit = base >> DEV_BSHIFT;
4492 last_bit = (base + size - 1) >> DEV_BSHIFT;
4494 return (((vm_page_bits_t)2 << last_bit) -
4495 ((vm_page_bits_t)1 << first_bit));
4499 * vm_page_set_valid_range:
4501 * Sets portions of a page valid. The arguments are expected
4502 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
4503 * of any partial chunks touched by the range. The invalid portion of
4504 * such chunks will be zeroed.
4506 * (base + size) must be less then or equal to PAGE_SIZE.
4509 vm_page_set_valid_range(vm_page_t m, int base, int size)
4513 VM_OBJECT_ASSERT_WLOCKED(m->object);
4514 if (size == 0) /* handle degenerate case */
4518 * If the base is not DEV_BSIZE aligned and the valid
4519 * bit is clear, we have to zero out a portion of the
4522 if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
4523 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
4524 pmap_zero_page_area(m, frag, base - frag);
4527 * If the ending offset is not DEV_BSIZE aligned and the
4528 * valid bit is clear, we have to zero out a portion of
4531 endoff = base + size;
4532 if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
4533 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
4534 pmap_zero_page_area(m, endoff,
4535 DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
4538 * Assert that no previously invalid block that is now being validated
4541 KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
4542 ("vm_page_set_valid_range: page %p is dirty", m));
4545 * Set valid bits inclusive of any overlap.
4547 m->valid |= vm_page_bits(base, size);
4551 * Clear the given bits from the specified page's dirty field.
4553 static __inline void
4554 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
4557 #if PAGE_SIZE < 16384
4562 * If the object is locked and the page is neither exclusive busy nor
4563 * write mapped, then the page's dirty field cannot possibly be
4564 * set by a concurrent pmap operation.
4566 VM_OBJECT_ASSERT_WLOCKED(m->object);
4567 if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
4568 m->dirty &= ~pagebits;
4571 * The pmap layer can call vm_page_dirty() without
4572 * holding a distinguished lock. The combination of
4573 * the object's lock and an atomic operation suffice
4574 * to guarantee consistency of the page dirty field.
4576 * For PAGE_SIZE == 32768 case, compiler already
4577 * properly aligns the dirty field, so no forcible
4578 * alignment is needed. Only require existence of
4579 * atomic_clear_64 when page size is 32768.
4581 addr = (uintptr_t)&m->dirty;
4582 #if PAGE_SIZE == 32768
4583 atomic_clear_64((uint64_t *)addr, pagebits);
4584 #elif PAGE_SIZE == 16384
4585 atomic_clear_32((uint32_t *)addr, pagebits);
4586 #else /* PAGE_SIZE <= 8192 */
4588 * Use a trick to perform a 32-bit atomic on the
4589 * containing aligned word, to not depend on the existence
4590 * of atomic_clear_{8, 16}.
4592 shift = addr & (sizeof(uint32_t) - 1);
4593 #if BYTE_ORDER == BIG_ENDIAN
4594 shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
4598 addr &= ~(sizeof(uint32_t) - 1);
4599 atomic_clear_32((uint32_t *)addr, pagebits << shift);
4600 #endif /* PAGE_SIZE */
4605 * vm_page_set_validclean:
4607 * Sets portions of a page valid and clean. The arguments are expected
4608 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
4609 * of any partial chunks touched by the range. The invalid portion of
4610 * such chunks will be zero'd.
4612 * (base + size) must be less then or equal to PAGE_SIZE.
4615 vm_page_set_validclean(vm_page_t m, int base, int size)
4617 vm_page_bits_t oldvalid, pagebits;
4620 VM_OBJECT_ASSERT_WLOCKED(m->object);
4621 if (size == 0) /* handle degenerate case */
4625 * If the base is not DEV_BSIZE aligned and the valid
4626 * bit is clear, we have to zero out a portion of the
4629 if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
4630 (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
4631 pmap_zero_page_area(m, frag, base - frag);
4634 * If the ending offset is not DEV_BSIZE aligned and the
4635 * valid bit is clear, we have to zero out a portion of
4638 endoff = base + size;
4639 if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
4640 (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
4641 pmap_zero_page_area(m, endoff,
4642 DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
4645 * Set valid, clear dirty bits. If validating the entire
4646 * page we can safely clear the pmap modify bit. We also
4647 * use this opportunity to clear the VPO_NOSYNC flag. If a process
4648 * takes a write fault on a MAP_NOSYNC memory area the flag will
4651 * We set valid bits inclusive of any overlap, but we can only
4652 * clear dirty bits for DEV_BSIZE chunks that are fully within
4655 oldvalid = m->valid;
4656 pagebits = vm_page_bits(base, size);
4657 m->valid |= pagebits;
4659 if ((frag = base & (DEV_BSIZE - 1)) != 0) {
4660 frag = DEV_BSIZE - frag;
4666 pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
4668 if (base == 0 && size == PAGE_SIZE) {
4670 * The page can only be modified within the pmap if it is
4671 * mapped, and it can only be mapped if it was previously
4674 if (oldvalid == VM_PAGE_BITS_ALL)
4676 * Perform the pmap_clear_modify() first. Otherwise,
4677 * a concurrent pmap operation, such as
4678 * pmap_protect(), could clear a modification in the
4679 * pmap and set the dirty field on the page before
4680 * pmap_clear_modify() had begun and after the dirty
4681 * field was cleared here.
4683 pmap_clear_modify(m);
4685 m->oflags &= ~VPO_NOSYNC;
4686 } else if (oldvalid != VM_PAGE_BITS_ALL)
4687 m->dirty &= ~pagebits;
4689 vm_page_clear_dirty_mask(m, pagebits);
4693 vm_page_clear_dirty(vm_page_t m, int base, int size)
4696 vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
4700 * vm_page_set_invalid:
4702 * Invalidates DEV_BSIZE'd chunks within a page. Both the
4703 * valid and dirty bits for the effected areas are cleared.
4706 vm_page_set_invalid(vm_page_t m, int base, int size)
4708 vm_page_bits_t bits;
4712 VM_OBJECT_ASSERT_WLOCKED(object);
4713 if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
4714 size >= object->un_pager.vnp.vnp_size)
4715 bits = VM_PAGE_BITS_ALL;
4717 bits = vm_page_bits(base, size);
4718 if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
4721 KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
4722 !pmap_page_is_mapped(m),
4723 ("vm_page_set_invalid: page %p is mapped", m));
4729 * vm_page_zero_invalid()
4731 * The kernel assumes that the invalid portions of a page contain
4732 * garbage, but such pages can be mapped into memory by user code.
4733 * When this occurs, we must zero out the non-valid portions of the
4734 * page so user code sees what it expects.
4736 * Pages are most often semi-valid when the end of a file is mapped
4737 * into memory and the file's size is not page aligned.
4740 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
4745 VM_OBJECT_ASSERT_WLOCKED(m->object);
4747 * Scan the valid bits looking for invalid sections that
4748 * must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the
4749 * valid bit may be set ) have already been zeroed by
4750 * vm_page_set_validclean().
4752 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
4753 if (i == (PAGE_SIZE / DEV_BSIZE) ||
4754 (m->valid & ((vm_page_bits_t)1 << i))) {
4756 pmap_zero_page_area(m,
4757 b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
4764 * setvalid is TRUE when we can safely set the zero'd areas
4765 * as being valid. We can do this if there are no cache consistancy
4766 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
4769 m->valid = VM_PAGE_BITS_ALL;
4775 * Is (partial) page valid? Note that the case where size == 0
4776 * will return FALSE in the degenerate case where the page is
4777 * entirely invalid, and TRUE otherwise.
4780 vm_page_is_valid(vm_page_t m, int base, int size)
4782 vm_page_bits_t bits;
4784 VM_OBJECT_ASSERT_LOCKED(m->object);
4785 bits = vm_page_bits(base, size);
4786 return (m->valid != 0 && (m->valid & bits) == bits);
4790 * Returns true if all of the specified predicates are true for the entire
4791 * (super)page and false otherwise.
4794 vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m)
4800 if (skip_m != NULL && skip_m->object != object)
4802 VM_OBJECT_ASSERT_LOCKED(object);
4803 npages = atop(pagesizes[m->psind]);
4806 * The physically contiguous pages that make up a superpage, i.e., a
4807 * page with a page size index ("psind") greater than zero, will
4808 * occupy adjacent entries in vm_page_array[].
4810 for (i = 0; i < npages; i++) {
4811 /* Always test object consistency, including "skip_m". */
4812 if (m[i].object != object)
4814 if (&m[i] == skip_m)
4816 if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i]))
4818 if ((flags & PS_ALL_DIRTY) != 0) {
4820 * Calling vm_page_test_dirty() or pmap_is_modified()
4821 * might stop this case from spuriously returning
4822 * "false". However, that would require a write lock
4823 * on the object containing "m[i]".
4825 if (m[i].dirty != VM_PAGE_BITS_ALL)
4828 if ((flags & PS_ALL_VALID) != 0 &&
4829 m[i].valid != VM_PAGE_BITS_ALL)
4836 * Set the page's dirty bits if the page is modified.
4839 vm_page_test_dirty(vm_page_t m)
4842 VM_OBJECT_ASSERT_WLOCKED(m->object);
4843 if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
4848 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
4851 mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
4855 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
4858 mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
4862 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
4865 return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
4868 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
4870 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
4873 vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
4877 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
4880 mtx_assert_(vm_page_lockptr(m), a, file, line);
4886 vm_page_object_lock_assert(vm_page_t m)
4890 * Certain of the page's fields may only be modified by the
4891 * holder of the containing object's lock or the exclusive busy.
4892 * holder. Unfortunately, the holder of the write busy is
4893 * not recorded, and thus cannot be checked here.
4895 if (m->object != NULL && !vm_page_xbusied(m))
4896 VM_OBJECT_ASSERT_WLOCKED(m->object);
4900 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
4903 if ((bits & PGA_WRITEABLE) == 0)
4907 * The PGA_WRITEABLE flag can only be set if the page is
4908 * managed, is exclusively busied or the object is locked.
4909 * Currently, this flag is only set by pmap_enter().
4911 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
4912 ("PGA_WRITEABLE on unmanaged page"));
4913 if (!vm_page_xbusied(m))
4914 VM_OBJECT_ASSERT_LOCKED(m->object);
4918 #include "opt_ddb.h"
4920 #include <sys/kernel.h>
4922 #include <ddb/ddb.h>
4924 DB_SHOW_COMMAND(page, vm_page_print_page_info)
4927 db_printf("vm_cnt.v_free_count: %d\n", vm_free_count());
4928 db_printf("vm_cnt.v_inactive_count: %d\n", vm_inactive_count());
4929 db_printf("vm_cnt.v_active_count: %d\n", vm_active_count());
4930 db_printf("vm_cnt.v_laundry_count: %d\n", vm_laundry_count());
4931 db_printf("vm_cnt.v_wire_count: %d\n", vm_wire_count());
4932 db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
4933 db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
4934 db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
4935 db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
4938 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
4942 db_printf("pq_free %d\n", vm_free_count());
4943 for (dom = 0; dom < vm_ndomains; dom++) {
4945 "dom %d page_cnt %d free %d pq_act %d pq_inact %d pq_laund %d pq_unsw %d\n",
4947 vm_dom[dom].vmd_page_count,
4948 vm_dom[dom].vmd_free_count,
4949 vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
4950 vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
4951 vm_dom[dom].vmd_pagequeues[PQ_LAUNDRY].pq_cnt,
4952 vm_dom[dom].vmd_pagequeues[PQ_UNSWAPPABLE].pq_cnt);
4956 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
4959 boolean_t phys, virt;
4962 db_printf("show pginfo addr\n");
4966 phys = strchr(modif, 'p') != NULL;
4967 virt = strchr(modif, 'v') != NULL;
4969 m = PHYS_TO_VM_PAGE(pmap_kextract(addr));
4971 m = PHYS_TO_VM_PAGE(addr);
4973 m = (vm_page_t)addr;
4975 "page %p obj %p pidx 0x%jx phys 0x%jx q %d ref %u\n"
4976 " af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
4977 m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
4978 m->queue, m->ref_count, m->aflags, m->oflags,
4979 m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);