kernel - Make VM fault waits in low memory the same as other low memory waits
[dragonfly.git] / sys / vm / vm_page.c
CommitLineData
984263bc 1/*
9ad0147b
MD
2 * (MPSAFE)
3 *
984263bc
MD
4 * Copyright (c) 1991 Regents of the University of California.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
984263bc
MD
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
35 * $FreeBSD: src/sys/vm/vm_page.c,v 1.147.2.18 2002/03/10 05:03:19 alc Exp $
36 */
37
38/*
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
41 *
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43 *
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.
49 *
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.
53 *
54 * Carnegie Mellon requests users of this software to return to
55 *
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
60 *
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
63 */
984263bc 64/*
de71fd3f
MD
65 * Resident memory management module. The module manipulates 'VM pages'.
66 * A VM page is the core building block for memory management.
984263bc
MD
67 */
68
69#include <sys/param.h>
70#include <sys/systm.h>
71#include <sys/malloc.h>
72#include <sys/proc.h>
73#include <sys/vmmeter.h>
74#include <sys/vnode.h>
cd3c66bd 75#include <sys/kernel.h>
79d182b0
MD
76#include <sys/alist.h>
77#include <sys/sysctl.h>
984263bc
MD
78
79#include <vm/vm.h>
80#include <vm/vm_param.h>
81#include <sys/lock.h>
82#include <vm/vm_kern.h>
83#include <vm/pmap.h>
84#include <vm/vm_map.h>
85#include <vm/vm_object.h>
86#include <vm/vm_page.h>
87#include <vm/vm_pageout.h>
88#include <vm/vm_pager.h>
89#include <vm/vm_extern.h>
096e95c0 90#include <vm/swap_pager.h>
984263bc 91
480c83b6 92#include <machine/inttypes.h>
8e5e6f1b
AH
93#include <machine/md_var.h>
94
bb6811be 95#include <vm/vm_page2.h>
b12defdc 96#include <sys/spinlock2.h>
bb6811be 97
906c754c
MD
98#define VMACTION_HSIZE 256
99#define VMACTION_HMASK (VMACTION_HSIZE - 1)
100
de71fd3f
MD
101static void vm_page_queue_init(void);
102static void vm_page_free_wakeup(void);
85946b6c 103static vm_page_t vm_page_select_cache(u_short pg_color);
74232d8e 104static vm_page_t _vm_page_list_find2(int basequeue, int index);
b12defdc 105static void _vm_page_deactivate_locked(vm_page_t m, int athead);
984263bc 106
b12defdc
MD
107/*
108 * Array of tailq lists
109 */
110__cachealign struct vpgqueues vm_page_queues[PQ_COUNT];
984263bc 111
906c754c
MD
112LIST_HEAD(vm_page_action_list, vm_page_action);
113struct vm_page_action_list action_list[VMACTION_HSIZE];
cd3c66bd 114static volatile int vm_pages_waiting;
906c754c 115
79d182b0
MD
116static struct alist vm_contig_alist;
117static struct almeta vm_contig_ameta[ALIST_RECORDS_65536];
118static struct spinlock vm_contig_spin = SPINLOCK_INITIALIZER(&vm_contig_spin);
119
120static u_long vm_dma_reserved = 0;
121TUNABLE_ULONG("vm.dma_reserved", &vm_dma_reserved);
122SYSCTL_ULONG(_vm, OID_AUTO, dma_reserved, CTLFLAG_RD, &vm_dma_reserved, 0,
123 "Memory reserved for DMA");
124SYSCTL_UINT(_vm, OID_AUTO, dma_free_pages, CTLFLAG_RD,
125 &vm_contig_alist.bl_free, 0, "Memory reserved for DMA");
906c754c 126
ef67e7a3
SZ
127static int vm_contig_verbose = 0;
128TUNABLE_INT("vm.contig_verbose", &vm_contig_verbose);
129
1f804340
MD
130RB_GENERATE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare,
131 vm_pindex_t, pindex);
132
984263bc 133static void
de71fd3f
MD
134vm_page_queue_init(void)
135{
984263bc
MD
136 int i;
137
de71fd3f 138 for (i = 0; i < PQ_L2_SIZE; i++)
12e4aaff 139 vm_page_queues[PQ_FREE+i].cnt = &vmstats.v_free_count;
de71fd3f
MD
140 for (i = 0; i < PQ_L2_SIZE; i++)
141 vm_page_queues[PQ_CACHE+i].cnt = &vmstats.v_cache_count;
027193eb
MD
142 for (i = 0; i < PQ_L2_SIZE; i++)
143 vm_page_queues[PQ_INACTIVE+i].cnt = &vmstats.v_inactive_count;
144 for (i = 0; i < PQ_L2_SIZE; i++)
145 vm_page_queues[PQ_ACTIVE+i].cnt = &vmstats.v_active_count;
146 for (i = 0; i < PQ_L2_SIZE; i++)
147 vm_page_queues[PQ_HOLD+i].cnt = &vmstats.v_active_count;
de71fd3f
MD
148 /* PQ_NONE has no queue */
149
b12defdc 150 for (i = 0; i < PQ_COUNT; i++) {
984263bc 151 TAILQ_INIT(&vm_page_queues[i].pl);
b12defdc
MD
152 spin_init(&vm_page_queues[i].spin);
153 }
906c754c
MD
154
155 for (i = 0; i < VMACTION_HSIZE; i++)
156 LIST_INIT(&action_list[i]);
984263bc
MD
157}
158
de71fd3f
MD
159/*
160 * note: place in initialized data section? Is this necessary?
161 */
984263bc 162long first_page = 0;
de71fd3f 163int vm_page_array_size = 0;
984263bc 164int vm_page_zero_count = 0;
79d182b0
MD
165vm_page_t vm_page_array = NULL;
166vm_paddr_t vm_low_phys_reserved;
984263bc
MD
167
168/*
de71fd3f 169 * (low level boot)
984263bc 170 *
de71fd3f
MD
171 * Sets the page size, perhaps based upon the memory size.
172 * Must be called before any use of page-size dependent functions.
984263bc
MD
173 */
174void
175vm_set_page_size(void)
176{
12e4aaff
MD
177 if (vmstats.v_page_size == 0)
178 vmstats.v_page_size = PAGE_SIZE;
179 if (((vmstats.v_page_size - 1) & vmstats.v_page_size) != 0)
984263bc
MD
180 panic("vm_set_page_size: page size not a power of two");
181}
182
183/*
de71fd3f 184 * (low level boot)
984263bc 185 *
de71fd3f
MD
186 * Add a new page to the freelist for use by the system. New pages
187 * are added to both the head and tail of the associated free page
188 * queue in a bottom-up fashion, so both zero'd and non-zero'd page
189 * requests pull 'recent' adds (higher physical addresses) first.
161399b3 190 *
bc3396b8
MD
191 * Beware that the page zeroing daemon will also be running soon after
192 * boot, moving pages from the head to the tail of the PQ_FREE queues.
193 *
654a39f0 194 * Must be called in a critical section.
984263bc 195 */
79d182b0 196static void
6ef943a3 197vm_add_new_page(vm_paddr_t pa)
984263bc 198{
161399b3 199 struct vpgqueues *vpq;
de71fd3f 200 vm_page_t m;
984263bc 201
984263bc
MD
202 m = PHYS_TO_VM_PAGE(pa);
203 m->phys_addr = pa;
204 m->flags = 0;
205 m->pc = (pa >> PAGE_SHIFT) & PQ_L2_MASK;
85946b6c 206 /*
bc3396b8
MD
207 * Twist for cpu localization in addition to page coloring, so
208 * different cpus selecting by m->queue get different page colors.
85946b6c
MD
209 */
210 m->pc ^= ((pa >> PAGE_SHIFT) / PQ_L2_SIZE) & PQ_L2_MASK;
211 m->pc ^= ((pa >> PAGE_SHIFT) / (PQ_L2_SIZE * PQ_L2_SIZE)) & PQ_L2_MASK;
79d182b0
MD
212 /*
213 * Reserve a certain number of contiguous low memory pages for
214 * contigmalloc() to use.
215 */
216 if (pa < vm_low_phys_reserved) {
217 atomic_add_int(&vmstats.v_page_count, 1);
218 atomic_add_int(&vmstats.v_dma_pages, 1);
219 m->queue = PQ_NONE;
220 m->wire_count = 1;
3ae0c654 221 atomic_add_int(&vmstats.v_wire_count, 1);
79d182b0
MD
222 alist_free(&vm_contig_alist, pa >> PAGE_SHIFT, 1);
223 return;
224 }
225
226 /*
227 * General page
228 */
984263bc 229 m->queue = m->pc + PQ_FREE;
26bcc0c0 230 KKASSERT(m->dirty == 0);
de71fd3f 231
b12defdc
MD
232 atomic_add_int(&vmstats.v_page_count, 1);
233 atomic_add_int(&vmstats.v_free_count, 1);
161399b3 234 vpq = &vm_page_queues[m->queue];
bc3396b8
MD
235 if ((vpq->flipflop & 15) == 0) {
236 pmap_zero_page(VM_PAGE_TO_PHYS(m));
237 m->flags |= PG_ZERO;
161399b3 238 TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
bc3396b8
MD
239 atomic_add_int(&vm_page_zero_count, 1);
240 } else {
161399b3 241 TAILQ_INSERT_HEAD(&vpq->pl, m, pageq);
bc3396b8
MD
242 }
243 ++vpq->flipflop;
b12defdc 244 ++vpq->lcnt;
984263bc
MD
245}
246
247/*
de71fd3f 248 * (low level boot)
984263bc 249 *
de71fd3f 250 * Initializes the resident memory module.
984263bc 251 *
da23a592
MD
252 * Preallocates memory for critical VM structures and arrays prior to
253 * kernel_map becoming available.
26bcc0c0 254 *
da23a592
MD
255 * Memory is allocated from (virtual2_start, virtual2_end) if available,
256 * otherwise memory is allocated from (virtual_start, virtual_end).
257 *
258 * On x86-64 (virtual_start, virtual_end) is only 2GB and may not be
259 * large enough to hold vm_page_array & other structures for machines with
260 * large amounts of ram, so we want to use virtual2* when available.
984263bc 261 */
da23a592
MD
262void
263vm_page_startup(void)
984263bc 264{
da23a592 265 vm_offset_t vaddr = virtual2_start ? virtual2_start : virtual_start;
984263bc 266 vm_offset_t mapped;
6ef943a3
MD
267 vm_size_t npages;
268 vm_paddr_t page_range;
269 vm_paddr_t new_end;
984263bc 270 int i;
6ef943a3 271 vm_paddr_t pa;
984263bc 272 int nblocks;
6ef943a3 273 vm_paddr_t last_pa;
6ef943a3
MD
274 vm_paddr_t end;
275 vm_paddr_t biggestone, biggestsize;
6ef943a3 276 vm_paddr_t total;
984263bc
MD
277
278 total = 0;
279 biggestsize = 0;
280 biggestone = 0;
281 nblocks = 0;
282 vaddr = round_page(vaddr);
283
284 for (i = 0; phys_avail[i + 1]; i += 2) {
aecf2182
MD
285 phys_avail[i] = round_page64(phys_avail[i]);
286 phys_avail[i + 1] = trunc_page64(phys_avail[i + 1]);
984263bc
MD
287 }
288
289 for (i = 0; phys_avail[i + 1]; i += 2) {
6ef943a3 290 vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
984263bc
MD
291
292 if (size > biggestsize) {
293 biggestone = i;
294 biggestsize = size;
295 }
296 ++nblocks;
297 total += size;
298 }
299
300 end = phys_avail[biggestone+1];
1f804340 301 end = trunc_page(end);
984263bc
MD
302
303 /*
304 * Initialize the queue headers for the free queue, the active queue
305 * and the inactive queue.
306 */
984263bc
MD
307 vm_page_queue_init();
308
6abe3bd0 309#if !defined(_KERNEL_VIRTUAL)
8e5e6f1b 310 /*
b12defdc
MD
311 * VKERNELs don't support minidumps and as such don't need
312 * vm_page_dump
313 *
8e5e6f1b
AH
314 * Allocate a bitmap to indicate that a random physical page
315 * needs to be included in a minidump.
316 *
317 * The amd64 port needs this to indicate which direct map pages
318 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
319 *
320 * However, i386 still needs this workspace internally within the
321 * minidump code. In theory, they are not needed on i386, but are
322 * included should the sf_buf code decide to use them.
323 */
324 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
325 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
326 end -= vm_page_dump_size;
327 vm_page_dump = (void *)pmap_map(&vaddr, end, end + vm_page_dump_size,
328 VM_PROT_READ | VM_PROT_WRITE);
329 bzero((void *)vm_page_dump, vm_page_dump_size);
6abe3bd0 330#endif
984263bc
MD
331 /*
332 * Compute the number of pages of memory that will be available for
333 * use (taking into account the overhead of a page structure per
334 * page).
335 */
984263bc 336 first_page = phys_avail[0] / PAGE_SIZE;
984263bc 337 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE - first_page;
1f804340 338 npages = (total - (page_range * sizeof(struct vm_page))) / PAGE_SIZE;
de71fd3f 339
79d182b0
MD
340#ifndef _KERNEL_VIRTUAL
341 /*
342 * (only applies to real kernels)
343 *
344 * Initialize the contiguous reserve map. We initially reserve up
345 * to 1/4 available physical memory or 65536 pages (~256MB), whichever
346 * is lower.
347 *
348 * Once device initialization is complete we return most of the
349 * reserved memory back to the normal page queues but leave some
350 * in reserve for things like usb attachments.
351 */
352 vm_low_phys_reserved = (vm_paddr_t)65536 << PAGE_SHIFT;
353 if (vm_low_phys_reserved > total / 4)
354 vm_low_phys_reserved = total / 4;
355 if (vm_dma_reserved == 0) {
356 vm_dma_reserved = 16 * 1024 * 1024; /* 16MB */
357 if (vm_dma_reserved > total / 16)
358 vm_dma_reserved = total / 16;
359 }
360#endif
361 alist_init(&vm_contig_alist, 65536, vm_contig_ameta,
362 ALIST_RECORDS_65536);
363
984263bc
MD
364 /*
365 * Initialize the mem entry structures now, and put them in the free
366 * queue.
367 */
984263bc 368 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
79d182b0 369 mapped = pmap_map(&vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE);
8e5e6f1b
AH
370 vm_page_array = (vm_page_t)mapped;
371
0e6594a8 372#if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
8e5e6f1b
AH
373 /*
374 * since pmap_map on amd64 returns stuff out of a direct-map region,
375 * we have to manually add these pages to the minidump tracking so
376 * that they can be dumped, including the vm_page_array.
377 */
378 for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
379 dump_add_page(pa);
8fdd3267 380#endif
984263bc
MD
381
382 /*
383 * Clear all of the page structures
384 */
385 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
386 vm_page_array_size = page_range;
387
388 /*
161399b3 389 * Construct the free queue(s) in ascending order (by physical
984263bc
MD
390 * address) so that the first 16MB of physical memory is allocated
391 * last rather than first. On large-memory machines, this avoids
392 * the exhaustion of low physical memory before isa_dmainit has run.
393 */
12e4aaff
MD
394 vmstats.v_page_count = 0;
395 vmstats.v_free_count = 0;
984263bc
MD
396 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) {
397 pa = phys_avail[i];
398 if (i == biggestone)
399 last_pa = new_end;
400 else
401 last_pa = phys_avail[i + 1];
402 while (pa < last_pa && npages-- > 0) {
403 vm_add_new_page(pa);
404 pa += PAGE_SIZE;
405 }
406 }
da23a592
MD
407 if (virtual2_start)
408 virtual2_start = vaddr;
409 else
410 virtual_start = vaddr;
984263bc
MD
411}
412
79d182b0
MD
413/*
414 * We tended to reserve a ton of memory for contigmalloc(). Now that most
415 * drivers have initialized we want to return most the remaining free
416 * reserve back to the VM page queues so they can be used for normal
417 * allocations.
418 *
419 * We leave vm_dma_reserved bytes worth of free pages in the reserve pool.
420 */
421static void
422vm_page_startup_finish(void *dummy __unused)
423{
424 alist_blk_t blk;
425 alist_blk_t rblk;
426 alist_blk_t count;
427 alist_blk_t xcount;
428 alist_blk_t bfree;
429 vm_page_t m;
430
431 spin_lock(&vm_contig_spin);
432 for (;;) {
433 bfree = alist_free_info(&vm_contig_alist, &blk, &count);
434 if (bfree <= vm_dma_reserved / PAGE_SIZE)
435 break;
436 if (count == 0)
437 break;
438
439 /*
440 * Figure out how much of the initial reserve we have to
441 * free in order to reach our target.
442 */
443 bfree -= vm_dma_reserved / PAGE_SIZE;
444 if (count > bfree) {
445 blk += count - bfree;
446 count = bfree;
447 }
448
449 /*
450 * Calculate the nearest power of 2 <= count.
451 */
452 for (xcount = 1; xcount <= count; xcount <<= 1)
453 ;
454 xcount >>= 1;
455 blk += count - xcount;
456 count = xcount;
457
458 /*
459 * Allocate the pages from the alist, then free them to
460 * the normal VM page queues.
461 *
462 * Pages allocated from the alist are wired. We have to
463 * busy, unwire, and free them. We must also adjust
464 * vm_low_phys_reserved before freeing any pages to prevent
465 * confusion.
466 */
467 rblk = alist_alloc(&vm_contig_alist, blk, count);
468 if (rblk != blk) {
469 kprintf("vm_page_startup_finish: Unable to return "
470 "dma space @0x%08x/%d -> 0x%08x\n",
471 blk, count, rblk);
472 break;
473 }
474 atomic_add_int(&vmstats.v_dma_pages, -count);
475 spin_unlock(&vm_contig_spin);
476
477 m = PHYS_TO_VM_PAGE((vm_paddr_t)blk << PAGE_SHIFT);
478 vm_low_phys_reserved = VM_PAGE_TO_PHYS(m);
479 while (count) {
480 vm_page_busy_wait(m, FALSE, "cpgfr");
481 vm_page_unwire(m, 0);
482 vm_page_free(m);
483 --count;
484 ++m;
485 }
486 spin_lock(&vm_contig_spin);
487 }
488 spin_unlock(&vm_contig_spin);
489
490 /*
491 * Print out how much DMA space drivers have already allocated and
492 * how much is left over.
493 */
494 kprintf("DMA space used: %jdk, remaining available: %jdk\n",
495 (intmax_t)(vmstats.v_dma_pages - vm_contig_alist.bl_free) *
496 (PAGE_SIZE / 1024),
497 (intmax_t)vm_contig_alist.bl_free * (PAGE_SIZE / 1024));
498}
499SYSINIT(vm_pgend, SI_SUB_PROC0_POST, SI_ORDER_ANY,
500 vm_page_startup_finish, NULL)
501
502
984263bc 503/*
1f804340
MD
504 * Scan comparison function for Red-Black tree scans. An inclusive
505 * (start,end) is expected. Other fields are not used.
984263bc 506 */
1f804340
MD
507int
508rb_vm_page_scancmp(struct vm_page *p, void *data)
984263bc 509{
1f804340 510 struct rb_vm_page_scan_info *info = data;
984263bc 511
1f804340
MD
512 if (p->pindex < info->start_pindex)
513 return(-1);
514 if (p->pindex > info->end_pindex)
515 return(1);
516 return(0);
517}
518
519int
520rb_vm_page_compare(struct vm_page *p1, struct vm_page *p2)
521{
522 if (p1->pindex < p2->pindex)
523 return(-1);
524 if (p1->pindex > p2->pindex)
525 return(1);
526 return(0);
984263bc
MD
527}
528
b12defdc
MD
529/*
530 * Each page queue has its own spin lock, which is fairly optimal for
531 * allocating and freeing pages at least.
532 *
533 * The caller must hold the vm_page_spin_lock() before locking a vm_page's
534 * queue spinlock via this function. Also note that m->queue cannot change
535 * unless both the page and queue are locked.
536 */
537static __inline
538void
539_vm_page_queue_spin_lock(vm_page_t m)
540{
541 u_short queue;
542
543 queue = m->queue;
544 if (queue != PQ_NONE) {
545 spin_lock(&vm_page_queues[queue].spin);
546 KKASSERT(queue == m->queue);
547 }
548}
549
550static __inline
551void
552_vm_page_queue_spin_unlock(vm_page_t m)
553{
554 u_short queue;
555
556 queue = m->queue;
557 cpu_ccfence();
558 if (queue != PQ_NONE)
559 spin_unlock(&vm_page_queues[queue].spin);
560}
561
562static __inline
563void
564_vm_page_queues_spin_lock(u_short queue)
565{
566 cpu_ccfence();
567 if (queue != PQ_NONE)
568 spin_lock(&vm_page_queues[queue].spin);
569}
570
571
572static __inline
573void
574_vm_page_queues_spin_unlock(u_short queue)
575{
576 cpu_ccfence();
577 if (queue != PQ_NONE)
578 spin_unlock(&vm_page_queues[queue].spin);
579}
580
581void
582vm_page_queue_spin_lock(vm_page_t m)
583{
584 _vm_page_queue_spin_lock(m);
585}
586
587void
588vm_page_queues_spin_lock(u_short queue)
589{
590 _vm_page_queues_spin_lock(queue);
591}
592
593void
594vm_page_queue_spin_unlock(vm_page_t m)
595{
596 _vm_page_queue_spin_unlock(m);
597}
598
599void
600vm_page_queues_spin_unlock(u_short queue)
601{
602 _vm_page_queues_spin_unlock(queue);
603}
604
605/*
606 * This locks the specified vm_page and its queue in the proper order
607 * (page first, then queue). The queue may change so the caller must
608 * recheck on return.
609 */
610static __inline
611void
612_vm_page_and_queue_spin_lock(vm_page_t m)
613{
614 vm_page_spin_lock(m);
615 _vm_page_queue_spin_lock(m);
616}
617
618static __inline
619void
620_vm_page_and_queue_spin_unlock(vm_page_t m)
621{
622 _vm_page_queues_spin_unlock(m->queue);
623 vm_page_spin_unlock(m);
624}
625
626void
627vm_page_and_queue_spin_unlock(vm_page_t m)
628{
629 _vm_page_and_queue_spin_unlock(m);
630}
631
632void
633vm_page_and_queue_spin_lock(vm_page_t m)
634{
635 _vm_page_and_queue_spin_lock(m);
636}
637
638/*
639 * Helper function removes vm_page from its current queue.
640 * Returns the base queue the page used to be on.
641 *
642 * The vm_page and the queue must be spinlocked.
643 * This function will unlock the queue but leave the page spinlocked.
644 */
645static __inline u_short
646_vm_page_rem_queue_spinlocked(vm_page_t m)
647{
648 struct vpgqueues *pq;
649 u_short queue;
650
651 queue = m->queue;
652 if (queue != PQ_NONE) {
653 pq = &vm_page_queues[queue];
654 TAILQ_REMOVE(&pq->pl, m, pageq);
655 atomic_add_int(pq->cnt, -1);
656 pq->lcnt--;
657 m->queue = PQ_NONE;
85946b6c 658 vm_page_queues_spin_unlock(queue);
b12defdc
MD
659 if ((queue - m->pc) == PQ_FREE && (m->flags & PG_ZERO))
660 atomic_subtract_int(&vm_page_zero_count, 1);
b12defdc
MD
661 if ((queue - m->pc) == PQ_CACHE || (queue - m->pc) == PQ_FREE)
662 return (queue - m->pc);
663 }
664 return queue;
665}
666
667/*
668 * Helper function places the vm_page on the specified queue.
669 *
670 * The vm_page must be spinlocked.
671 * This function will return with both the page and the queue locked.
672 */
673static __inline void
674_vm_page_add_queue_spinlocked(vm_page_t m, u_short queue, int athead)
675{
676 struct vpgqueues *pq;
677
678 KKASSERT(m->queue == PQ_NONE);
679
680 if (queue != PQ_NONE) {
681 vm_page_queues_spin_lock(queue);
682 pq = &vm_page_queues[queue];
683 ++pq->lcnt;
684 atomic_add_int(pq->cnt, 1);
685 m->queue = queue;
686
687 /*
688 * Put zero'd pages on the end ( where we look for zero'd pages
689 * first ) and non-zerod pages at the head.
690 */
691 if (queue - m->pc == PQ_FREE) {
692 if (m->flags & PG_ZERO) {
693 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
694 atomic_add_int(&vm_page_zero_count, 1);
695 } else {
696 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
697 }
698 } else if (athead) {
699 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
700 } else {
701 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
702 }
703 /* leave the queue spinlocked */
704 }
705}
706
707/*
708 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
709 * m->busy is zero. Returns TRUE if it had to sleep, FALSE if we
710 * did not. Only one sleep call will be made before returning.
711 *
712 * This function does NOT busy the page and on return the page is not
713 * guaranteed to be available.
714 */
715void
716vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
717{
718 u_int32_t flags;
719
720 for (;;) {
721 flags = m->flags;
722 cpu_ccfence();
723
724 if ((flags & PG_BUSY) == 0 &&
725 (also_m_busy == 0 || (flags & PG_SBUSY) == 0)) {
726 break;
727 }
728 tsleep_interlock(m, 0);
729 if (atomic_cmpset_int(&m->flags, flags,
730 flags | PG_WANTED | PG_REFERENCED)) {
731 tsleep(m, PINTERLOCKED, msg, 0);
732 break;
733 }
734 }
735}
736
737/*
738 * Wait until PG_BUSY can be set, then set it. If also_m_busy is TRUE we
739 * also wait for m->busy to become 0 before setting PG_BUSY.
740 */
741void
742VM_PAGE_DEBUG_EXT(vm_page_busy_wait)(vm_page_t m,
743 int also_m_busy, const char *msg
744 VM_PAGE_DEBUG_ARGS)
745{
746 u_int32_t flags;
747
748 for (;;) {
749 flags = m->flags;
750 cpu_ccfence();
751 if (flags & PG_BUSY) {
752 tsleep_interlock(m, 0);
753 if (atomic_cmpset_int(&m->flags, flags,
754 flags | PG_WANTED | PG_REFERENCED)) {
755 tsleep(m, PINTERLOCKED, msg, 0);
756 }
757 } else if (also_m_busy && (flags & PG_SBUSY)) {
758 tsleep_interlock(m, 0);
759 if (atomic_cmpset_int(&m->flags, flags,
760 flags | PG_WANTED | PG_REFERENCED)) {
761 tsleep(m, PINTERLOCKED, msg, 0);
762 }
763 } else {
764 if (atomic_cmpset_int(&m->flags, flags,
765 flags | PG_BUSY)) {
766#ifdef VM_PAGE_DEBUG
767 m->busy_func = func;
768 m->busy_line = lineno;
769#endif
770 break;
771 }
772 }
773 }
774}
775
776/*
777 * Attempt to set PG_BUSY. If also_m_busy is TRUE we only succeed if m->busy
778 * is also 0.
779 *
780 * Returns non-zero on failure.
781 */
782int
783VM_PAGE_DEBUG_EXT(vm_page_busy_try)(vm_page_t m, int also_m_busy
784 VM_PAGE_DEBUG_ARGS)
785{
786 u_int32_t flags;
787
788 for (;;) {
789 flags = m->flags;
790 cpu_ccfence();
791 if (flags & PG_BUSY)
792 return TRUE;
793 if (also_m_busy && (flags & PG_SBUSY))
794 return TRUE;
795 if (atomic_cmpset_int(&m->flags, flags, flags | PG_BUSY)) {
796#ifdef VM_PAGE_DEBUG
797 m->busy_func = func;
798 m->busy_line = lineno;
799#endif
800 return FALSE;
801 }
802 }
803}
804
805/*
806 * Clear the PG_BUSY flag and return non-zero to indicate to the caller
807 * that a wakeup() should be performed.
808 *
809 * The vm_page must be spinlocked and will remain spinlocked on return.
810 * The related queue must NOT be spinlocked (which could deadlock us).
811 *
812 * (inline version)
813 */
814static __inline
815int
816_vm_page_wakeup(vm_page_t m)
817{
818 u_int32_t flags;
819
820 for (;;) {
821 flags = m->flags;
822 cpu_ccfence();
823 if (atomic_cmpset_int(&m->flags, flags,
824 flags & ~(PG_BUSY | PG_WANTED))) {
825 break;
826 }
827 }
828 return(flags & PG_WANTED);
829}
830
831/*
832 * Clear the PG_BUSY flag and wakeup anyone waiting for the page. This
833 * is typically the last call you make on a page before moving onto
834 * other things.
835 */
836void
837vm_page_wakeup(vm_page_t m)
838{
839 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
840 vm_page_spin_lock(m);
841 if (_vm_page_wakeup(m)) {
842 vm_page_spin_unlock(m);
843 wakeup(m);
844 } else {
845 vm_page_spin_unlock(m);
846 }
847}
848
573fb415
MD
849/*
850 * Holding a page keeps it from being reused. Other parts of the system
851 * can still disassociate the page from its current object and free it, or
852 * perform read or write I/O on it and/or otherwise manipulate the page,
853 * but if the page is held the VM system will leave the page and its data
854 * intact and not reuse the page for other purposes until the last hold
855 * reference is released. (see vm_page_wire() if you want to prevent the
856 * page from being disassociated from its object too).
857 *
573fb415
MD
858 * The caller must still validate the contents of the page and, if necessary,
859 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
860 * before manipulating the page.
b12defdc
MD
861 *
862 * XXX get vm_page_spin_lock() here and move FREE->HOLD if necessary
573fb415
MD
863 */
864void
865vm_page_hold(vm_page_t m)
866{
b12defdc
MD
867 vm_page_spin_lock(m);
868 atomic_add_int(&m->hold_count, 1);
869 if (m->queue - m->pc == PQ_FREE) {
870 _vm_page_queue_spin_lock(m);
871 _vm_page_rem_queue_spinlocked(m);
027193eb 872 _vm_page_add_queue_spinlocked(m, PQ_HOLD + m->pc, 0);
b12defdc
MD
873 _vm_page_queue_spin_unlock(m);
874 }
875 vm_page_spin_unlock(m);
573fb415
MD
876}
877
de71fd3f
MD
878/*
879 * The opposite of vm_page_hold(). A page can be freed while being held,
b12defdc
MD
880 * which places it on the PQ_HOLD queue. If we are able to busy the page
881 * after the hold count drops to zero we will move the page to the
882 * appropriate PQ_FREE queue by calling vm_page_free_toq().
de71fd3f 883 */
984263bc 884void
573fb415 885vm_page_unhold(vm_page_t m)
984263bc 886{
b12defdc
MD
887 vm_page_spin_lock(m);
888 atomic_add_int(&m->hold_count, -1);
027193eb 889 if (m->hold_count == 0 && m->queue - m->pc == PQ_HOLD) {
b12defdc
MD
890 _vm_page_queue_spin_lock(m);
891 _vm_page_rem_queue_spinlocked(m);
892 _vm_page_add_queue_spinlocked(m, PQ_FREE + m->pc, 0);
893 _vm_page_queue_spin_unlock(m);
97edb3b6 894 }
b12defdc 895 vm_page_spin_unlock(m);
984263bc
MD
896}
897
898/*
573fb415 899 * Inserts the given vm_page into the object and object list.
984263bc 900 *
de71fd3f
MD
901 * The pagetables are not updated but will presumably fault the page
902 * in if necessary, or if a kernel page the caller will at some point
903 * enter the page into the kernel's pmap. We are not allowed to block
904 * here so we *can't* do this anyway.
984263bc 905 *
de71fd3f 906 * This routine may not block.
398c240d 907 * This routine must be called with the vm_object held.
654a39f0 908 * This routine must be called with a critical section held.
d2d8515b
MD
909 *
910 * This routine returns TRUE if the page was inserted into the object
911 * successfully, and FALSE if the page already exists in the object.
984263bc 912 */
d2d8515b 913int
984263bc
MD
914vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
915{
ce94514e 916 ASSERT_LWKT_TOKEN_HELD_EXCL(vm_object_token(object));
984263bc
MD
917 if (m->object != NULL)
918 panic("vm_page_insert: already inserted");
919
b12defdc 920 object->generation++;
b12defdc 921
984263bc 922 /*
b12defdc
MD
923 * Record the object/offset pair in this page and add the
924 * pv_list_count of the page to the object.
925 *
926 * The vm_page spin lock is required for interactions with the pmap.
984263bc 927 */
b12defdc 928 vm_page_spin_lock(m);
984263bc
MD
929 m->object = object;
930 m->pindex = pindex;
d2d8515b
MD
931 if (vm_page_rb_tree_RB_INSERT(&object->rb_memq, m)) {
932 m->object = NULL;
933 m->pindex = 0;
934 vm_page_spin_unlock(m);
935 return FALSE;
936 }
937 object->resident_page_count++;
85946b6c 938 /* atomic_add_int(&object->agg_pv_list_count, m->md.pv_list_count); */
b12defdc 939 vm_page_spin_unlock(m);
50a55c46 940
984263bc
MD
941 /*
942 * Since we are inserting a new and possibly dirty page,
943 * update the object's OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY flags.
944 */
d86d27a8
MD
945 if ((m->valid & m->dirty) ||
946 (m->flags & (PG_WRITEABLE | PG_NEED_COMMIT)))
984263bc 947 vm_object_set_writeable_dirty(object);
096e95c0
MD
948
949 /*
950 * Checks for a swap assignment and sets PG_SWAPPED if appropriate.
951 */
952 swap_pager_page_inserted(m);
d2d8515b 953 return TRUE;
984263bc
MD
954}
955
956/*
b12defdc 957 * Removes the given vm_page_t from the (object,index) table
984263bc 958 *
de71fd3f
MD
959 * The underlying pmap entry (if any) is NOT removed here.
960 * This routine may not block.
9765affa 961 *
9ad0147b
MD
962 * The page must be BUSY and will remain BUSY on return.
963 * No other requirements.
9765affa 964 *
9ad0147b
MD
965 * NOTE: FreeBSD side effect was to unbusy the page on return. We leave
966 * it busy.
984263bc 967 */
984263bc
MD
968void
969vm_page_remove(vm_page_t m)
970{
971 vm_object_t object;
972
654a39f0 973 if (m->object == NULL) {
984263bc 974 return;
654a39f0 975 }
984263bc 976
de71fd3f 977 if ((m->flags & PG_BUSY) == 0)
984263bc 978 panic("vm_page_remove: page not busy");
984263bc 979
984263bc
MD
980 object = m->object;
981
398c240d
VS
982 vm_object_hold(object);
983
984263bc 984 /*
1f804340 985 * Remove the page from the object and update the object.
b12defdc
MD
986 *
987 * The vm_page spin lock is required for interactions with the pmap.
984263bc 988 */
b12defdc 989 vm_page_spin_lock(m);
1f804340 990 vm_page_rb_tree_RB_REMOVE(&object->rb_memq, m);
984263bc 991 object->resident_page_count--;
85946b6c 992 /* atomic_add_int(&object->agg_pv_list_count, -m->md.pv_list_count); */
984263bc 993 m->object = NULL;
b12defdc 994 vm_page_spin_unlock(m);
1f804340 995
b12defdc 996 object->generation++;
398c240d 997
b12defdc 998 vm_object_drop(object);
984263bc
MD
999}
1000
1001/*
de71fd3f
MD
1002 * Locate and return the page at (object, pindex), or NULL if the
1003 * page could not be found.
1004 *
b12defdc 1005 * The caller must hold the vm_object token.
984263bc 1006 */
984263bc
MD
1007vm_page_t
1008vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1009{
1010 vm_page_t m;
984263bc
MD
1011
1012 /*
1013 * Search the hash table for this object/offset pair
1014 */
b12defdc 1015 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
1f804340 1016 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
1f804340
MD
1017 KKASSERT(m == NULL || (m->object == object && m->pindex == pindex));
1018 return(m);
984263bc
MD
1019}
1020
b12defdc
MD
1021vm_page_t
1022VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_wait)(struct vm_object *object,
1023 vm_pindex_t pindex,
1024 int also_m_busy, const char *msg
1025 VM_PAGE_DEBUG_ARGS)
1026{
1027 u_int32_t flags;
1028 vm_page_t m;
1029
1030 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
1031 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
1032 while (m) {
1033 KKASSERT(m->object == object && m->pindex == pindex);
1034 flags = m->flags;
1035 cpu_ccfence();
1036 if (flags & PG_BUSY) {
1037 tsleep_interlock(m, 0);
1038 if (atomic_cmpset_int(&m->flags, flags,
1039 flags | PG_WANTED | PG_REFERENCED)) {
1040 tsleep(m, PINTERLOCKED, msg, 0);
1041 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq,
1042 pindex);
1043 }
1044 } else if (also_m_busy && (flags & PG_SBUSY)) {
1045 tsleep_interlock(m, 0);
1046 if (atomic_cmpset_int(&m->flags, flags,
1047 flags | PG_WANTED | PG_REFERENCED)) {
1048 tsleep(m, PINTERLOCKED, msg, 0);
1049 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq,
1050 pindex);
1051 }
1052 } else if (atomic_cmpset_int(&m->flags, flags,
1053 flags | PG_BUSY)) {
1054#ifdef VM_PAGE_DEBUG
1055 m->busy_func = func;
1056 m->busy_line = lineno;
1057#endif
1058 break;
1059 }
1060 }
1061 return m;
1062}
1063
984263bc 1064/*
b12defdc 1065 * Attempt to lookup and busy a page.
984263bc 1066 *
b12defdc 1067 * Returns NULL if the page could not be found
984263bc 1068 *
b12defdc
MD
1069 * Returns a vm_page and error == TRUE if the page exists but could not
1070 * be busied.
984263bc 1071 *
b12defdc
MD
1072 * Returns a vm_page and error == FALSE on success.
1073 */
1074vm_page_t
1075VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_try)(struct vm_object *object,
1076 vm_pindex_t pindex,
1077 int also_m_busy, int *errorp
1078 VM_PAGE_DEBUG_ARGS)
1079{
1080 u_int32_t flags;
1081 vm_page_t m;
1082
1083 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
1084 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
1085 *errorp = FALSE;
1086 while (m) {
1087 KKASSERT(m->object == object && m->pindex == pindex);
1088 flags = m->flags;
1089 cpu_ccfence();
1090 if (flags & PG_BUSY) {
1091 *errorp = TRUE;
1092 break;
1093 }
1094 if (also_m_busy && (flags & PG_SBUSY)) {
1095 *errorp = TRUE;
1096 break;
1097 }
1098 if (atomic_cmpset_int(&m->flags, flags, flags | PG_BUSY)) {
1099#ifdef VM_PAGE_DEBUG
1100 m->busy_func = func;
1101 m->busy_line = lineno;
1102#endif
1103 break;
1104 }
1105 }
1106 return m;
1107}
1108
1109/*
1110 * Caller must hold the related vm_object
1111 */
1112vm_page_t
1113vm_page_next(vm_page_t m)
1114{
1115 vm_page_t next;
1116
1117 next = vm_page_rb_tree_RB_NEXT(m);
1118 if (next && next->pindex != m->pindex + 1)
1119 next = NULL;
1120 return (next);
1121}
1122
1123/*
1124 * vm_page_rename()
1125 *
1126 * Move the given vm_page from its current object to the specified
1127 * target object/offset. The page must be busy and will remain so
1128 * on return.
984263bc 1129 *
b12defdc
MD
1130 * new_object must be held.
1131 * This routine might block. XXX ?
1132 *
1133 * NOTE: Swap associated with the page must be invalidated by the move. We
de71fd3f
MD
1134 * have to do this for several reasons: (1) we aren't freeing the
1135 * page, (2) we are dirtying the page, (3) the VM system is probably
1136 * moving the page from object A to B, and will then later move
1137 * the backing store from A to B and we can't have a conflict.
984263bc 1138 *
b12defdc 1139 * NOTE: We *always* dirty the page. It is necessary both for the
de71fd3f
MD
1140 * fact that we moved it, and because we may be invalidating
1141 * swap. If the page is on the cache, we have to deactivate it
1142 * or vm_page_dirty() will panic. Dirty pages are not allowed
1143 * on the cache.
984263bc 1144 */
984263bc
MD
1145void
1146vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1147{
b12defdc 1148 KKASSERT(m->flags & PG_BUSY);
ce94514e 1149 ASSERT_LWKT_TOKEN_HELD_EXCL(vm_object_token(new_object));
b12defdc 1150 if (m->object) {
ce94514e 1151 ASSERT_LWKT_TOKEN_HELD_EXCL(vm_object_token(m->object));
b12defdc
MD
1152 vm_page_remove(m);
1153 }
d2d8515b 1154 if (vm_page_insert(m, new_object, new_pindex) == FALSE) {
480c83b6 1155 panic("vm_page_rename: target exists (%p,%"PRIu64")",
d2d8515b
MD
1156 new_object, new_pindex);
1157 }
984263bc
MD
1158 if (m->queue - m->pc == PQ_CACHE)
1159 vm_page_deactivate(m);
1160 vm_page_dirty(m);
984263bc
MD
1161}
1162
1163/*
de71fd3f
MD
1164 * vm_page_unqueue() without any wakeup. This routine is used when a page
1165 * is being moved between queues or otherwise is to remain BUSYied by the
1166 * caller.
984263bc 1167 *
de71fd3f 1168 * This routine may not block.
984263bc 1169 */
984263bc
MD
1170void
1171vm_page_unqueue_nowakeup(vm_page_t m)
1172{
b12defdc
MD
1173 vm_page_and_queue_spin_lock(m);
1174 (void)_vm_page_rem_queue_spinlocked(m);
1175 vm_page_spin_unlock(m);
984263bc
MD
1176}
1177
1178/*
de71fd3f
MD
1179 * vm_page_unqueue() - Remove a page from its queue, wakeup the pagedemon
1180 * if necessary.
984263bc 1181 *
de71fd3f 1182 * This routine may not block.
984263bc 1183 */
984263bc
MD
1184void
1185vm_page_unqueue(vm_page_t m)
1186{
b12defdc 1187 u_short queue;
de71fd3f 1188
b12defdc
MD
1189 vm_page_and_queue_spin_lock(m);
1190 queue = _vm_page_rem_queue_spinlocked(m);
1191 if (queue == PQ_FREE || queue == PQ_CACHE) {
1192 vm_page_spin_unlock(m);
1193 pagedaemon_wakeup();
1194 } else {
1195 vm_page_spin_unlock(m);
984263bc
MD
1196 }
1197}
1198
984263bc 1199/*
de71fd3f 1200 * vm_page_list_find()
984263bc 1201 *
de71fd3f 1202 * Find a page on the specified queue with color optimization.
984263bc 1203 *
de71fd3f
MD
1204 * The page coloring optimization attempts to locate a page that does
1205 * not overload other nearby pages in the object in the cpu's L1 or L2
1206 * caches. We need this optimization because cpu caches tend to be
85946b6c
MD
1207 * physical caches, while object spaces tend to be virtual.
1208 *
1209 * On MP systems each PQ_FREE and PQ_CACHE color queue has its own spinlock
1210 * and the algorithm is adjusted to localize allocations on a per-core basis.
1211 * This is done by 'twisting' the colors.
984263bc 1212 *
b12defdc
MD
1213 * The page is returned spinlocked and removed from its queue (it will
1214 * be on PQ_NONE), or NULL. The page is not PG_BUSY'd. The caller
1215 * is responsible for dealing with the busy-page case (usually by
1216 * deactivating the page and looping).
1217 *
1218 * NOTE: This routine is carefully inlined. A non-inlined version
1219 * is available for outside callers but the only critical path is
1220 * from within this source file.
984263bc 1221 *
b12defdc
MD
1222 * NOTE: This routine assumes that the vm_pages found in PQ_CACHE and PQ_FREE
1223 * represent stable storage, allowing us to order our locks vm_page
1224 * first, then queue.
984263bc 1225 */
74232d8e 1226static __inline
984263bc 1227vm_page_t
74232d8e
MD
1228_vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
1229{
1230 vm_page_t m;
1231
b12defdc
MD
1232 for (;;) {
1233 if (prefer_zero)
1234 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
1235 else
1236 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
1237 if (m == NULL) {
1238 m = _vm_page_list_find2(basequeue, index);
1239 return(m);
1240 }
1241 vm_page_and_queue_spin_lock(m);
1242 if (m->queue == basequeue + index) {
1243 _vm_page_rem_queue_spinlocked(m);
1244 /* vm_page_t spin held, no queue spin */
1245 break;
1246 }
1247 vm_page_and_queue_spin_unlock(m);
1248 }
74232d8e
MD
1249 return(m);
1250}
1251
1252static vm_page_t
1253_vm_page_list_find2(int basequeue, int index)
984263bc
MD
1254{
1255 int i;
1256 vm_page_t m = NULL;
1257 struct vpgqueues *pq;
1258
1259 pq = &vm_page_queues[basequeue];
1260
1261 /*
1262 * Note that for the first loop, index+i and index-i wind up at the
1263 * same place. Even though this is not totally optimal, we've already
1264 * blown it by missing the cache case so we do not care.
1265 */
b12defdc
MD
1266 for (i = PQ_L2_SIZE / 2; i > 0; --i) {
1267 for (;;) {
1268 m = TAILQ_FIRST(&pq[(index + i) & PQ_L2_MASK].pl);
1269 if (m) {
1270 _vm_page_and_queue_spin_lock(m);
1271 if (m->queue ==
1272 basequeue + ((index + i) & PQ_L2_MASK)) {
1273 _vm_page_rem_queue_spinlocked(m);
1274 return(m);
1275 }
1276 _vm_page_and_queue_spin_unlock(m);
1277 continue;
1278 }
1279 m = TAILQ_FIRST(&pq[(index - i) & PQ_L2_MASK].pl);
1280 if (m) {
1281 _vm_page_and_queue_spin_lock(m);
1282 if (m->queue ==
1283 basequeue + ((index - i) & PQ_L2_MASK)) {
1284 _vm_page_rem_queue_spinlocked(m);
1285 return(m);
1286 }
1287 _vm_page_and_queue_spin_unlock(m);
1288 continue;
1289 }
1290 break; /* next i */
1291 }
984263bc
MD
1292 }
1293 return(m);
1294}
1295
573fb415 1296/*
b12defdc
MD
1297 * Returns a vm_page candidate for allocation. The page is not busied so
1298 * it can move around. The caller must busy the page (and typically
1299 * deactivate it if it cannot be busied!)
1300 *
1301 * Returns a spinlocked vm_page that has been removed from its queue.
573fb415 1302 */
74232d8e
MD
1303vm_page_t
1304vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
1305{
1306 return(_vm_page_list_find(basequeue, index, prefer_zero));
1307}
1308
984263bc 1309/*
b12defdc
MD
1310 * Find a page on the cache queue with color optimization, remove it
1311 * from the queue, and busy it. The returned page will not be spinlocked.
1312 *
1313 * A candidate failure will be deactivated. Candidates can fail due to
1314 * being busied by someone else, in which case they will be deactivated.
984263bc 1315 *
de71fd3f 1316 * This routine may not block.
b12defdc 1317 *
984263bc 1318 */
b12defdc 1319static vm_page_t
85946b6c 1320vm_page_select_cache(u_short pg_color)
984263bc
MD
1321{
1322 vm_page_t m;
1323
b12defdc 1324 for (;;) {
85946b6c 1325 m = _vm_page_list_find(PQ_CACHE, pg_color & PQ_L2_MASK, FALSE);
b12defdc
MD
1326 if (m == NULL)
1327 break;
1328 /*
1329 * (m) has been removed from its queue and spinlocked
1330 */
1331 if (vm_page_busy_try(m, TRUE)) {
1332 _vm_page_deactivate_locked(m, 0);
1333 vm_page_spin_unlock(m);
a491077e
MD
1334#ifdef INVARIANTS
1335 kprintf("Warning: busy page %p found in cache\n", m);
1336#endif
b12defdc
MD
1337 } else {
1338 /*
1339 * We successfully busied the page
1340 */
9bf025db 1341 if ((m->flags & (PG_UNMANAGED | PG_NEED_COMMIT)) == 0 &&
b12defdc 1342 m->hold_count == 0 &&
9bf025db
MD
1343 m->wire_count == 0 &&
1344 (m->dirty & m->valid) == 0) {
b12defdc
MD
1345 vm_page_spin_unlock(m);
1346 pagedaemon_wakeup();
1347 return(m);
1348 }
9bf025db
MD
1349
1350 /*
1351 * The page cannot be recycled, deactivate it.
1352 */
b12defdc
MD
1353 _vm_page_deactivate_locked(m, 0);
1354 if (_vm_page_wakeup(m)) {
1355 vm_page_spin_unlock(m);
1356 wakeup(m);
1357 } else {
1358 vm_page_spin_unlock(m);
1359 }
984263bc 1360 }
984263bc 1361 }
b12defdc 1362 return (m);
984263bc
MD
1363}
1364
1365/*
de71fd3f
MD
1366 * Find a free or zero page, with specified preference. We attempt to
1367 * inline the nominal case and fall back to _vm_page_select_free()
b12defdc 1368 * otherwise. A busied page is removed from the queue and returned.
984263bc 1369 *
de71fd3f 1370 * This routine may not block.
984263bc 1371 */
984263bc 1372static __inline vm_page_t
85946b6c 1373vm_page_select_free(u_short pg_color, boolean_t prefer_zero)
984263bc
MD
1374{
1375 vm_page_t m;
1376
b12defdc 1377 for (;;) {
85946b6c 1378 m = _vm_page_list_find(PQ_FREE, pg_color & PQ_L2_MASK,
b12defdc
MD
1379 prefer_zero);
1380 if (m == NULL)
1381 break;
1382 if (vm_page_busy_try(m, TRUE)) {
90244566
MD
1383 /*
1384 * Various mechanisms such as a pmap_collect can
1385 * result in a busy page on the free queue. We
1386 * have to move the page out of the way so we can
1387 * retry the allocation. If the other thread is not
1388 * allocating the page then m->valid will remain 0 and
1389 * the pageout daemon will free the page later on.
1390 *
1391 * Since we could not busy the page, however, we
1392 * cannot make assumptions as to whether the page
1393 * will be allocated by the other thread or not,
1394 * so all we can do is deactivate it to move it out
1395 * of the way. In particular, if the other thread
1396 * wires the page it may wind up on the inactive
1397 * queue and the pageout daemon will have to deal
1398 * with that case too.
1399 */
b12defdc
MD
1400 _vm_page_deactivate_locked(m, 0);
1401 vm_page_spin_unlock(m);
1402#ifdef INVARIANTS
1403 kprintf("Warning: busy page %p found in cache\n", m);
1404#endif
1405 } else {
90244566
MD
1406 /*
1407 * Theoretically if we are able to busy the page
1408 * atomic with the queue removal (using the vm_page
1409 * lock) nobody else should be able to mess with the
1410 * page before us.
1411 */
9bf025db
MD
1412 KKASSERT((m->flags & (PG_UNMANAGED |
1413 PG_NEED_COMMIT)) == 0);
b12defdc
MD
1414 KKASSERT(m->hold_count == 0);
1415 KKASSERT(m->wire_count == 0);
1416 vm_page_spin_unlock(m);
1417 pagedaemon_wakeup();
1418
1419 /* return busied and removed page */
1420 return(m);
1421 }
1422 }
984263bc
MD
1423 return(m);
1424}
1425
54341a3b
MD
1426/*
1427 * This implements a per-cpu cache of free, zero'd, ready-to-go pages.
1428 * The idea is to populate this cache prior to acquiring any locks so
1429 * we don't wind up potentially zeroing VM pages (under heavy loads) while
1430 * holding potentialy contending locks.
1431 *
1432 * Note that we allocate the page uninserted into anything and use a pindex
1433 * of 0, the vm_page_alloc() will effectively add gd_cpuid so these
1434 * allocations should wind up being uncontended. However, we still want
1435 * to rove across PQ_L2_SIZE.
1436 */
1437void
1438vm_page_pcpu_cache(void)
1439{
1440#if 0
1441 globaldata_t gd = mycpu;
1442 vm_page_t m;
1443
1444 if (gd->gd_vmpg_count < GD_MINVMPG) {
1445 crit_enter_gd(gd);
1446 while (gd->gd_vmpg_count < GD_MAXVMPG) {
1447 m = vm_page_alloc(NULL, ticks & ~ncpus2_mask,
1448 VM_ALLOC_NULL_OK | VM_ALLOC_NORMAL |
1449 VM_ALLOC_NULL_OK | VM_ALLOC_ZERO);
1450 if (gd->gd_vmpg_count < GD_MAXVMPG) {
1451 if ((m->flags & PG_ZERO) == 0) {
1452 pmap_zero_page(VM_PAGE_TO_PHYS(m));
1453 vm_page_flag_set(m, PG_ZERO);
1454 }
1455 gd->gd_vmpg_array[gd->gd_vmpg_count++] = m;
1456 } else {
1457 vm_page_free(m);
1458 }
1459 }
1460 crit_exit_gd(gd);
1461 }
1462#endif
1463}
1464
984263bc 1465/*
de71fd3f 1466 * vm_page_alloc()
984263bc 1467 *
de71fd3f 1468 * Allocate and return a memory cell associated with this VM object/offset
85946b6c 1469 * pair. If object is NULL an unassociated page will be allocated.
984263bc 1470 *
d2d8515b
MD
1471 * The returned page will be busied and removed from its queues. This
1472 * routine can block and may return NULL if a race occurs and the page
1473 * is found to already exist at the specified (object, pindex).
de71fd3f 1474 *
dc1fd4b3 1475 * VM_ALLOC_NORMAL allow use of cache pages, nominal free drain
39208dbe 1476 * VM_ALLOC_QUICK like normal but cannot use cache
dc1fd4b3
MD
1477 * VM_ALLOC_SYSTEM greater free drain
1478 * VM_ALLOC_INTERRUPT allow free list to be completely drained
d2d8515b
MD
1479 * VM_ALLOC_ZERO advisory request for pre-zero'd page only
1480 * VM_ALLOC_FORCE_ZERO advisory request for pre-zero'd page only
1481 * VM_ALLOC_NULL_OK ok to return NULL on insertion collision
1482 * (see vm_page_grab())
54341a3b
MD
1483 * VM_ALLOC_USE_GD ok to use per-gd cache
1484 *
d2d8515b 1485 * The object must be held if not NULL
85946b6c 1486 * This routine may not block
984263bc 1487 *
de71fd3f
MD
1488 * Additional special handling is required when called from an interrupt
1489 * (VM_ALLOC_INTERRUPT). We are not allowed to mess with the page cache
1490 * in this case.
984263bc 1491 */
984263bc
MD
1492vm_page_t
1493vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int page_req)
1494{
54341a3b 1495 globaldata_t gd = mycpu;
9d494b34 1496 vm_object_t obj;
54341a3b 1497 vm_page_t m;
85946b6c 1498 u_short pg_color;
984263bc 1499
54341a3b
MD
1500#if 0
1501 /*
1502 * Special per-cpu free VM page cache. The pages are pre-busied
1503 * and pre-zerod for us.
1504 */
1505 if (gd->gd_vmpg_count && (page_req & VM_ALLOC_USE_GD)) {
1506 crit_enter_gd(gd);
1507 if (gd->gd_vmpg_count) {
1508 m = gd->gd_vmpg_array[--gd->gd_vmpg_count];
1509 crit_exit_gd(gd);
1510 goto done;
1511 }
1512 crit_exit_gd(gd);
1513 }
1514#endif
1515 m = NULL;
1516
85946b6c
MD
1517 /*
1518 * Cpu twist - cpu localization algorithm
1519 */
1520 if (object) {
54341a3b 1521 pg_color = gd->gd_cpuid + (pindex & ~ncpus_fit_mask) +
85946b6c 1522 (object->pg_color & ~ncpus_fit_mask);
85946b6c 1523 } else {
54341a3b 1524 pg_color = gd->gd_cpuid + (pindex & ~ncpus_fit_mask);
85946b6c 1525 }
dc1fd4b3 1526 KKASSERT(page_req &
39208dbe
MD
1527 (VM_ALLOC_NORMAL|VM_ALLOC_QUICK|
1528 VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
984263bc
MD
1529
1530 /*
4ecf7cc9
MD
1531 * Certain system threads (pageout daemon, buf_daemon's) are
1532 * allowed to eat deeper into the free page list.
984263bc 1533 */
4ecf7cc9 1534 if (curthread->td_flags & TDF_SYSTHREAD)
dc1fd4b3 1535 page_req |= VM_ALLOC_SYSTEM;
984263bc 1536
984263bc 1537loop:
dc1fd4b3
MD
1538 if (vmstats.v_free_count > vmstats.v_free_reserved ||
1539 ((page_req & VM_ALLOC_INTERRUPT) && vmstats.v_free_count > 0) ||
1540 ((page_req & VM_ALLOC_SYSTEM) && vmstats.v_cache_count == 0 &&
1541 vmstats.v_free_count > vmstats.v_interrupt_free_min)
1542 ) {
984263bc 1543 /*
dc1fd4b3 1544 * The free queue has sufficient free pages to take one out.
984263bc 1545 */
d2d8515b 1546 if (page_req & (VM_ALLOC_ZERO | VM_ALLOC_FORCE_ZERO))
85946b6c 1547 m = vm_page_select_free(pg_color, TRUE);
984263bc 1548 else
85946b6c 1549 m = vm_page_select_free(pg_color, FALSE);
dc1fd4b3 1550 } else if (page_req & VM_ALLOC_NORMAL) {
984263bc 1551 /*
dc1fd4b3
MD
1552 * Allocatable from the cache (non-interrupt only). On
1553 * success, we must free the page and try again, thus
1554 * ensuring that vmstats.v_*_free_min counters are replenished.
984263bc 1555 */
dc1fd4b3
MD
1556#ifdef INVARIANTS
1557 if (curthread->td_preempted) {
086c1d7e 1558 kprintf("vm_page_alloc(): warning, attempt to allocate"
dc1fd4b3
MD
1559 " cache page from preempting interrupt\n");
1560 m = NULL;
1561 } else {
85946b6c 1562 m = vm_page_select_cache(pg_color);
dc1fd4b3
MD
1563 }
1564#else
85946b6c 1565 m = vm_page_select_cache(pg_color);
dc1fd4b3 1566#endif
984263bc 1567 /*
9765affa 1568 * On success move the page into the free queue and loop.
bdea739c
MD
1569 *
1570 * Only do this if we can safely acquire the vm_object lock,
1571 * because this is effectively a random page and the caller
1572 * might be holding the lock shared, we don't want to
1573 * deadlock.
984263bc 1574 */
dc1fd4b3
MD
1575 if (m != NULL) {
1576 KASSERT(m->dirty == 0,
d2d8515b 1577 ("Found dirty cache page %p", m));
9d494b34
MD
1578 if ((obj = m->object) != NULL) {
1579 if (vm_object_hold_try(obj)) {
bdea739c
MD
1580 vm_page_protect(m, VM_PROT_NONE);
1581 vm_page_free(m);
9d494b34
MD
1582 /* m->object NULL here */
1583 vm_object_drop(obj);
bdea739c
MD
1584 } else {
1585 vm_page_deactivate(m);
1586 vm_page_wakeup(m);
1587 }
1588 } else {
1589 vm_page_protect(m, VM_PROT_NONE);
1590 vm_page_free(m);
1591 }
dc1fd4b3
MD
1592 goto loop;
1593 }
1594
1595 /*
1596 * On failure return NULL
1597 */
984263bc 1598#if defined(DIAGNOSTIC)
dc1fd4b3 1599 if (vmstats.v_cache_count > 0)
086c1d7e 1600 kprintf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", vmstats.v_cache_count);
984263bc 1601#endif
dc1fd4b3
MD
1602 vm_pageout_deficit++;
1603 pagedaemon_wakeup();
1604 return (NULL);
984263bc
MD
1605 } else {
1606 /*
dc1fd4b3 1607 * No pages available, wakeup the pageout daemon and give up.
984263bc 1608 */
984263bc
MD
1609 vm_pageout_deficit++;
1610 pagedaemon_wakeup();
1611 return (NULL);
1612 }
1613
1614 /*
b12defdc
MD
1615 * v_free_count can race so loop if we don't find the expected
1616 * page.
984263bc 1617 */
b12defdc
MD
1618 if (m == NULL)
1619 goto loop;
984263bc
MD
1620
1621 /*
d2d8515b
MD
1622 * Good page found. The page has already been busied for us and
1623 * removed from its queues.
984263bc 1624 */
d2d8515b
MD
1625 KASSERT(m->dirty == 0,
1626 ("vm_page_alloc: free/cache page %p was dirty", m));
b12defdc 1627 KKASSERT(m->queue == PQ_NONE);
984263bc 1628
54341a3b
MD
1629#if 0
1630done:
1631#endif
984263bc 1632 /*
d2d8515b
MD
1633 * Initialize the structure, inheriting some flags but clearing
1634 * all the rest. The page has already been busied for us.
984263bc 1635 */
d2d8515b 1636 vm_page_flag_clear(m, ~(PG_ZERO | PG_BUSY | PG_SBUSY));
b12defdc
MD
1637 KKASSERT(m->wire_count == 0);
1638 KKASSERT(m->busy == 0);
984263bc 1639 m->act_count = 0;
984263bc 1640 m->valid = 0;
984263bc
MD
1641
1642 /*
b12defdc
MD
1643 * Caller must be holding the object lock (asserted by
1644 * vm_page_insert()).
1645 *
1646 * NOTE: Inserting a page here does not insert it into any pmaps
1647 * (which could cause us to block allocating memory).
85946b6c
MD
1648 *
1649 * NOTE: If no object an unassociated page is allocated, m->pindex
1650 * can be used by the caller for any purpose.
984263bc 1651 */
d2d8515b
MD
1652 if (object) {
1653 if (vm_page_insert(m, object, pindex) == FALSE) {
d2d8515b 1654 vm_page_free(m);
d2d8515b 1655 if ((page_req & VM_ALLOC_NULL_OK) == 0)
f7de9d7f
MD
1656 panic("PAGE RACE %p[%ld]/%p",
1657 object, (long)pindex, m);
1658 m = NULL;
d2d8515b
MD
1659 }
1660 } else {
85946b6c 1661 m->pindex = pindex;
d2d8515b 1662 }
984263bc
MD
1663
1664 /*
1665 * Don't wakeup too often - wakeup the pageout daemon when
1666 * we would be nearly out of memory.
1667 */
20479584 1668 pagedaemon_wakeup();
984263bc 1669
9765affa
MD
1670 /*
1671 * A PG_BUSY page is returned.
1672 */
984263bc
MD
1673 return (m);
1674}
1675
79d182b0
MD
1676/*
1677 * Attempt to allocate contiguous physical memory with the specified
1678 * requirements.
1679 */
1680vm_page_t
1681vm_page_alloc_contig(vm_paddr_t low, vm_paddr_t high,
1682 unsigned long alignment, unsigned long boundary,
1683 unsigned long size)
1684{
1685 alist_blk_t blk;
1686
1687 alignment >>= PAGE_SHIFT;
1688 if (alignment == 0)
1689 alignment = 1;
1690 boundary >>= PAGE_SHIFT;
1691 if (boundary == 0)
1692 boundary = 1;
1693 size = (size + PAGE_MASK) >> PAGE_SHIFT;
1694
1695 spin_lock(&vm_contig_spin);
1696 blk = alist_alloc(&vm_contig_alist, 0, size);
1697 if (blk == ALIST_BLOCK_NONE) {
1698 spin_unlock(&vm_contig_spin);
1699 if (bootverbose) {
1700 kprintf("vm_page_alloc_contig: %ldk nospace\n",
1701 (size + PAGE_MASK) * (PAGE_SIZE / 1024));
1702 }
1703 return(NULL);
1704 }
1705 if (high && ((vm_paddr_t)(blk + size) << PAGE_SHIFT) > high) {
1706 alist_free(&vm_contig_alist, blk, size);
1707 spin_unlock(&vm_contig_spin);
1708 if (bootverbose) {
1709 kprintf("vm_page_alloc_contig: %ldk high "
1710 "%016jx failed\n",
1711 (size + PAGE_MASK) * (PAGE_SIZE / 1024),
1712 (intmax_t)high);
1713 }
1714 return(NULL);
1715 }
1716 spin_unlock(&vm_contig_spin);
ef67e7a3 1717 if (vm_contig_verbose) {
79d182b0
MD
1718 kprintf("vm_page_alloc_contig: %016jx/%ldk\n",
1719 (intmax_t)(vm_paddr_t)blk << PAGE_SHIFT,
1720 (size + PAGE_MASK) * (PAGE_SIZE / 1024));
1721 }
1722 return (PHYS_TO_VM_PAGE((vm_paddr_t)blk << PAGE_SHIFT));
1723}
1724
1725/*
1726 * Free contiguously allocated pages. The pages will be wired but not busy.
1727 * When freeing to the alist we leave them wired and not busy.
1728 */
1729void
1730vm_page_free_contig(vm_page_t m, unsigned long size)
1731{
1732 vm_paddr_t pa = VM_PAGE_TO_PHYS(m);
1733 vm_pindex_t start = pa >> PAGE_SHIFT;
1734 vm_pindex_t pages = (size + PAGE_MASK) >> PAGE_SHIFT;
1735
ef67e7a3 1736 if (vm_contig_verbose) {
79d182b0
MD
1737 kprintf("vm_page_free_contig: %016jx/%ldk\n",
1738 (intmax_t)pa, size / 1024);
1739 }
1740 if (pa < vm_low_phys_reserved) {
1741 KKASSERT(pa + size <= vm_low_phys_reserved);
1742 spin_lock(&vm_contig_spin);
1743 alist_free(&vm_contig_alist, start, pages);
1744 spin_unlock(&vm_contig_spin);
1745 } else {
1746 while (pages) {
1747 vm_page_busy_wait(m, FALSE, "cpgfr");
1748 vm_page_unwire(m, 0);
1749 vm_page_free(m);
1750 --pages;
1751 ++m;
1752 }
1753
1754 }
1755}
1756
1757
163f8d24
MD
1758/*
1759 * Wait for sufficient free memory for nominal heavy memory use kernel
1760 * operations.
55b50bd5
MD
1761 *
1762 * WARNING! Be sure never to call this in any vm_pageout code path, which
1763 * will trivially deadlock the system.
163f8d24
MD
1764 */
1765void
1766vm_wait_nominal(void)
1767{
1768 while (vm_page_count_min(0))
1769 vm_wait(0);
1770}
1771
12052253
MD
1772/*
1773 * Test if vm_wait_nominal() would block.
1774 */
1775int
1776vm_test_nominal(void)
1777{
1778 if (vm_page_count_min(0))
1779 return(1);
1780 return(0);
1781}
1782
984263bc 1783/*
de71fd3f
MD
1784 * Block until free pages are available for allocation, called in various
1785 * places before memory allocations.
cd3c66bd
MD
1786 *
1787 * The caller may loop if vm_page_count_min() == FALSE so we cannot be
1788 * more generous then that.
984263bc 1789 */
984263bc 1790void
4ecf7cc9 1791vm_wait(int timo)
984263bc 1792{
cd3c66bd
MD
1793 /*
1794 * never wait forever
1795 */
1796 if (timo == 0)
1797 timo = hz;
9ad0147b 1798 lwkt_gettoken(&vm_token);
cd3c66bd 1799
bc6dffab 1800 if (curthread == pagethread) {
cd3c66bd
MD
1801 /*
1802 * The pageout daemon itself needs pages, this is bad.
1803 */
1804 if (vm_page_count_min(0)) {
1805 vm_pageout_pages_needed = 1;
1806 tsleep(&vm_pageout_pages_needed, 0, "VMWait", timo);
1807 }
984263bc 1808 } else {
cd3c66bd
MD
1809 /*
1810 * Wakeup the pageout daemon if necessary and wait.
1811 */
1812 if (vm_page_count_target()) {
1813 if (vm_pages_needed == 0) {
1814 vm_pages_needed = 1;
1815 wakeup(&vm_pages_needed);
1816 }
1817 ++vm_pages_waiting; /* SMP race ok */
1818 tsleep(&vmstats.v_free_count, 0, "vmwait", timo);
984263bc 1819 }
984263bc 1820 }
9ad0147b 1821 lwkt_reltoken(&vm_token);
984263bc
MD
1822}
1823
1824/*
de71fd3f
MD
1825 * Block until free pages are available for allocation
1826 *
cd3c66bd 1827 * Called only from vm_fault so that processes page faulting can be
de71fd3f 1828 * easily tracked.
984263bc 1829 */
984263bc 1830void
3b47bfc4 1831vm_wait_pfault(void)
984263bc 1832{
cd3c66bd
MD
1833 /*
1834 * Wakeup the pageout daemon if necessary and wait.
1835 */
3b47bfc4 1836 if (vm_page_count_min(0)) {
cd3c66bd 1837 lwkt_gettoken(&vm_token);
3b47bfc4
MD
1838 while (vm_page_count_severe()) {
1839 if (vm_page_count_target()) {
1840 if (vm_pages_needed == 0) {
1841 vm_pages_needed = 1;
1842 wakeup(&vm_pages_needed);
1843 }
1844 ++vm_pages_waiting; /* SMP race ok */
1845 tsleep(&vmstats.v_free_count, 0, "pfault", hz);
cd3c66bd 1846 }
cd3c66bd
MD
1847 }
1848 lwkt_reltoken(&vm_token);
984263bc 1849 }
984263bc
MD
1850}
1851
984263bc 1852/*
de71fd3f
MD
1853 * Put the specified page on the active list (if appropriate). Ensure
1854 * that act_count is at least ACT_INIT but do not otherwise mess with it.
984263bc 1855 *
b12defdc 1856 * The caller should be holding the page busied ? XXX
de71fd3f 1857 * This routine may not block.
984263bc
MD
1858 */
1859void
1860vm_page_activate(vm_page_t m)
1861{
b12defdc 1862 u_short oqueue;
984263bc 1863
b12defdc 1864 vm_page_spin_lock(m);
027193eb 1865 if (m->queue - m->pc != PQ_ACTIVE) {
b12defdc
MD
1866 _vm_page_queue_spin_lock(m);
1867 oqueue = _vm_page_rem_queue_spinlocked(m);
1868 /* page is left spinlocked, queue is unlocked */
984263bc 1869
b12defdc
MD
1870 if (oqueue == PQ_CACHE)
1871 mycpu->gd_cnt.v_reactivated++;
984263bc 1872 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
984263bc
MD
1873 if (m->act_count < ACT_INIT)
1874 m->act_count = ACT_INIT;
027193eb 1875 _vm_page_add_queue_spinlocked(m, PQ_ACTIVE + m->pc, 0);
984263bc 1876 }
b12defdc
MD
1877 _vm_page_and_queue_spin_unlock(m);
1878 if (oqueue == PQ_CACHE || oqueue == PQ_FREE)
1879 pagedaemon_wakeup();
984263bc
MD
1880 } else {
1881 if (m->act_count < ACT_INIT)
1882 m->act_count = ACT_INIT;
b12defdc 1883 vm_page_spin_unlock(m);
984263bc 1884 }
984263bc
MD
1885}
1886
1887/*
de71fd3f
MD
1888 * Helper routine for vm_page_free_toq() and vm_page_cache(). This
1889 * routine is called when a page has been added to the cache or free
1890 * queues.
984263bc 1891 *
de71fd3f 1892 * This routine may not block.
984263bc
MD
1893 */
1894static __inline void
1895vm_page_free_wakeup(void)
1896{
1897 /*
cd3c66bd
MD
1898 * If the pageout daemon itself needs pages, then tell it that
1899 * there are some free.
984263bc
MD
1900 */
1901 if (vm_pageout_pages_needed &&
de71fd3f
MD
1902 vmstats.v_cache_count + vmstats.v_free_count >=
1903 vmstats.v_pageout_free_min
1904 ) {
984263bc
MD
1905 wakeup(&vm_pageout_pages_needed);
1906 vm_pageout_pages_needed = 0;
1907 }
de71fd3f 1908
984263bc 1909 /*
cd3c66bd
MD
1910 * Wakeup processes that are waiting on memory.
1911 *
1912 * NOTE: vm_paging_target() is the pageout daemon's target, while
1913 * vm_page_count_target() is somewhere inbetween. We want
1914 * to wake processes up prior to the pageout daemon reaching
1915 * its target to provide some hysteresis.
984263bc 1916 */
cd3c66bd
MD
1917 if (vm_pages_waiting) {
1918 if (!vm_page_count_target()) {
1919 /*
1920 * Plenty of pages are free, wakeup everyone.
1921 */
1922 vm_pages_waiting = 0;
1923 wakeup(&vmstats.v_free_count);
1924 ++mycpu->gd_cnt.v_ppwakeups;
1925 } else if (!vm_page_count_min(0)) {
1926 /*
1927 * Some pages are free, wakeup someone.
1928 */
1929 int wcount = vm_pages_waiting;
1930 if (wcount > 0)
1931 --wcount;
1932 vm_pages_waiting = wcount;
1933 wakeup_one(&vmstats.v_free_count);
1934 ++mycpu->gd_cnt.v_ppwakeups;
1935 }
984263bc
MD
1936 }
1937}
1938
1939/*
b12defdc
MD
1940 * Returns the given page to the PQ_FREE or PQ_HOLD list and disassociates
1941 * it from its VM object.
984263bc 1942 *
b12defdc
MD
1943 * The vm_page must be PG_BUSY on entry. PG_BUSY will be released on
1944 * return (the page will have been freed).
984263bc 1945 */
984263bc
MD
1946void
1947vm_page_free_toq(vm_page_t m)
1948{
12e4aaff 1949 mycpu->gd_cnt.v_tfree++;
17cde63e 1950 KKASSERT((m->flags & PG_MAPPED) == 0);
b12defdc 1951 KKASSERT(m->flags & PG_BUSY);
17cde63e 1952
984263bc 1953 if (m->busy || ((m->queue - m->pc) == PQ_FREE)) {
79d182b0
MD
1954 kprintf("vm_page_free: pindex(%lu), busy(%d), "
1955 "PG_BUSY(%d), hold(%d)\n",
1956 (u_long)m->pindex, m->busy,
1957 ((m->flags & PG_BUSY) ? 1 : 0), m->hold_count);
984263bc
MD
1958 if ((m->queue - m->pc) == PQ_FREE)
1959 panic("vm_page_free: freeing free page");
1960 else
1961 panic("vm_page_free: freeing busy page");
1962 }
1963
1964 /*
b12defdc
MD
1965 * Remove from object, spinlock the page and its queues and
1966 * remove from any queue. No queue spinlock will be held
1967 * after this section (because the page was removed from any
1968 * queue).
984263bc 1969 */
984263bc 1970 vm_page_remove(m);
b12defdc
MD
1971 vm_page_and_queue_spin_lock(m);
1972 _vm_page_rem_queue_spinlocked(m);
984263bc
MD
1973
1974 /*
f2d22ebf
MD
1975 * No further management of fictitious pages occurs beyond object
1976 * and queue removal.
984263bc 1977 */
984263bc 1978 if ((m->flags & PG_FICTITIOUS) != 0) {
b12defdc 1979 vm_page_spin_unlock(m);
9765affa 1980 vm_page_wakeup(m);
984263bc
MD
1981 return;
1982 }
1983
1984 m->valid = 0;
1985 vm_page_undirty(m);
1986
1987 if (m->wire_count != 0) {
1988 if (m->wire_count > 1) {
de71fd3f
MD
1989 panic(
1990 "vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1991 m->wire_count, (long)m->pindex);
984263bc 1992 }
73c351d1 1993 panic("vm_page_free: freeing wired page");
984263bc
MD
1994 }
1995
984263bc
MD
1996 /*
1997 * Clear the UNMANAGED flag when freeing an unmanaged page.
9bf025db 1998 * Clear the NEED_COMMIT flag
984263bc 1999 */
9bf025db 2000 if (m->flags & PG_UNMANAGED)
b12defdc 2001 vm_page_flag_clear(m, PG_UNMANAGED);
9bf025db
MD
2002 if (m->flags & PG_NEED_COMMIT)
2003 vm_page_flag_clear(m, PG_NEED_COMMIT);
984263bc
MD
2004
2005 if (m->hold_count != 0) {
d0aa00e8 2006 vm_page_flag_clear(m, PG_ZERO);
027193eb 2007 _vm_page_add_queue_spinlocked(m, PQ_HOLD + m->pc, 0);
de71fd3f 2008 } else {
b12defdc 2009 _vm_page_add_queue_spinlocked(m, PQ_FREE + m->pc, 0);
de71fd3f 2010 }
984263bc
MD
2011
2012 /*
b12defdc
MD
2013 * This sequence allows us to clear PG_BUSY while still holding
2014 * its spin lock, which reduces contention vs allocators. We
2015 * must not leave the queue locked or _vm_page_wakeup() may
2016 * deadlock.
984263bc 2017 */
b12defdc
MD
2018 _vm_page_queue_spin_unlock(m);
2019 if (_vm_page_wakeup(m)) {
2020 vm_page_spin_unlock(m);
2021 wakeup(m);
984263bc 2022 } else {
b12defdc 2023 vm_page_spin_unlock(m);
984263bc 2024 }
984263bc 2025 vm_page_free_wakeup();
984263bc
MD
2026}
2027
bb6811be
MD
2028/*
2029 * vm_page_free_fromq_fast()
2030 *
2031 * Remove a non-zero page from one of the free queues; the page is removed for
2032 * zeroing, so do not issue a wakeup.
bb6811be
MD
2033 */
2034vm_page_t
2035vm_page_free_fromq_fast(void)
2036{
2037 static int qi;
2038 vm_page_t m;
2039 int i;
2040
bb6811be
MD
2041 for (i = 0; i < PQ_L2_SIZE; ++i) {
2042 m = vm_page_list_find(PQ_FREE, qi, FALSE);
b12defdc
MD
2043 /* page is returned spinlocked and removed from its queue */
2044 if (m) {
2045 if (vm_page_busy_try(m, TRUE)) {
2046 /*
2047 * We were unable to busy the page, deactivate
2048 * it and loop.
2049 */
2050 _vm_page_deactivate_locked(m, 0);
2051 vm_page_spin_unlock(m);
90244566 2052 } else if (m->flags & PG_ZERO) {
b12defdc
MD
2053 /*
2054 * The page is PG_ZERO, requeue it and loop
2055 */
2056 _vm_page_add_queue_spinlocked(m,
2057 PQ_FREE + m->pc,
2058 0);
2059 vm_page_queue_spin_unlock(m);
2060 if (_vm_page_wakeup(m)) {
2061 vm_page_spin_unlock(m);
2062 wakeup(m);
2063 } else {
2064 vm_page_spin_unlock(m);
2065 }
90244566
MD
2066 } else {
2067 /*
2068 * The page is not PG_ZERO'd so return it.
2069 */
2070 vm_page_spin_unlock(m);
9bf025db
MD
2071 KKASSERT((m->flags & (PG_UNMANAGED |
2072 PG_NEED_COMMIT)) == 0);
90244566
MD
2073 KKASSERT(m->hold_count == 0);
2074 KKASSERT(m->wire_count == 0);
2075 break;
b12defdc
MD
2076 }
2077 m = NULL;
bb6811be 2078 }
b12defdc 2079 qi = (qi + PQ_PRIME2) & PQ_L2_MASK;
bb6811be 2080 }
bb6811be
MD
2081 return (m);
2082}
2083
984263bc 2084/*
de71fd3f
MD
2085 * vm_page_unmanage()
2086 *
2087 * Prevent PV management from being done on the page. The page is
2088 * removed from the paging queues as if it were wired, and as a
2089 * consequence of no longer being managed the pageout daemon will not
2090 * touch it (since there is no way to locate the pte mappings for the
2091 * page). madvise() calls that mess with the pmap will also no longer
2092 * operate on the page.
2093 *
2094 * Beyond that the page is still reasonably 'normal'. Freeing the page
2095 * will clear the flag.
2096 *
2097 * This routine is used by OBJT_PHYS objects - objects using unswappable
2098 * physical memory as backing store rather then swap-backed memory and
2099 * will eventually be extended to support 4MB unmanaged physical
2100 * mappings.
654a39f0 2101 *
b12defdc 2102 * Caller must be holding the page busy.
984263bc 2103 */
984263bc
MD
2104void
2105vm_page_unmanage(vm_page_t m)
2106{
b12defdc 2107 KKASSERT(m->flags & PG_BUSY);
984263bc
MD
2108 if ((m->flags & PG_UNMANAGED) == 0) {
2109 if (m->wire_count == 0)
2110 vm_page_unqueue(m);
2111 }
2112 vm_page_flag_set(m, PG_UNMANAGED);
984263bc
MD
2113}
2114
2115/*
de71fd3f
MD
2116 * Mark this page as wired down by yet another map, removing it from
2117 * paging queues as necessary.
984263bc 2118 *
b12defdc 2119 * Caller must be holding the page busy.
984263bc
MD
2120 */
2121void
2122vm_page_wire(vm_page_t m)
2123{
984263bc
MD
2124 /*
2125 * Only bump the wire statistics if the page is not already wired,
2126 * and only unqueue the page if it is on some queue (if it is unmanaged
f2d22ebf
MD
2127 * it is already off the queues). Don't do anything with fictitious
2128 * pages because they are always wired.
984263bc 2129 */
b12defdc 2130 KKASSERT(m->flags & PG_BUSY);
f2d22ebf 2131 if ((m->flags & PG_FICTITIOUS) == 0) {
b12defdc 2132 if (atomic_fetchadd_int(&m->wire_count, 1) == 0) {
f2d22ebf
MD
2133 if ((m->flags & PG_UNMANAGED) == 0)
2134 vm_page_unqueue(m);
b12defdc 2135 atomic_add_int(&vmstats.v_wire_count, 1);
f2d22ebf 2136 }
f2d22ebf 2137 KASSERT(m->wire_count != 0,
17cde63e 2138 ("vm_page_wire: wire_count overflow m=%p", m));
984263bc 2139 }
984263bc
MD
2140}
2141
2142/*
de71fd3f
MD
2143 * Release one wiring of this page, potentially enabling it to be paged again.
2144 *
2145 * Many pages placed on the inactive queue should actually go
2146 * into the cache, but it is difficult to figure out which. What
2147 * we do instead, if the inactive target is well met, is to put
2148 * clean pages at the head of the inactive queue instead of the tail.
2149 * This will cause them to be moved to the cache more quickly and
2150 * if not actively re-referenced, freed more quickly. If we just
2151 * stick these pages at the end of the inactive queue, heavy filesystem
2152 * meta-data accesses can cause an unnecessary paging load on memory bound
2153 * processes. This optimization causes one-time-use metadata to be
2154 * reused more quickly.
2155 *
f84f7e81
MD
2156 * Pages marked PG_NEED_COMMIT are always activated and never placed on
2157 * the inactive queue. This helps the pageout daemon determine memory
2158 * pressure and act on out-of-memory situations more quickly.
2159 *
de71fd3f
MD
2160 * BUT, if we are in a low-memory situation we have no choice but to
2161 * put clean pages on the cache queue.
2162 *
2163 * A number of routines use vm_page_unwire() to guarantee that the page
2164 * will go into either the inactive or active queues, and will NEVER
2165 * be placed in the cache - for example, just after dirtying a page.
2166 * dirty pages in the cache are not allowed.
2167 *
2168 * The page queues must be locked.
2169 * This routine may not block.
984263bc
MD
2170 */
2171void
2172vm_page_unwire(vm_page_t m, int activate)
2173{
b12defdc 2174 KKASSERT(m->flags & PG_BUSY);
f2d22ebf
MD
2175 if (m->flags & PG_FICTITIOUS) {
2176 /* do nothing */
2177 } else if (m->wire_count <= 0) {
2178 panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
2179 } else {
b12defdc
MD
2180 if (atomic_fetchadd_int(&m->wire_count, -1) == 1) {
2181 atomic_add_int(&vmstats.v_wire_count, -1);
984263bc
MD
2182 if (m->flags & PG_UNMANAGED) {
2183 ;
f84f7e81 2184 } else if (activate || (m->flags & PG_NEED_COMMIT)) {
b12defdc 2185 vm_page_spin_lock(m);
027193eb
MD
2186 _vm_page_add_queue_spinlocked(m,
2187 PQ_ACTIVE + m->pc, 0);
b12defdc 2188 _vm_page_and_queue_spin_unlock(m);
984263bc 2189 } else {
b12defdc 2190 vm_page_spin_lock(m);
984263bc 2191 vm_page_flag_clear(m, PG_WINATCFLS);
027193eb
MD
2192 _vm_page_add_queue_spinlocked(m,
2193 PQ_INACTIVE + m->pc, 0);
e527fb6b 2194 ++vm_swapcache_inactive_heuristic;
b12defdc 2195 _vm_page_and_queue_spin_unlock(m);
984263bc
MD
2196 }
2197 }
984263bc 2198 }
984263bc
MD
2199}
2200
984263bc
MD
2201/*
2202 * Move the specified page to the inactive queue. If the page has
2203 * any associated swap, the swap is deallocated.
2204 *
2205 * Normally athead is 0 resulting in LRU operation. athead is set
2206 * to 1 if we want this page to be 'as if it were placed in the cache',
2207 * except without unmapping it from the process address space.
2208 *
b12defdc 2209 * vm_page's spinlock must be held on entry and will remain held on return.
984263bc
MD
2210 * This routine may not block.
2211 */
b12defdc
MD
2212static void
2213_vm_page_deactivate_locked(vm_page_t m, int athead)
984263bc 2214{
b12defdc
MD
2215 u_short oqueue;
2216
984263bc
MD
2217 /*
2218 * Ignore if already inactive.
2219 */
027193eb 2220 if (m->queue - m->pc == PQ_INACTIVE)
984263bc 2221 return;
b12defdc
MD
2222 _vm_page_queue_spin_lock(m);
2223 oqueue = _vm_page_rem_queue_spinlocked(m);
984263bc 2224
984263bc 2225 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
b12defdc 2226 if (oqueue == PQ_CACHE)
12e4aaff 2227 mycpu->gd_cnt.v_reactivated++;
984263bc 2228 vm_page_flag_clear(m, PG_WINATCFLS);
027193eb 2229 _vm_page_add_queue_spinlocked(m, PQ_INACTIVE + m->pc, athead);
b12defdc 2230 if (athead == 0)
e527fb6b 2231 ++vm_swapcache_inactive_heuristic;
984263bc 2232 }
b12defdc
MD
2233 _vm_page_queue_spin_unlock(m);
2234 /* leaves vm_page spinlocked */
984263bc
MD
2235}
2236
573fb415
MD
2237/*
2238 * Attempt to deactivate a page.
2239 *
2240 * No requirements.
2241 */
984263bc
MD
2242void
2243vm_page_deactivate(vm_page_t m)
2244{
b12defdc
MD
2245 vm_page_spin_lock(m);
2246 _vm_page_deactivate_locked(m, 0);
2247 vm_page_spin_unlock(m);
2248}
2249
2250void
2251vm_page_deactivate_locked(vm_page_t m)
2252{
2253 _vm_page_deactivate_locked(m, 0);
984263bc
MD
2254}
2255
2256/*
573fb415 2257 * Attempt to move a page to PQ_CACHE.
b12defdc 2258 *
984263bc 2259 * Returns 0 on failure, 1 on success
573fb415 2260 *
b12defdc
MD
2261 * The page should NOT be busied by the caller. This function will validate
2262 * whether the page can be safely moved to the cache.
984263bc
MD
2263 */
2264int
2265vm_page_try_to_cache(vm_page_t m)
2266{
b12defdc
MD
2267 vm_page_spin_lock(m);
2268 if (vm_page_busy_try(m, TRUE)) {
2269 vm_page_spin_unlock(m);
2270 return(0);
2271 }
2272 if (m->dirty || m->hold_count || m->wire_count ||
9bf025db 2273 (m->flags & (PG_UNMANAGED | PG_NEED_COMMIT))) {
b12defdc
MD
2274 if (_vm_page_wakeup(m)) {
2275 vm_page_spin_unlock(m);
2276 wakeup(m);
2277 } else {
2278 vm_page_spin_unlock(m);
2279 }
984263bc
MD
2280 return(0);
2281 }
b12defdc
MD
2282 vm_page_spin_unlock(m);
2283
2284 /*
2285 * Page busied by us and no longer spinlocked. Dirty pages cannot
2286 * be moved to the cache.
2287 */
984263bc 2288 vm_page_test_dirty(m);
d86d27a8 2289 if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
50e32333 2290 vm_page_wakeup(m);
984263bc 2291 return(0);
654a39f0 2292 }
984263bc
MD
2293 vm_page_cache(m);
2294 return(1);
2295}
2296
2297/*
de71fd3f
MD
2298 * Attempt to free the page. If we cannot free it, we do nothing.
2299 * 1 is returned on success, 0 on failure.
573fb415
MD
2300 *
2301 * No requirements.
984263bc 2302 */
984263bc
MD
2303int
2304vm_page_try_to_free(vm_page_t m)
2305{
b12defdc
MD
2306 vm_page_spin_lock(m);
2307 if (vm_page_busy_try(m, TRUE)) {
2308 vm_page_spin_unlock(m);
2309 return(0);
2310 }
82034c53
MD
2311
2312 /*
2313 * The page can be in any state, including already being on the free
2314 * queue. Check to see if it really can be freed.
2315 */
2316 if (m->dirty || /* can't free if it is dirty */
2317 m->hold_count || /* or held (XXX may be wrong) */
2318 m->wire_count || /* or wired */
9bf025db
MD
2319 (m->flags & (PG_UNMANAGED | /* or unmanaged */
2320 PG_NEED_COMMIT)) || /* or needs a commit */
82034c53
MD
2321 m->queue - m->pc == PQ_FREE || /* already on PQ_FREE */
2322 m->queue - m->pc == PQ_HOLD) { /* already on PQ_HOLD */
b12defdc
MD
2323 if (_vm_page_wakeup(m)) {
2324 vm_page_spin_unlock(m);
2325 wakeup(m);
2326 } else {
2327 vm_page_spin_unlock(m);
2328 }
984263bc
MD
2329 return(0);
2330 }
b12defdc
MD
2331 vm_page_spin_unlock(m);
2332
2333 /*
82034c53
MD
2334 * We can probably free the page.
2335 *
b12defdc
MD
2336 * Page busied by us and no longer spinlocked. Dirty pages will
2337 * not be freed by this function. We have to re-test the
2338 * dirty bit after cleaning out the pmaps.
2339 */
984263bc 2340 vm_page_test_dirty(m);
d86d27a8 2341 if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
b12defdc 2342 vm_page_wakeup(m);
984263bc 2343 return(0);
654a39f0 2344 }
984263bc 2345 vm_page_protect(m, VM_PROT_NONE);
d86d27a8 2346 if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
b12defdc
MD
2347 vm_page_wakeup(m);
2348 return(0);
2349 }
984263bc
MD
2350 vm_page_free(m);
2351 return(1);
2352}
2353
984263bc
MD
2354/*
2355 * vm_page_cache
2356 *
2357 * Put the specified page onto the page cache queue (if appropriate).
2358 *
a491077e
MD
2359 * The page must be busy, and this routine will release the busy and
2360 * possibly even free the page.
984263bc
MD
2361 */
2362void
2363vm_page_cache(vm_page_t m)
2364{
9bf025db
MD
2365 if ((m->flags & (PG_UNMANAGED | PG_NEED_COMMIT)) ||
2366 m->busy || m->wire_count || m->hold_count) {
086c1d7e 2367 kprintf("vm_page_cache: attempting to cache busy/held page\n");
a491077e 2368 vm_page_wakeup(m);
984263bc
MD
2369 return;
2370 }
c9ec86b3
MD
2371
2372 /*
2373 * Already in the cache (and thus not mapped)
2374 */
17cde63e
MD
2375 if ((m->queue - m->pc) == PQ_CACHE) {
2376 KKASSERT((m->flags & PG_MAPPED) == 0);
a491077e 2377 vm_page_wakeup(m);
984263bc 2378 return;
17cde63e 2379 }
984263bc
MD
2380
2381 /*
c9ec86b3
MD
2382 * Caller is required to test m->dirty, but note that the act of
2383 * removing the page from its maps can cause it to become dirty
2384 * on an SMP system due to another cpu running in usermode.
984263bc 2385 */
c9ec86b3 2386 if (m->dirty) {
984263bc
MD
2387 panic("vm_page_cache: caching a dirty page, pindex: %ld",
2388 (long)m->pindex);
2389 }
c9ec86b3
MD
2390
2391 /*
2392 * Remove all pmaps and indicate that the page is not
17cde63e
MD
2393 * writeable or mapped. Our vm_page_protect() call may
2394 * have blocked (especially w/ VM_PROT_NONE), so recheck
2395 * everything.
c9ec86b3
MD
2396 */
2397 vm_page_protect(m, VM_PROT_NONE);
9bf025db
MD
2398 if ((m->flags & (PG_UNMANAGED | PG_MAPPED)) ||
2399 m->busy || m->wire_count || m->hold_count) {
a491077e 2400 vm_page_wakeup(m);
9bf025db 2401 } else if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
c9ec86b3 2402 vm_page_deactivate(m);
a491077e 2403 vm_page_wakeup(m);
c9ec86b3 2404 } else {
b12defdc
MD
2405 _vm_page_and_queue_spin_lock(m);
2406 _vm_page_rem_queue_spinlocked(m);
2407 _vm_page_add_queue_spinlocked(m, PQ_CACHE + m->pc, 0);
2408 _vm_page_queue_spin_unlock(m);
2409 if (_vm_page_wakeup(m)) {
2410 vm_page_spin_unlock(m);
2411 wakeup(m);
2412 } else {
2413 vm_page_spin_unlock(m);
2414 }
c9ec86b3
MD
2415 vm_page_free_wakeup();
2416 }
984263bc
MD
2417}
2418
2419/*
de71fd3f
MD
2420 * vm_page_dontneed()
2421 *
2422 * Cache, deactivate, or do nothing as appropriate. This routine
2423 * is typically used by madvise() MADV_DONTNEED.
2424 *
2425 * Generally speaking we want to move the page into the cache so
2426 * it gets reused quickly. However, this can result in a silly syndrome
2427 * due to the page recycling too quickly. Small objects will not be
2428 * fully cached. On the otherhand, if we move the page to the inactive
2429 * queue we wind up with a problem whereby very large objects
2430 * unnecessarily blow away our inactive and cache queues.
2431 *
2432 * The solution is to move the pages based on a fixed weighting. We
2433 * either leave them alone, deactivate them, or move them to the cache,
2434 * where moving them to the cache has the highest weighting.
2435 * By forcing some pages into other queues we eventually force the
2436 * system to balance the queues, potentially recovering other unrelated
2437 * space from active. The idea is to not force this to happen too
2438 * often.
573fb415 2439 *
b12defdc 2440 * The page must be busied.
984263bc 2441 */
984263bc
MD
2442void
2443vm_page_dontneed(vm_page_t m)
2444{
2445 static int dnweight;
2446 int dnw;
2447 int head;
2448
2449 dnw = ++dnweight;
2450
2451 /*
2452 * occassionally leave the page alone
2453 */
984263bc 2454 if ((dnw & 0x01F0) == 0 ||
027193eb 2455 m->queue - m->pc == PQ_INACTIVE ||
984263bc
MD
2456 m->queue - m->pc == PQ_CACHE
2457 ) {
2458 if (m->act_count >= ACT_INIT)
2459 --m->act_count;
2460 return;
2461 }
2462
31da5e4d
VS
2463 /*
2464 * If vm_page_dontneed() is inactivating a page, it must clear
2465 * the referenced flag; otherwise the pagedaemon will see references
2466 * on the page in the inactive queue and reactivate it. Until the
2467 * page can move to the cache queue, madvise's job is not done.
2468 */
2469 vm_page_flag_clear(m, PG_REFERENCED);
2470 pmap_clear_reference(m);
2471
984263bc
MD
2472 if (m->dirty == 0)
2473 vm_page_test_dirty(m);
2474
2475 if (m->dirty || (dnw & 0x0070) == 0) {
2476 /*
2477 * Deactivate the page 3 times out of 32.
2478 */
2479 head = 0;
2480 } else {
2481 /*
2482 * Cache the page 28 times out of every 32. Note that
2483 * the page is deactivated instead of cached, but placed
2484 * at the head of the queue instead of the tail.
2485 */
2486 head = 1;
2487 }
b12defdc
MD
2488 vm_page_spin_lock(m);
2489 _vm_page_deactivate_locked(m, head);
2490 vm_page_spin_unlock(m);
2491}
2492
2493/*
2494 * These routines manipulate the 'soft busy' count for a page. A soft busy
2495 * is almost like PG_BUSY except that it allows certain compatible operations
2496 * to occur on the page while it is busy. For example, a page undergoing a
2497 * write can still be mapped read-only.
2498 *
2499 * Because vm_pages can overlap buffers m->busy can be > 1. m->busy is only
2500 * adjusted while the vm_page is PG_BUSY so the flash will occur when the
2501 * busy bit is cleared.
2502 */
2503void
2504vm_page_io_start(vm_page_t m)
2505{
2506 KASSERT(m->flags & PG_BUSY, ("vm_page_io_start: page not busy!!!"));
2507 atomic_add_char(&m->busy, 1);
2508 vm_page_flag_set(m, PG_SBUSY);
2509}
2510
2511void
2512vm_page_io_finish(vm_page_t m)
2513{
2514 KASSERT(m->flags & PG_BUSY, ("vm_page_io_finish: page not busy!!!"));
2515 atomic_subtract_char(&m->busy, 1);
2516 if (m->busy == 0)
2517 vm_page_flag_clear(m, PG_SBUSY);
984263bc
MD
2518}
2519
9bf025db
MD
2520/*
2521 * Indicate that a clean VM page requires a filesystem commit and cannot
2522 * be reused. Used by tmpfs.
2523 */
2524void
2525vm_page_need_commit(vm_page_t m)
2526{
2527 vm_page_flag_set(m, PG_NEED_COMMIT);
d86d27a8 2528 vm_object_set_writeable_dirty(m->object);
9bf025db
MD
2529}
2530
2531void
2532vm_page_clear_commit(vm_page_t m)
2533{
2534 vm_page_flag_clear(m, PG_NEED_COMMIT);
2535}
2536
984263bc 2537/*
06ecca5a 2538 * Grab a page, blocking if it is busy and allocating a page if necessary.
d2d8515b
MD
2539 * A busy page is returned or NULL. The page may or may not be valid and
2540 * might not be on a queue (the caller is responsible for the disposition of
2541 * the page).
984263bc 2542 *
d2d8515b
MD
2543 * If VM_ALLOC_ZERO is specified and the grab must allocate a new page, the
2544 * page will be zero'd and marked valid.
b12defdc 2545 *
d2d8515b
MD
2546 * If VM_ALLOC_FORCE_ZERO is specified the page will be zero'd and marked
2547 * valid even if it already exists.
2548 *
2549 * If VM_ALLOC_RETRY is specified this routine will never return NULL. Also
2550 * note that VM_ALLOC_NORMAL must be specified if VM_ALLOC_RETRY is specified.
d149178e 2551 * VM_ALLOC_NULL_OK is implied when VM_ALLOC_RETRY is specified.
dc1fd4b3 2552 *
06ecca5a
MD
2553 * This routine may block, but if VM_ALLOC_RETRY is not set then NULL is
2554 * always returned if we had blocked.
d2d8515b 2555 *
06ecca5a 2556 * This routine may not be called from an interrupt.
06ecca5a 2557 *
d2d8515b 2558 * PG_ZERO is *ALWAYS* cleared by this routine.
573fb415 2559 *
d2d8515b 2560 * No other requirements.
984263bc
MD
2561 */
2562vm_page_t
2563vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
2564{
984263bc 2565 vm_page_t m;
b12defdc 2566 int error;
984263bc 2567
dc1fd4b3
MD
2568 KKASSERT(allocflags &
2569 (VM_ALLOC_NORMAL|VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
398c240d 2570 vm_object_hold(object);
b12defdc
MD
2571 for (;;) {
2572 m = vm_page_lookup_busy_try(object, pindex, TRUE, &error);
2573 if (error) {
2574 vm_page_sleep_busy(m, TRUE, "pgrbwt");
2575 if ((allocflags & VM_ALLOC_RETRY) == 0) {
2576 m = NULL;
2577 break;
984263bc 2578 }
d2d8515b 2579 /* retry */
b12defdc 2580 } else if (m == NULL) {
d149178e
MD
2581 if (allocflags & VM_ALLOC_RETRY)
2582 allocflags |= VM_ALLOC_NULL_OK;
b12defdc
MD
2583 m = vm_page_alloc(object, pindex,
2584 allocflags & ~VM_ALLOC_RETRY);
2585 if (m)
2586 break;
2587 vm_wait(0);
2588 if ((allocflags & VM_ALLOC_RETRY) == 0)
d2d8515b 2589 goto failed;
984263bc 2590 } else {
b12defdc
MD
2591 /* m found */
2592 break;
984263bc
MD
2593 }
2594 }
d2d8515b
MD
2595
2596 /*
2597 * If VM_ALLOC_ZERO an invalid page will be zero'd and set valid.
2598 *
2599 * If VM_ALLOC_FORCE_ZERO the page is unconditionally zero'd and set
2600 * valid even if already valid.
2601 */
2602 if (m->valid == 0) {
2603 if (allocflags & (VM_ALLOC_ZERO | VM_ALLOC_FORCE_ZERO)) {
2604 if ((m->flags & PG_ZERO) == 0)
2605 pmap_zero_page(VM_PAGE_TO_PHYS(m));
2606 m->valid = VM_PAGE_BITS_ALL;
2607 }
2608 } else if (allocflags & VM_ALLOC_FORCE_ZERO) {
2609 pmap_zero_page(VM_PAGE_TO_PHYS(m));
2610 m->valid = VM_PAGE_BITS_ALL;
2611 }
2612 vm_page_flag_clear(m, PG_ZERO);
2613failed:
398c240d 2614 vm_object_drop(object);
06ecca5a 2615 return(m);
984263bc
MD
2616}
2617
2618/*
2619 * Mapping function for valid bits or for dirty bits in
2620 * a page. May not block.
2621 *
2622 * Inputs are required to range within a page.
573fb415
MD
2623 *
2624 * No requirements.
2625 * Non blocking.
984263bc 2626 */
573fb415 2627int
984263bc
MD
2628vm_page_bits(int base, int size)
2629{
2630 int first_bit;
2631 int last_bit;
2632
2633 KASSERT(
2634 base + size <= PAGE_SIZE,
2635 ("vm_page_bits: illegal base/size %d/%d", base, size)
2636 );
2637
2638 if (size == 0) /* handle degenerate case */
2639 return(0);
2640
2641 first_bit = base >> DEV_BSHIFT;
2642 last_bit = (base + size - 1) >> DEV_BSHIFT;
2643
2644 return ((2 << last_bit) - (1 << first_bit));
2645}
2646
2647/*
de71fd3f
MD
2648 * Sets portions of a page valid and clean. The arguments are expected
2649 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
2650 * of any partial chunks touched by the range. The invalid portion of
2651 * such chunks will be zero'd.
984263bc 2652 *
c7841cbe
MD
2653 * NOTE: When truncating a buffer vnode_pager_setsize() will automatically
2654 * align base to DEV_BSIZE so as not to mark clean a partially
2655 * truncated device block. Otherwise the dirty page status might be
2656 * lost.
2657 *
de71fd3f 2658 * This routine may not block.
984263bc 2659 *
de71fd3f 2660 * (base + size) must be less then or equal to PAGE_SIZE.
984263bc 2661 */
1a54183b
MD
2662static void
2663_vm_page_zero_valid(vm_page_t m, int base, int size)
984263bc 2664{
984263bc
MD
2665 int frag;
2666 int endoff;
2667
2668 if (size == 0) /* handle degenerate case */
2669 return;
2670
2671 /*
2672 * If the base is not DEV_BSIZE aligned and the valid
2673 * bit is clear, we have to zero out a portion of the
2674 * first block.
2675 */
2676
2677 if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
2678 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0
2679 ) {
2680 pmap_zero_page_area(
2681 VM_PAGE_TO_PHYS(m),
2682 frag,
2683 base - frag
2684 );
2685 }
2686
2687 /*
2688 * If the ending offset is not DEV_BSIZE aligned and the
2689 * valid bit is clear, we have to zero out a portion of
2690 * the last block.
2691 */
2692
2693 endoff = base + size;
2694
2695 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
2696 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0
2697 ) {
2698 pmap_zero_page_area(
2699 VM_PAGE_TO_PHYS(m),
2700 endoff,
2701 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))
2702 );
2703 }
1a54183b 2704}
984263bc 2705
1a54183b
MD
2706/*
2707 * Set valid, clear dirty bits. If validating the entire
2708 * page we can safely clear the pmap modify bit. We also
2709 * use this opportunity to clear the PG_NOSYNC flag. If a process
2710 * takes a write fault on a MAP_NOSYNC memory area the flag will
2711 * be set again.
2712 *
2713 * We set valid bits inclusive of any overlap, but we can only
2714 * clear dirty bits for DEV_BSIZE chunks that are fully within
2715 * the range.
573fb415
MD
2716 *
2717 * Page must be busied?
2718 * No other requirements.
1a54183b
MD
2719 */
2720void
2721vm_page_set_valid(vm_page_t m, int base, int size)
2722{
2723 _vm_page_zero_valid(m, base, size);
2724 m->valid |= vm_page_bits(base, size);
2725}
984263bc 2726
cb1cf930
MD
2727
2728/*
2729 * Set valid bits and clear dirty bits.
2730 *
2731 * NOTE: This function does not clear the pmap modified bit.
2732 * Also note that e.g. NFS may use a byte-granular base
2733 * and size.
573fb415 2734 *
9a0cb7b1
MD
2735 * WARNING: Page must be busied? But vfs_clean_one_page() will call
2736 * this without necessarily busying the page (via bdwrite()).
2737 * So for now vm_token must also be held.
2738 *
573fb415 2739 * No other requirements.
cb1cf930 2740 */
1a54183b
MD
2741void
2742vm_page_set_validclean(vm_page_t m, int base, int size)
2743{
2744 int pagebits;
2745
2746 _vm_page_zero_valid(m, base, size);
984263bc
MD
2747 pagebits = vm_page_bits(base, size);
2748 m->valid |= pagebits;
984263bc
MD
2749 m->dirty &= ~pagebits;
2750 if (base == 0 && size == PAGE_SIZE) {
cb1cf930 2751 /*pmap_clear_modify(m);*/
984263bc
MD
2752 vm_page_flag_clear(m, PG_NOSYNC);
2753 }
2754}
2755
0a8aee15
MD
2756/*
2757 * Set valid & dirty. Used by buwrite()
573fb415 2758 *
9a0cb7b1
MD
2759 * WARNING: Page must be busied? But vfs_dirty_one_page() will
2760 * call this function in buwrite() so for now vm_token must
9bf025db 2761 * be held.
9a0cb7b1 2762 *
573fb415 2763 * No other requirements.
0a8aee15
MD
2764 */
2765void
2766vm_page_set_validdirty(vm_page_t m, int base, int size)
2767{
2768 int pagebits;
2769
2770 pagebits = vm_page_bits(base, size);
2771 m->valid |= pagebits;
2772 m->dirty |= pagebits;
d89ce96a 2773 if (m->object)
9bf025db 2774 vm_object_set_writeable_dirty(m->object);
0a8aee15
MD
2775}
2776
cb1cf930
MD
2777/*
2778 * Clear dirty bits.
2779 *
2780 * NOTE: This function does not clear the pmap modified bit.
2781 * Also note that e.g. NFS may use a byte-granular base
2782 * and size.
573fb415
MD
2783 *
2784 * Page must be busied?
2785 * No other requirements.
cb1cf930 2786 */
984263bc
MD
2787void
2788vm_page_clear_dirty(vm_page_t m, int base, int size)
2789{
2790 m->dirty &= ~vm_page_bits(base, size);
1a54183b 2791 if (base == 0 && size == PAGE_SIZE) {
cb1cf930 2792 /*pmap_clear_modify(m);*/
1a54183b
MD
2793 vm_page_flag_clear(m, PG_NOSYNC);
2794 }
984263bc
MD
2795}
2796
17cde63e
MD
2797/*
2798 * Make the page all-dirty.
2799 *
2800 * Also make sure the related object and vnode reflect the fact that the
2801 * object may now contain a dirty page.
573fb415
MD
2802 *
2803 * Page must be busied?
2804 * No other requirements.
17cde63e
MD
2805 */
2806void
2807vm_page_dirty(vm_page_t m)
2808{
2809#ifdef INVARIANTS
2810 int pqtype = m->queue - m->pc;
2811#endif
2812 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
2813 ("vm_page_dirty: page in free/cache queue!"));
2814 if (m->dirty != VM_PAGE_BITS_ALL) {
2815 m->dirty = VM_PAGE_BITS_ALL;
2816 if (m->object)
2817 vm_object_set_writeable_dirty(m->object);
2818 }
2819}
2820
984263bc 2821/*
de71fd3f
MD
2822 * Invalidates DEV_BSIZE'd chunks within a page. Both the
2823 * valid and dirty bits for the effected areas are cleared.
984263bc 2824 *
573fb415
MD
2825 * Page must be busied?
2826 * Does not block.
2827 * No other requirements.
984263bc
MD
2828 */
2829void
2830vm_page_set_invalid(vm_page_t m, int base, int size)
2831{
2832 int bits;
2833
2834 bits = vm_page_bits(base, size);
2835 m->valid &= ~bits;
2836 m->dirty &= ~bits;
2837 m->object->generation++;
2838}
2839
2840/*
de71fd3f
MD
2841 * The kernel assumes that the invalid portions of a page contain
2842 * garbage, but such pages can be mapped into memory by user code.
2843 * When this occurs, we must zero out the non-valid portions of the
2844 * page so user code sees what it expects.
984263bc 2845 *
de71fd3f
MD
2846 * Pages are most often semi-valid when the end of a file is mapped
2847 * into memory and the file's size is not page aligned.
573fb415
MD
2848 *
2849 * Page must be busied?
2850 * No other requirements.
984263bc 2851 */
984263bc
MD
2852void
2853vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
2854{
2855 int b;
2856 int i;
2857
2858 /*
2859 * Scan the valid bits looking for invalid sections that
2860 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
2861 * valid bit may be set ) have already been zerod by
2862 * vm_page_set_validclean().
2863 */
984263bc
MD
2864 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
2865 if (i == (PAGE_SIZE / DEV_BSIZE) ||
2866 (m->valid & (1 << i))
2867 ) {
2868 if (i > b) {
2869 pmap_zero_page_area(
2870 VM_PAGE_TO_PHYS(m),
2871 b << DEV_BSHIFT,
2872 (i - b) << DEV_BSHIFT
2873 );
2874 }
2875 b = i + 1;
2876 }
2877 }
2878
2879 /*
2880 * setvalid is TRUE when we can safely set the zero'd areas
2881 * as being valid. We can do this if there are no cache consistency
2882 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
2883 */
984263bc
MD
2884 if (setvalid)
2885 m->valid = VM_PAGE_BITS_ALL;
2886}
2887
2888/*
de71fd3f
MD
2889 * Is a (partial) page valid? Note that the case where size == 0
2890 * will return FALSE in the degenerate case where the page is entirely
2891 * invalid, and TRUE otherwise.
984263bc 2892 *
573fb415
MD
2893 * Does not block.
2894 * No other requirements.
984263bc 2895 */
984263bc
MD
2896int
2897vm_page_is_valid(vm_page_t m, int base, int size)
2898{
2899 int bits = vm_page_bits(base, size);
2900
2901 if (m->valid && ((m->valid & bits) == bits))
2902 return 1;
2903 else
2904 return 0;
2905}
2906
2907/*
2908 * update dirty bits from pmap/mmu. May not block.
573fb415 2909 *
b12defdc 2910 * Caller must hold the page busy
984263bc 2911 */
984263bc
MD
2912void
2913vm_page_test_dirty(vm_page_t m)
2914{
2915 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
2916 vm_page_dirty(m);
2917 }
2918}
2919
906c754c
MD
2920/*
2921 * Register an action, associating it with its vm_page
2922 */
2923void
2924vm_page_register_action(vm_page_action_t action, vm_page_event_t event)
2925{
2926 struct vm_page_action_list *list;
2927 int hv;
2928
2929 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
2930 list = &action_list[hv];
2931
2932 lwkt_gettoken(&vm_token);
2933 vm_page_flag_set(action->m, PG_ACTIONLIST);
2934 action->event = event;
2935 LIST_INSERT_HEAD(list, action, entry);
2936 lwkt_reltoken(&vm_token);
2937}
2938
2939/*
2940 * Unregister an action, disassociating it from its related vm_page
2941 */
2942void
2943vm_page_unregister_action(vm_page_action_t action)
2944{
2945 struct vm_page_action_list *list;
2946 int hv;
2947
2948 lwkt_gettoken(&vm_token);
2949 if (action->event != VMEVENT_NONE) {
2950 action->event = VMEVENT_NONE;
2951 LIST_REMOVE(action, entry);
2952
2953 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
2954 list = &action_list[hv];
2955 if (LIST_EMPTY(list))
2956 vm_page_flag_clear(action->m, PG_ACTIONLIST);
2957 }
2958 lwkt_reltoken(&vm_token);
2959}
2960
10192bae
MD
2961/*
2962 * Issue an event on a VM page. Corresponding action structures are
2963 * removed from the page's list and called.
906c754c
MD
2964 *
2965 * If the vm_page has no more pending action events we clear its
2966 * PG_ACTIONLIST flag.
10192bae
MD
2967 */
2968void
2969vm_page_event_internal(vm_page_t m, vm_page_event_t event)
2970{
906c754c
MD
2971 struct vm_page_action_list *list;
2972 struct vm_page_action *scan;
2973 struct vm_page_action *next;
2974 int hv;
2975 int all;
10192bae 2976
906c754c
MD
2977 hv = (int)((intptr_t)m >> 8) & VMACTION_HMASK;
2978 list = &action_list[hv];
2979 all = 1;
2980
2981 lwkt_gettoken(&vm_token);
2982 LIST_FOREACH_MUTABLE(scan, list, entry, next) {
2983 if (scan->m == m) {
2984 if (scan->event == event) {
2985 scan->event = VMEVENT_NONE;
2986 LIST_REMOVE(scan, entry);
2987 scan->func(m, scan);
2988 /* XXX */
2989 } else {
2990 all = 0;
2991 }
10192bae
MD
2992 }
2993 }
906c754c
MD
2994 if (all)
2995 vm_page_flag_clear(m, PG_ACTIONLIST);
2996 lwkt_reltoken(&vm_token);
10192bae
MD
2997}
2998
984263bc
MD
2999#include "opt_ddb.h"
3000#ifdef DDB
3001#include <sys/kernel.h>
3002
3003#include <ddb/ddb.h>
3004
3005DB_SHOW_COMMAND(page, vm_page_print_page_info)
3006{
12e4aaff
MD
3007 db_printf("vmstats.v_free_count: %d\n", vmstats.v_free_count);
3008 db_printf("vmstats.v_cache_count: %d\n", vmstats.v_cache_count);
3009 db_printf("vmstats.v_inactive_count: %d\n", vmstats.v_inactive_count);
3010 db_printf("vmstats.v_active_count: %d\n", vmstats.v_active_count);
3011 db_printf("vmstats.v_wire_count: %d\n", vmstats.v_wire_count);
3012 db_printf("vmstats.v_free_reserved: %d\n", vmstats.v_free_reserved);
3013 db_printf("vmstats.v_free_min: %d\n", vmstats.v_free_min);
3014 db_printf("vmstats.v_free_target: %d\n", vmstats.v_free_target);
3015 db_printf("vmstats.v_cache_min: %d\n", vmstats.v_cache_min);
3016 db_printf("vmstats.v_inactive_target: %d\n", vmstats.v_inactive_target);
984263bc
MD
3017}
3018
3019DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3020{
3021 int i;
3022 db_printf("PQ_FREE:");
3023 for(i=0;i<PQ_L2_SIZE;i++) {
3024 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt);
3025 }
3026 db_printf("\n");
3027
3028 db_printf("PQ_CACHE:");
3029 for(i=0;i<PQ_L2_SIZE;i++) {
3030 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt);
3031 }
3032 db_printf("\n");
3033
027193eb
MD
3034 db_printf("PQ_ACTIVE:");
3035 for(i=0;i<PQ_L2_SIZE;i++) {
3036 db_printf(" %d", vm_page_queues[PQ_ACTIVE + i].lcnt);
3037 }
3038 db_printf("\n");
3039
3040 db_printf("PQ_INACTIVE:");
3041 for(i=0;i<PQ_L2_SIZE;i++) {
3042 db_printf(" %d", vm_page_queues[PQ_INACTIVE + i].lcnt);
3043 }
3044 db_printf("\n");
984263bc
MD
3045}
3046#endif /* DDB */