vm: Put vm_page_alloc_contig debug prints under vm_contig_verbose
[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
MD
206#ifdef SMP
207 /*
bc3396b8
MD
208 * Twist for cpu localization in addition to page coloring, so
209 * different cpus selecting by m->queue get different page colors.
85946b6c
MD
210 */
211 m->pc ^= ((pa >> PAGE_SHIFT) / PQ_L2_SIZE) & PQ_L2_MASK;
212 m->pc ^= ((pa >> PAGE_SHIFT) / (PQ_L2_SIZE * PQ_L2_SIZE)) & PQ_L2_MASK;
213#endif
79d182b0
MD
214 /*
215 * Reserve a certain number of contiguous low memory pages for
216 * contigmalloc() to use.
217 */
218 if (pa < vm_low_phys_reserved) {
219 atomic_add_int(&vmstats.v_page_count, 1);
220 atomic_add_int(&vmstats.v_dma_pages, 1);
221 m->queue = PQ_NONE;
222 m->wire_count = 1;
3ae0c654 223 atomic_add_int(&vmstats.v_wire_count, 1);
79d182b0
MD
224 alist_free(&vm_contig_alist, pa >> PAGE_SHIFT, 1);
225 return;
226 }
227
228 /*
229 * General page
230 */
984263bc 231 m->queue = m->pc + PQ_FREE;
26bcc0c0 232 KKASSERT(m->dirty == 0);
de71fd3f 233
b12defdc
MD
234 atomic_add_int(&vmstats.v_page_count, 1);
235 atomic_add_int(&vmstats.v_free_count, 1);
161399b3 236 vpq = &vm_page_queues[m->queue];
bc3396b8
MD
237 if ((vpq->flipflop & 15) == 0) {
238 pmap_zero_page(VM_PAGE_TO_PHYS(m));
239 m->flags |= PG_ZERO;
161399b3 240 TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
bc3396b8
MD
241 atomic_add_int(&vm_page_zero_count, 1);
242 } else {
161399b3 243 TAILQ_INSERT_HEAD(&vpq->pl, m, pageq);
bc3396b8
MD
244 }
245 ++vpq->flipflop;
b12defdc 246 ++vpq->lcnt;
984263bc
MD
247}
248
249/*
de71fd3f 250 * (low level boot)
984263bc 251 *
de71fd3f 252 * Initializes the resident memory module.
984263bc 253 *
da23a592
MD
254 * Preallocates memory for critical VM structures and arrays prior to
255 * kernel_map becoming available.
26bcc0c0 256 *
da23a592
MD
257 * Memory is allocated from (virtual2_start, virtual2_end) if available,
258 * otherwise memory is allocated from (virtual_start, virtual_end).
259 *
260 * On x86-64 (virtual_start, virtual_end) is only 2GB and may not be
261 * large enough to hold vm_page_array & other structures for machines with
262 * large amounts of ram, so we want to use virtual2* when available.
984263bc 263 */
da23a592
MD
264void
265vm_page_startup(void)
984263bc 266{
da23a592 267 vm_offset_t vaddr = virtual2_start ? virtual2_start : virtual_start;
984263bc 268 vm_offset_t mapped;
6ef943a3
MD
269 vm_size_t npages;
270 vm_paddr_t page_range;
271 vm_paddr_t new_end;
984263bc 272 int i;
6ef943a3 273 vm_paddr_t pa;
984263bc 274 int nblocks;
6ef943a3 275 vm_paddr_t last_pa;
6ef943a3
MD
276 vm_paddr_t end;
277 vm_paddr_t biggestone, biggestsize;
6ef943a3 278 vm_paddr_t total;
984263bc
MD
279
280 total = 0;
281 biggestsize = 0;
282 biggestone = 0;
283 nblocks = 0;
284 vaddr = round_page(vaddr);
285
286 for (i = 0; phys_avail[i + 1]; i += 2) {
aecf2182
MD
287 phys_avail[i] = round_page64(phys_avail[i]);
288 phys_avail[i + 1] = trunc_page64(phys_avail[i + 1]);
984263bc
MD
289 }
290
291 for (i = 0; phys_avail[i + 1]; i += 2) {
6ef943a3 292 vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
984263bc
MD
293
294 if (size > biggestsize) {
295 biggestone = i;
296 biggestsize = size;
297 }
298 ++nblocks;
299 total += size;
300 }
301
302 end = phys_avail[biggestone+1];
1f804340 303 end = trunc_page(end);
984263bc
MD
304
305 /*
306 * Initialize the queue headers for the free queue, the active queue
307 * and the inactive queue.
308 */
984263bc
MD
309 vm_page_queue_init();
310
6abe3bd0 311#if !defined(_KERNEL_VIRTUAL)
984263bc 312 /*
b12defdc
MD
313 * VKERNELs don't support minidumps and as such don't need
314 * vm_page_dump
315 *
8e5e6f1b
AH
316 * Allocate a bitmap to indicate that a random physical page
317 * needs to be included in a minidump.
318 *
319 * The amd64 port needs this to indicate which direct map pages
320 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
321 *
322 * However, i386 still needs this workspace internally within the
323 * minidump code. In theory, they are not needed on i386, but are
324 * included should the sf_buf code decide to use them.
325 */
326 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
327 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
328 end -= vm_page_dump_size;
329 vm_page_dump = (void *)pmap_map(&vaddr, end, end + vm_page_dump_size,
330 VM_PROT_READ | VM_PROT_WRITE);
331 bzero((void *)vm_page_dump, vm_page_dump_size);
6abe3bd0 332#endif
8e5e6f1b 333 /*
984263bc
MD
334 * Compute the number of pages of memory that will be available for
335 * use (taking into account the overhead of a page structure per
336 * page).
337 */
984263bc 338 first_page = phys_avail[0] / PAGE_SIZE;
984263bc 339 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE - first_page;
1f804340 340 npages = (total - (page_range * sizeof(struct vm_page))) / PAGE_SIZE;
de71fd3f 341
79d182b0
MD
342#ifndef _KERNEL_VIRTUAL
343 /*
344 * (only applies to real kernels)
345 *
346 * Initialize the contiguous reserve map. We initially reserve up
347 * to 1/4 available physical memory or 65536 pages (~256MB), whichever
348 * is lower.
349 *
350 * Once device initialization is complete we return most of the
351 * reserved memory back to the normal page queues but leave some
352 * in reserve for things like usb attachments.
353 */
354 vm_low_phys_reserved = (vm_paddr_t)65536 << PAGE_SHIFT;
355 if (vm_low_phys_reserved > total / 4)
356 vm_low_phys_reserved = total / 4;
357 if (vm_dma_reserved == 0) {
358 vm_dma_reserved = 16 * 1024 * 1024; /* 16MB */
359 if (vm_dma_reserved > total / 16)
360 vm_dma_reserved = total / 16;
361 }
362#endif
363 alist_init(&vm_contig_alist, 65536, vm_contig_ameta,
364 ALIST_RECORDS_65536);
365
984263bc
MD
366 /*
367 * Initialize the mem entry structures now, and put them in the free
368 * queue.
369 */
984263bc 370 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
79d182b0 371 mapped = pmap_map(&vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE);
8e5e6f1b
AH
372 vm_page_array = (vm_page_t)mapped;
373
0e6594a8 374#if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
8e5e6f1b
AH
375 /*
376 * since pmap_map on amd64 returns stuff out of a direct-map region,
377 * we have to manually add these pages to the minidump tracking so
378 * that they can be dumped, including the vm_page_array.
379 */
380 for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
381 dump_add_page(pa);
8fdd3267 382#endif
984263bc
MD
383
384 /*
385 * Clear all of the page structures
386 */
387 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
388 vm_page_array_size = page_range;
389
390 /*
161399b3 391 * Construct the free queue(s) in ascending order (by physical
984263bc
MD
392 * address) so that the first 16MB of physical memory is allocated
393 * last rather than first. On large-memory machines, this avoids
394 * the exhaustion of low physical memory before isa_dmainit has run.
395 */
12e4aaff
MD
396 vmstats.v_page_count = 0;
397 vmstats.v_free_count = 0;
984263bc
MD
398 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) {
399 pa = phys_avail[i];
400 if (i == biggestone)
401 last_pa = new_end;
402 else
403 last_pa = phys_avail[i + 1];
404 while (pa < last_pa && npages-- > 0) {
405 vm_add_new_page(pa);
406 pa += PAGE_SIZE;
407 }
408 }
da23a592
MD
409 if (virtual2_start)
410 virtual2_start = vaddr;
411 else
412 virtual_start = vaddr;
984263bc
MD
413}
414
415/*
79d182b0
MD
416 * We tended to reserve a ton of memory for contigmalloc(). Now that most
417 * drivers have initialized we want to return most the remaining free
418 * reserve back to the VM page queues so they can be used for normal
419 * allocations.
420 *
421 * We leave vm_dma_reserved bytes worth of free pages in the reserve pool.
422 */
423static void
424vm_page_startup_finish(void *dummy __unused)
425{
426 alist_blk_t blk;
427 alist_blk_t rblk;
428 alist_blk_t count;
429 alist_blk_t xcount;
430 alist_blk_t bfree;
431 vm_page_t m;
432
433 spin_lock(&vm_contig_spin);
434 for (;;) {
435 bfree = alist_free_info(&vm_contig_alist, &blk, &count);
436 if (bfree <= vm_dma_reserved / PAGE_SIZE)
437 break;
438 if (count == 0)
439 break;
440
441 /*
442 * Figure out how much of the initial reserve we have to
443 * free in order to reach our target.
444 */
445 bfree -= vm_dma_reserved / PAGE_SIZE;
446 if (count > bfree) {
447 blk += count - bfree;
448 count = bfree;
449 }
450
451 /*
452 * Calculate the nearest power of 2 <= count.
453 */
454 for (xcount = 1; xcount <= count; xcount <<= 1)
455 ;
456 xcount >>= 1;
457 blk += count - xcount;
458 count = xcount;
459
460 /*
461 * Allocate the pages from the alist, then free them to
462 * the normal VM page queues.
463 *
464 * Pages allocated from the alist are wired. We have to
465 * busy, unwire, and free them. We must also adjust
466 * vm_low_phys_reserved before freeing any pages to prevent
467 * confusion.
468 */
469 rblk = alist_alloc(&vm_contig_alist, blk, count);
470 if (rblk != blk) {
471 kprintf("vm_page_startup_finish: Unable to return "
472 "dma space @0x%08x/%d -> 0x%08x\n",
473 blk, count, rblk);
474 break;
475 }
476 atomic_add_int(&vmstats.v_dma_pages, -count);
477 spin_unlock(&vm_contig_spin);
478
479 m = PHYS_TO_VM_PAGE((vm_paddr_t)blk << PAGE_SHIFT);
480 vm_low_phys_reserved = VM_PAGE_TO_PHYS(m);
481 while (count) {
482 vm_page_busy_wait(m, FALSE, "cpgfr");
483 vm_page_unwire(m, 0);
484 vm_page_free(m);
485 --count;
486 ++m;
487 }
488 spin_lock(&vm_contig_spin);
489 }
490 spin_unlock(&vm_contig_spin);
491
492 /*
493 * Print out how much DMA space drivers have already allocated and
494 * how much is left over.
495 */
496 kprintf("DMA space used: %jdk, remaining available: %jdk\n",
497 (intmax_t)(vmstats.v_dma_pages - vm_contig_alist.bl_free) *
498 (PAGE_SIZE / 1024),
499 (intmax_t)vm_contig_alist.bl_free * (PAGE_SIZE / 1024));
500}
501SYSINIT(vm_pgend, SI_SUB_PROC0_POST, SI_ORDER_ANY,
502 vm_page_startup_finish, NULL)
503
504
505/*
1f804340
MD
506 * Scan comparison function for Red-Black tree scans. An inclusive
507 * (start,end) is expected. Other fields are not used.
984263bc 508 */
1f804340
MD
509int
510rb_vm_page_scancmp(struct vm_page *p, void *data)
984263bc 511{
1f804340 512 struct rb_vm_page_scan_info *info = data;
984263bc 513
1f804340
MD
514 if (p->pindex < info->start_pindex)
515 return(-1);
516 if (p->pindex > info->end_pindex)
517 return(1);
518 return(0);
519}
520
521int
522rb_vm_page_compare(struct vm_page *p1, struct vm_page *p2)
523{
524 if (p1->pindex < p2->pindex)
525 return(-1);
526 if (p1->pindex > p2->pindex)
527 return(1);
528 return(0);
984263bc
MD
529}
530
de71fd3f 531/*
b12defdc
MD
532 * Each page queue has its own spin lock, which is fairly optimal for
533 * allocating and freeing pages at least.
534 *
535 * The caller must hold the vm_page_spin_lock() before locking a vm_page's
536 * queue spinlock via this function. Also note that m->queue cannot change
537 * unless both the page and queue are locked.
538 */
539static __inline
540void
541_vm_page_queue_spin_lock(vm_page_t m)
542{
543 u_short queue;
544
545 queue = m->queue;
546 if (queue != PQ_NONE) {
547 spin_lock(&vm_page_queues[queue].spin);
548 KKASSERT(queue == m->queue);
549 }
550}
551
552static __inline
553void
554_vm_page_queue_spin_unlock(vm_page_t m)
555{
556 u_short queue;
557
558 queue = m->queue;
559 cpu_ccfence();
560 if (queue != PQ_NONE)
561 spin_unlock(&vm_page_queues[queue].spin);
562}
563
564static __inline
565void
566_vm_page_queues_spin_lock(u_short queue)
567{
568 cpu_ccfence();
569 if (queue != PQ_NONE)
570 spin_lock(&vm_page_queues[queue].spin);
571}
572
573
574static __inline
575void
576_vm_page_queues_spin_unlock(u_short queue)
577{
578 cpu_ccfence();
579 if (queue != PQ_NONE)
580 spin_unlock(&vm_page_queues[queue].spin);
581}
582
583void
584vm_page_queue_spin_lock(vm_page_t m)
585{
586 _vm_page_queue_spin_lock(m);
587}
588
589void
590vm_page_queues_spin_lock(u_short queue)
591{
592 _vm_page_queues_spin_lock(queue);
593}
594
595void
596vm_page_queue_spin_unlock(vm_page_t m)
597{
598 _vm_page_queue_spin_unlock(m);
599}
600
601void
602vm_page_queues_spin_unlock(u_short queue)
603{
604 _vm_page_queues_spin_unlock(queue);
605}
606
607/*
608 * This locks the specified vm_page and its queue in the proper order
609 * (page first, then queue). The queue may change so the caller must
610 * recheck on return.
611 */
612static __inline
613void
614_vm_page_and_queue_spin_lock(vm_page_t m)
615{
616 vm_page_spin_lock(m);
617 _vm_page_queue_spin_lock(m);
618}
619
620static __inline
621void
622_vm_page_and_queue_spin_unlock(vm_page_t m)
623{
624 _vm_page_queues_spin_unlock(m->queue);
625 vm_page_spin_unlock(m);
626}
627
628void
629vm_page_and_queue_spin_unlock(vm_page_t m)
630{
631 _vm_page_and_queue_spin_unlock(m);
632}
633
634void
635vm_page_and_queue_spin_lock(vm_page_t m)
636{
637 _vm_page_and_queue_spin_lock(m);
638}
639
640/*
641 * Helper function removes vm_page from its current queue.
642 * Returns the base queue the page used to be on.
643 *
644 * The vm_page and the queue must be spinlocked.
645 * This function will unlock the queue but leave the page spinlocked.
646 */
647static __inline u_short
648_vm_page_rem_queue_spinlocked(vm_page_t m)
649{
650 struct vpgqueues *pq;
651 u_short queue;
652
653 queue = m->queue;
654 if (queue != PQ_NONE) {
655 pq = &vm_page_queues[queue];
656 TAILQ_REMOVE(&pq->pl, m, pageq);
657 atomic_add_int(pq->cnt, -1);
658 pq->lcnt--;
659 m->queue = PQ_NONE;
85946b6c 660 vm_page_queues_spin_unlock(queue);
b12defdc
MD
661 if ((queue - m->pc) == PQ_FREE && (m->flags & PG_ZERO))
662 atomic_subtract_int(&vm_page_zero_count, 1);
b12defdc
MD
663 if ((queue - m->pc) == PQ_CACHE || (queue - m->pc) == PQ_FREE)
664 return (queue - m->pc);
665 }
666 return queue;
667}
668
669/*
670 * Helper function places the vm_page on the specified queue.
671 *
672 * The vm_page must be spinlocked.
673 * This function will return with both the page and the queue locked.
674 */
675static __inline void
676_vm_page_add_queue_spinlocked(vm_page_t m, u_short queue, int athead)
677{
678 struct vpgqueues *pq;
679
680 KKASSERT(m->queue == PQ_NONE);
681
682 if (queue != PQ_NONE) {
683 vm_page_queues_spin_lock(queue);
684 pq = &vm_page_queues[queue];
685 ++pq->lcnt;
686 atomic_add_int(pq->cnt, 1);
687 m->queue = queue;
688
689 /*
690 * Put zero'd pages on the end ( where we look for zero'd pages
691 * first ) and non-zerod pages at the head.
692 */
693 if (queue - m->pc == PQ_FREE) {
694 if (m->flags & PG_ZERO) {
695 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
696 atomic_add_int(&vm_page_zero_count, 1);
697 } else {
698 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
699 }
700 } else if (athead) {
701 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
702 } else {
703 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
704 }
705 /* leave the queue spinlocked */
706 }
707}
708
709/*
710 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
711 * m->busy is zero. Returns TRUE if it had to sleep, FALSE if we
712 * did not. Only one sleep call will be made before returning.
713 *
714 * This function does NOT busy the page and on return the page is not
715 * guaranteed to be available.
716 */
717void
718vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
719{
720 u_int32_t flags;
721
722 for (;;) {
723 flags = m->flags;
724 cpu_ccfence();
725
726 if ((flags & PG_BUSY) == 0 &&
727 (also_m_busy == 0 || (flags & PG_SBUSY) == 0)) {
728 break;
729 }
730 tsleep_interlock(m, 0);
731 if (atomic_cmpset_int(&m->flags, flags,
732 flags | PG_WANTED | PG_REFERENCED)) {
733 tsleep(m, PINTERLOCKED, msg, 0);
734 break;
735 }
736 }
737}
738
739/*
740 * Wait until PG_BUSY can be set, then set it. If also_m_busy is TRUE we
741 * also wait for m->busy to become 0 before setting PG_BUSY.
742 */
743void
744VM_PAGE_DEBUG_EXT(vm_page_busy_wait)(vm_page_t m,
745 int also_m_busy, const char *msg
746 VM_PAGE_DEBUG_ARGS)
747{
748 u_int32_t flags;
749
750 for (;;) {
751 flags = m->flags;
752 cpu_ccfence();
753 if (flags & PG_BUSY) {
754 tsleep_interlock(m, 0);
755 if (atomic_cmpset_int(&m->flags, flags,
756 flags | PG_WANTED | PG_REFERENCED)) {
757 tsleep(m, PINTERLOCKED, msg, 0);
758 }
759 } else if (also_m_busy && (flags & PG_SBUSY)) {
760 tsleep_interlock(m, 0);
761 if (atomic_cmpset_int(&m->flags, flags,
762 flags | PG_WANTED | PG_REFERENCED)) {
763 tsleep(m, PINTERLOCKED, msg, 0);
764 }
765 } else {
766 if (atomic_cmpset_int(&m->flags, flags,
767 flags | PG_BUSY)) {
768#ifdef VM_PAGE_DEBUG
769 m->busy_func = func;
770 m->busy_line = lineno;
771#endif
772 break;
773 }
774 }
775 }
776}
777
778/*
779 * Attempt to set PG_BUSY. If also_m_busy is TRUE we only succeed if m->busy
780 * is also 0.
781 *
782 * Returns non-zero on failure.
783 */
784int
785VM_PAGE_DEBUG_EXT(vm_page_busy_try)(vm_page_t m, int also_m_busy
786 VM_PAGE_DEBUG_ARGS)
787{
788 u_int32_t flags;
789
790 for (;;) {
791 flags = m->flags;
792 cpu_ccfence();
793 if (flags & PG_BUSY)
794 return TRUE;
795 if (also_m_busy && (flags & PG_SBUSY))
796 return TRUE;
797 if (atomic_cmpset_int(&m->flags, flags, flags | PG_BUSY)) {
798#ifdef VM_PAGE_DEBUG
799 m->busy_func = func;
800 m->busy_line = lineno;
801#endif
802 return FALSE;
803 }
804 }
805}
806
807/*
808 * Clear the PG_BUSY flag and return non-zero to indicate to the caller
809 * that a wakeup() should be performed.
810 *
811 * The vm_page must be spinlocked and will remain spinlocked on return.
812 * The related queue must NOT be spinlocked (which could deadlock us).
813 *
814 * (inline version)
815 */
816static __inline
817int
818_vm_page_wakeup(vm_page_t m)
819{
820 u_int32_t flags;
821
822 for (;;) {
823 flags = m->flags;
824 cpu_ccfence();
825 if (atomic_cmpset_int(&m->flags, flags,
826 flags & ~(PG_BUSY | PG_WANTED))) {
827 break;
828 }
829 }
830 return(flags & PG_WANTED);
831}
832
833/*
834 * Clear the PG_BUSY flag and wakeup anyone waiting for the page. This
835 * is typically the last call you make on a page before moving onto
836 * other things.
837 */
838void
839vm_page_wakeup(vm_page_t m)
840{
841 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
842 vm_page_spin_lock(m);
843 if (_vm_page_wakeup(m)) {
844 vm_page_spin_unlock(m);
845 wakeup(m);
846 } else {
847 vm_page_spin_unlock(m);
848 }
849}
850
851/*
573fb415
MD
852 * Holding a page keeps it from being reused. Other parts of the system
853 * can still disassociate the page from its current object and free it, or
854 * perform read or write I/O on it and/or otherwise manipulate the page,
855 * but if the page is held the VM system will leave the page and its data
856 * intact and not reuse the page for other purposes until the last hold
857 * reference is released. (see vm_page_wire() if you want to prevent the
858 * page from being disassociated from its object too).
859 *
573fb415
MD
860 * The caller must still validate the contents of the page and, if necessary,
861 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
862 * before manipulating the page.
b12defdc
MD
863 *
864 * XXX get vm_page_spin_lock() here and move FREE->HOLD if necessary
573fb415
MD
865 */
866void
867vm_page_hold(vm_page_t m)
868{
b12defdc
MD
869 vm_page_spin_lock(m);
870 atomic_add_int(&m->hold_count, 1);
871 if (m->queue - m->pc == PQ_FREE) {
872 _vm_page_queue_spin_lock(m);
873 _vm_page_rem_queue_spinlocked(m);
027193eb 874 _vm_page_add_queue_spinlocked(m, PQ_HOLD + m->pc, 0);
b12defdc
MD
875 _vm_page_queue_spin_unlock(m);
876 }
877 vm_page_spin_unlock(m);
573fb415
MD
878}
879
880/*
de71fd3f 881 * The opposite of vm_page_hold(). A page can be freed while being held,
b12defdc
MD
882 * which places it on the PQ_HOLD queue. If we are able to busy the page
883 * after the hold count drops to zero we will move the page to the
884 * appropriate PQ_FREE queue by calling vm_page_free_toq().
de71fd3f 885 */
984263bc 886void
573fb415 887vm_page_unhold(vm_page_t m)
984263bc 888{
b12defdc
MD
889 vm_page_spin_lock(m);
890 atomic_add_int(&m->hold_count, -1);
027193eb 891 if (m->hold_count == 0 && m->queue - m->pc == PQ_HOLD) {
b12defdc
MD
892 _vm_page_queue_spin_lock(m);
893 _vm_page_rem_queue_spinlocked(m);
894 _vm_page_add_queue_spinlocked(m, PQ_FREE + m->pc, 0);
895 _vm_page_queue_spin_unlock(m);
97edb3b6 896 }
b12defdc 897 vm_page_spin_unlock(m);
984263bc
MD
898}
899
900/*
573fb415 901 * Inserts the given vm_page into the object and object list.
984263bc 902 *
de71fd3f
MD
903 * The pagetables are not updated but will presumably fault the page
904 * in if necessary, or if a kernel page the caller will at some point
905 * enter the page into the kernel's pmap. We are not allowed to block
906 * here so we *can't* do this anyway.
984263bc 907 *
de71fd3f 908 * This routine may not block.
398c240d 909 * This routine must be called with the vm_object held.
654a39f0 910 * This routine must be called with a critical section held.
d2d8515b
MD
911 *
912 * This routine returns TRUE if the page was inserted into the object
913 * successfully, and FALSE if the page already exists in the object.
984263bc 914 */
d2d8515b 915int
984263bc
MD
916vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
917{
b12defdc 918 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
984263bc
MD
919 if (m->object != NULL)
920 panic("vm_page_insert: already inserted");
921
b12defdc 922 object->generation++;
b12defdc 923
984263bc 924 /*
b12defdc
MD
925 * Record the object/offset pair in this page and add the
926 * pv_list_count of the page to the object.
927 *
928 * The vm_page spin lock is required for interactions with the pmap.
984263bc 929 */
b12defdc 930 vm_page_spin_lock(m);
984263bc
MD
931 m->object = object;
932 m->pindex = pindex;
d2d8515b
MD
933 if (vm_page_rb_tree_RB_INSERT(&object->rb_memq, m)) {
934 m->object = NULL;
935 m->pindex = 0;
936 vm_page_spin_unlock(m);
937 return FALSE;
938 }
939 object->resident_page_count++;
85946b6c 940 /* atomic_add_int(&object->agg_pv_list_count, m->md.pv_list_count); */
b12defdc 941 vm_page_spin_unlock(m);
50a55c46
MD
942
943 /*
984263bc
MD
944 * Since we are inserting a new and possibly dirty page,
945 * update the object's OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY flags.
946 */
17cde63e 947 if ((m->valid & m->dirty) || (m->flags & PG_WRITEABLE))
984263bc 948 vm_object_set_writeable_dirty(object);
096e95c0
MD
949
950 /*
951 * Checks for a swap assignment and sets PG_SWAPPED if appropriate.
952 */
953 swap_pager_page_inserted(m);
d2d8515b 954 return TRUE;
984263bc
MD
955}
956
957/*
b12defdc 958 * Removes the given vm_page_t from the (object,index) table
984263bc 959 *
de71fd3f
MD
960 * The underlying pmap entry (if any) is NOT removed here.
961 * This routine may not block.
9765affa 962 *
9ad0147b
MD
963 * The page must be BUSY and will remain BUSY on return.
964 * No other requirements.
9765affa 965 *
9ad0147b
MD
966 * NOTE: FreeBSD side effect was to unbusy the page on return. We leave
967 * it busy.
984263bc 968 */
984263bc
MD
969void
970vm_page_remove(vm_page_t m)
971{
972 vm_object_t object;
973
654a39f0 974 if (m->object == NULL) {
984263bc 975 return;
654a39f0 976 }
984263bc 977
de71fd3f 978 if ((m->flags & PG_BUSY) == 0)
984263bc 979 panic("vm_page_remove: page not busy");
984263bc 980
984263bc
MD
981 object = m->object;
982
398c240d
VS
983 vm_object_hold(object);
984
984263bc 985 /*
1f804340 986 * Remove the page from the object and update the object.
b12defdc
MD
987 *
988 * The vm_page spin lock is required for interactions with the pmap.
984263bc 989 */
b12defdc 990 vm_page_spin_lock(m);
1f804340 991 vm_page_rb_tree_RB_REMOVE(&object->rb_memq, m);
984263bc 992 object->resident_page_count--;
85946b6c 993 /* atomic_add_int(&object->agg_pv_list_count, -m->md.pv_list_count); */
984263bc 994 m->object = NULL;
b12defdc 995 vm_page_spin_unlock(m);
1f804340 996
b12defdc 997 object->generation++;
398c240d 998
b12defdc 999 vm_object_drop(object);
984263bc
MD
1000}
1001
1002/*
de71fd3f
MD
1003 * Locate and return the page at (object, pindex), or NULL if the
1004 * page could not be found.
1005 *
b12defdc 1006 * The caller must hold the vm_object token.
984263bc 1007 */
984263bc
MD
1008vm_page_t
1009vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1010{
1011 vm_page_t m;
984263bc
MD
1012
1013 /*
1014 * Search the hash table for this object/offset pair
1015 */
b12defdc 1016 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
1f804340 1017 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
1f804340
MD
1018 KKASSERT(m == NULL || (m->object == object && m->pindex == pindex));
1019 return(m);
984263bc
MD
1020}
1021
b12defdc
MD
1022vm_page_t
1023VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_wait)(struct vm_object *object,
1024 vm_pindex_t pindex,
1025 int also_m_busy, const char *msg
1026 VM_PAGE_DEBUG_ARGS)
1027{
1028 u_int32_t flags;
1029 vm_page_t m;
1030
1031 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
1032 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
1033 while (m) {
1034 KKASSERT(m->object == object && m->pindex == pindex);
1035 flags = m->flags;
1036 cpu_ccfence();
1037 if (flags & PG_BUSY) {
1038 tsleep_interlock(m, 0);
1039 if (atomic_cmpset_int(&m->flags, flags,
1040 flags | PG_WANTED | PG_REFERENCED)) {
1041 tsleep(m, PINTERLOCKED, msg, 0);
1042 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq,
1043 pindex);
1044 }
1045 } else if (also_m_busy && (flags & PG_SBUSY)) {
1046 tsleep_interlock(m, 0);
1047 if (atomic_cmpset_int(&m->flags, flags,
1048 flags | PG_WANTED | PG_REFERENCED)) {
1049 tsleep(m, PINTERLOCKED, msg, 0);
1050 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq,
1051 pindex);
1052 }
1053 } else if (atomic_cmpset_int(&m->flags, flags,
1054 flags | PG_BUSY)) {
1055#ifdef VM_PAGE_DEBUG
1056 m->busy_func = func;
1057 m->busy_line = lineno;
1058#endif
1059 break;
1060 }
1061 }
1062 return m;
1063}
1064
984263bc 1065/*
b12defdc 1066 * Attempt to lookup and busy a page.
984263bc 1067 *
b12defdc 1068 * Returns NULL if the page could not be found
984263bc 1069 *
b12defdc
MD
1070 * Returns a vm_page and error == TRUE if the page exists but could not
1071 * be busied.
984263bc 1072 *
b12defdc
MD
1073 * Returns a vm_page and error == FALSE on success.
1074 */
1075vm_page_t
1076VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_try)(struct vm_object *object,
1077 vm_pindex_t pindex,
1078 int also_m_busy, int *errorp
1079 VM_PAGE_DEBUG_ARGS)
1080{
1081 u_int32_t flags;
1082 vm_page_t m;
1083
1084 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
1085 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
1086 *errorp = FALSE;
1087 while (m) {
1088 KKASSERT(m->object == object && m->pindex == pindex);
1089 flags = m->flags;
1090 cpu_ccfence();
1091 if (flags & PG_BUSY) {
1092 *errorp = TRUE;
1093 break;
1094 }
1095 if (also_m_busy && (flags & PG_SBUSY)) {
1096 *errorp = TRUE;
1097 break;
1098 }
1099 if (atomic_cmpset_int(&m->flags, flags, flags | PG_BUSY)) {
1100#ifdef VM_PAGE_DEBUG
1101 m->busy_func = func;
1102 m->busy_line = lineno;
1103#endif
1104 break;
1105 }
1106 }
1107 return m;
1108}
1109
1110/*
1111 * Caller must hold the related vm_object
1112 */
1113vm_page_t
1114vm_page_next(vm_page_t m)
1115{
1116 vm_page_t next;
1117
1118 next = vm_page_rb_tree_RB_NEXT(m);
1119 if (next && next->pindex != m->pindex + 1)
1120 next = NULL;
1121 return (next);
1122}
1123
1124/*
1125 * vm_page_rename()
1126 *
1127 * Move the given vm_page from its current object to the specified
1128 * target object/offset. The page must be busy and will remain so
1129 * on return.
984263bc 1130 *
b12defdc
MD
1131 * new_object must be held.
1132 * This routine might block. XXX ?
1133 *
1134 * NOTE: Swap associated with the page must be invalidated by the move. We
de71fd3f
MD
1135 * have to do this for several reasons: (1) we aren't freeing the
1136 * page, (2) we are dirtying the page, (3) the VM system is probably
1137 * moving the page from object A to B, and will then later move
1138 * the backing store from A to B and we can't have a conflict.
984263bc 1139 *
b12defdc 1140 * NOTE: We *always* dirty the page. It is necessary both for the
de71fd3f
MD
1141 * fact that we moved it, and because we may be invalidating
1142 * swap. If the page is on the cache, we have to deactivate it
1143 * or vm_page_dirty() will panic. Dirty pages are not allowed
1144 * on the cache.
984263bc 1145 */
984263bc
MD
1146void
1147vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1148{
b12defdc
MD
1149 KKASSERT(m->flags & PG_BUSY);
1150 ASSERT_LWKT_TOKEN_HELD(vm_object_token(new_object));
1151 if (m->object) {
1152 ASSERT_LWKT_TOKEN_HELD(vm_object_token(m->object));
1153 vm_page_remove(m);
1154 }
d2d8515b 1155 if (vm_page_insert(m, new_object, new_pindex) == FALSE) {
480c83b6 1156 panic("vm_page_rename: target exists (%p,%"PRIu64")",
d2d8515b
MD
1157 new_object, new_pindex);
1158 }
984263bc
MD
1159 if (m->queue - m->pc == PQ_CACHE)
1160 vm_page_deactivate(m);
1161 vm_page_dirty(m);
984263bc
MD
1162}
1163
1164/*
de71fd3f
MD
1165 * vm_page_unqueue() without any wakeup. This routine is used when a page
1166 * is being moved between queues or otherwise is to remain BUSYied by the
1167 * caller.
984263bc 1168 *
de71fd3f 1169 * This routine may not block.
984263bc 1170 */
984263bc
MD
1171void
1172vm_page_unqueue_nowakeup(vm_page_t m)
1173{
b12defdc
MD
1174 vm_page_and_queue_spin_lock(m);
1175 (void)_vm_page_rem_queue_spinlocked(m);
1176 vm_page_spin_unlock(m);
984263bc
MD
1177}
1178
1179/*
de71fd3f
MD
1180 * vm_page_unqueue() - Remove a page from its queue, wakeup the pagedemon
1181 * if necessary.
984263bc 1182 *
de71fd3f 1183 * This routine may not block.
984263bc 1184 */
984263bc
MD
1185void
1186vm_page_unqueue(vm_page_t m)
1187{
b12defdc 1188 u_short queue;
de71fd3f 1189
b12defdc
MD
1190 vm_page_and_queue_spin_lock(m);
1191 queue = _vm_page_rem_queue_spinlocked(m);
1192 if (queue == PQ_FREE || queue == PQ_CACHE) {
1193 vm_page_spin_unlock(m);
1194 pagedaemon_wakeup();
1195 } else {
1196 vm_page_spin_unlock(m);
984263bc
MD
1197 }
1198}
1199
984263bc 1200/*
de71fd3f 1201 * vm_page_list_find()
984263bc 1202 *
de71fd3f 1203 * Find a page on the specified queue with color optimization.
984263bc 1204 *
de71fd3f
MD
1205 * The page coloring optimization attempts to locate a page that does
1206 * not overload other nearby pages in the object in the cpu's L1 or L2
1207 * caches. We need this optimization because cpu caches tend to be
85946b6c
MD
1208 * physical caches, while object spaces tend to be virtual.
1209 *
1210 * On MP systems each PQ_FREE and PQ_CACHE color queue has its own spinlock
1211 * and the algorithm is adjusted to localize allocations on a per-core basis.
1212 * This is done by 'twisting' the colors.
984263bc 1213 *
b12defdc
MD
1214 * The page is returned spinlocked and removed from its queue (it will
1215 * be on PQ_NONE), or NULL. The page is not PG_BUSY'd. The caller
1216 * is responsible for dealing with the busy-page case (usually by
1217 * deactivating the page and looping).
1218 *
1219 * NOTE: This routine is carefully inlined. A non-inlined version
1220 * is available for outside callers but the only critical path is
1221 * from within this source file.
984263bc 1222 *
b12defdc
MD
1223 * NOTE: This routine assumes that the vm_pages found in PQ_CACHE and PQ_FREE
1224 * represent stable storage, allowing us to order our locks vm_page
1225 * first, then queue.
984263bc 1226 */
74232d8e 1227static __inline
984263bc 1228vm_page_t
74232d8e
MD
1229_vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
1230{
1231 vm_page_t m;
1232
b12defdc
MD
1233 for (;;) {
1234 if (prefer_zero)
1235 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
1236 else
1237 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
1238 if (m == NULL) {
1239 m = _vm_page_list_find2(basequeue, index);
1240 return(m);
1241 }
1242 vm_page_and_queue_spin_lock(m);
1243 if (m->queue == basequeue + index) {
1244 _vm_page_rem_queue_spinlocked(m);
1245 /* vm_page_t spin held, no queue spin */
1246 break;
1247 }
1248 vm_page_and_queue_spin_unlock(m);
1249 }
74232d8e
MD
1250 return(m);
1251}
1252
1253static vm_page_t
1254_vm_page_list_find2(int basequeue, int index)
984263bc
MD
1255{
1256 int i;
1257 vm_page_t m = NULL;
1258 struct vpgqueues *pq;
1259
1260 pq = &vm_page_queues[basequeue];
1261
1262 /*
1263 * Note that for the first loop, index+i and index-i wind up at the
1264 * same place. Even though this is not totally optimal, we've already
1265 * blown it by missing the cache case so we do not care.
1266 */
b12defdc
MD
1267 for (i = PQ_L2_SIZE / 2; i > 0; --i) {
1268 for (;;) {
1269 m = TAILQ_FIRST(&pq[(index + i) & PQ_L2_MASK].pl);
1270 if (m) {
1271 _vm_page_and_queue_spin_lock(m);
1272 if (m->queue ==
1273 basequeue + ((index + i) & PQ_L2_MASK)) {
1274 _vm_page_rem_queue_spinlocked(m);
1275 return(m);
1276 }
1277 _vm_page_and_queue_spin_unlock(m);
1278 continue;
1279 }
1280 m = TAILQ_FIRST(&pq[(index - i) & PQ_L2_MASK].pl);
1281 if (m) {
1282 _vm_page_and_queue_spin_lock(m);
1283 if (m->queue ==
1284 basequeue + ((index - i) & PQ_L2_MASK)) {
1285 _vm_page_rem_queue_spinlocked(m);
1286 return(m);
1287 }
1288 _vm_page_and_queue_spin_unlock(m);
1289 continue;
1290 }
1291 break; /* next i */
1292 }
984263bc
MD
1293 }
1294 return(m);
1295}
1296
573fb415 1297/*
b12defdc
MD
1298 * Returns a vm_page candidate for allocation. The page is not busied so
1299 * it can move around. The caller must busy the page (and typically
1300 * deactivate it if it cannot be busied!)
1301 *
1302 * Returns a spinlocked vm_page that has been removed from its queue.
573fb415 1303 */
74232d8e
MD
1304vm_page_t
1305vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
1306{
1307 return(_vm_page_list_find(basequeue, index, prefer_zero));
1308}
1309
984263bc 1310/*
b12defdc
MD
1311 * Find a page on the cache queue with color optimization, remove it
1312 * from the queue, and busy it. The returned page will not be spinlocked.
1313 *
1314 * A candidate failure will be deactivated. Candidates can fail due to
1315 * being busied by someone else, in which case they will be deactivated.
984263bc 1316 *
de71fd3f 1317 * This routine may not block.
b12defdc 1318 *
984263bc 1319 */
b12defdc 1320static vm_page_t
85946b6c 1321vm_page_select_cache(u_short pg_color)
984263bc
MD
1322{
1323 vm_page_t m;
1324
b12defdc 1325 for (;;) {
85946b6c 1326 m = _vm_page_list_find(PQ_CACHE, pg_color & PQ_L2_MASK, FALSE);
b12defdc
MD
1327 if (m == NULL)
1328 break;
1329 /*
1330 * (m) has been removed from its queue and spinlocked
1331 */
1332 if (vm_page_busy_try(m, TRUE)) {
1333 _vm_page_deactivate_locked(m, 0);
1334 vm_page_spin_unlock(m);
a491077e
MD
1335#ifdef INVARIANTS
1336 kprintf("Warning: busy page %p found in cache\n", m);
1337#endif
b12defdc
MD
1338 } else {
1339 /*
1340 * We successfully busied the page
1341 */
9bf025db 1342 if ((m->flags & (PG_UNMANAGED | PG_NEED_COMMIT)) == 0 &&
b12defdc 1343 m->hold_count == 0 &&
9bf025db
MD
1344 m->wire_count == 0 &&
1345 (m->dirty & m->valid) == 0) {
b12defdc
MD
1346 vm_page_spin_unlock(m);
1347 pagedaemon_wakeup();
1348 return(m);
1349 }
9bf025db
MD
1350
1351 /*
1352 * The page cannot be recycled, deactivate it.
1353 */
b12defdc
MD
1354 _vm_page_deactivate_locked(m, 0);
1355 if (_vm_page_wakeup(m)) {
1356 vm_page_spin_unlock(m);
1357 wakeup(m);
1358 } else {
1359 vm_page_spin_unlock(m);
1360 }
984263bc 1361 }
984263bc 1362 }
b12defdc 1363 return (m);
984263bc
MD
1364}
1365
1366/*
de71fd3f
MD
1367 * Find a free or zero page, with specified preference. We attempt to
1368 * inline the nominal case and fall back to _vm_page_select_free()
b12defdc 1369 * otherwise. A busied page is removed from the queue and returned.
984263bc 1370 *
de71fd3f 1371 * This routine may not block.
984263bc 1372 */
984263bc 1373static __inline vm_page_t
85946b6c 1374vm_page_select_free(u_short pg_color, boolean_t prefer_zero)
984263bc
MD
1375{
1376 vm_page_t m;
1377
b12defdc 1378 for (;;) {
85946b6c 1379 m = _vm_page_list_find(PQ_FREE, pg_color & PQ_L2_MASK,
b12defdc
MD
1380 prefer_zero);
1381 if (m == NULL)
1382 break;
1383 if (vm_page_busy_try(m, TRUE)) {
90244566
MD
1384 /*
1385 * Various mechanisms such as a pmap_collect can
1386 * result in a busy page on the free queue. We
1387 * have to move the page out of the way so we can
1388 * retry the allocation. If the other thread is not
1389 * allocating the page then m->valid will remain 0 and
1390 * the pageout daemon will free the page later on.
1391 *
1392 * Since we could not busy the page, however, we
1393 * cannot make assumptions as to whether the page
1394 * will be allocated by the other thread or not,
1395 * so all we can do is deactivate it to move it out
1396 * of the way. In particular, if the other thread
1397 * wires the page it may wind up on the inactive
1398 * queue and the pageout daemon will have to deal
1399 * with that case too.
1400 */
b12defdc
MD
1401 _vm_page_deactivate_locked(m, 0);
1402 vm_page_spin_unlock(m);
1403#ifdef INVARIANTS
1404 kprintf("Warning: busy page %p found in cache\n", m);
1405#endif
1406 } else {
90244566
MD
1407 /*
1408 * Theoretically if we are able to busy the page
1409 * atomic with the queue removal (using the vm_page
1410 * lock) nobody else should be able to mess with the
1411 * page before us.
1412 */
9bf025db
MD
1413 KKASSERT((m->flags & (PG_UNMANAGED |
1414 PG_NEED_COMMIT)) == 0);
b12defdc
MD
1415 KKASSERT(m->hold_count == 0);
1416 KKASSERT(m->wire_count == 0);
1417 vm_page_spin_unlock(m);
1418 pagedaemon_wakeup();
1419
1420 /* return busied and removed page */
1421 return(m);
1422 }
1423 }
984263bc
MD
1424 return(m);
1425}
1426
1427/*
54341a3b
MD
1428 * This implements a per-cpu cache of free, zero'd, ready-to-go pages.
1429 * The idea is to populate this cache prior to acquiring any locks so
1430 * we don't wind up potentially zeroing VM pages (under heavy loads) while
1431 * holding potentialy contending locks.
1432 *
1433 * Note that we allocate the page uninserted into anything and use a pindex
1434 * of 0, the vm_page_alloc() will effectively add gd_cpuid so these
1435 * allocations should wind up being uncontended. However, we still want
1436 * to rove across PQ_L2_SIZE.
1437 */
1438void
1439vm_page_pcpu_cache(void)
1440{
1441#if 0
1442 globaldata_t gd = mycpu;
1443 vm_page_t m;
1444
1445 if (gd->gd_vmpg_count < GD_MINVMPG) {
1446 crit_enter_gd(gd);
1447 while (gd->gd_vmpg_count < GD_MAXVMPG) {
1448 m = vm_page_alloc(NULL, ticks & ~ncpus2_mask,
1449 VM_ALLOC_NULL_OK | VM_ALLOC_NORMAL |
1450 VM_ALLOC_NULL_OK | VM_ALLOC_ZERO);
1451 if (gd->gd_vmpg_count < GD_MAXVMPG) {
1452 if ((m->flags & PG_ZERO) == 0) {
1453 pmap_zero_page(VM_PAGE_TO_PHYS(m));
1454 vm_page_flag_set(m, PG_ZERO);
1455 }
1456 gd->gd_vmpg_array[gd->gd_vmpg_count++] = m;
1457 } else {
1458 vm_page_free(m);
1459 }
1460 }
1461 crit_exit_gd(gd);
1462 }
1463#endif
1464}
1465
1466/*
de71fd3f 1467 * vm_page_alloc()
984263bc 1468 *
de71fd3f 1469 * Allocate and return a memory cell associated with this VM object/offset
85946b6c 1470 * pair. If object is NULL an unassociated page will be allocated.
984263bc 1471 *
d2d8515b
MD
1472 * The returned page will be busied and removed from its queues. This
1473 * routine can block and may return NULL if a race occurs and the page
1474 * is found to already exist at the specified (object, pindex).
de71fd3f 1475 *
dc1fd4b3 1476 * VM_ALLOC_NORMAL allow use of cache pages, nominal free drain
39208dbe 1477 * VM_ALLOC_QUICK like normal but cannot use cache
dc1fd4b3
MD
1478 * VM_ALLOC_SYSTEM greater free drain
1479 * VM_ALLOC_INTERRUPT allow free list to be completely drained
d2d8515b
MD
1480 * VM_ALLOC_ZERO advisory request for pre-zero'd page only
1481 * VM_ALLOC_FORCE_ZERO advisory request for pre-zero'd page only
1482 * VM_ALLOC_NULL_OK ok to return NULL on insertion collision
1483 * (see vm_page_grab())
54341a3b
MD
1484 * VM_ALLOC_USE_GD ok to use per-gd cache
1485 *
d2d8515b 1486 * The object must be held if not NULL
85946b6c 1487 * This routine may not block
984263bc 1488 *
de71fd3f
MD
1489 * Additional special handling is required when called from an interrupt
1490 * (VM_ALLOC_INTERRUPT). We are not allowed to mess with the page cache
1491 * in this case.
984263bc 1492 */
984263bc
MD
1493vm_page_t
1494vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int page_req)
1495{
54341a3b
MD
1496#ifdef SMP
1497 globaldata_t gd = mycpu;
1498#endif
9d494b34 1499 vm_object_t obj;
54341a3b 1500 vm_page_t m;
85946b6c 1501 u_short pg_color;
984263bc 1502
54341a3b
MD
1503#if 0
1504 /*
1505 * Special per-cpu free VM page cache. The pages are pre-busied
1506 * and pre-zerod for us.
1507 */
1508 if (gd->gd_vmpg_count && (page_req & VM_ALLOC_USE_GD)) {
1509 crit_enter_gd(gd);
1510 if (gd->gd_vmpg_count) {
1511 m = gd->gd_vmpg_array[--gd->gd_vmpg_count];
1512 crit_exit_gd(gd);
1513 goto done;
1514 }
1515 crit_exit_gd(gd);
1516 }
1517#endif
1518 m = NULL;
1519
85946b6c
MD
1520#ifdef SMP
1521 /*
1522 * Cpu twist - cpu localization algorithm
1523 */
1524 if (object) {
54341a3b 1525 pg_color = gd->gd_cpuid + (pindex & ~ncpus_fit_mask) +
85946b6c 1526 (object->pg_color & ~ncpus_fit_mask);
85946b6c 1527 } else {
54341a3b 1528 pg_color = gd->gd_cpuid + (pindex & ~ncpus_fit_mask);
85946b6c
MD
1529 }
1530#else
1531 /*
1532 * Normal page coloring algorithm
1533 */
1534 if (object) {
1535 pg_color = object->pg_color + pindex;
85946b6c
MD
1536 } else {
1537 pg_color = pindex;
1538 }
1539#endif
dc1fd4b3 1540 KKASSERT(page_req &
39208dbe
MD
1541 (VM_ALLOC_NORMAL|VM_ALLOC_QUICK|
1542 VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
984263bc
MD
1543
1544 /*
4ecf7cc9
MD
1545 * Certain system threads (pageout daemon, buf_daemon's) are
1546 * allowed to eat deeper into the free page list.
984263bc 1547 */
4ecf7cc9 1548 if (curthread->td_flags & TDF_SYSTHREAD)
dc1fd4b3 1549 page_req |= VM_ALLOC_SYSTEM;
984263bc 1550
984263bc 1551loop:
dc1fd4b3
MD
1552 if (vmstats.v_free_count > vmstats.v_free_reserved ||
1553 ((page_req & VM_ALLOC_INTERRUPT) && vmstats.v_free_count > 0) ||
1554 ((page_req & VM_ALLOC_SYSTEM) && vmstats.v_cache_count == 0 &&
1555 vmstats.v_free_count > vmstats.v_interrupt_free_min)
1556 ) {
984263bc 1557 /*
dc1fd4b3 1558 * The free queue has sufficient free pages to take one out.
984263bc 1559 */
d2d8515b 1560 if (page_req & (VM_ALLOC_ZERO | VM_ALLOC_FORCE_ZERO))
85946b6c 1561 m = vm_page_select_free(pg_color, TRUE);
984263bc 1562 else
85946b6c 1563 m = vm_page_select_free(pg_color, FALSE);
dc1fd4b3 1564 } else if (page_req & VM_ALLOC_NORMAL) {
984263bc 1565 /*
dc1fd4b3
MD
1566 * Allocatable from the cache (non-interrupt only). On
1567 * success, we must free the page and try again, thus
1568 * ensuring that vmstats.v_*_free_min counters are replenished.
984263bc 1569 */
dc1fd4b3
MD
1570#ifdef INVARIANTS
1571 if (curthread->td_preempted) {
086c1d7e 1572 kprintf("vm_page_alloc(): warning, attempt to allocate"
dc1fd4b3
MD
1573 " cache page from preempting interrupt\n");
1574 m = NULL;
1575 } else {
85946b6c 1576 m = vm_page_select_cache(pg_color);
dc1fd4b3
MD
1577 }
1578#else
85946b6c 1579 m = vm_page_select_cache(pg_color);
dc1fd4b3 1580#endif
984263bc 1581 /*
9765affa 1582 * On success move the page into the free queue and loop.
bdea739c
MD
1583 *
1584 * Only do this if we can safely acquire the vm_object lock,
1585 * because this is effectively a random page and the caller
1586 * might be holding the lock shared, we don't want to
1587 * deadlock.
984263bc 1588 */
dc1fd4b3
MD
1589 if (m != NULL) {
1590 KASSERT(m->dirty == 0,
d2d8515b 1591 ("Found dirty cache page %p", m));
9d494b34
MD
1592 if ((obj = m->object) != NULL) {
1593 if (vm_object_hold_try(obj)) {
bdea739c
MD
1594 vm_page_protect(m, VM_PROT_NONE);
1595 vm_page_free(m);
9d494b34
MD
1596 /* m->object NULL here */
1597 vm_object_drop(obj);
bdea739c
MD
1598 } else {
1599 vm_page_deactivate(m);
1600 vm_page_wakeup(m);
1601 }
1602 } else {
1603 vm_page_protect(m, VM_PROT_NONE);
1604 vm_page_free(m);
1605 }
dc1fd4b3
MD
1606 goto loop;
1607 }
1608
1609 /*
1610 * On failure return NULL
1611 */
984263bc 1612#if defined(DIAGNOSTIC)
dc1fd4b3 1613 if (vmstats.v_cache_count > 0)
086c1d7e 1614 kprintf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", vmstats.v_cache_count);
984263bc 1615#endif
dc1fd4b3
MD
1616 vm_pageout_deficit++;
1617 pagedaemon_wakeup();
1618 return (NULL);
984263bc
MD
1619 } else {
1620 /*
dc1fd4b3 1621 * No pages available, wakeup the pageout daemon and give up.
984263bc 1622 */
984263bc
MD
1623 vm_pageout_deficit++;
1624 pagedaemon_wakeup();
1625 return (NULL);
1626 }
1627
1628 /*
b12defdc
MD
1629 * v_free_count can race so loop if we don't find the expected
1630 * page.
984263bc 1631 */
b12defdc
MD
1632 if (m == NULL)
1633 goto loop;
984263bc
MD
1634
1635 /*
d2d8515b
MD
1636 * Good page found. The page has already been busied for us and
1637 * removed from its queues.
984263bc 1638 */
d2d8515b
MD
1639 KASSERT(m->dirty == 0,
1640 ("vm_page_alloc: free/cache page %p was dirty", m));
b12defdc 1641 KKASSERT(m->queue == PQ_NONE);
984263bc 1642
54341a3b
MD
1643#if 0
1644done:
1645#endif
984263bc 1646 /*
d2d8515b
MD
1647 * Initialize the structure, inheriting some flags but clearing
1648 * all the rest. The page has already been busied for us.
984263bc 1649 */
d2d8515b 1650 vm_page_flag_clear(m, ~(PG_ZERO | PG_BUSY | PG_SBUSY));
b12defdc
MD
1651 KKASSERT(m->wire_count == 0);
1652 KKASSERT(m->busy == 0);
984263bc 1653 m->act_count = 0;
984263bc 1654 m->valid = 0;
984263bc
MD
1655
1656 /*
b12defdc
MD
1657 * Caller must be holding the object lock (asserted by
1658 * vm_page_insert()).
1659 *
1660 * NOTE: Inserting a page here does not insert it into any pmaps
1661 * (which could cause us to block allocating memory).
85946b6c
MD
1662 *
1663 * NOTE: If no object an unassociated page is allocated, m->pindex
1664 * can be used by the caller for any purpose.
984263bc 1665 */
d2d8515b
MD
1666 if (object) {
1667 if (vm_page_insert(m, object, pindex) == FALSE) {
480c83b6 1668 kprintf("PAGE RACE (%p:%d,%"PRIu64")\n",
d2d8515b
MD
1669 object, object->type, pindex);
1670 vm_page_free(m);
1671 m = NULL;
1672 if ((page_req & VM_ALLOC_NULL_OK) == 0)
1673 panic("PAGE RACE");
1674 }
1675 } else {
85946b6c 1676 m->pindex = pindex;
d2d8515b 1677 }
984263bc
MD
1678
1679 /*
1680 * Don't wakeup too often - wakeup the pageout daemon when
1681 * we would be nearly out of memory.
1682 */
20479584 1683 pagedaemon_wakeup();
984263bc 1684
9765affa
MD
1685 /*
1686 * A PG_BUSY page is returned.
1687 */
984263bc
MD
1688 return (m);
1689}
1690
1691/*
79d182b0
MD
1692 * Attempt to allocate contiguous physical memory with the specified
1693 * requirements.
1694 */
1695vm_page_t
1696vm_page_alloc_contig(vm_paddr_t low, vm_paddr_t high,
1697 unsigned long alignment, unsigned long boundary,
1698 unsigned long size)
1699{
1700 alist_blk_t blk;
1701
1702 alignment >>= PAGE_SHIFT;
1703 if (alignment == 0)
1704 alignment = 1;
1705 boundary >>= PAGE_SHIFT;
1706 if (boundary == 0)
1707 boundary = 1;
1708 size = (size + PAGE_MASK) >> PAGE_SHIFT;
1709
1710 spin_lock(&vm_contig_spin);
1711 blk = alist_alloc(&vm_contig_alist, 0, size);
1712 if (blk == ALIST_BLOCK_NONE) {
1713 spin_unlock(&vm_contig_spin);
1714 if (bootverbose) {
1715 kprintf("vm_page_alloc_contig: %ldk nospace\n",
1716 (size + PAGE_MASK) * (PAGE_SIZE / 1024));
1717 }
1718 return(NULL);
1719 }
1720 if (high && ((vm_paddr_t)(blk + size) << PAGE_SHIFT) > high) {
1721 alist_free(&vm_contig_alist, blk, size);
1722 spin_unlock(&vm_contig_spin);
1723 if (bootverbose) {
1724 kprintf("vm_page_alloc_contig: %ldk high "
1725 "%016jx failed\n",
1726 (size + PAGE_MASK) * (PAGE_SIZE / 1024),
1727 (intmax_t)high);
1728 }
1729 return(NULL);
1730 }
1731 spin_unlock(&vm_contig_spin);
ef67e7a3 1732 if (vm_contig_verbose) {
79d182b0
MD
1733 kprintf("vm_page_alloc_contig: %016jx/%ldk\n",
1734 (intmax_t)(vm_paddr_t)blk << PAGE_SHIFT,
1735 (size + PAGE_MASK) * (PAGE_SIZE / 1024));
1736 }
1737 return (PHYS_TO_VM_PAGE((vm_paddr_t)blk << PAGE_SHIFT));
1738}
1739
1740/*
1741 * Free contiguously allocated pages. The pages will be wired but not busy.
1742 * When freeing to the alist we leave them wired and not busy.
1743 */
1744void
1745vm_page_free_contig(vm_page_t m, unsigned long size)
1746{
1747 vm_paddr_t pa = VM_PAGE_TO_PHYS(m);
1748 vm_pindex_t start = pa >> PAGE_SHIFT;
1749 vm_pindex_t pages = (size + PAGE_MASK) >> PAGE_SHIFT;
1750
ef67e7a3 1751 if (vm_contig_verbose) {
79d182b0
MD
1752 kprintf("vm_page_free_contig: %016jx/%ldk\n",
1753 (intmax_t)pa, size / 1024);
1754 }
1755 if (pa < vm_low_phys_reserved) {
1756 KKASSERT(pa + size <= vm_low_phys_reserved);
1757 spin_lock(&vm_contig_spin);
1758 alist_free(&vm_contig_alist, start, pages);
1759 spin_unlock(&vm_contig_spin);
1760 } else {
1761 while (pages) {
1762 vm_page_busy_wait(m, FALSE, "cpgfr");
1763 vm_page_unwire(m, 0);
1764 vm_page_free(m);
1765 --pages;
1766 ++m;
1767 }
1768
1769 }
1770}
1771
1772
1773/*
163f8d24
MD
1774 * Wait for sufficient free memory for nominal heavy memory use kernel
1775 * operations.
55b50bd5
MD
1776 *
1777 * WARNING! Be sure never to call this in any vm_pageout code path, which
1778 * will trivially deadlock the system.
163f8d24
MD
1779 */
1780void
1781vm_wait_nominal(void)
1782{
1783 while (vm_page_count_min(0))
1784 vm_wait(0);
1785}
1786
1787/*
12052253
MD
1788 * Test if vm_wait_nominal() would block.
1789 */
1790int
1791vm_test_nominal(void)
1792{
1793 if (vm_page_count_min(0))
1794 return(1);
1795 return(0);
1796}
1797
1798/*
de71fd3f
MD
1799 * Block until free pages are available for allocation, called in various
1800 * places before memory allocations.
cd3c66bd
MD
1801 *
1802 * The caller may loop if vm_page_count_min() == FALSE so we cannot be
1803 * more generous then that.
984263bc 1804 */
984263bc 1805void
4ecf7cc9 1806vm_wait(int timo)
984263bc 1807{
cd3c66bd
MD
1808 /*
1809 * never wait forever
1810 */
1811 if (timo == 0)
1812 timo = hz;
9ad0147b 1813 lwkt_gettoken(&vm_token);
cd3c66bd 1814
bc6dffab 1815 if (curthread == pagethread) {
cd3c66bd
MD
1816 /*
1817 * The pageout daemon itself needs pages, this is bad.
1818 */
1819 if (vm_page_count_min(0)) {
1820 vm_pageout_pages_needed = 1;
1821 tsleep(&vm_pageout_pages_needed, 0, "VMWait", timo);
1822 }
984263bc 1823 } else {
cd3c66bd
MD
1824 /*
1825 * Wakeup the pageout daemon if necessary and wait.
1826 */
1827 if (vm_page_count_target()) {
1828 if (vm_pages_needed == 0) {
1829 vm_pages_needed = 1;
1830 wakeup(&vm_pages_needed);
1831 }
1832 ++vm_pages_waiting; /* SMP race ok */
1833 tsleep(&vmstats.v_free_count, 0, "vmwait", timo);
984263bc 1834 }
984263bc 1835 }
9ad0147b 1836 lwkt_reltoken(&vm_token);
984263bc
MD
1837}
1838
1839/*
de71fd3f
MD
1840 * Block until free pages are available for allocation
1841 *
cd3c66bd 1842 * Called only from vm_fault so that processes page faulting can be
de71fd3f 1843 * easily tracked.
984263bc 1844 */
984263bc
MD
1845void
1846vm_waitpfault(void)
1847{
cd3c66bd
MD
1848 /*
1849 * Wakeup the pageout daemon if necessary and wait.
1850 */
1851 if (vm_page_count_target()) {
1852 lwkt_gettoken(&vm_token);
1853 if (vm_page_count_target()) {
1854 if (vm_pages_needed == 0) {
1855 vm_pages_needed = 1;
1856 wakeup(&vm_pages_needed);
1857 }
1858 ++vm_pages_waiting; /* SMP race ok */
1859 tsleep(&vmstats.v_free_count, 0, "pfault", hz);
1860 }
1861 lwkt_reltoken(&vm_token);
984263bc 1862 }
984263bc
MD
1863}
1864
1865/*
de71fd3f
MD
1866 * Put the specified page on the active list (if appropriate). Ensure
1867 * that act_count is at least ACT_INIT but do not otherwise mess with it.
984263bc 1868 *
b12defdc 1869 * The caller should be holding the page busied ? XXX
de71fd3f 1870 * This routine may not block.
984263bc
MD
1871 */
1872void
1873vm_page_activate(vm_page_t m)
1874{
b12defdc 1875 u_short oqueue;
984263bc 1876
b12defdc 1877 vm_page_spin_lock(m);
027193eb 1878 if (m->queue - m->pc != PQ_ACTIVE) {
b12defdc
MD
1879 _vm_page_queue_spin_lock(m);
1880 oqueue = _vm_page_rem_queue_spinlocked(m);
1881 /* page is left spinlocked, queue is unlocked */
984263bc 1882
b12defdc
MD
1883 if (oqueue == PQ_CACHE)
1884 mycpu->gd_cnt.v_reactivated++;
984263bc 1885 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
984263bc
MD
1886 if (m->act_count < ACT_INIT)
1887 m->act_count = ACT_INIT;
027193eb 1888 _vm_page_add_queue_spinlocked(m, PQ_ACTIVE + m->pc, 0);
984263bc 1889 }
b12defdc
MD
1890 _vm_page_and_queue_spin_unlock(m);
1891 if (oqueue == PQ_CACHE || oqueue == PQ_FREE)
1892 pagedaemon_wakeup();
984263bc
MD
1893 } else {
1894 if (m->act_count < ACT_INIT)
1895 m->act_count = ACT_INIT;
b12defdc 1896 vm_page_spin_unlock(m);
984263bc 1897 }
984263bc
MD
1898}
1899
1900/*
de71fd3f
MD
1901 * Helper routine for vm_page_free_toq() and vm_page_cache(). This
1902 * routine is called when a page has been added to the cache or free
1903 * queues.
984263bc 1904 *
de71fd3f 1905 * This routine may not block.
984263bc
MD
1906 */
1907static __inline void
1908vm_page_free_wakeup(void)
1909{
1910 /*
cd3c66bd
MD
1911 * If the pageout daemon itself needs pages, then tell it that
1912 * there are some free.
984263bc
MD
1913 */
1914 if (vm_pageout_pages_needed &&
de71fd3f
MD
1915 vmstats.v_cache_count + vmstats.v_free_count >=
1916 vmstats.v_pageout_free_min
1917 ) {
984263bc
MD
1918 wakeup(&vm_pageout_pages_needed);
1919 vm_pageout_pages_needed = 0;
1920 }
de71fd3f 1921
984263bc 1922 /*
cd3c66bd
MD
1923 * Wakeup processes that are waiting on memory.
1924 *
1925 * NOTE: vm_paging_target() is the pageout daemon's target, while
1926 * vm_page_count_target() is somewhere inbetween. We want
1927 * to wake processes up prior to the pageout daemon reaching
1928 * its target to provide some hysteresis.
984263bc 1929 */
cd3c66bd
MD
1930 if (vm_pages_waiting) {
1931 if (!vm_page_count_target()) {
1932 /*
1933 * Plenty of pages are free, wakeup everyone.
1934 */
1935 vm_pages_waiting = 0;
1936 wakeup(&vmstats.v_free_count);
1937 ++mycpu->gd_cnt.v_ppwakeups;
1938 } else if (!vm_page_count_min(0)) {
1939 /*
1940 * Some pages are free, wakeup someone.
1941 */
1942 int wcount = vm_pages_waiting;
1943 if (wcount > 0)
1944 --wcount;
1945 vm_pages_waiting = wcount;
1946 wakeup_one(&vmstats.v_free_count);
1947 ++mycpu->gd_cnt.v_ppwakeups;
1948 }
984263bc
MD
1949 }
1950}
1951
1952/*
b12defdc
MD
1953 * Returns the given page to the PQ_FREE or PQ_HOLD list and disassociates
1954 * it from its VM object.
984263bc 1955 *
b12defdc
MD
1956 * The vm_page must be PG_BUSY on entry. PG_BUSY will be released on
1957 * return (the page will have been freed).
984263bc 1958 */
984263bc
MD
1959void
1960vm_page_free_toq(vm_page_t m)
1961{
12e4aaff 1962 mycpu->gd_cnt.v_tfree++;
17cde63e 1963 KKASSERT((m->flags & PG_MAPPED) == 0);
b12defdc 1964 KKASSERT(m->flags & PG_BUSY);
17cde63e 1965
984263bc 1966 if (m->busy || ((m->queue - m->pc) == PQ_FREE)) {
79d182b0
MD
1967 kprintf("vm_page_free: pindex(%lu), busy(%d), "
1968 "PG_BUSY(%d), hold(%d)\n",
1969 (u_long)m->pindex, m->busy,
1970 ((m->flags & PG_BUSY) ? 1 : 0), m->hold_count);
984263bc
MD
1971 if ((m->queue - m->pc) == PQ_FREE)
1972 panic("vm_page_free: freeing free page");
1973 else
1974 panic("vm_page_free: freeing busy page");
1975 }
1976
1977 /*
b12defdc
MD
1978 * Remove from object, spinlock the page and its queues and
1979 * remove from any queue. No queue spinlock will be held
1980 * after this section (because the page was removed from any
1981 * queue).
984263bc 1982 */
984263bc 1983 vm_page_remove(m);
b12defdc
MD
1984 vm_page_and_queue_spin_lock(m);
1985 _vm_page_rem_queue_spinlocked(m);
984263bc
MD
1986
1987 /*
f2d22ebf
MD
1988 * No further management of fictitious pages occurs beyond object
1989 * and queue removal.
984263bc 1990 */
984263bc 1991 if ((m->flags & PG_FICTITIOUS) != 0) {
b12defdc 1992 vm_page_spin_unlock(m);
9765affa 1993 vm_page_wakeup(m);
984263bc
MD
1994 return;
1995 }
1996
1997 m->valid = 0;
1998 vm_page_undirty(m);
1999
2000 if (m->wire_count != 0) {
2001 if (m->wire_count > 1) {
de71fd3f
MD
2002 panic(
2003 "vm_page_free: invalid wire count (%d), pindex: 0x%lx",
2004 m->wire_count, (long)m->pindex);
984263bc 2005 }
73c351d1 2006 panic("vm_page_free: freeing wired page");
984263bc
MD
2007 }
2008
2009 /*
984263bc 2010 * Clear the UNMANAGED flag when freeing an unmanaged page.
9bf025db 2011 * Clear the NEED_COMMIT flag
984263bc 2012 */
9bf025db 2013 if (m->flags & PG_UNMANAGED)
b12defdc 2014 vm_page_flag_clear(m, PG_UNMANAGED);
9bf025db
MD
2015 if (m->flags & PG_NEED_COMMIT)
2016 vm_page_flag_clear(m, PG_NEED_COMMIT);
984263bc
MD
2017
2018 if (m->hold_count != 0) {
d0aa00e8 2019 vm_page_flag_clear(m, PG_ZERO);
027193eb 2020 _vm_page_add_queue_spinlocked(m, PQ_HOLD + m->pc, 0);
de71fd3f 2021 } else {
b12defdc 2022 _vm_page_add_queue_spinlocked(m, PQ_FREE + m->pc, 0);
de71fd3f 2023 }
984263bc
MD
2024
2025 /*
b12defdc
MD
2026 * This sequence allows us to clear PG_BUSY while still holding
2027 * its spin lock, which reduces contention vs allocators. We
2028 * must not leave the queue locked or _vm_page_wakeup() may
2029 * deadlock.
984263bc 2030 */
b12defdc
MD
2031 _vm_page_queue_spin_unlock(m);
2032 if (_vm_page_wakeup(m)) {
2033 vm_page_spin_unlock(m);
2034 wakeup(m);
984263bc 2035 } else {
b12defdc 2036 vm_page_spin_unlock(m);
984263bc 2037 }
984263bc 2038 vm_page_free_wakeup();
984263bc
MD
2039}
2040
2041/*
bb6811be
MD
2042 * vm_page_free_fromq_fast()
2043 *
2044 * Remove a non-zero page from one of the free queues; the page is removed for
2045 * zeroing, so do not issue a wakeup.
bb6811be
MD
2046 */
2047vm_page_t
2048vm_page_free_fromq_fast(void)
2049{
2050 static int qi;
2051 vm_page_t m;
2052 int i;
2053
bb6811be
MD
2054 for (i = 0; i < PQ_L2_SIZE; ++i) {
2055 m = vm_page_list_find(PQ_FREE, qi, FALSE);
b12defdc
MD
2056 /* page is returned spinlocked and removed from its queue */
2057 if (m) {
2058 if (vm_page_busy_try(m, TRUE)) {
2059 /*
2060 * We were unable to busy the page, deactivate
2061 * it and loop.
2062 */
2063 _vm_page_deactivate_locked(m, 0);
2064 vm_page_spin_unlock(m);
90244566 2065 } else if (m->flags & PG_ZERO) {
b12defdc
MD
2066 /*
2067 * The page is PG_ZERO, requeue it and loop
2068 */
2069 _vm_page_add_queue_spinlocked(m,
2070 PQ_FREE + m->pc,
2071 0);
2072 vm_page_queue_spin_unlock(m);
2073 if (_vm_page_wakeup(m)) {
2074 vm_page_spin_unlock(m);
2075 wakeup(m);
2076 } else {
2077 vm_page_spin_unlock(m);
2078 }
90244566
MD
2079 } else {
2080 /*
2081 * The page is not PG_ZERO'd so return it.
2082 */
2083 vm_page_spin_unlock(m);
9bf025db
MD
2084 KKASSERT((m->flags & (PG_UNMANAGED |
2085 PG_NEED_COMMIT)) == 0);
90244566
MD
2086 KKASSERT(m->hold_count == 0);
2087 KKASSERT(m->wire_count == 0);
2088 break;
b12defdc
MD
2089 }
2090 m = NULL;
bb6811be 2091 }
b12defdc 2092 qi = (qi + PQ_PRIME2) & PQ_L2_MASK;
bb6811be 2093 }
bb6811be
MD
2094 return (m);
2095}
2096
2097/*
de71fd3f
MD
2098 * vm_page_unmanage()
2099 *
2100 * Prevent PV management from being done on the page. The page is
2101 * removed from the paging queues as if it were wired, and as a
2102 * consequence of no longer being managed the pageout daemon will not
2103 * touch it (since there is no way to locate the pte mappings for the
2104 * page). madvise() calls that mess with the pmap will also no longer
2105 * operate on the page.
2106 *
2107 * Beyond that the page is still reasonably 'normal'. Freeing the page
2108 * will clear the flag.
2109 *
2110 * This routine is used by OBJT_PHYS objects - objects using unswappable
2111 * physical memory as backing store rather then swap-backed memory and
2112 * will eventually be extended to support 4MB unmanaged physical
2113 * mappings.
654a39f0 2114 *
b12defdc 2115 * Caller must be holding the page busy.
984263bc 2116 */
984263bc
MD
2117void
2118vm_page_unmanage(vm_page_t m)
2119{
b12defdc 2120 KKASSERT(m->flags & PG_BUSY);
984263bc
MD
2121 if ((m->flags & PG_UNMANAGED) == 0) {
2122 if (m->wire_count == 0)
2123 vm_page_unqueue(m);
2124 }
2125 vm_page_flag_set(m, PG_UNMANAGED);
984263bc
MD
2126}
2127
2128/*
de71fd3f
MD
2129 * Mark this page as wired down by yet another map, removing it from
2130 * paging queues as necessary.
984263bc 2131 *
b12defdc 2132 * Caller must be holding the page busy.
984263bc
MD
2133 */
2134void
2135vm_page_wire(vm_page_t m)
2136{
984263bc
MD
2137 /*
2138 * Only bump the wire statistics if the page is not already wired,
2139 * and only unqueue the page if it is on some queue (if it is unmanaged
f2d22ebf
MD
2140 * it is already off the queues). Don't do anything with fictitious
2141 * pages because they are always wired.
984263bc 2142 */
b12defdc 2143 KKASSERT(m->flags & PG_BUSY);
f2d22ebf 2144 if ((m->flags & PG_FICTITIOUS) == 0) {
b12defdc 2145 if (atomic_fetchadd_int(&m->wire_count, 1) == 0) {
f2d22ebf
MD
2146 if ((m->flags & PG_UNMANAGED) == 0)
2147 vm_page_unqueue(m);
b12defdc 2148 atomic_add_int(&vmstats.v_wire_count, 1);
f2d22ebf 2149 }
f2d22ebf 2150 KASSERT(m->wire_count != 0,
17cde63e 2151 ("vm_page_wire: wire_count overflow m=%p", m));
984263bc 2152 }
984263bc
MD
2153}
2154
2155/*
de71fd3f
MD
2156 * Release one wiring of this page, potentially enabling it to be paged again.
2157 *
2158 * Many pages placed on the inactive queue should actually go
2159 * into the cache, but it is difficult to figure out which. What
2160 * we do instead, if the inactive target is well met, is to put
2161 * clean pages at the head of the inactive queue instead of the tail.
2162 * This will cause them to be moved to the cache more quickly and
2163 * if not actively re-referenced, freed more quickly. If we just
2164 * stick these pages at the end of the inactive queue, heavy filesystem
2165 * meta-data accesses can cause an unnecessary paging load on memory bound
2166 * processes. This optimization causes one-time-use metadata to be
2167 * reused more quickly.
2168 *
f84f7e81
MD
2169 * Pages marked PG_NEED_COMMIT are always activated and never placed on
2170 * the inactive queue. This helps the pageout daemon determine memory
2171 * pressure and act on out-of-memory situations more quickly.
2172 *
de71fd3f
MD
2173 * BUT, if we are in a low-memory situation we have no choice but to
2174 * put clean pages on the cache queue.
2175 *
2176 * A number of routines use vm_page_unwire() to guarantee that the page
2177 * will go into either the inactive or active queues, and will NEVER
2178 * be placed in the cache - for example, just after dirtying a page.
2179 * dirty pages in the cache are not allowed.
2180 *
2181 * The page queues must be locked.
2182 * This routine may not block.
984263bc
MD
2183 */
2184void
2185vm_page_unwire(vm_page_t m, int activate)
2186{
b12defdc 2187 KKASSERT(m->flags & PG_BUSY);
f2d22ebf
MD
2188 if (m->flags & PG_FICTITIOUS) {
2189 /* do nothing */
2190 } else if (m->wire_count <= 0) {
2191 panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
2192 } else {
b12defdc
MD
2193 if (atomic_fetchadd_int(&m->wire_count, -1) == 1) {
2194 atomic_add_int(&vmstats.v_wire_count, -1);
984263bc
MD
2195 if (m->flags & PG_UNMANAGED) {
2196 ;
f84f7e81 2197 } else if (activate || (m->flags & PG_NEED_COMMIT)) {
b12defdc 2198 vm_page_spin_lock(m);
027193eb
MD
2199 _vm_page_add_queue_spinlocked(m,
2200 PQ_ACTIVE + m->pc, 0);
b12defdc 2201 _vm_page_and_queue_spin_unlock(m);
984263bc 2202 } else {
b12defdc 2203 vm_page_spin_lock(m);
984263bc 2204 vm_page_flag_clear(m, PG_WINATCFLS);
027193eb
MD
2205 _vm_page_add_queue_spinlocked(m,
2206 PQ_INACTIVE + m->pc, 0);
e527fb6b 2207 ++vm_swapcache_inactive_heuristic;
b12defdc 2208 _vm_page_and_queue_spin_unlock(m);
984263bc
MD
2209 }
2210 }
984263bc 2211 }
984263bc
MD
2212}
2213
984263bc
MD
2214/*
2215 * Move the specified page to the inactive queue. If the page has
2216 * any associated swap, the swap is deallocated.
2217 *
2218 * Normally athead is 0 resulting in LRU operation. athead is set
2219 * to 1 if we want this page to be 'as if it were placed in the cache',
2220 * except without unmapping it from the process address space.
2221 *
b12defdc 2222 * vm_page's spinlock must be held on entry and will remain held on return.
984263bc
MD
2223 * This routine may not block.
2224 */
b12defdc
MD
2225static void
2226_vm_page_deactivate_locked(vm_page_t m, int athead)
984263bc 2227{
b12defdc
MD
2228 u_short oqueue;
2229
984263bc
MD
2230 /*
2231 * Ignore if already inactive.
2232 */
027193eb 2233 if (m->queue - m->pc == PQ_INACTIVE)
984263bc 2234 return;
b12defdc
MD
2235 _vm_page_queue_spin_lock(m);
2236 oqueue = _vm_page_rem_queue_spinlocked(m);
984263bc 2237
984263bc 2238 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
b12defdc 2239 if (oqueue == PQ_CACHE)
12e4aaff 2240 mycpu->gd_cnt.v_reactivated++;
984263bc 2241 vm_page_flag_clear(m, PG_WINATCFLS);
027193eb 2242 _vm_page_add_queue_spinlocked(m, PQ_INACTIVE + m->pc, athead);
b12defdc 2243 if (athead == 0)
e527fb6b 2244 ++vm_swapcache_inactive_heuristic;
984263bc 2245 }
b12defdc
MD
2246 _vm_page_queue_spin_unlock(m);
2247 /* leaves vm_page spinlocked */
984263bc
MD
2248}
2249
573fb415
MD
2250/*
2251 * Attempt to deactivate a page.
2252 *
2253 * No requirements.
2254 */
984263bc
MD
2255void
2256vm_page_deactivate(vm_page_t m)
2257{
b12defdc
MD
2258 vm_page_spin_lock(m);
2259 _vm_page_deactivate_locked(m, 0);
2260 vm_page_spin_unlock(m);
2261}
2262
2263void
2264vm_page_deactivate_locked(vm_page_t m)
2265{
2266 _vm_page_deactivate_locked(m, 0);
984263bc
MD
2267}
2268
2269/*
573fb415 2270 * Attempt to move a page to PQ_CACHE.
b12defdc 2271 *
984263bc 2272 * Returns 0 on failure, 1 on success
573fb415 2273 *
b12defdc
MD
2274 * The page should NOT be busied by the caller. This function will validate
2275 * whether the page can be safely moved to the cache.
984263bc
MD
2276 */
2277int
2278vm_page_try_to_cache(vm_page_t m)
2279{
b12defdc
MD
2280 vm_page_spin_lock(m);
2281 if (vm_page_busy_try(m, TRUE)) {
2282 vm_page_spin_unlock(m);
2283 return(0);
2284 }
2285 if (m->dirty || m->hold_count || m->wire_count ||
9bf025db 2286 (m->flags & (PG_UNMANAGED | PG_NEED_COMMIT))) {
b12defdc
MD
2287 if (_vm_page_wakeup(m)) {
2288 vm_page_spin_unlock(m);
2289 wakeup(m);
2290 } else {
2291 vm_page_spin_unlock(m);
2292 }
984263bc
MD
2293 return(0);
2294 }
b12defdc
MD
2295 vm_page_spin_unlock(m);
2296
2297 /*
2298 * Page busied by us and no longer spinlocked. Dirty pages cannot
2299 * be moved to the cache.
2300 */
984263bc 2301 vm_page_test_dirty(m);
654a39f0 2302 if (m->dirty) {
50e32333 2303 vm_page_wakeup(m);
984263bc 2304 return(0);
654a39f0 2305 }
984263bc
MD
2306 vm_page_cache(m);
2307 return(1);
2308}
2309
2310/*
de71fd3f
MD
2311 * Attempt to free the page. If we cannot free it, we do nothing.
2312 * 1 is returned on success, 0 on failure.
573fb415
MD
2313 *
2314 * No requirements.
984263bc 2315 */
984263bc
MD
2316int
2317vm_page_try_to_free(vm_page_t m)
2318{
b12defdc
MD
2319 vm_page_spin_lock(m);
2320 if (vm_page_busy_try(m, TRUE)) {
2321 vm_page_spin_unlock(m);
2322 return(0);
2323 }
82034c53
MD
2324
2325 /*
2326 * The page can be in any state, including already being on the free
2327 * queue. Check to see if it really can be freed.
2328 */
2329 if (m->dirty || /* can't free if it is dirty */
2330 m->hold_count || /* or held (XXX may be wrong) */
2331 m->wire_count || /* or wired */
9bf025db
MD
2332 (m->flags & (PG_UNMANAGED | /* or unmanaged */
2333 PG_NEED_COMMIT)) || /* or needs a commit */
82034c53
MD
2334 m->queue - m->pc == PQ_FREE || /* already on PQ_FREE */
2335 m->queue - m->pc == PQ_HOLD) { /* already on PQ_HOLD */
b12defdc
MD
2336 if (_vm_page_wakeup(m)) {
2337 vm_page_spin_unlock(m);
2338 wakeup(m);
2339 } else {
2340 vm_page_spin_unlock(m);
2341 }
984263bc
MD
2342 return(0);
2343 }
b12defdc
MD
2344 vm_page_spin_unlock(m);
2345
2346 /*
82034c53
MD
2347 * We can probably free the page.
2348 *
b12defdc
MD
2349 * Page busied by us and no longer spinlocked. Dirty pages will
2350 * not be freed by this function. We have to re-test the
2351 * dirty bit after cleaning out the pmaps.
2352 */
984263bc 2353 vm_page_test_dirty(m);
654a39f0 2354 if (m->dirty) {
b12defdc 2355 vm_page_wakeup(m);
984263bc 2356 return(0);
654a39f0 2357 }
984263bc 2358 vm_page_protect(m, VM_PROT_NONE);
b12defdc
MD
2359 if (m->dirty) {
2360 vm_page_wakeup(m);
2361 return(0);
2362 }
984263bc
MD
2363 vm_page_free(m);
2364 return(1);
2365}
2366
984263bc
MD
2367/*
2368 * vm_page_cache
2369 *
2370 * Put the specified page onto the page cache queue (if appropriate).
2371 *
a491077e
MD
2372 * The page must be busy, and this routine will release the busy and
2373 * possibly even free the page.
984263bc
MD
2374 */
2375void
2376vm_page_cache(vm_page_t m)
2377{
9bf025db
MD
2378 if ((m->flags & (PG_UNMANAGED | PG_NEED_COMMIT)) ||
2379 m->busy || m->wire_count || m->hold_count) {
086c1d7e 2380 kprintf("vm_page_cache: attempting to cache busy/held page\n");
a491077e 2381 vm_page_wakeup(m);
984263bc
MD
2382 return;
2383 }
c9ec86b3
MD
2384
2385 /*
2386 * Already in the cache (and thus not mapped)
2387 */
17cde63e
MD
2388 if ((m->queue - m->pc) == PQ_CACHE) {
2389 KKASSERT((m->flags & PG_MAPPED) == 0);
a491077e 2390 vm_page_wakeup(m);
984263bc 2391 return;
17cde63e 2392 }
984263bc
MD
2393
2394 /*
c9ec86b3
MD
2395 * Caller is required to test m->dirty, but note that the act of
2396 * removing the page from its maps can cause it to become dirty
2397 * on an SMP system due to another cpu running in usermode.
984263bc 2398 */
c9ec86b3 2399 if (m->dirty) {
984263bc
MD
2400 panic("vm_page_cache: caching a dirty page, pindex: %ld",
2401 (long)m->pindex);
2402 }
c9ec86b3
MD
2403
2404 /*
2405 * Remove all pmaps and indicate that the page is not
17cde63e
MD
2406 * writeable or mapped. Our vm_page_protect() call may
2407 * have blocked (especially w/ VM_PROT_NONE), so recheck
2408 * everything.
c9ec86b3
MD
2409 */
2410 vm_page_protect(m, VM_PROT_NONE);
9bf025db
MD
2411 if ((m->flags & (PG_UNMANAGED | PG_MAPPED)) ||
2412 m->busy || m->wire_count || m->hold_count) {
a491077e 2413 vm_page_wakeup(m);
9bf025db 2414 } else if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
c9ec86b3 2415 vm_page_deactivate(m);
a491077e 2416 vm_page_wakeup(m);
c9ec86b3 2417 } else {
b12defdc
MD
2418 _vm_page_and_queue_spin_lock(m);
2419 _vm_page_rem_queue_spinlocked(m);
2420 _vm_page_add_queue_spinlocked(m, PQ_CACHE + m->pc, 0);
2421 _vm_page_queue_spin_unlock(m);
2422 if (_vm_page_wakeup(m)) {
2423 vm_page_spin_unlock(m);
2424 wakeup(m);
2425 } else {
2426 vm_page_spin_unlock(m);
2427 }
c9ec86b3
MD
2428 vm_page_free_wakeup();
2429 }
984263bc
MD
2430}
2431
2432/*
de71fd3f
MD
2433 * vm_page_dontneed()
2434 *
2435 * Cache, deactivate, or do nothing as appropriate. This routine
2436 * is typically used by madvise() MADV_DONTNEED.
2437 *
2438 * Generally speaking we want to move the page into the cache so
2439 * it gets reused quickly. However, this can result in a silly syndrome
2440 * due to the page recycling too quickly. Small objects will not be
2441 * fully cached. On the otherhand, if we move the page to the inactive
2442 * queue we wind up with a problem whereby very large objects
2443 * unnecessarily blow away our inactive and cache queues.
2444 *
2445 * The solution is to move the pages based on a fixed weighting. We
2446 * either leave them alone, deactivate them, or move them to the cache,
2447 * where moving them to the cache has the highest weighting.
2448 * By forcing some pages into other queues we eventually force the
2449 * system to balance the queues, potentially recovering other unrelated
2450 * space from active. The idea is to not force this to happen too
2451 * often.
573fb415 2452 *
b12defdc 2453 * The page must be busied.
984263bc 2454 */
984263bc
MD
2455void
2456vm_page_dontneed(vm_page_t m)
2457{
2458 static int dnweight;
2459 int dnw;
2460 int head;
2461
2462 dnw = ++dnweight;
2463
2464 /*
2465 * occassionally leave the page alone
2466 */
984263bc 2467 if ((dnw & 0x01F0) == 0 ||
027193eb 2468 m->queue - m->pc == PQ_INACTIVE ||
984263bc
MD
2469 m->queue - m->pc == PQ_CACHE
2470 ) {
2471 if (m->act_count >= ACT_INIT)
2472 --m->act_count;
2473 return;
2474 }
2475
31da5e4d
VS
2476 /*
2477 * If vm_page_dontneed() is inactivating a page, it must clear
2478 * the referenced flag; otherwise the pagedaemon will see references
2479 * on the page in the inactive queue and reactivate it. Until the
2480 * page can move to the cache queue, madvise's job is not done.
2481 */
2482 vm_page_flag_clear(m, PG_REFERENCED);
2483 pmap_clear_reference(m);
2484
984263bc
MD
2485 if (m->dirty == 0)
2486 vm_page_test_dirty(m);
2487
2488 if (m->dirty || (dnw & 0x0070) == 0) {
2489 /*
2490 * Deactivate the page 3 times out of 32.
2491 */
2492 head = 0;
2493 } else {
2494 /*
2495 * Cache the page 28 times out of every 32. Note that
2496 * the page is deactivated instead of cached, but placed
2497 * at the head of the queue instead of the tail.
2498 */
2499 head = 1;
2500 }
b12defdc
MD
2501 vm_page_spin_lock(m);
2502 _vm_page_deactivate_locked(m, head);
2503 vm_page_spin_unlock(m);
2504}
2505
2506/*
2507 * These routines manipulate the 'soft busy' count for a page. A soft busy
2508 * is almost like PG_BUSY except that it allows certain compatible operations
2509 * to occur on the page while it is busy. For example, a page undergoing a
2510 * write can still be mapped read-only.
2511 *
2512 * Because vm_pages can overlap buffers m->busy can be > 1. m->busy is only
2513 * adjusted while the vm_page is PG_BUSY so the flash will occur when the
2514 * busy bit is cleared.
2515 */
2516void
2517vm_page_io_start(vm_page_t m)
2518{
2519 KASSERT(m->flags & PG_BUSY, ("vm_page_io_start: page not busy!!!"));
2520 atomic_add_char(&m->busy, 1);
2521 vm_page_flag_set(m, PG_SBUSY);
2522}
2523
2524void
2525vm_page_io_finish(vm_page_t m)
2526{
2527 KASSERT(m->flags & PG_BUSY, ("vm_page_io_finish: page not busy!!!"));
2528 atomic_subtract_char(&m->busy, 1);
2529 if (m->busy == 0)
2530 vm_page_flag_clear(m, PG_SBUSY);
984263bc
MD
2531}
2532
2533/*
9bf025db
MD
2534 * Indicate that a clean VM page requires a filesystem commit and cannot
2535 * be reused. Used by tmpfs.
2536 */
2537void
2538vm_page_need_commit(vm_page_t m)
2539{
2540 vm_page_flag_set(m, PG_NEED_COMMIT);
2541}
2542
2543void
2544vm_page_clear_commit(vm_page_t m)
2545{
2546 vm_page_flag_clear(m, PG_NEED_COMMIT);
2547}
2548
2549/*
06ecca5a 2550 * Grab a page, blocking if it is busy and allocating a page if necessary.
d2d8515b
MD
2551 * A busy page is returned or NULL. The page may or may not be valid and
2552 * might not be on a queue (the caller is responsible for the disposition of
2553 * the page).
984263bc 2554 *
d2d8515b
MD
2555 * If VM_ALLOC_ZERO is specified and the grab must allocate a new page, the
2556 * page will be zero'd and marked valid.
b12defdc 2557 *
d2d8515b
MD
2558 * If VM_ALLOC_FORCE_ZERO is specified the page will be zero'd and marked
2559 * valid even if it already exists.
2560 *
2561 * If VM_ALLOC_RETRY is specified this routine will never return NULL. Also
2562 * note that VM_ALLOC_NORMAL must be specified if VM_ALLOC_RETRY is specified.
d149178e 2563 * VM_ALLOC_NULL_OK is implied when VM_ALLOC_RETRY is specified.
dc1fd4b3 2564 *
06ecca5a
MD
2565 * This routine may block, but if VM_ALLOC_RETRY is not set then NULL is
2566 * always returned if we had blocked.
d2d8515b 2567 *
06ecca5a 2568 * This routine may not be called from an interrupt.
06ecca5a 2569 *
d2d8515b 2570 * PG_ZERO is *ALWAYS* cleared by this routine.
573fb415 2571 *
d2d8515b 2572 * No other requirements.
984263bc
MD
2573 */
2574vm_page_t
2575vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
2576{
984263bc 2577 vm_page_t m;
b12defdc 2578 int error;
984263bc 2579
dc1fd4b3
MD
2580 KKASSERT(allocflags &
2581 (VM_ALLOC_NORMAL|VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
398c240d 2582 vm_object_hold(object);
b12defdc
MD
2583 for (;;) {
2584 m = vm_page_lookup_busy_try(object, pindex, TRUE, &error);
2585 if (error) {
2586 vm_page_sleep_busy(m, TRUE, "pgrbwt");
2587 if ((allocflags & VM_ALLOC_RETRY) == 0) {
2588 m = NULL;
2589 break;
984263bc 2590 }
d2d8515b 2591 /* retry */
b12defdc 2592 } else if (m == NULL) {
d149178e
MD
2593 if (allocflags & VM_ALLOC_RETRY)
2594 allocflags |= VM_ALLOC_NULL_OK;
b12defdc
MD
2595 m = vm_page_alloc(object, pindex,
2596 allocflags & ~VM_ALLOC_RETRY);
2597 if (m)
2598 break;
2599 vm_wait(0);
2600 if ((allocflags & VM_ALLOC_RETRY) == 0)
d2d8515b 2601 goto failed;
984263bc 2602 } else {
b12defdc
MD
2603 /* m found */
2604 break;
984263bc
MD
2605 }
2606 }
d2d8515b
MD
2607
2608 /*
2609 * If VM_ALLOC_ZERO an invalid page will be zero'd and set valid.
2610 *
2611 * If VM_ALLOC_FORCE_ZERO the page is unconditionally zero'd and set
2612 * valid even if already valid.
2613 */
2614 if (m->valid == 0) {
2615 if (allocflags & (VM_ALLOC_ZERO | VM_ALLOC_FORCE_ZERO)) {
2616 if ((m->flags & PG_ZERO) == 0)
2617 pmap_zero_page(VM_PAGE_TO_PHYS(m));
2618 m->valid = VM_PAGE_BITS_ALL;
2619 }
2620 } else if (allocflags & VM_ALLOC_FORCE_ZERO) {
2621 pmap_zero_page(VM_PAGE_TO_PHYS(m));
2622 m->valid = VM_PAGE_BITS_ALL;
2623 }
2624 vm_page_flag_clear(m, PG_ZERO);
2625failed:
398c240d 2626 vm_object_drop(object);
06ecca5a 2627 return(m);
984263bc
MD
2628}
2629
2630/*
2631 * Mapping function for valid bits or for dirty bits in
2632 * a page. May not block.
2633 *
2634 * Inputs are required to range within a page.
573fb415
MD
2635 *
2636 * No requirements.
2637 * Non blocking.
984263bc 2638 */
573fb415 2639int
984263bc
MD
2640vm_page_bits(int base, int size)
2641{
2642 int first_bit;
2643 int last_bit;
2644
2645 KASSERT(
2646 base + size <= PAGE_SIZE,
2647 ("vm_page_bits: illegal base/size %d/%d", base, size)
2648 );
2649
2650 if (size == 0) /* handle degenerate case */
2651 return(0);
2652
2653 first_bit = base >> DEV_BSHIFT;
2654 last_bit = (base + size - 1) >> DEV_BSHIFT;
2655
2656 return ((2 << last_bit) - (1 << first_bit));
2657}
2658
2659/*
de71fd3f
MD
2660 * Sets portions of a page valid and clean. The arguments are expected
2661 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
2662 * of any partial chunks touched by the range. The invalid portion of
2663 * such chunks will be zero'd.
984263bc 2664 *
c7841cbe
MD
2665 * NOTE: When truncating a buffer vnode_pager_setsize() will automatically
2666 * align base to DEV_BSIZE so as not to mark clean a partially
2667 * truncated device block. Otherwise the dirty page status might be
2668 * lost.
2669 *
de71fd3f 2670 * This routine may not block.
984263bc 2671 *
de71fd3f 2672 * (base + size) must be less then or equal to PAGE_SIZE.
984263bc 2673 */
1a54183b
MD
2674static void
2675_vm_page_zero_valid(vm_page_t m, int base, int size)
984263bc 2676{
984263bc
MD
2677 int frag;
2678 int endoff;
2679
2680 if (size == 0) /* handle degenerate case */
2681 return;
2682
2683 /*
2684 * If the base is not DEV_BSIZE aligned and the valid
2685 * bit is clear, we have to zero out a portion of the
2686 * first block.
2687 */
2688
2689 if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
2690 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0
2691 ) {
2692 pmap_zero_page_area(
2693 VM_PAGE_TO_PHYS(m),
2694 frag,
2695 base - frag
2696 );
2697 }
2698
2699 /*
2700 * If the ending offset is not DEV_BSIZE aligned and the
2701 * valid bit is clear, we have to zero out a portion of
2702 * the last block.
2703 */
2704
2705 endoff = base + size;
2706
2707 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
2708 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0
2709 ) {
2710 pmap_zero_page_area(
2711 VM_PAGE_TO_PHYS(m),
2712 endoff,
2713 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))
2714 );
2715 }
1a54183b 2716}
984263bc 2717
1a54183b
MD
2718/*
2719 * Set valid, clear dirty bits. If validating the entire
2720 * page we can safely clear the pmap modify bit. We also
2721 * use this opportunity to clear the PG_NOSYNC flag. If a process
2722 * takes a write fault on a MAP_NOSYNC memory area the flag will
2723 * be set again.
2724 *
2725 * We set valid bits inclusive of any overlap, but we can only
2726 * clear dirty bits for DEV_BSIZE chunks that are fully within
2727 * the range.
573fb415
MD
2728 *
2729 * Page must be busied?
2730 * No other requirements.
1a54183b
MD
2731 */
2732void
2733vm_page_set_valid(vm_page_t m, int base, int size)
2734{
2735 _vm_page_zero_valid(m, base, size);
2736 m->valid |= vm_page_bits(base, size);
2737}
984263bc 2738
cb1cf930
MD
2739
2740/*
2741 * Set valid bits and clear dirty bits.
2742 *
2743 * NOTE: This function does not clear the pmap modified bit.
2744 * Also note that e.g. NFS may use a byte-granular base
2745 * and size.
573fb415 2746 *
9a0cb7b1
MD
2747 * WARNING: Page must be busied? But vfs_clean_one_page() will call
2748 * this without necessarily busying the page (via bdwrite()).
2749 * So for now vm_token must also be held.
2750 *
573fb415 2751 * No other requirements.
cb1cf930 2752 */
1a54183b
MD
2753void
2754vm_page_set_validclean(vm_page_t m, int base, int size)
2755{
2756 int pagebits;
2757
2758 _vm_page_zero_valid(m, base, size);
984263bc
MD
2759 pagebits = vm_page_bits(base, size);
2760 m->valid |= pagebits;
984263bc
MD
2761 m->dirty &= ~pagebits;
2762 if (base == 0 && size == PAGE_SIZE) {
cb1cf930 2763 /*pmap_clear_modify(m);*/
984263bc
MD
2764 vm_page_flag_clear(m, PG_NOSYNC);
2765 }
2766}
2767
cb1cf930 2768/*
0a8aee15 2769 * Set valid & dirty. Used by buwrite()
573fb415 2770 *
9a0cb7b1
MD
2771 * WARNING: Page must be busied? But vfs_dirty_one_page() will
2772 * call this function in buwrite() so for now vm_token must
9bf025db 2773 * be held.
9a0cb7b1 2774 *
573fb415 2775 * No other requirements.
0a8aee15
MD
2776 */
2777void
2778vm_page_set_validdirty(vm_page_t m, int base, int size)
2779{
2780 int pagebits;
2781
2782 pagebits = vm_page_bits(base, size);
2783 m->valid |= pagebits;
2784 m->dirty |= pagebits;
d89ce96a 2785 if (m->object)
9bf025db 2786 vm_object_set_writeable_dirty(m->object);
0a8aee15
MD
2787}
2788
2789/*
cb1cf930
MD
2790 * Clear dirty bits.
2791 *
2792 * NOTE: This function does not clear the pmap modified bit.
2793 * Also note that e.g. NFS may use a byte-granular base
2794 * and size.
573fb415
MD
2795 *
2796 * Page must be busied?
2797 * No other requirements.
cb1cf930 2798 */
984263bc
MD
2799void
2800vm_page_clear_dirty(vm_page_t m, int base, int size)
2801{
2802 m->dirty &= ~vm_page_bits(base, size);
1a54183b 2803 if (base == 0 && size == PAGE_SIZE) {
cb1cf930 2804 /*pmap_clear_modify(m);*/
1a54183b
MD
2805 vm_page_flag_clear(m, PG_NOSYNC);
2806 }
984263bc
MD
2807}
2808
2809/*
17cde63e
MD
2810 * Make the page all-dirty.
2811 *
2812 * Also make sure the related object and vnode reflect the fact that the
2813 * object may now contain a dirty page.
573fb415
MD
2814 *
2815 * Page must be busied?
2816 * No other requirements.
17cde63e
MD
2817 */
2818void
2819vm_page_dirty(vm_page_t m)
2820{
2821#ifdef INVARIANTS
2822 int pqtype = m->queue - m->pc;
2823#endif
2824 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
2825 ("vm_page_dirty: page in free/cache queue!"));
2826 if (m->dirty != VM_PAGE_BITS_ALL) {
2827 m->dirty = VM_PAGE_BITS_ALL;
2828 if (m->object)
2829 vm_object_set_writeable_dirty(m->object);
2830 }
2831}
2832
2833/*
de71fd3f
MD
2834 * Invalidates DEV_BSIZE'd chunks within a page. Both the
2835 * valid and dirty bits for the effected areas are cleared.
984263bc 2836 *
573fb415
MD
2837 * Page must be busied?
2838 * Does not block.
2839 * No other requirements.
984263bc
MD
2840 */
2841void
2842vm_page_set_invalid(vm_page_t m, int base, int size)
2843{
2844 int bits;
2845
2846 bits = vm_page_bits(base, size);
2847 m->valid &= ~bits;
2848 m->dirty &= ~bits;
2849 m->object->generation++;
2850}
2851
2852/*
de71fd3f
MD
2853 * The kernel assumes that the invalid portions of a page contain
2854 * garbage, but such pages can be mapped into memory by user code.
2855 * When this occurs, we must zero out the non-valid portions of the
2856 * page so user code sees what it expects.
984263bc 2857 *
de71fd3f
MD
2858 * Pages are most often semi-valid when the end of a file is mapped
2859 * into memory and the file's size is not page aligned.
573fb415
MD
2860 *
2861 * Page must be busied?
2862 * No other requirements.
984263bc 2863 */
984263bc
MD
2864void
2865vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
2866{
2867 int b;
2868 int i;
2869
2870 /*
2871 * Scan the valid bits looking for invalid sections that
2872 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
2873 * valid bit may be set ) have already been zerod by
2874 * vm_page_set_validclean().
2875 */
984263bc
MD
2876 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
2877 if (i == (PAGE_SIZE / DEV_BSIZE) ||
2878 (m->valid & (1 << i))
2879 ) {
2880 if (i > b) {
2881 pmap_zero_page_area(
2882 VM_PAGE_TO_PHYS(m),
2883 b << DEV_BSHIFT,
2884 (i - b) << DEV_BSHIFT
2885 );
2886 }
2887 b = i + 1;
2888 }
2889 }
2890
2891 /*
2892 * setvalid is TRUE when we can safely set the zero'd areas
2893 * as being valid. We can do this if there are no cache consistency
2894 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
2895 */
984263bc
MD
2896 if (setvalid)
2897 m->valid = VM_PAGE_BITS_ALL;
2898}
2899
2900/*
de71fd3f
MD
2901 * Is a (partial) page valid? Note that the case where size == 0
2902 * will return FALSE in the degenerate case where the page is entirely
2903 * invalid, and TRUE otherwise.
984263bc 2904 *
573fb415
MD
2905 * Does not block.
2906 * No other requirements.
984263bc 2907 */
984263bc
MD
2908int
2909vm_page_is_valid(vm_page_t m, int base, int size)
2910{
2911 int bits = vm_page_bits(base, size);
2912
2913 if (m->valid && ((m->valid & bits) == bits))
2914 return 1;
2915 else
2916 return 0;
2917}
2918
2919/*
2920 * update dirty bits from pmap/mmu. May not block.
573fb415 2921 *
b12defdc 2922 * Caller must hold the page busy
984263bc 2923 */
984263bc
MD
2924void
2925vm_page_test_dirty(vm_page_t m)
2926{
2927 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
2928 vm_page_dirty(m);
2929 }
2930}
2931
10192bae 2932/*
906c754c
MD
2933 * Register an action, associating it with its vm_page
2934 */
2935void
2936vm_page_register_action(vm_page_action_t action, vm_page_event_t event)
2937{
2938 struct vm_page_action_list *list;
2939 int hv;
2940
2941 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
2942 list = &action_list[hv];
2943
2944 lwkt_gettoken(&vm_token);
2945 vm_page_flag_set(action->m, PG_ACTIONLIST);
2946 action->event = event;
2947 LIST_INSERT_HEAD(list, action, entry);
2948 lwkt_reltoken(&vm_token);
2949}
2950
2951/*
2952 * Unregister an action, disassociating it from its related vm_page
2953 */
2954void
2955vm_page_unregister_action(vm_page_action_t action)
2956{
2957 struct vm_page_action_list *list;
2958 int hv;
2959
2960 lwkt_gettoken(&vm_token);
2961 if (action->event != VMEVENT_NONE) {
2962 action->event = VMEVENT_NONE;
2963 LIST_REMOVE(action, entry);
2964
2965 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
2966 list = &action_list[hv];
2967 if (LIST_EMPTY(list))
2968 vm_page_flag_clear(action->m, PG_ACTIONLIST);
2969 }
2970 lwkt_reltoken(&vm_token);
2971}
2972
2973/*
10192bae
MD
2974 * Issue an event on a VM page. Corresponding action structures are
2975 * removed from the page's list and called.
906c754c
MD
2976 *
2977 * If the vm_page has no more pending action events we clear its
2978 * PG_ACTIONLIST flag.
10192bae
MD
2979 */
2980void
2981vm_page_event_internal(vm_page_t m, vm_page_event_t event)
2982{
906c754c
MD
2983 struct vm_page_action_list *list;
2984 struct vm_page_action *scan;
2985 struct vm_page_action *next;
2986 int hv;
2987 int all;
10192bae 2988
906c754c
MD
2989 hv = (int)((intptr_t)m >> 8) & VMACTION_HMASK;
2990 list = &action_list[hv];
2991 all = 1;
2992
2993 lwkt_gettoken(&vm_token);
2994 LIST_FOREACH_MUTABLE(scan, list, entry, next) {
2995 if (scan->m == m) {
2996 if (scan->event == event) {
2997 scan->event = VMEVENT_NONE;
2998 LIST_REMOVE(scan, entry);
2999 scan->func(m, scan);
3000 /* XXX */
3001 } else {
3002 all = 0;
3003 }
10192bae
MD
3004 }
3005 }
906c754c
MD
3006 if (all)
3007 vm_page_flag_clear(m, PG_ACTIONLIST);
3008 lwkt_reltoken(&vm_token);
10192bae
MD
3009}
3010
984263bc
MD
3011#include "opt_ddb.h"
3012#ifdef DDB
3013#include <sys/kernel.h>
3014
3015#include <ddb/ddb.h>
3016
3017DB_SHOW_COMMAND(page, vm_page_print_page_info)
3018{
12e4aaff
MD
3019 db_printf("vmstats.v_free_count: %d\n", vmstats.v_free_count);
3020 db_printf("vmstats.v_cache_count: %d\n", vmstats.v_cache_count);
3021 db_printf("vmstats.v_inactive_count: %d\n", vmstats.v_inactive_count);
3022 db_printf("vmstats.v_active_count: %d\n", vmstats.v_active_count);
3023 db_printf("vmstats.v_wire_count: %d\n", vmstats.v_wire_count);
3024 db_printf("vmstats.v_free_reserved: %d\n", vmstats.v_free_reserved);
3025 db_printf("vmstats.v_free_min: %d\n", vmstats.v_free_min);
3026 db_printf("vmstats.v_free_target: %d\n", vmstats.v_free_target);
3027 db_printf("vmstats.v_cache_min: %d\n", vmstats.v_cache_min);
3028 db_printf("vmstats.v_inactive_target: %d\n", vmstats.v_inactive_target);
984263bc
MD
3029}
3030
3031DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3032{
3033 int i;
3034 db_printf("PQ_FREE:");
3035 for(i=0;i<PQ_L2_SIZE;i++) {
3036 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt);
3037 }
3038 db_printf("\n");
3039
3040 db_printf("PQ_CACHE:");
3041 for(i=0;i<PQ_L2_SIZE;i++) {
3042 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt);
3043 }
3044 db_printf("\n");
3045
027193eb
MD
3046 db_printf("PQ_ACTIVE:");
3047 for(i=0;i<PQ_L2_SIZE;i++) {
3048 db_printf(" %d", vm_page_queues[PQ_ACTIVE + i].lcnt);
3049 }
3050 db_printf("\n");
3051
3052 db_printf("PQ_INACTIVE:");
3053 for(i=0;i<PQ_L2_SIZE;i++) {
3054 db_printf(" %d", vm_page_queues[PQ_INACTIVE + i].lcnt);
3055 }
3056 db_printf("\n");
984263bc
MD
3057}
3058#endif /* DDB */