kernel - Add missing vm_page_wakeup()
[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.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
39 * $FreeBSD: src/sys/vm/vm_page.c,v 1.147.2.18 2002/03/10 05:03:19 alc Exp $
cfd17028 40 * $DragonFly: src/sys/vm/vm_page.c,v 1.40 2008/08/25 17:01:42 dillon Exp $
984263bc
MD
41 */
42
43/*
44 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
45 * All rights reserved.
46 *
47 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
48 *
49 * Permission to use, copy, modify and distribute this software and
50 * its documentation is hereby granted, provided that both the copyright
51 * notice and this permission notice appear in all copies of the
52 * software, derivative works or modified versions, and any portions
53 * thereof, and that both notices appear in supporting documentation.
54 *
55 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
58 *
59 * Carnegie Mellon requests users of this software to return to
60 *
61 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
62 * School of Computer Science
63 * Carnegie Mellon University
64 * Pittsburgh PA 15213-3890
65 *
66 * any improvements or extensions that they make and grant Carnegie the
67 * rights to redistribute these changes.
68 */
984263bc 69/*
de71fd3f
MD
70 * Resident memory management module. The module manipulates 'VM pages'.
71 * A VM page is the core building block for memory management.
984263bc
MD
72 */
73
74#include <sys/param.h>
75#include <sys/systm.h>
76#include <sys/malloc.h>
77#include <sys/proc.h>
78#include <sys/vmmeter.h>
79#include <sys/vnode.h>
cd3c66bd 80#include <sys/kernel.h>
984263bc
MD
81
82#include <vm/vm.h>
83#include <vm/vm_param.h>
84#include <sys/lock.h>
85#include <vm/vm_kern.h>
86#include <vm/pmap.h>
87#include <vm/vm_map.h>
88#include <vm/vm_object.h>
89#include <vm/vm_page.h>
90#include <vm/vm_pageout.h>
91#include <vm/vm_pager.h>
92#include <vm/vm_extern.h>
096e95c0 93#include <vm/swap_pager.h>
984263bc 94
8e5e6f1b
AH
95#include <machine/md_var.h>
96
bb6811be 97#include <vm/vm_page2.h>
bb6811be 98
906c754c
MD
99#define VMACTION_HSIZE 256
100#define VMACTION_HMASK (VMACTION_HSIZE - 1)
101
de71fd3f
MD
102static void vm_page_queue_init(void);
103static void vm_page_free_wakeup(void);
104static vm_page_t vm_page_select_cache(vm_object_t, vm_pindex_t);
74232d8e 105static vm_page_t _vm_page_list_find2(int basequeue, int index);
984263bc 106
de71fd3f 107struct vpgqueues vm_page_queues[PQ_COUNT]; /* Array of tailq lists */
984263bc 108
906c754c
MD
109LIST_HEAD(vm_page_action_list, vm_page_action);
110struct vm_page_action_list action_list[VMACTION_HSIZE];
cd3c66bd 111static volatile int vm_pages_waiting;
906c754c
MD
112
113
1f804340
MD
114RB_GENERATE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare,
115 vm_pindex_t, pindex);
116
984263bc 117static void
de71fd3f
MD
118vm_page_queue_init(void)
119{
984263bc
MD
120 int i;
121
de71fd3f 122 for (i = 0; i < PQ_L2_SIZE; i++)
12e4aaff 123 vm_page_queues[PQ_FREE+i].cnt = &vmstats.v_free_count;
de71fd3f
MD
124 for (i = 0; i < PQ_L2_SIZE; i++)
125 vm_page_queues[PQ_CACHE+i].cnt = &vmstats.v_cache_count;
984263bc 126
de71fd3f 127 vm_page_queues[PQ_INACTIVE].cnt = &vmstats.v_inactive_count;
12e4aaff
MD
128 vm_page_queues[PQ_ACTIVE].cnt = &vmstats.v_active_count;
129 vm_page_queues[PQ_HOLD].cnt = &vmstats.v_active_count;
de71fd3f
MD
130 /* PQ_NONE has no queue */
131
132 for (i = 0; i < PQ_COUNT; i++)
984263bc 133 TAILQ_INIT(&vm_page_queues[i].pl);
906c754c
MD
134
135 for (i = 0; i < VMACTION_HSIZE; i++)
136 LIST_INIT(&action_list[i]);
984263bc
MD
137}
138
de71fd3f
MD
139/*
140 * note: place in initialized data section? Is this necessary?
141 */
984263bc 142long first_page = 0;
de71fd3f 143int vm_page_array_size = 0;
984263bc 144int vm_page_zero_count = 0;
de71fd3f 145vm_page_t vm_page_array = 0;
984263bc
MD
146
147/*
de71fd3f 148 * (low level boot)
984263bc 149 *
de71fd3f
MD
150 * Sets the page size, perhaps based upon the memory size.
151 * Must be called before any use of page-size dependent functions.
984263bc
MD
152 */
153void
154vm_set_page_size(void)
155{
12e4aaff
MD
156 if (vmstats.v_page_size == 0)
157 vmstats.v_page_size = PAGE_SIZE;
158 if (((vmstats.v_page_size - 1) & vmstats.v_page_size) != 0)
984263bc
MD
159 panic("vm_set_page_size: page size not a power of two");
160}
161
162/*
de71fd3f 163 * (low level boot)
984263bc 164 *
de71fd3f
MD
165 * Add a new page to the freelist for use by the system. New pages
166 * are added to both the head and tail of the associated free page
167 * queue in a bottom-up fashion, so both zero'd and non-zero'd page
168 * requests pull 'recent' adds (higher physical addresses) first.
161399b3 169 *
654a39f0 170 * Must be called in a critical section.
984263bc
MD
171 */
172vm_page_t
6ef943a3 173vm_add_new_page(vm_paddr_t pa)
984263bc 174{
161399b3 175 struct vpgqueues *vpq;
de71fd3f 176 vm_page_t m;
984263bc 177
12e4aaff
MD
178 ++vmstats.v_page_count;
179 ++vmstats.v_free_count;
984263bc
MD
180 m = PHYS_TO_VM_PAGE(pa);
181 m->phys_addr = pa;
182 m->flags = 0;
183 m->pc = (pa >> PAGE_SHIFT) & PQ_L2_MASK;
184 m->queue = m->pc + PQ_FREE;
26bcc0c0 185 KKASSERT(m->dirty == 0);
de71fd3f 186
161399b3
MD
187 vpq = &vm_page_queues[m->queue];
188 if (vpq->flipflop)
189 TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
190 else
191 TAILQ_INSERT_HEAD(&vpq->pl, m, pageq);
192 vpq->flipflop = 1 - vpq->flipflop;
de71fd3f 193
984263bc
MD
194 vm_page_queues[m->queue].lcnt++;
195 return (m);
196}
197
198/*
de71fd3f 199 * (low level boot)
984263bc 200 *
de71fd3f 201 * Initializes the resident memory module.
984263bc 202 *
da23a592
MD
203 * Preallocates memory for critical VM structures and arrays prior to
204 * kernel_map becoming available.
26bcc0c0 205 *
da23a592
MD
206 * Memory is allocated from (virtual2_start, virtual2_end) if available,
207 * otherwise memory is allocated from (virtual_start, virtual_end).
208 *
209 * On x86-64 (virtual_start, virtual_end) is only 2GB and may not be
210 * large enough to hold vm_page_array & other structures for machines with
211 * large amounts of ram, so we want to use virtual2* when available.
984263bc 212 */
da23a592
MD
213void
214vm_page_startup(void)
984263bc 215{
da23a592 216 vm_offset_t vaddr = virtual2_start ? virtual2_start : virtual_start;
984263bc 217 vm_offset_t mapped;
6ef943a3
MD
218 vm_size_t npages;
219 vm_paddr_t page_range;
220 vm_paddr_t new_end;
984263bc 221 int i;
6ef943a3 222 vm_paddr_t pa;
984263bc 223 int nblocks;
6ef943a3 224 vm_paddr_t last_pa;
6ef943a3
MD
225 vm_paddr_t end;
226 vm_paddr_t biggestone, biggestsize;
6ef943a3 227 vm_paddr_t total;
984263bc
MD
228
229 total = 0;
230 biggestsize = 0;
231 biggestone = 0;
232 nblocks = 0;
233 vaddr = round_page(vaddr);
234
235 for (i = 0; phys_avail[i + 1]; i += 2) {
aecf2182
MD
236 phys_avail[i] = round_page64(phys_avail[i]);
237 phys_avail[i + 1] = trunc_page64(phys_avail[i + 1]);
984263bc
MD
238 }
239
240 for (i = 0; phys_avail[i + 1]; i += 2) {
6ef943a3 241 vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
984263bc
MD
242
243 if (size > biggestsize) {
244 biggestone = i;
245 biggestsize = size;
246 }
247 ++nblocks;
248 total += size;
249 }
250
251 end = phys_avail[biggestone+1];
1f804340 252 end = trunc_page(end);
984263bc
MD
253
254 /*
255 * Initialize the queue headers for the free queue, the active queue
256 * and the inactive queue.
257 */
258
259 vm_page_queue_init();
260
6abe3bd0
AH
261 /* VKERNELs don't support minidumps and as such don't need vm_page_dump */
262#if !defined(_KERNEL_VIRTUAL)
8e5e6f1b
AH
263 /*
264 * Allocate a bitmap to indicate that a random physical page
265 * needs to be included in a minidump.
266 *
267 * The amd64 port needs this to indicate which direct map pages
268 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
269 *
270 * However, i386 still needs this workspace internally within the
271 * minidump code. In theory, they are not needed on i386, but are
272 * included should the sf_buf code decide to use them.
273 */
274 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
275 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
276 end -= vm_page_dump_size;
277 vm_page_dump = (void *)pmap_map(&vaddr, end, end + vm_page_dump_size,
278 VM_PROT_READ | VM_PROT_WRITE);
279 bzero((void *)vm_page_dump, vm_page_dump_size);
6abe3bd0 280#endif
8e5e6f1b 281
984263bc
MD
282 /*
283 * Compute the number of pages of memory that will be available for
284 * use (taking into account the overhead of a page structure per
285 * page).
286 */
984263bc 287 first_page = phys_avail[0] / PAGE_SIZE;
984263bc 288 page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE - first_page;
1f804340 289 npages = (total - (page_range * sizeof(struct vm_page))) / PAGE_SIZE;
de71fd3f 290
984263bc
MD
291 /*
292 * Initialize the mem entry structures now, and put them in the free
293 * queue.
294 */
984263bc 295 new_end = trunc_page(end - page_range * sizeof(struct vm_page));
8e5e6f1b 296 mapped = pmap_map(&vaddr, new_end, end,
984263bc 297 VM_PROT_READ | VM_PROT_WRITE);
8e5e6f1b
AH
298 vm_page_array = (vm_page_t)mapped;
299
0e6594a8 300#if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
8e5e6f1b
AH
301 /*
302 * since pmap_map on amd64 returns stuff out of a direct-map region,
303 * we have to manually add these pages to the minidump tracking so
304 * that they can be dumped, including the vm_page_array.
305 */
306 for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
307 dump_add_page(pa);
8fdd3267 308#endif
984263bc
MD
309
310 /*
311 * Clear all of the page structures
312 */
313 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
314 vm_page_array_size = page_range;
315
316 /*
161399b3 317 * Construct the free queue(s) in ascending order (by physical
984263bc
MD
318 * address) so that the first 16MB of physical memory is allocated
319 * last rather than first. On large-memory machines, this avoids
320 * the exhaustion of low physical memory before isa_dmainit has run.
321 */
12e4aaff
MD
322 vmstats.v_page_count = 0;
323 vmstats.v_free_count = 0;
984263bc
MD
324 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) {
325 pa = phys_avail[i];
326 if (i == biggestone)
327 last_pa = new_end;
328 else
329 last_pa = phys_avail[i + 1];
330 while (pa < last_pa && npages-- > 0) {
331 vm_add_new_page(pa);
332 pa += PAGE_SIZE;
333 }
334 }
da23a592
MD
335 if (virtual2_start)
336 virtual2_start = vaddr;
337 else
338 virtual_start = vaddr;
984263bc
MD
339}
340
341/*
1f804340
MD
342 * Scan comparison function for Red-Black tree scans. An inclusive
343 * (start,end) is expected. Other fields are not used.
984263bc 344 */
1f804340
MD
345int
346rb_vm_page_scancmp(struct vm_page *p, void *data)
984263bc 347{
1f804340 348 struct rb_vm_page_scan_info *info = data;
984263bc 349
1f804340
MD
350 if (p->pindex < info->start_pindex)
351 return(-1);
352 if (p->pindex > info->end_pindex)
353 return(1);
354 return(0);
355}
356
357int
358rb_vm_page_compare(struct vm_page *p1, struct vm_page *p2)
359{
360 if (p1->pindex < p2->pindex)
361 return(-1);
362 if (p1->pindex > p2->pindex)
363 return(1);
364 return(0);
984263bc
MD
365}
366
573fb415
MD
367/*
368 * Holding a page keeps it from being reused. Other parts of the system
369 * can still disassociate the page from its current object and free it, or
370 * perform read or write I/O on it and/or otherwise manipulate the page,
371 * but if the page is held the VM system will leave the page and its data
372 * intact and not reuse the page for other purposes until the last hold
373 * reference is released. (see vm_page_wire() if you want to prevent the
374 * page from being disassociated from its object too).
375 *
376 * The caller must hold vm_token.
377 *
378 * The caller must still validate the contents of the page and, if necessary,
379 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
380 * before manipulating the page.
381 */
382void
383vm_page_hold(vm_page_t m)
384{
385 ASSERT_LWKT_TOKEN_HELD(&vm_token);
386 ++m->hold_count;
387}
388
de71fd3f
MD
389/*
390 * The opposite of vm_page_hold(). A page can be freed while being held,
391 * which places it on the PQ_HOLD queue. We must call vm_page_free_toq()
392 * in this case to actually free it once the hold count drops to 0.
393 *
573fb415
MD
394 * The caller must hold vm_token if non-blocking operation is desired,
395 * but otherwise does not need to.
de71fd3f 396 */
984263bc 397void
573fb415 398vm_page_unhold(vm_page_t m)
984263bc 399{
573fb415
MD
400 lwkt_gettoken(&vm_token);
401 --m->hold_count;
402 KASSERT(m->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
403 if (m->hold_count == 0 && m->queue == PQ_HOLD) {
404 vm_page_busy(m);
405 vm_page_free_toq(m);
97edb3b6 406 }
573fb415 407 lwkt_reltoken(&vm_token);
984263bc
MD
408}
409
410/*
573fb415 411 * Inserts the given vm_page into the object and object list.
984263bc 412 *
de71fd3f
MD
413 * The pagetables are not updated but will presumably fault the page
414 * in if necessary, or if a kernel page the caller will at some point
415 * enter the page into the kernel's pmap. We are not allowed to block
416 * here so we *can't* do this anyway.
984263bc 417 *
de71fd3f 418 * This routine may not block.
573fb415 419 * This routine must be called with the vm_token held.
654a39f0 420 * This routine must be called with a critical section held.
984263bc 421 */
984263bc
MD
422void
423vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
424{
573fb415 425 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc
MD
426 if (m->object != NULL)
427 panic("vm_page_insert: already inserted");
428
429 /*
430 * Record the object/offset pair in this page
431 */
984263bc
MD
432 m->object = object;
433 m->pindex = pindex;
434
435 /*
1f804340 436 * Insert it into the object.
984263bc 437 */
1f804340 438 vm_page_rb_tree_RB_INSERT(&object->rb_memq, m);
984263bc
MD
439 object->generation++;
440
441 /*
442 * show that the object has one more resident page.
443 */
984263bc
MD
444 object->resident_page_count++;
445
50a55c46
MD
446 /*
447 * Add the pv_list_cout of the page when its inserted in
448 * the object
449 */
450 object->agg_pv_list_count = object->agg_pv_list_count + m->md.pv_list_count;
451
984263bc
MD
452 /*
453 * Since we are inserting a new and possibly dirty page,
454 * update the object's OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY flags.
455 */
17cde63e 456 if ((m->valid & m->dirty) || (m->flags & PG_WRITEABLE))
984263bc 457 vm_object_set_writeable_dirty(object);
096e95c0
MD
458
459 /*
460 * Checks for a swap assignment and sets PG_SWAPPED if appropriate.
461 */
462 swap_pager_page_inserted(m);
984263bc
MD
463}
464
465/*
9765affa
MD
466 * Removes the given vm_page_t from the global (object,index) hash table
467 * and from the object's memq.
984263bc 468 *
de71fd3f
MD
469 * The underlying pmap entry (if any) is NOT removed here.
470 * This routine may not block.
9765affa 471 *
9ad0147b
MD
472 * The page must be BUSY and will remain BUSY on return.
473 * No other requirements.
9765affa 474 *
9ad0147b
MD
475 * NOTE: FreeBSD side effect was to unbusy the page on return. We leave
476 * it busy.
984263bc 477 */
984263bc
MD
478void
479vm_page_remove(vm_page_t m)
480{
481 vm_object_t object;
482
9ad0147b 483 lwkt_gettoken(&vm_token);
654a39f0 484 if (m->object == NULL) {
9ad0147b 485 lwkt_reltoken(&vm_token);
984263bc 486 return;
654a39f0 487 }
984263bc 488
de71fd3f 489 if ((m->flags & PG_BUSY) == 0)
984263bc 490 panic("vm_page_remove: page not busy");
984263bc 491
984263bc
MD
492 object = m->object;
493
494 /*
1f804340 495 * Remove the page from the object and update the object.
984263bc 496 */
1f804340 497 vm_page_rb_tree_RB_REMOVE(&object->rb_memq, m);
984263bc 498 object->resident_page_count--;
50a55c46 499 object->agg_pv_list_count = object->agg_pv_list_count - m->md.pv_list_count;
984263bc 500 object->generation++;
984263bc 501 m->object = NULL;
1f804340 502
9ad0147b 503 lwkt_reltoken(&vm_token);
984263bc
MD
504}
505
506/*
de71fd3f
MD
507 * Locate and return the page at (object, pindex), or NULL if the
508 * page could not be found.
509 *
77912481 510 * The caller must hold vm_token.
984263bc 511 */
984263bc
MD
512vm_page_t
513vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
514{
515 vm_page_t m;
984263bc
MD
516
517 /*
518 * Search the hash table for this object/offset pair
519 */
77912481 520 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1f804340 521 m = vm_page_rb_tree_RB_LOOKUP(&object->rb_memq, pindex);
1f804340
MD
522 KKASSERT(m == NULL || (m->object == object && m->pindex == pindex));
523 return(m);
984263bc
MD
524}
525
526/*
de71fd3f 527 * vm_page_rename()
984263bc 528 *
de71fd3f
MD
529 * Move the given memory entry from its current object to the specified
530 * target object/offset.
984263bc 531 *
de71fd3f
MD
532 * The object must be locked.
533 * This routine may not block.
984263bc 534 *
de71fd3f 535 * Note: This routine will raise itself to splvm(), the caller need not.
984263bc 536 *
de71fd3f
MD
537 * Note: Swap associated with the page must be invalidated by the move. We
538 * have to do this for several reasons: (1) we aren't freeing the
539 * page, (2) we are dirtying the page, (3) the VM system is probably
540 * moving the page from object A to B, and will then later move
541 * the backing store from A to B and we can't have a conflict.
984263bc 542 *
de71fd3f
MD
543 * Note: We *always* dirty the page. It is necessary both for the
544 * fact that we moved it, and because we may be invalidating
545 * swap. If the page is on the cache, we have to deactivate it
546 * or vm_page_dirty() will panic. Dirty pages are not allowed
547 * on the cache.
984263bc 548 */
984263bc
MD
549void
550vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
551{
9ad0147b 552 lwkt_gettoken(&vm_token);
984263bc
MD
553 vm_page_remove(m);
554 vm_page_insert(m, new_object, new_pindex);
555 if (m->queue - m->pc == PQ_CACHE)
556 vm_page_deactivate(m);
557 vm_page_dirty(m);
9765affa 558 vm_page_wakeup(m);
9ad0147b 559 lwkt_reltoken(&vm_token);
984263bc
MD
560}
561
562/*
de71fd3f
MD
563 * vm_page_unqueue() without any wakeup. This routine is used when a page
564 * is being moved between queues or otherwise is to remain BUSYied by the
565 * caller.
984263bc 566 *
573fb415 567 * The caller must hold vm_token
de71fd3f 568 * This routine may not block.
984263bc 569 */
984263bc
MD
570void
571vm_page_unqueue_nowakeup(vm_page_t m)
572{
573 int queue = m->queue;
574 struct vpgqueues *pq;
de71fd3f 575
573fb415 576 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc
MD
577 if (queue != PQ_NONE) {
578 pq = &vm_page_queues[queue];
579 m->queue = PQ_NONE;
580 TAILQ_REMOVE(&pq->pl, m, pageq);
581 (*pq->cnt)--;
582 pq->lcnt--;
583 }
584}
585
586/*
de71fd3f
MD
587 * vm_page_unqueue() - Remove a page from its queue, wakeup the pagedemon
588 * if necessary.
984263bc 589 *
573fb415 590 * The caller must hold vm_token
de71fd3f 591 * This routine may not block.
984263bc 592 */
984263bc
MD
593void
594vm_page_unqueue(vm_page_t m)
595{
596 int queue = m->queue;
597 struct vpgqueues *pq;
de71fd3f 598
573fb415 599 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc
MD
600 if (queue != PQ_NONE) {
601 m->queue = PQ_NONE;
602 pq = &vm_page_queues[queue];
603 TAILQ_REMOVE(&pq->pl, m, pageq);
604 (*pq->cnt)--;
605 pq->lcnt--;
20479584
MD
606 if ((queue - m->pc) == PQ_CACHE || (queue - m->pc) == PQ_FREE)
607 pagedaemon_wakeup();
984263bc
MD
608 }
609}
610
984263bc 611/*
de71fd3f 612 * vm_page_list_find()
984263bc 613 *
de71fd3f 614 * Find a page on the specified queue with color optimization.
984263bc 615 *
de71fd3f
MD
616 * The page coloring optimization attempts to locate a page that does
617 * not overload other nearby pages in the object in the cpu's L1 or L2
618 * caches. We need this optimization because cpu caches tend to be
619 * physical caches, while object spaces tend to be virtual.
984263bc 620 *
573fb415 621 * Must be called with vm_token held.
de71fd3f 622 * This routine may not block.
984263bc 623 *
de71fd3f
MD
624 * Note that this routine is carefully inlined. A non-inlined version
625 * is available for outside callers but the only critical path is
626 * from within this source file.
984263bc 627 */
74232d8e 628static __inline
984263bc 629vm_page_t
74232d8e
MD
630_vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
631{
632 vm_page_t m;
633
634 if (prefer_zero)
635 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
636 else
637 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
638 if (m == NULL)
639 m = _vm_page_list_find2(basequeue, index);
640 return(m);
641}
642
643static vm_page_t
644_vm_page_list_find2(int basequeue, int index)
984263bc
MD
645{
646 int i;
647 vm_page_t m = NULL;
648 struct vpgqueues *pq;
649
650 pq = &vm_page_queues[basequeue];
651
652 /*
653 * Note that for the first loop, index+i and index-i wind up at the
654 * same place. Even though this is not totally optimal, we've already
655 * blown it by missing the cache case so we do not care.
656 */
657
658 for(i = PQ_L2_SIZE / 2; i > 0; --i) {
659 if ((m = TAILQ_FIRST(&pq[(index + i) & PQ_L2_MASK].pl)) != NULL)
660 break;
661
662 if ((m = TAILQ_FIRST(&pq[(index - i) & PQ_L2_MASK].pl)) != NULL)
663 break;
664 }
665 return(m);
666}
667
573fb415
MD
668/*
669 * Must be called with vm_token held if the caller desired non-blocking
670 * operation and a stable result.
671 */
74232d8e
MD
672vm_page_t
673vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
674{
675 return(_vm_page_list_find(basequeue, index, prefer_zero));
676}
677
984263bc 678/*
de71fd3f
MD
679 * Find a page on the cache queue with color optimization. As pages
680 * might be found, but not applicable, they are deactivated. This
681 * keeps us from using potentially busy cached pages.
984263bc 682 *
de71fd3f 683 * This routine may not block.
573fb415 684 * Must be called with vm_token held.
984263bc
MD
685 */
686vm_page_t
687vm_page_select_cache(vm_object_t object, vm_pindex_t pindex)
688{
689 vm_page_t m;
690
573fb415 691 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc 692 while (TRUE) {
659c6a07 693 m = _vm_page_list_find(
984263bc
MD
694 PQ_CACHE,
695 (pindex + object->pg_color) & PQ_L2_MASK,
696 FALSE
697 );
698 if (m && ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
699 m->hold_count || m->wire_count)) {
a491077e 700 /* cache page found busy */
984263bc 701 vm_page_deactivate(m);
a491077e
MD
702#ifdef INVARIANTS
703 kprintf("Warning: busy page %p found in cache\n", m);
704#endif
984263bc
MD
705 continue;
706 }
707 return m;
708 }
de71fd3f 709 /* not reached */
984263bc
MD
710}
711
712/*
de71fd3f
MD
713 * Find a free or zero page, with specified preference. We attempt to
714 * inline the nominal case and fall back to _vm_page_select_free()
715 * otherwise.
984263bc 716 *
654a39f0 717 * This routine must be called with a critical section held.
de71fd3f 718 * This routine may not block.
984263bc 719 */
984263bc
MD
720static __inline vm_page_t
721vm_page_select_free(vm_object_t object, vm_pindex_t pindex, boolean_t prefer_zero)
722{
723 vm_page_t m;
724
659c6a07 725 m = _vm_page_list_find(
984263bc
MD
726 PQ_FREE,
727 (pindex + object->pg_color) & PQ_L2_MASK,
728 prefer_zero
729 );
730 return(m);
731}
732
733/*
de71fd3f 734 * vm_page_alloc()
984263bc 735 *
de71fd3f
MD
736 * Allocate and return a memory cell associated with this VM object/offset
737 * pair.
984263bc
MD
738 *
739 * page_req classes:
de71fd3f 740 *
dc1fd4b3 741 * VM_ALLOC_NORMAL allow use of cache pages, nominal free drain
39208dbe 742 * VM_ALLOC_QUICK like normal but cannot use cache
dc1fd4b3
MD
743 * VM_ALLOC_SYSTEM greater free drain
744 * VM_ALLOC_INTERRUPT allow free list to be completely drained
745 * VM_ALLOC_ZERO advisory request for pre-zero'd page
984263bc 746 *
de71fd3f
MD
747 * The object must be locked.
748 * This routine may not block.
9765affa 749 * The returned page will be marked PG_BUSY
984263bc 750 *
de71fd3f
MD
751 * Additional special handling is required when called from an interrupt
752 * (VM_ALLOC_INTERRUPT). We are not allowed to mess with the page cache
753 * in this case.
984263bc 754 */
984263bc
MD
755vm_page_t
756vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int page_req)
757{
758 vm_page_t m = NULL;
984263bc 759
ba9d3e52
AH
760 lwkt_gettoken(&vm_token);
761
cfd17028 762 KKASSERT(object != NULL);
984263bc
MD
763 KASSERT(!vm_page_lookup(object, pindex),
764 ("vm_page_alloc: page already allocated"));
dc1fd4b3 765 KKASSERT(page_req &
39208dbe
MD
766 (VM_ALLOC_NORMAL|VM_ALLOC_QUICK|
767 VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
984263bc
MD
768
769 /*
4ecf7cc9
MD
770 * Certain system threads (pageout daemon, buf_daemon's) are
771 * allowed to eat deeper into the free page list.
984263bc 772 */
4ecf7cc9 773 if (curthread->td_flags & TDF_SYSTHREAD)
dc1fd4b3 774 page_req |= VM_ALLOC_SYSTEM;
984263bc 775
984263bc 776loop:
dc1fd4b3
MD
777 if (vmstats.v_free_count > vmstats.v_free_reserved ||
778 ((page_req & VM_ALLOC_INTERRUPT) && vmstats.v_free_count > 0) ||
779 ((page_req & VM_ALLOC_SYSTEM) && vmstats.v_cache_count == 0 &&
780 vmstats.v_free_count > vmstats.v_interrupt_free_min)
781 ) {
984263bc 782 /*
dc1fd4b3 783 * The free queue has sufficient free pages to take one out.
984263bc 784 */
dc1fd4b3 785 if (page_req & VM_ALLOC_ZERO)
984263bc
MD
786 m = vm_page_select_free(object, pindex, TRUE);
787 else
788 m = vm_page_select_free(object, pindex, FALSE);
dc1fd4b3 789 } else if (page_req & VM_ALLOC_NORMAL) {
984263bc 790 /*
dc1fd4b3
MD
791 * Allocatable from the cache (non-interrupt only). On
792 * success, we must free the page and try again, thus
793 * ensuring that vmstats.v_*_free_min counters are replenished.
984263bc 794 */
dc1fd4b3
MD
795#ifdef INVARIANTS
796 if (curthread->td_preempted) {
086c1d7e 797 kprintf("vm_page_alloc(): warning, attempt to allocate"
dc1fd4b3
MD
798 " cache page from preempting interrupt\n");
799 m = NULL;
800 } else {
801 m = vm_page_select_cache(object, pindex);
802 }
803#else
804 m = vm_page_select_cache(object, pindex);
805#endif
984263bc 806 /*
9765affa 807 * On success move the page into the free queue and loop.
984263bc 808 */
dc1fd4b3
MD
809 if (m != NULL) {
810 KASSERT(m->dirty == 0,
811 ("Found dirty cache page %p", m));
812 vm_page_busy(m);
813 vm_page_protect(m, VM_PROT_NONE);
814 vm_page_free(m);
815 goto loop;
816 }
817
818 /*
819 * On failure return NULL
820 */
9ad0147b 821 lwkt_reltoken(&vm_token);
984263bc 822#if defined(DIAGNOSTIC)
dc1fd4b3 823 if (vmstats.v_cache_count > 0)
086c1d7e 824 kprintf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", vmstats.v_cache_count);
984263bc 825#endif
dc1fd4b3
MD
826 vm_pageout_deficit++;
827 pagedaemon_wakeup();
828 return (NULL);
984263bc
MD
829 } else {
830 /*
dc1fd4b3 831 * No pages available, wakeup the pageout daemon and give up.
984263bc 832 */
9ad0147b 833 lwkt_reltoken(&vm_token);
984263bc
MD
834 vm_pageout_deficit++;
835 pagedaemon_wakeup();
836 return (NULL);
837 }
838
839 /*
9765affa
MD
840 * Good page found. The page has not yet been busied. We are in
841 * a critical section.
984263bc 842 */
dc1fd4b3 843 KASSERT(m != NULL, ("vm_page_alloc(): missing page on free queue\n"));
26bcc0c0
MD
844 KASSERT(m->dirty == 0,
845 ("vm_page_alloc: free/cache page %p was dirty", m));
984263bc
MD
846
847 /*
848 * Remove from free queue
849 */
984263bc
MD
850 vm_page_unqueue_nowakeup(m);
851
852 /*
9765affa
MD
853 * Initialize structure. Only the PG_ZERO flag is inherited. Set
854 * the page PG_BUSY
984263bc 855 */
984263bc
MD
856 if (m->flags & PG_ZERO) {
857 vm_page_zero_count--;
858 m->flags = PG_ZERO | PG_BUSY;
859 } else {
860 m->flags = PG_BUSY;
861 }
862 m->wire_count = 0;
863 m->hold_count = 0;
864 m->act_count = 0;
865 m->busy = 0;
866 m->valid = 0;
984263bc
MD
867
868 /*
080c00e6 869 * vm_page_insert() is safe while holding vm_token. Note also that
984263bc
MD
870 * inserting a page here does not insert it into the pmap (which
871 * could cause us to block allocating memory). We cannot block
872 * anywhere.
873 */
984263bc
MD
874 vm_page_insert(m, object, pindex);
875
876 /*
877 * Don't wakeup too often - wakeup the pageout daemon when
878 * we would be nearly out of memory.
879 */
20479584 880 pagedaemon_wakeup();
984263bc 881
9ad0147b 882 lwkt_reltoken(&vm_token);
9765affa
MD
883
884 /*
885 * A PG_BUSY page is returned.
886 */
984263bc
MD
887 return (m);
888}
889
163f8d24
MD
890/*
891 * Wait for sufficient free memory for nominal heavy memory use kernel
892 * operations.
893 */
894void
895vm_wait_nominal(void)
896{
897 while (vm_page_count_min(0))
898 vm_wait(0);
899}
900
12052253
MD
901/*
902 * Test if vm_wait_nominal() would block.
903 */
904int
905vm_test_nominal(void)
906{
907 if (vm_page_count_min(0))
908 return(1);
909 return(0);
910}
911
984263bc 912/*
de71fd3f
MD
913 * Block until free pages are available for allocation, called in various
914 * places before memory allocations.
cd3c66bd
MD
915 *
916 * The caller may loop if vm_page_count_min() == FALSE so we cannot be
917 * more generous then that.
984263bc 918 */
984263bc 919void
4ecf7cc9 920vm_wait(int timo)
984263bc 921{
cd3c66bd
MD
922 /*
923 * never wait forever
924 */
925 if (timo == 0)
926 timo = hz;
9ad0147b 927 lwkt_gettoken(&vm_token);
cd3c66bd 928
bc6dffab 929 if (curthread == pagethread) {
cd3c66bd
MD
930 /*
931 * The pageout daemon itself needs pages, this is bad.
932 */
933 if (vm_page_count_min(0)) {
934 vm_pageout_pages_needed = 1;
935 tsleep(&vm_pageout_pages_needed, 0, "VMWait", timo);
936 }
984263bc 937 } else {
cd3c66bd
MD
938 /*
939 * Wakeup the pageout daemon if necessary and wait.
940 */
941 if (vm_page_count_target()) {
942 if (vm_pages_needed == 0) {
943 vm_pages_needed = 1;
944 wakeup(&vm_pages_needed);
945 }
946 ++vm_pages_waiting; /* SMP race ok */
947 tsleep(&vmstats.v_free_count, 0, "vmwait", timo);
984263bc 948 }
984263bc 949 }
9ad0147b 950 lwkt_reltoken(&vm_token);
984263bc
MD
951}
952
953/*
de71fd3f
MD
954 * Block until free pages are available for allocation
955 *
cd3c66bd 956 * Called only from vm_fault so that processes page faulting can be
de71fd3f 957 * easily tracked.
984263bc 958 */
984263bc
MD
959void
960vm_waitpfault(void)
961{
cd3c66bd
MD
962 /*
963 * Wakeup the pageout daemon if necessary and wait.
964 */
965 if (vm_page_count_target()) {
966 lwkt_gettoken(&vm_token);
967 if (vm_page_count_target()) {
968 if (vm_pages_needed == 0) {
969 vm_pages_needed = 1;
970 wakeup(&vm_pages_needed);
971 }
972 ++vm_pages_waiting; /* SMP race ok */
973 tsleep(&vmstats.v_free_count, 0, "pfault", hz);
974 }
975 lwkt_reltoken(&vm_token);
984263bc 976 }
984263bc
MD
977}
978
984263bc 979/*
de71fd3f
MD
980 * Put the specified page on the active list (if appropriate). Ensure
981 * that act_count is at least ACT_INIT but do not otherwise mess with it.
984263bc 982 *
de71fd3f
MD
983 * The page queues must be locked.
984 * This routine may not block.
984263bc
MD
985 */
986void
987vm_page_activate(vm_page_t m)
988{
9ad0147b 989 lwkt_gettoken(&vm_token);
984263bc
MD
990 if (m->queue != PQ_ACTIVE) {
991 if ((m->queue - m->pc) == PQ_CACHE)
12e4aaff 992 mycpu->gd_cnt.v_reactivated++;
984263bc
MD
993
994 vm_page_unqueue(m);
995
996 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
997 m->queue = PQ_ACTIVE;
998 vm_page_queues[PQ_ACTIVE].lcnt++;
de71fd3f
MD
999 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_ACTIVE].pl,
1000 m, pageq);
984263bc
MD
1001 if (m->act_count < ACT_INIT)
1002 m->act_count = ACT_INIT;
12e4aaff 1003 vmstats.v_active_count++;
984263bc
MD
1004 }
1005 } else {
1006 if (m->act_count < ACT_INIT)
1007 m->act_count = ACT_INIT;
1008 }
9ad0147b 1009 lwkt_reltoken(&vm_token);
984263bc
MD
1010}
1011
1012/*
de71fd3f
MD
1013 * Helper routine for vm_page_free_toq() and vm_page_cache(). This
1014 * routine is called when a page has been added to the cache or free
1015 * queues.
984263bc 1016 *
de71fd3f
MD
1017 * This routine may not block.
1018 * This routine must be called at splvm()
984263bc
MD
1019 */
1020static __inline void
1021vm_page_free_wakeup(void)
1022{
1023 /*
cd3c66bd
MD
1024 * If the pageout daemon itself needs pages, then tell it that
1025 * there are some free.
984263bc
MD
1026 */
1027 if (vm_pageout_pages_needed &&
de71fd3f
MD
1028 vmstats.v_cache_count + vmstats.v_free_count >=
1029 vmstats.v_pageout_free_min
1030 ) {
984263bc
MD
1031 wakeup(&vm_pageout_pages_needed);
1032 vm_pageout_pages_needed = 0;
1033 }
de71fd3f 1034
984263bc 1035 /*
cd3c66bd
MD
1036 * Wakeup processes that are waiting on memory.
1037 *
1038 * NOTE: vm_paging_target() is the pageout daemon's target, while
1039 * vm_page_count_target() is somewhere inbetween. We want
1040 * to wake processes up prior to the pageout daemon reaching
1041 * its target to provide some hysteresis.
984263bc 1042 */
cd3c66bd
MD
1043 if (vm_pages_waiting) {
1044 if (!vm_page_count_target()) {
1045 /*
1046 * Plenty of pages are free, wakeup everyone.
1047 */
1048 vm_pages_waiting = 0;
1049 wakeup(&vmstats.v_free_count);
1050 ++mycpu->gd_cnt.v_ppwakeups;
1051 } else if (!vm_page_count_min(0)) {
1052 /*
1053 * Some pages are free, wakeup someone.
1054 */
1055 int wcount = vm_pages_waiting;
1056 if (wcount > 0)
1057 --wcount;
1058 vm_pages_waiting = wcount;
1059 wakeup_one(&vmstats.v_free_count);
1060 ++mycpu->gd_cnt.v_ppwakeups;
1061 }
984263bc
MD
1062 }
1063}
1064
1065/*
1066 * vm_page_free_toq:
1067 *
9765affa
MD
1068 * Returns the given page to the PQ_FREE list, disassociating it with
1069 * any VM object.
1070 *
1071 * The vm_page must be PG_BUSY on entry. PG_BUSY will be released on
1072 * return (the page will have been freed). No particular spl is required
1073 * on entry.
984263bc 1074 *
984263bc
MD
1075 * This routine may not block.
1076 */
984263bc
MD
1077void
1078vm_page_free_toq(vm_page_t m)
1079{
984263bc 1080 struct vpgqueues *pq;
984263bc 1081
9ad0147b 1082 lwkt_gettoken(&vm_token);
12e4aaff 1083 mycpu->gd_cnt.v_tfree++;
984263bc 1084
17cde63e
MD
1085 KKASSERT((m->flags & PG_MAPPED) == 0);
1086
984263bc 1087 if (m->busy || ((m->queue - m->pc) == PQ_FREE)) {
086c1d7e 1088 kprintf(
984263bc
MD
1089 "vm_page_free: pindex(%lu), busy(%d), PG_BUSY(%d), hold(%d)\n",
1090 (u_long)m->pindex, m->busy, (m->flags & PG_BUSY) ? 1 : 0,
1091 m->hold_count);
1092 if ((m->queue - m->pc) == PQ_FREE)
1093 panic("vm_page_free: freeing free page");
1094 else
1095 panic("vm_page_free: freeing busy page");
1096 }
1097
1098 /*
1099 * unqueue, then remove page. Note that we cannot destroy
1100 * the page here because we do not want to call the pager's
1101 * callback routine until after we've put the page on the
1102 * appropriate free queue.
1103 */
984263bc
MD
1104 vm_page_unqueue_nowakeup(m);
1105 vm_page_remove(m);
1106
1107 /*
f2d22ebf
MD
1108 * No further management of fictitious pages occurs beyond object
1109 * and queue removal.
984263bc 1110 */
984263bc 1111 if ((m->flags & PG_FICTITIOUS) != 0) {
9765affa 1112 vm_page_wakeup(m);
9ad0147b 1113 lwkt_reltoken(&vm_token);
984263bc
MD
1114 return;
1115 }
1116
1117 m->valid = 0;
1118 vm_page_undirty(m);
1119
1120 if (m->wire_count != 0) {
1121 if (m->wire_count > 1) {
de71fd3f
MD
1122 panic(
1123 "vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1124 m->wire_count, (long)m->pindex);
984263bc 1125 }
73c351d1 1126 panic("vm_page_free: freeing wired page");
984263bc
MD
1127 }
1128
984263bc
MD
1129 /*
1130 * Clear the UNMANAGED flag when freeing an unmanaged page.
1131 */
984263bc 1132 if (m->flags & PG_UNMANAGED) {
d0aa00e8 1133 vm_page_flag_clear(m, PG_UNMANAGED);
984263bc
MD
1134 }
1135
1136 if (m->hold_count != 0) {
d0aa00e8 1137 vm_page_flag_clear(m, PG_ZERO);
984263bc 1138 m->queue = PQ_HOLD;
de71fd3f 1139 } else {
984263bc 1140 m->queue = PQ_FREE + m->pc;
de71fd3f 1141 }
984263bc
MD
1142 pq = &vm_page_queues[m->queue];
1143 pq->lcnt++;
1144 ++(*pq->cnt);
1145
1146 /*
1147 * Put zero'd pages on the end ( where we look for zero'd pages
1148 * first ) and non-zerod pages at the head.
1149 */
984263bc
MD
1150 if (m->flags & PG_ZERO) {
1151 TAILQ_INSERT_TAIL(&pq->pl, m, pageq);
1152 ++vm_page_zero_count;
1153 } else {
1154 TAILQ_INSERT_HEAD(&pq->pl, m, pageq);
1155 }
9765affa 1156 vm_page_wakeup(m);
984263bc 1157 vm_page_free_wakeup();
9ad0147b 1158 lwkt_reltoken(&vm_token);
984263bc
MD
1159}
1160
bb6811be
MD
1161/*
1162 * vm_page_free_fromq_fast()
1163 *
1164 * Remove a non-zero page from one of the free queues; the page is removed for
1165 * zeroing, so do not issue a wakeup.
1166 *
1167 * MPUNSAFE
1168 */
1169vm_page_t
1170vm_page_free_fromq_fast(void)
1171{
1172 static int qi;
1173 vm_page_t m;
1174 int i;
1175
9ad0147b 1176 lwkt_gettoken(&vm_token);
bb6811be
MD
1177 for (i = 0; i < PQ_L2_SIZE; ++i) {
1178 m = vm_page_list_find(PQ_FREE, qi, FALSE);
1179 qi = (qi + PQ_PRIME2) & PQ_L2_MASK;
1180 if (m && (m->flags & PG_ZERO) == 0) {
080c00e6 1181 KKASSERT(m->busy == 0 && (m->flags & PG_BUSY) == 0);
bb6811be
MD
1182 vm_page_unqueue_nowakeup(m);
1183 vm_page_busy(m);
1184 break;
1185 }
1186 m = NULL;
1187 }
9ad0147b 1188 lwkt_reltoken(&vm_token);
bb6811be
MD
1189 return (m);
1190}
1191
984263bc 1192/*
de71fd3f
MD
1193 * vm_page_unmanage()
1194 *
1195 * Prevent PV management from being done on the page. The page is
1196 * removed from the paging queues as if it were wired, and as a
1197 * consequence of no longer being managed the pageout daemon will not
1198 * touch it (since there is no way to locate the pte mappings for the
1199 * page). madvise() calls that mess with the pmap will also no longer
1200 * operate on the page.
1201 *
1202 * Beyond that the page is still reasonably 'normal'. Freeing the page
1203 * will clear the flag.
1204 *
1205 * This routine is used by OBJT_PHYS objects - objects using unswappable
1206 * physical memory as backing store rather then swap-backed memory and
1207 * will eventually be extended to support 4MB unmanaged physical
1208 * mappings.
654a39f0
MD
1209 *
1210 * Must be called with a critical section held.
573fb415 1211 * Must be called with vm_token held.
984263bc 1212 */
984263bc
MD
1213void
1214vm_page_unmanage(vm_page_t m)
1215{
573fb415 1216 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc
MD
1217 if ((m->flags & PG_UNMANAGED) == 0) {
1218 if (m->wire_count == 0)
1219 vm_page_unqueue(m);
1220 }
1221 vm_page_flag_set(m, PG_UNMANAGED);
984263bc
MD
1222}
1223
1224/*
de71fd3f
MD
1225 * Mark this page as wired down by yet another map, removing it from
1226 * paging queues as necessary.
984263bc 1227 *
de71fd3f
MD
1228 * The page queues must be locked.
1229 * This routine may not block.
984263bc
MD
1230 */
1231void
1232vm_page_wire(vm_page_t m)
1233{
984263bc
MD
1234 /*
1235 * Only bump the wire statistics if the page is not already wired,
1236 * and only unqueue the page if it is on some queue (if it is unmanaged
f2d22ebf
MD
1237 * it is already off the queues). Don't do anything with fictitious
1238 * pages because they are always wired.
984263bc 1239 */
9ad0147b 1240 lwkt_gettoken(&vm_token);
f2d22ebf
MD
1241 if ((m->flags & PG_FICTITIOUS) == 0) {
1242 if (m->wire_count == 0) {
1243 if ((m->flags & PG_UNMANAGED) == 0)
1244 vm_page_unqueue(m);
1245 vmstats.v_wire_count++;
1246 }
1247 m->wire_count++;
1248 KASSERT(m->wire_count != 0,
17cde63e 1249 ("vm_page_wire: wire_count overflow m=%p", m));
984263bc 1250 }
9ad0147b 1251 lwkt_reltoken(&vm_token);
984263bc
MD
1252}
1253
1254/*
de71fd3f
MD
1255 * Release one wiring of this page, potentially enabling it to be paged again.
1256 *
1257 * Many pages placed on the inactive queue should actually go
1258 * into the cache, but it is difficult to figure out which. What
1259 * we do instead, if the inactive target is well met, is to put
1260 * clean pages at the head of the inactive queue instead of the tail.
1261 * This will cause them to be moved to the cache more quickly and
1262 * if not actively re-referenced, freed more quickly. If we just
1263 * stick these pages at the end of the inactive queue, heavy filesystem
1264 * meta-data accesses can cause an unnecessary paging load on memory bound
1265 * processes. This optimization causes one-time-use metadata to be
1266 * reused more quickly.
1267 *
1268 * BUT, if we are in a low-memory situation we have no choice but to
1269 * put clean pages on the cache queue.
1270 *
1271 * A number of routines use vm_page_unwire() to guarantee that the page
1272 * will go into either the inactive or active queues, and will NEVER
1273 * be placed in the cache - for example, just after dirtying a page.
1274 * dirty pages in the cache are not allowed.
1275 *
1276 * The page queues must be locked.
1277 * This routine may not block.
984263bc
MD
1278 */
1279void
1280vm_page_unwire(vm_page_t m, int activate)
1281{
9ad0147b 1282 lwkt_gettoken(&vm_token);
f2d22ebf
MD
1283 if (m->flags & PG_FICTITIOUS) {
1284 /* do nothing */
1285 } else if (m->wire_count <= 0) {
1286 panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
1287 } else {
1288 if (--m->wire_count == 0) {
1289 --vmstats.v_wire_count;
984263bc
MD
1290 if (m->flags & PG_UNMANAGED) {
1291 ;
1292 } else if (activate) {
f2d22ebf
MD
1293 TAILQ_INSERT_TAIL(
1294 &vm_page_queues[PQ_ACTIVE].pl, m, pageq);
984263bc
MD
1295 m->queue = PQ_ACTIVE;
1296 vm_page_queues[PQ_ACTIVE].lcnt++;
12e4aaff 1297 vmstats.v_active_count++;
984263bc
MD
1298 } else {
1299 vm_page_flag_clear(m, PG_WINATCFLS);
f2d22ebf
MD
1300 TAILQ_INSERT_TAIL(
1301 &vm_page_queues[PQ_INACTIVE].pl, m, pageq);
984263bc
MD
1302 m->queue = PQ_INACTIVE;
1303 vm_page_queues[PQ_INACTIVE].lcnt++;
12e4aaff 1304 vmstats.v_inactive_count++;
e527fb6b 1305 ++vm_swapcache_inactive_heuristic;
984263bc
MD
1306 }
1307 }
984263bc 1308 }
9ad0147b 1309 lwkt_reltoken(&vm_token);
984263bc
MD
1310}
1311
1312
1313/*
1314 * Move the specified page to the inactive queue. If the page has
1315 * any associated swap, the swap is deallocated.
1316 *
1317 * Normally athead is 0 resulting in LRU operation. athead is set
1318 * to 1 if we want this page to be 'as if it were placed in the cache',
1319 * except without unmapping it from the process address space.
1320 *
1321 * This routine may not block.
573fb415 1322 * The caller must hold vm_token.
984263bc
MD
1323 */
1324static __inline void
1325_vm_page_deactivate(vm_page_t m, int athead)
1326{
984263bc
MD
1327 /*
1328 * Ignore if already inactive.
1329 */
1330 if (m->queue == PQ_INACTIVE)
1331 return;
1332
984263bc
MD
1333 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1334 if ((m->queue - m->pc) == PQ_CACHE)
12e4aaff 1335 mycpu->gd_cnt.v_reactivated++;
984263bc
MD
1336 vm_page_flag_clear(m, PG_WINATCFLS);
1337 vm_page_unqueue(m);
e527fb6b
MD
1338 if (athead) {
1339 TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl,
1340 m, pageq);
1341 } else {
1342 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl,
1343 m, pageq);
1344 ++vm_swapcache_inactive_heuristic;
1345 }
984263bc
MD
1346 m->queue = PQ_INACTIVE;
1347 vm_page_queues[PQ_INACTIVE].lcnt++;
12e4aaff 1348 vmstats.v_inactive_count++;
984263bc 1349 }
984263bc
MD
1350}
1351
573fb415
MD
1352/*
1353 * Attempt to deactivate a page.
1354 *
1355 * No requirements.
1356 */
984263bc
MD
1357void
1358vm_page_deactivate(vm_page_t m)
1359{
573fb415
MD
1360 lwkt_gettoken(&vm_token);
1361 _vm_page_deactivate(m, 0);
1362 lwkt_reltoken(&vm_token);
984263bc
MD
1363}
1364
1365/*
573fb415 1366 * Attempt to move a page to PQ_CACHE.
984263bc 1367 * Returns 0 on failure, 1 on success
573fb415
MD
1368 *
1369 * No requirements.
984263bc
MD
1370 */
1371int
1372vm_page_try_to_cache(vm_page_t m)
1373{
9ad0147b 1374 lwkt_gettoken(&vm_token);
984263bc
MD
1375 if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1376 (m->flags & (PG_BUSY|PG_UNMANAGED))) {
9ad0147b 1377 lwkt_reltoken(&vm_token);
984263bc
MD
1378 return(0);
1379 }
a491077e 1380 vm_page_busy(m);
984263bc 1381 vm_page_test_dirty(m);
654a39f0 1382 if (m->dirty) {
50e32333 1383 vm_page_wakeup(m);
9ad0147b 1384 lwkt_reltoken(&vm_token);
984263bc 1385 return(0);
654a39f0 1386 }
984263bc 1387 vm_page_cache(m);
9ad0147b 1388 lwkt_reltoken(&vm_token);
984263bc
MD
1389 return(1);
1390}
1391
1392/*
de71fd3f
MD
1393 * Attempt to free the page. If we cannot free it, we do nothing.
1394 * 1 is returned on success, 0 on failure.
573fb415
MD
1395 *
1396 * No requirements.
984263bc 1397 */
984263bc
MD
1398int
1399vm_page_try_to_free(vm_page_t m)
1400{
9ad0147b 1401 lwkt_gettoken(&vm_token);
984263bc
MD
1402 if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1403 (m->flags & (PG_BUSY|PG_UNMANAGED))) {
9ad0147b 1404 lwkt_reltoken(&vm_token);
984263bc
MD
1405 return(0);
1406 }
1407 vm_page_test_dirty(m);
654a39f0 1408 if (m->dirty) {
9ad0147b 1409 lwkt_reltoken(&vm_token);
984263bc 1410 return(0);
654a39f0 1411 }
984263bc
MD
1412 vm_page_busy(m);
1413 vm_page_protect(m, VM_PROT_NONE);
1414 vm_page_free(m);
9ad0147b 1415 lwkt_reltoken(&vm_token);
984263bc
MD
1416 return(1);
1417}
1418
984263bc
MD
1419/*
1420 * vm_page_cache
1421 *
1422 * Put the specified page onto the page cache queue (if appropriate).
1423 *
573fb415 1424 * The caller must hold vm_token.
984263bc 1425 * This routine may not block.
a491077e
MD
1426 * The page must be busy, and this routine will release the busy and
1427 * possibly even free the page.
984263bc
MD
1428 */
1429void
1430vm_page_cache(vm_page_t m)
1431{
573fb415 1432 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc 1433
a491077e
MD
1434 if ((m->flags & PG_UNMANAGED) || m->busy ||
1435 m->wire_count || m->hold_count) {
086c1d7e 1436 kprintf("vm_page_cache: attempting to cache busy/held page\n");
a491077e 1437 vm_page_wakeup(m);
984263bc
MD
1438 return;
1439 }
c9ec86b3
MD
1440
1441 /*
1442 * Already in the cache (and thus not mapped)
1443 */
17cde63e
MD
1444 if ((m->queue - m->pc) == PQ_CACHE) {
1445 KKASSERT((m->flags & PG_MAPPED) == 0);
a491077e 1446 vm_page_wakeup(m);
984263bc 1447 return;
17cde63e 1448 }
984263bc
MD
1449
1450 /*
c9ec86b3
MD
1451 * Caller is required to test m->dirty, but note that the act of
1452 * removing the page from its maps can cause it to become dirty
1453 * on an SMP system due to another cpu running in usermode.
984263bc 1454 */
c9ec86b3 1455 if (m->dirty) {
984263bc
MD
1456 panic("vm_page_cache: caching a dirty page, pindex: %ld",
1457 (long)m->pindex);
1458 }
c9ec86b3
MD
1459
1460 /*
1461 * Remove all pmaps and indicate that the page is not
17cde63e
MD
1462 * writeable or mapped. Our vm_page_protect() call may
1463 * have blocked (especially w/ VM_PROT_NONE), so recheck
1464 * everything.
c9ec86b3
MD
1465 */
1466 vm_page_protect(m, VM_PROT_NONE);
a491077e 1467 if ((m->flags & (PG_UNMANAGED|PG_MAPPED)) || m->busy ||
17cde63e 1468 m->wire_count || m->hold_count) {
a491077e 1469 vm_page_wakeup(m);
17cde63e 1470 } else if (m->dirty) {
c9ec86b3 1471 vm_page_deactivate(m);
a491077e 1472 vm_page_wakeup(m);
c9ec86b3
MD
1473 } else {
1474 vm_page_unqueue_nowakeup(m);
1475 m->queue = PQ_CACHE + m->pc;
1476 vm_page_queues[m->queue].lcnt++;
1477 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, pageq);
1478 vmstats.v_cache_count++;
a491077e 1479 vm_page_wakeup(m);
c9ec86b3
MD
1480 vm_page_free_wakeup();
1481 }
984263bc
MD
1482}
1483
1484/*
de71fd3f
MD
1485 * vm_page_dontneed()
1486 *
1487 * Cache, deactivate, or do nothing as appropriate. This routine
1488 * is typically used by madvise() MADV_DONTNEED.
1489 *
1490 * Generally speaking we want to move the page into the cache so
1491 * it gets reused quickly. However, this can result in a silly syndrome
1492 * due to the page recycling too quickly. Small objects will not be
1493 * fully cached. On the otherhand, if we move the page to the inactive
1494 * queue we wind up with a problem whereby very large objects
1495 * unnecessarily blow away our inactive and cache queues.
1496 *
1497 * The solution is to move the pages based on a fixed weighting. We
1498 * either leave them alone, deactivate them, or move them to the cache,
1499 * where moving them to the cache has the highest weighting.
1500 * By forcing some pages into other queues we eventually force the
1501 * system to balance the queues, potentially recovering other unrelated
1502 * space from active. The idea is to not force this to happen too
1503 * often.
573fb415
MD
1504 *
1505 * No requirements.
984263bc 1506 */
984263bc
MD
1507void
1508vm_page_dontneed(vm_page_t m)
1509{
1510 static int dnweight;
1511 int dnw;
1512 int head;
1513
1514 dnw = ++dnweight;
1515
1516 /*
1517 * occassionally leave the page alone
1518 */
9ad0147b 1519 lwkt_gettoken(&vm_token);
984263bc
MD
1520 if ((dnw & 0x01F0) == 0 ||
1521 m->queue == PQ_INACTIVE ||
1522 m->queue - m->pc == PQ_CACHE
1523 ) {
1524 if (m->act_count >= ACT_INIT)
1525 --m->act_count;
9ad0147b 1526 lwkt_reltoken(&vm_token);
984263bc
MD
1527 return;
1528 }
1529
1530 if (m->dirty == 0)
1531 vm_page_test_dirty(m);
1532
1533 if (m->dirty || (dnw & 0x0070) == 0) {
1534 /*
1535 * Deactivate the page 3 times out of 32.
1536 */
1537 head = 0;
1538 } else {
1539 /*
1540 * Cache the page 28 times out of every 32. Note that
1541 * the page is deactivated instead of cached, but placed
1542 * at the head of the queue instead of the tail.
1543 */
1544 head = 1;
1545 }
1546 _vm_page_deactivate(m, head);
9ad0147b 1547 lwkt_reltoken(&vm_token);
984263bc
MD
1548}
1549
1550/*
06ecca5a
MD
1551 * Grab a page, blocking if it is busy and allocating a page if necessary.
1552 * A busy page is returned or NULL.
984263bc 1553 *
dc1fd4b3 1554 * If VM_ALLOC_RETRY is specified VM_ALLOC_NORMAL must also be specified.
06ecca5a 1555 * If VM_ALLOC_RETRY is not specified
dc1fd4b3 1556 *
06ecca5a
MD
1557 * This routine may block, but if VM_ALLOC_RETRY is not set then NULL is
1558 * always returned if we had blocked.
1559 * This routine will never return NULL if VM_ALLOC_RETRY is set.
1560 * This routine may not be called from an interrupt.
1561 * The returned page may not be entirely valid.
1562 *
1563 * This routine may be called from mainline code without spl protection and
1564 * be guarenteed a busied page associated with the object at the specified
1565 * index.
573fb415
MD
1566 *
1567 * No requirements.
984263bc
MD
1568 */
1569vm_page_t
1570vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
1571{
984263bc 1572 vm_page_t m;
654a39f0 1573 int generation;
984263bc 1574
dc1fd4b3
MD
1575 KKASSERT(allocflags &
1576 (VM_ALLOC_NORMAL|VM_ALLOC_INTERRUPT|VM_ALLOC_SYSTEM));
9ad0147b 1577 lwkt_gettoken(&vm_token);
984263bc
MD
1578retrylookup:
1579 if ((m = vm_page_lookup(object, pindex)) != NULL) {
1580 if (m->busy || (m->flags & PG_BUSY)) {
1581 generation = object->generation;
1582
984263bc
MD
1583 while ((object->generation == generation) &&
1584 (m->busy || (m->flags & PG_BUSY))) {
1585 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
377d4740 1586 tsleep(m, 0, "pgrbwt", 0);
984263bc 1587 if ((allocflags & VM_ALLOC_RETRY) == 0) {
06ecca5a
MD
1588 m = NULL;
1589 goto done;
984263bc
MD
1590 }
1591 }
984263bc
MD
1592 goto retrylookup;
1593 } else {
1594 vm_page_busy(m);
06ecca5a 1595 goto done;
984263bc
MD
1596 }
1597 }
984263bc
MD
1598 m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
1599 if (m == NULL) {
4ecf7cc9 1600 vm_wait(0);
984263bc 1601 if ((allocflags & VM_ALLOC_RETRY) == 0)
06ecca5a 1602 goto done;
984263bc
MD
1603 goto retrylookup;
1604 }
06ecca5a 1605done:
9ad0147b 1606 lwkt_reltoken(&vm_token);
06ecca5a 1607 return(m);
984263bc
MD
1608}
1609
1610/*
1611 * Mapping function for valid bits or for dirty bits in
1612 * a page. May not block.
1613 *
1614 * Inputs are required to range within a page.
573fb415
MD
1615 *
1616 * No requirements.
1617 * Non blocking.
984263bc 1618 */
573fb415 1619int
984263bc
MD
1620vm_page_bits(int base, int size)
1621{
1622 int first_bit;
1623 int last_bit;
1624
1625 KASSERT(
1626 base + size <= PAGE_SIZE,
1627 ("vm_page_bits: illegal base/size %d/%d", base, size)
1628 );
1629
1630 if (size == 0) /* handle degenerate case */
1631 return(0);
1632
1633 first_bit = base >> DEV_BSHIFT;
1634 last_bit = (base + size - 1) >> DEV_BSHIFT;
1635
1636 return ((2 << last_bit) - (1 << first_bit));
1637}
1638
1639/*
de71fd3f
MD
1640 * Sets portions of a page valid and clean. The arguments are expected
1641 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1642 * of any partial chunks touched by the range. The invalid portion of
1643 * such chunks will be zero'd.
984263bc 1644 *
c7841cbe
MD
1645 * NOTE: When truncating a buffer vnode_pager_setsize() will automatically
1646 * align base to DEV_BSIZE so as not to mark clean a partially
1647 * truncated device block. Otherwise the dirty page status might be
1648 * lost.
1649 *
de71fd3f 1650 * This routine may not block.
984263bc 1651 *
de71fd3f 1652 * (base + size) must be less then or equal to PAGE_SIZE.
984263bc 1653 */
1a54183b
MD
1654static void
1655_vm_page_zero_valid(vm_page_t m, int base, int size)
984263bc 1656{
984263bc
MD
1657 int frag;
1658 int endoff;
1659
1660 if (size == 0) /* handle degenerate case */
1661 return;
1662
1663 /*
1664 * If the base is not DEV_BSIZE aligned and the valid
1665 * bit is clear, we have to zero out a portion of the
1666 * first block.
1667 */
1668
1669 if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1670 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0
1671 ) {
1672 pmap_zero_page_area(
1673 VM_PAGE_TO_PHYS(m),
1674 frag,
1675 base - frag
1676 );
1677 }
1678
1679 /*
1680 * If the ending offset is not DEV_BSIZE aligned and the
1681 * valid bit is clear, we have to zero out a portion of
1682 * the last block.
1683 */
1684
1685 endoff = base + size;
1686
1687 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1688 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0
1689 ) {
1690 pmap_zero_page_area(
1691 VM_PAGE_TO_PHYS(m),
1692 endoff,
1693 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))
1694 );
1695 }
1a54183b 1696}
984263bc 1697
1a54183b
MD
1698/*
1699 * Set valid, clear dirty bits. If validating the entire
1700 * page we can safely clear the pmap modify bit. We also
1701 * use this opportunity to clear the PG_NOSYNC flag. If a process
1702 * takes a write fault on a MAP_NOSYNC memory area the flag will
1703 * be set again.
1704 *
1705 * We set valid bits inclusive of any overlap, but we can only
1706 * clear dirty bits for DEV_BSIZE chunks that are fully within
1707 * the range.
573fb415
MD
1708 *
1709 * Page must be busied?
1710 * No other requirements.
1a54183b
MD
1711 */
1712void
1713vm_page_set_valid(vm_page_t m, int base, int size)
1714{
1715 _vm_page_zero_valid(m, base, size);
1716 m->valid |= vm_page_bits(base, size);
1717}
984263bc 1718
cb1cf930
MD
1719
1720/*
1721 * Set valid bits and clear dirty bits.
1722 *
1723 * NOTE: This function does not clear the pmap modified bit.
1724 * Also note that e.g. NFS may use a byte-granular base
1725 * and size.
573fb415 1726 *
9a0cb7b1
MD
1727 * WARNING: Page must be busied? But vfs_clean_one_page() will call
1728 * this without necessarily busying the page (via bdwrite()).
1729 * So for now vm_token must also be held.
1730 *
573fb415 1731 * No other requirements.
cb1cf930 1732 */
1a54183b
MD
1733void
1734vm_page_set_validclean(vm_page_t m, int base, int size)
1735{
1736 int pagebits;
1737
1738 _vm_page_zero_valid(m, base, size);
984263bc
MD
1739 pagebits = vm_page_bits(base, size);
1740 m->valid |= pagebits;
984263bc
MD
1741 m->dirty &= ~pagebits;
1742 if (base == 0 && size == PAGE_SIZE) {
cb1cf930 1743 /*pmap_clear_modify(m);*/
984263bc
MD
1744 vm_page_flag_clear(m, PG_NOSYNC);
1745 }
1746}
1747
0a8aee15
MD
1748/*
1749 * Set valid & dirty. Used by buwrite()
573fb415 1750 *
9a0cb7b1
MD
1751 * WARNING: Page must be busied? But vfs_dirty_one_page() will
1752 * call this function in buwrite() so for now vm_token must
1753 * be held.
1754 *
573fb415 1755 * No other requirements.
0a8aee15
MD
1756 */
1757void
1758vm_page_set_validdirty(vm_page_t m, int base, int size)
1759{
1760 int pagebits;
1761
1762 pagebits = vm_page_bits(base, size);
1763 m->valid |= pagebits;
1764 m->dirty |= pagebits;
d89ce96a
MD
1765 if (m->object)
1766 vm_object_set_writeable_dirty(m->object);
0a8aee15
MD
1767}
1768
cb1cf930
MD
1769/*
1770 * Clear dirty bits.
1771 *
1772 * NOTE: This function does not clear the pmap modified bit.
1773 * Also note that e.g. NFS may use a byte-granular base
1774 * and size.
573fb415
MD
1775 *
1776 * Page must be busied?
1777 * No other requirements.
cb1cf930 1778 */
984263bc
MD
1779void
1780vm_page_clear_dirty(vm_page_t m, int base, int size)
1781{
1782 m->dirty &= ~vm_page_bits(base, size);
1a54183b 1783 if (base == 0 && size == PAGE_SIZE) {
cb1cf930 1784 /*pmap_clear_modify(m);*/
1a54183b
MD
1785 vm_page_flag_clear(m, PG_NOSYNC);
1786 }
984263bc
MD
1787}
1788
17cde63e
MD
1789/*
1790 * Make the page all-dirty.
1791 *
1792 * Also make sure the related object and vnode reflect the fact that the
1793 * object may now contain a dirty page.
573fb415
MD
1794 *
1795 * Page must be busied?
1796 * No other requirements.
17cde63e
MD
1797 */
1798void
1799vm_page_dirty(vm_page_t m)
1800{
1801#ifdef INVARIANTS
1802 int pqtype = m->queue - m->pc;
1803#endif
1804 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
1805 ("vm_page_dirty: page in free/cache queue!"));
1806 if (m->dirty != VM_PAGE_BITS_ALL) {
1807 m->dirty = VM_PAGE_BITS_ALL;
1808 if (m->object)
1809 vm_object_set_writeable_dirty(m->object);
1810 }
1811}
1812
984263bc 1813/*
de71fd3f
MD
1814 * Invalidates DEV_BSIZE'd chunks within a page. Both the
1815 * valid and dirty bits for the effected areas are cleared.
984263bc 1816 *
573fb415
MD
1817 * Page must be busied?
1818 * Does not block.
1819 * No other requirements.
984263bc
MD
1820 */
1821void
1822vm_page_set_invalid(vm_page_t m, int base, int size)
1823{
1824 int bits;
1825
1826 bits = vm_page_bits(base, size);
1827 m->valid &= ~bits;
1828 m->dirty &= ~bits;
1829 m->object->generation++;
1830}
1831
1832/*
de71fd3f
MD
1833 * The kernel assumes that the invalid portions of a page contain
1834 * garbage, but such pages can be mapped into memory by user code.
1835 * When this occurs, we must zero out the non-valid portions of the
1836 * page so user code sees what it expects.
984263bc 1837 *
de71fd3f
MD
1838 * Pages are most often semi-valid when the end of a file is mapped
1839 * into memory and the file's size is not page aligned.
573fb415
MD
1840 *
1841 * Page must be busied?
1842 * No other requirements.
984263bc 1843 */
984263bc
MD
1844void
1845vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
1846{
1847 int b;
1848 int i;
1849
1850 /*
1851 * Scan the valid bits looking for invalid sections that
1852 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
1853 * valid bit may be set ) have already been zerod by
1854 * vm_page_set_validclean().
1855 */
984263bc
MD
1856 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
1857 if (i == (PAGE_SIZE / DEV_BSIZE) ||
1858 (m->valid & (1 << i))
1859 ) {
1860 if (i > b) {
1861 pmap_zero_page_area(
1862 VM_PAGE_TO_PHYS(m),
1863 b << DEV_BSHIFT,
1864 (i - b) << DEV_BSHIFT
1865 );
1866 }
1867 b = i + 1;
1868 }
1869 }
1870
1871 /*
1872 * setvalid is TRUE when we can safely set the zero'd areas
1873 * as being valid. We can do this if there are no cache consistency
1874 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
1875 */
984263bc
MD
1876 if (setvalid)
1877 m->valid = VM_PAGE_BITS_ALL;
1878}
1879
1880/*
de71fd3f
MD
1881 * Is a (partial) page valid? Note that the case where size == 0
1882 * will return FALSE in the degenerate case where the page is entirely
1883 * invalid, and TRUE otherwise.
984263bc 1884 *
573fb415
MD
1885 * Does not block.
1886 * No other requirements.
984263bc 1887 */
984263bc
MD
1888int
1889vm_page_is_valid(vm_page_t m, int base, int size)
1890{
1891 int bits = vm_page_bits(base, size);
1892
1893 if (m->valid && ((m->valid & bits) == bits))
1894 return 1;
1895 else
1896 return 0;
1897}
1898
1899/*
1900 * update dirty bits from pmap/mmu. May not block.
573fb415
MD
1901 *
1902 * Caller must hold vm_token if non-blocking operation desired.
1903 * No other requirements.
984263bc 1904 */
984263bc
MD
1905void
1906vm_page_test_dirty(vm_page_t m)
1907{
1908 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
1909 vm_page_dirty(m);
1910 }
1911}
1912
906c754c
MD
1913/*
1914 * Register an action, associating it with its vm_page
1915 */
1916void
1917vm_page_register_action(vm_page_action_t action, vm_page_event_t event)
1918{
1919 struct vm_page_action_list *list;
1920 int hv;
1921
1922 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1923 list = &action_list[hv];
1924
1925 lwkt_gettoken(&vm_token);
1926 vm_page_flag_set(action->m, PG_ACTIONLIST);
1927 action->event = event;
1928 LIST_INSERT_HEAD(list, action, entry);
1929 lwkt_reltoken(&vm_token);
1930}
1931
1932/*
1933 * Unregister an action, disassociating it from its related vm_page
1934 */
1935void
1936vm_page_unregister_action(vm_page_action_t action)
1937{
1938 struct vm_page_action_list *list;
1939 int hv;
1940
1941 lwkt_gettoken(&vm_token);
1942 if (action->event != VMEVENT_NONE) {
1943 action->event = VMEVENT_NONE;
1944 LIST_REMOVE(action, entry);
1945
1946 hv = (int)((intptr_t)action->m >> 8) & VMACTION_HMASK;
1947 list = &action_list[hv];
1948 if (LIST_EMPTY(list))
1949 vm_page_flag_clear(action->m, PG_ACTIONLIST);
1950 }
1951 lwkt_reltoken(&vm_token);
1952}
1953
10192bae
MD
1954/*
1955 * Issue an event on a VM page. Corresponding action structures are
1956 * removed from the page's list and called.
906c754c
MD
1957 *
1958 * If the vm_page has no more pending action events we clear its
1959 * PG_ACTIONLIST flag.
10192bae
MD
1960 */
1961void
1962vm_page_event_internal(vm_page_t m, vm_page_event_t event)
1963{
906c754c
MD
1964 struct vm_page_action_list *list;
1965 struct vm_page_action *scan;
1966 struct vm_page_action *next;
1967 int hv;
1968 int all;
10192bae 1969
906c754c
MD
1970 hv = (int)((intptr_t)m >> 8) & VMACTION_HMASK;
1971 list = &action_list[hv];
1972 all = 1;
1973
1974 lwkt_gettoken(&vm_token);
1975 LIST_FOREACH_MUTABLE(scan, list, entry, next) {
1976 if (scan->m == m) {
1977 if (scan->event == event) {
1978 scan->event = VMEVENT_NONE;
1979 LIST_REMOVE(scan, entry);
1980 scan->func(m, scan);
1981 /* XXX */
1982 } else {
1983 all = 0;
1984 }
10192bae
MD
1985 }
1986 }
906c754c
MD
1987 if (all)
1988 vm_page_flag_clear(m, PG_ACTIONLIST);
1989 lwkt_reltoken(&vm_token);
10192bae
MD
1990}
1991
bb6811be 1992
984263bc
MD
1993#include "opt_ddb.h"
1994#ifdef DDB
1995#include <sys/kernel.h>
1996
1997#include <ddb/ddb.h>
1998
1999DB_SHOW_COMMAND(page, vm_page_print_page_info)
2000{
12e4aaff
MD
2001 db_printf("vmstats.v_free_count: %d\n", vmstats.v_free_count);
2002 db_printf("vmstats.v_cache_count: %d\n", vmstats.v_cache_count);
2003 db_printf("vmstats.v_inactive_count: %d\n", vmstats.v_inactive_count);
2004 db_printf("vmstats.v_active_count: %d\n", vmstats.v_active_count);
2005 db_printf("vmstats.v_wire_count: %d\n", vmstats.v_wire_count);
2006 db_printf("vmstats.v_free_reserved: %d\n", vmstats.v_free_reserved);
2007 db_printf("vmstats.v_free_min: %d\n", vmstats.v_free_min);
2008 db_printf("vmstats.v_free_target: %d\n", vmstats.v_free_target);
2009 db_printf("vmstats.v_cache_min: %d\n", vmstats.v_cache_min);
2010 db_printf("vmstats.v_inactive_target: %d\n", vmstats.v_inactive_target);
984263bc
MD
2011}
2012
2013DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
2014{
2015 int i;
2016 db_printf("PQ_FREE:");
2017 for(i=0;i<PQ_L2_SIZE;i++) {
2018 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt);
2019 }
2020 db_printf("\n");
2021
2022 db_printf("PQ_CACHE:");
2023 for(i=0;i<PQ_L2_SIZE;i++) {
2024 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt);
2025 }
2026 db_printf("\n");
2027
2028 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n",
2029 vm_page_queues[PQ_ACTIVE].lcnt,
2030 vm_page_queues[PQ_INACTIVE].lcnt);
2031}
2032#endif /* DDB */