xref: /freebsd/sys/vm/vm_page.c (revision 195ebc7e9e4b129de810833791a19dfb4349d6a9)
1 /*-
2  * Copyright (c) 1991 Regents of the University of California.
3  * All rights reserved.
4  * Copyright (c) 1998 Matthew Dillon.  All Rights Reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * The Mach Operating System project at Carnegie-Mellon University.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	from: @(#)vm_page.c	7.4 (Berkeley) 5/7/91
34  */
35 
36 /*-
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  */
62 
63 /*
64  *			GENERAL RULES ON VM_PAGE MANIPULATION
65  *
66  *	- a pageq mutex is required when adding or removing a page from a
67  *	  page queue (vm_page_queue[]), regardless of other mutexes or the
68  *	  busy state of a page.
69  *
70  *	- a hash chain mutex is required when associating or disassociating
71  *	  a page from the VM PAGE CACHE hash table (vm_page_buckets),
72  *	  regardless of other mutexes or the busy state of a page.
73  *
74  *	- either a hash chain mutex OR a busied page is required in order
75  *	  to modify the page flags.  A hash chain mutex must be obtained in
76  *	  order to busy a page.  A page's flags cannot be modified by a
77  *	  hash chain mutex if the page is marked busy.
78  *
79  *	- The object memq mutex is held when inserting or removing
80  *	  pages from an object (vm_page_insert() or vm_page_remove()).  This
81  *	  is different from the object's main mutex.
82  *
83  *	Generally speaking, you have to be aware of side effects when running
84  *	vm_page ops.  A vm_page_lookup() will return with the hash chain
85  *	locked, whether it was able to lookup the page or not.  vm_page_free(),
86  *	vm_page_cache(), vm_page_activate(), and a number of other routines
87  *	will release the hash chain mutex for you.  Intermediate manipulation
88  *	routines such as vm_page_flag_set() expect the hash chain to be held
89  *	on entry and the hash chain will remain held on return.
90  *
91  *	pageq scanning can only occur with the pageq in question locked.
92  *	We have a known bottleneck with the active queue, but the cache
93  *	and free queues are actually arrays already.
94  */
95 
96 /*
97  *	Resident memory management module.
98  */
99 
100 #include <sys/cdefs.h>
101 __FBSDID("$FreeBSD$");
102 
103 #include "opt_vm.h"
104 
105 #include <sys/param.h>
106 #include <sys/systm.h>
107 #include <sys/lock.h>
108 #include <sys/kernel.h>
109 #include <sys/limits.h>
110 #include <sys/malloc.h>
111 #include <sys/mutex.h>
112 #include <sys/proc.h>
113 #include <sys/sysctl.h>
114 #include <sys/vmmeter.h>
115 #include <sys/vnode.h>
116 
117 #include <vm/vm.h>
118 #include <vm/vm_param.h>
119 #include <vm/vm_kern.h>
120 #include <vm/vm_object.h>
121 #include <vm/vm_page.h>
122 #include <vm/vm_pageout.h>
123 #include <vm/vm_pager.h>
124 #include <vm/vm_phys.h>
125 #include <vm/vm_reserv.h>
126 #include <vm/vm_extern.h>
127 #include <vm/uma.h>
128 #include <vm/uma_int.h>
129 
130 #include <machine/md_var.h>
131 
132 /*
133  *	Associated with page of user-allocatable memory is a
134  *	page structure.
135  */
136 
137 struct vpgqueues vm_page_queues[PQ_COUNT];
138 struct mtx vm_page_queue_mtx;
139 struct mtx vm_page_queue_free_mtx;
140 
141 vm_page_t vm_page_array = 0;
142 int vm_page_array_size = 0;
143 long first_page = 0;
144 int vm_page_zero_count = 0;
145 
146 static int boot_pages = UMA_BOOT_PAGES;
147 TUNABLE_INT("vm.boot_pages", &boot_pages);
148 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RD, &boot_pages, 0,
149 	"number of pages allocated for bootstrapping the VM system");
150 
151 static void vm_page_enqueue(int queue, vm_page_t m);
152 
153 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
154 #if PAGE_SIZE == 32768
155 #ifdef CTASSERT
156 CTASSERT(sizeof(u_long) >= 8);
157 #endif
158 #endif
159 
160 /*
161  *	vm_set_page_size:
162  *
163  *	Sets the page size, perhaps based upon the memory
164  *	size.  Must be called before any use of page-size
165  *	dependent functions.
166  */
167 void
168 vm_set_page_size(void)
169 {
170 	if (cnt.v_page_size == 0)
171 		cnt.v_page_size = PAGE_SIZE;
172 	if (((cnt.v_page_size - 1) & cnt.v_page_size) != 0)
173 		panic("vm_set_page_size: page size not a power of two");
174 }
175 
176 /*
177  *	vm_page_blacklist_lookup:
178  *
179  *	See if a physical address in this page has been listed
180  *	in the blacklist tunable.  Entries in the tunable are
181  *	separated by spaces or commas.  If an invalid integer is
182  *	encountered then the rest of the string is skipped.
183  */
184 static int
185 vm_page_blacklist_lookup(char *list, vm_paddr_t pa)
186 {
187 	vm_paddr_t bad;
188 	char *cp, *pos;
189 
190 	for (pos = list; *pos != '\0'; pos = cp) {
191 		bad = strtoq(pos, &cp, 0);
192 		if (*cp != '\0') {
193 			if (*cp == ' ' || *cp == ',') {
194 				cp++;
195 				if (cp == pos)
196 					continue;
197 			} else
198 				break;
199 		}
200 		if (pa == trunc_page(bad))
201 			return (1);
202 	}
203 	return (0);
204 }
205 
206 /*
207  *	vm_page_startup:
208  *
209  *	Initializes the resident memory module.
210  *
211  *	Allocates memory for the page cells, and
212  *	for the object/offset-to-page hash table headers.
213  *	Each page cell is initialized and placed on the free list.
214  */
215 vm_offset_t
216 vm_page_startup(vm_offset_t vaddr)
217 {
218 	vm_offset_t mapped;
219 	vm_paddr_t page_range;
220 	vm_paddr_t new_end;
221 	int i;
222 	vm_paddr_t pa;
223 	int nblocks;
224 	vm_paddr_t last_pa;
225 	char *list;
226 
227 	/* the biggest memory array is the second group of pages */
228 	vm_paddr_t end;
229 	vm_paddr_t biggestsize;
230 	vm_paddr_t low_water, high_water;
231 	int biggestone;
232 
233 	biggestsize = 0;
234 	biggestone = 0;
235 	nblocks = 0;
236 	vaddr = round_page(vaddr);
237 
238 	for (i = 0; phys_avail[i + 1]; i += 2) {
239 		phys_avail[i] = round_page(phys_avail[i]);
240 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
241 	}
242 
243 	low_water = phys_avail[0];
244 	high_water = phys_avail[1];
245 
246 	for (i = 0; phys_avail[i + 1]; i += 2) {
247 		vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
248 
249 		if (size > biggestsize) {
250 			biggestone = i;
251 			biggestsize = size;
252 		}
253 		if (phys_avail[i] < low_water)
254 			low_water = phys_avail[i];
255 		if (phys_avail[i + 1] > high_water)
256 			high_water = phys_avail[i + 1];
257 		++nblocks;
258 	}
259 
260 #ifdef XEN
261 	low_water = 0;
262 #endif
263 
264 	end = phys_avail[biggestone+1];
265 
266 	/*
267 	 * Initialize the locks.
268 	 */
269 	mtx_init(&vm_page_queue_mtx, "vm page queue mutex", NULL, MTX_DEF |
270 	    MTX_RECURSE);
271 	mtx_init(&vm_page_queue_free_mtx, "vm page queue free mutex", NULL,
272 	    MTX_DEF);
273 
274 	/*
275 	 * Initialize the queue headers for the hold queue, the active queue,
276 	 * and the inactive queue.
277 	 */
278 	for (i = 0; i < PQ_COUNT; i++)
279 		TAILQ_INIT(&vm_page_queues[i].pl);
280 	vm_page_queues[PQ_INACTIVE].cnt = &cnt.v_inactive_count;
281 	vm_page_queues[PQ_ACTIVE].cnt = &cnt.v_active_count;
282 	vm_page_queues[PQ_HOLD].cnt = &cnt.v_active_count;
283 
284 	/*
285 	 * Allocate memory for use when boot strapping the kernel memory
286 	 * allocator.
287 	 */
288 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
289 	new_end = trunc_page(new_end);
290 	mapped = pmap_map(&vaddr, new_end, end,
291 	    VM_PROT_READ | VM_PROT_WRITE);
292 	bzero((void *)mapped, end - new_end);
293 	uma_startup((void *)mapped, boot_pages);
294 
295 #if defined(__amd64__) || defined(__i386__) || defined(__arm__)
296 	/*
297 	 * Allocate a bitmap to indicate that a random physical page
298 	 * needs to be included in a minidump.
299 	 *
300 	 * The amd64 port needs this to indicate which direct map pages
301 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
302 	 *
303 	 * However, i386 still needs this workspace internally within the
304 	 * minidump code.  In theory, they are not needed on i386, but are
305 	 * included should the sf_buf code decide to use them.
306 	 */
307 	page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
308 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
309 	new_end -= vm_page_dump_size;
310 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
311 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
312 	bzero((void *)vm_page_dump, vm_page_dump_size);
313 #endif
314 	/*
315 	 * Compute the number of pages of memory that will be available for
316 	 * use (taking into account the overhead of a page structure per
317 	 * page).
318 	 */
319 	first_page = low_water / PAGE_SIZE;
320 #ifdef VM_PHYSSEG_SPARSE
321 	page_range = 0;
322 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
323 		page_range += atop(phys_avail[i + 1] - phys_avail[i]);
324 #elif defined(VM_PHYSSEG_DENSE)
325 	page_range = high_water / PAGE_SIZE - first_page;
326 #else
327 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
328 #endif
329 	end = new_end;
330 
331 	/*
332 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
333 	 */
334 	vaddr += PAGE_SIZE;
335 
336 	/*
337 	 * Initialize the mem entry structures now, and put them in the free
338 	 * queue.
339 	 */
340 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
341 	mapped = pmap_map(&vaddr, new_end, end,
342 	    VM_PROT_READ | VM_PROT_WRITE);
343 	vm_page_array = (vm_page_t) mapped;
344 #if VM_NRESERVLEVEL > 0
345 	/*
346 	 * Allocate memory for the reservation management system's data
347 	 * structures.
348 	 */
349 	new_end = vm_reserv_startup(&vaddr, new_end, high_water);
350 #endif
351 #ifdef __amd64__
352 	/*
353 	 * pmap_map on amd64 comes out of the direct-map, not kvm like i386,
354 	 * so the pages must be tracked for a crashdump to include this data.
355 	 * This includes the vm_page_array and the early UMA bootstrap pages.
356 	 */
357 	for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
358 		dump_add_page(pa);
359 #endif
360 	phys_avail[biggestone + 1] = new_end;
361 
362 	/*
363 	 * Clear all of the page structures
364 	 */
365 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
366 	for (i = 0; i < page_range; i++)
367 		vm_page_array[i].order = VM_NFREEORDER;
368 	vm_page_array_size = page_range;
369 
370 	/*
371 	 * Initialize the physical memory allocator.
372 	 */
373 	vm_phys_init();
374 
375 	/*
376 	 * Add every available physical page that is not blacklisted to
377 	 * the free lists.
378 	 */
379 	cnt.v_page_count = 0;
380 	cnt.v_free_count = 0;
381 	list = getenv("vm.blacklist");
382 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
383 		pa = phys_avail[i];
384 		last_pa = phys_avail[i + 1];
385 		while (pa < last_pa) {
386 			if (list != NULL &&
387 			    vm_page_blacklist_lookup(list, pa))
388 				printf("Skipping page with pa 0x%jx\n",
389 				    (uintmax_t)pa);
390 			else
391 				vm_phys_add_page(pa);
392 			pa += PAGE_SIZE;
393 		}
394 	}
395 	freeenv(list);
396 #if VM_NRESERVLEVEL > 0
397 	/*
398 	 * Initialize the reservation management system.
399 	 */
400 	vm_reserv_init();
401 #endif
402 	return (vaddr);
403 }
404 
405 void
406 vm_page_flag_set(vm_page_t m, unsigned short bits)
407 {
408 
409 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
410 	m->flags |= bits;
411 }
412 
413 void
414 vm_page_flag_clear(vm_page_t m, unsigned short bits)
415 {
416 
417 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
418 	m->flags &= ~bits;
419 }
420 
421 void
422 vm_page_busy(vm_page_t m)
423 {
424 
425 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
426 	KASSERT((m->oflags & VPO_BUSY) == 0,
427 	    ("vm_page_busy: page already busy!!!"));
428 	m->oflags |= VPO_BUSY;
429 }
430 
431 /*
432  *      vm_page_flash:
433  *
434  *      wakeup anyone waiting for the page.
435  */
436 void
437 vm_page_flash(vm_page_t m)
438 {
439 
440 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
441 	if (m->oflags & VPO_WANTED) {
442 		m->oflags &= ~VPO_WANTED;
443 		wakeup(m);
444 	}
445 }
446 
447 /*
448  *      vm_page_wakeup:
449  *
450  *      clear the VPO_BUSY flag and wakeup anyone waiting for the
451  *      page.
452  *
453  */
454 void
455 vm_page_wakeup(vm_page_t m)
456 {
457 
458 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
459 	KASSERT(m->oflags & VPO_BUSY, ("vm_page_wakeup: page not busy!!!"));
460 	m->oflags &= ~VPO_BUSY;
461 	vm_page_flash(m);
462 }
463 
464 void
465 vm_page_io_start(vm_page_t m)
466 {
467 
468 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
469 	m->busy++;
470 }
471 
472 void
473 vm_page_io_finish(vm_page_t m)
474 {
475 
476 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
477 	m->busy--;
478 	if (m->busy == 0)
479 		vm_page_flash(m);
480 }
481 
482 /*
483  * Keep page from being freed by the page daemon
484  * much of the same effect as wiring, except much lower
485  * overhead and should be used only for *very* temporary
486  * holding ("wiring").
487  */
488 void
489 vm_page_hold(vm_page_t mem)
490 {
491 
492 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
493         mem->hold_count++;
494 }
495 
496 void
497 vm_page_unhold(vm_page_t mem)
498 {
499 
500 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
501 	--mem->hold_count;
502 	KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
503 	if (mem->hold_count == 0 && VM_PAGE_INQUEUE2(mem, PQ_HOLD))
504 		vm_page_free_toq(mem);
505 }
506 
507 /*
508  *	vm_page_free:
509  *
510  *	Free a page.
511  */
512 void
513 vm_page_free(vm_page_t m)
514 {
515 
516 	m->flags &= ~PG_ZERO;
517 	vm_page_free_toq(m);
518 }
519 
520 /*
521  *	vm_page_free_zero:
522  *
523  *	Free a page to the zerod-pages queue
524  */
525 void
526 vm_page_free_zero(vm_page_t m)
527 {
528 
529 	m->flags |= PG_ZERO;
530 	vm_page_free_toq(m);
531 }
532 
533 /*
534  *	vm_page_sleep:
535  *
536  *	Sleep and release the page queues lock.
537  *
538  *	The object containing the given page must be locked.
539  */
540 void
541 vm_page_sleep(vm_page_t m, const char *msg)
542 {
543 
544 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
545 	if (!mtx_owned(&vm_page_queue_mtx))
546 		vm_page_lock_queues();
547 	vm_page_flag_set(m, PG_REFERENCED);
548 	vm_page_unlock_queues();
549 
550 	/*
551 	 * It's possible that while we sleep, the page will get
552 	 * unbusied and freed.  If we are holding the object
553 	 * lock, we will assume we hold a reference to the object
554 	 * such that even if m->object changes, we can re-lock
555 	 * it.
556 	 */
557 	m->oflags |= VPO_WANTED;
558 	msleep(m, VM_OBJECT_MTX(m->object), PVM, msg, 0);
559 }
560 
561 /*
562  *	vm_page_dirty:
563  *
564  *	make page all dirty
565  */
566 void
567 vm_page_dirty(vm_page_t m)
568 {
569 
570 	KASSERT((m->flags & PG_CACHED) == 0,
571 	    ("vm_page_dirty: page in cache!"));
572 	KASSERT(!VM_PAGE_IS_FREE(m),
573 	    ("vm_page_dirty: page is free!"));
574 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
575 	    ("vm_page_dirty: page is invalid!"));
576 	m->dirty = VM_PAGE_BITS_ALL;
577 }
578 
579 /*
580  *	vm_page_splay:
581  *
582  *	Implements Sleator and Tarjan's top-down splay algorithm.  Returns
583  *	the vm_page containing the given pindex.  If, however, that
584  *	pindex is not found in the vm_object, returns a vm_page that is
585  *	adjacent to the pindex, coming before or after it.
586  */
587 vm_page_t
588 vm_page_splay(vm_pindex_t pindex, vm_page_t root)
589 {
590 	struct vm_page dummy;
591 	vm_page_t lefttreemax, righttreemin, y;
592 
593 	if (root == NULL)
594 		return (root);
595 	lefttreemax = righttreemin = &dummy;
596 	for (;; root = y) {
597 		if (pindex < root->pindex) {
598 			if ((y = root->left) == NULL)
599 				break;
600 			if (pindex < y->pindex) {
601 				/* Rotate right. */
602 				root->left = y->right;
603 				y->right = root;
604 				root = y;
605 				if ((y = root->left) == NULL)
606 					break;
607 			}
608 			/* Link into the new root's right tree. */
609 			righttreemin->left = root;
610 			righttreemin = root;
611 		} else if (pindex > root->pindex) {
612 			if ((y = root->right) == NULL)
613 				break;
614 			if (pindex > y->pindex) {
615 				/* Rotate left. */
616 				root->right = y->left;
617 				y->left = root;
618 				root = y;
619 				if ((y = root->right) == NULL)
620 					break;
621 			}
622 			/* Link into the new root's left tree. */
623 			lefttreemax->right = root;
624 			lefttreemax = root;
625 		} else
626 			break;
627 	}
628 	/* Assemble the new root. */
629 	lefttreemax->right = root->left;
630 	righttreemin->left = root->right;
631 	root->left = dummy.right;
632 	root->right = dummy.left;
633 	return (root);
634 }
635 
636 /*
637  *	vm_page_insert:		[ internal use only ]
638  *
639  *	Inserts the given mem entry into the object and object list.
640  *
641  *	The pagetables are not updated but will presumably fault the page
642  *	in if necessary, or if a kernel page the caller will at some point
643  *	enter the page into the kernel's pmap.  We are not allowed to block
644  *	here so we *can't* do this anyway.
645  *
646  *	The object and page must be locked.
647  *	This routine may not block.
648  */
649 void
650 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
651 {
652 	vm_page_t root;
653 
654 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
655 	if (m->object != NULL)
656 		panic("vm_page_insert: page already inserted");
657 
658 	/*
659 	 * Record the object/offset pair in this page
660 	 */
661 	m->object = object;
662 	m->pindex = pindex;
663 
664 	/*
665 	 * Now link into the object's ordered list of backed pages.
666 	 */
667 	root = object->root;
668 	if (root == NULL) {
669 		m->left = NULL;
670 		m->right = NULL;
671 		TAILQ_INSERT_TAIL(&object->memq, m, listq);
672 	} else {
673 		root = vm_page_splay(pindex, root);
674 		if (pindex < root->pindex) {
675 			m->left = root->left;
676 			m->right = root;
677 			root->left = NULL;
678 			TAILQ_INSERT_BEFORE(root, m, listq);
679 		} else if (pindex == root->pindex)
680 			panic("vm_page_insert: offset already allocated");
681 		else {
682 			m->right = root->right;
683 			m->left = root;
684 			root->right = NULL;
685 			TAILQ_INSERT_AFTER(&object->memq, root, m, listq);
686 		}
687 	}
688 	object->root = m;
689 	object->generation++;
690 
691 	/*
692 	 * show that the object has one more resident page.
693 	 */
694 	object->resident_page_count++;
695 	/*
696 	 * Hold the vnode until the last page is released.
697 	 */
698 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
699 		vhold((struct vnode *)object->handle);
700 
701 	/*
702 	 * Since we are inserting a new and possibly dirty page,
703 	 * update the object's OBJ_MIGHTBEDIRTY flag.
704 	 */
705 	if (m->flags & PG_WRITEABLE)
706 		vm_object_set_writeable_dirty(object);
707 }
708 
709 /*
710  *	vm_page_remove:
711  *				NOTE: used by device pager as well -wfj
712  *
713  *	Removes the given mem entry from the object/offset-page
714  *	table and the object page list, but do not invalidate/terminate
715  *	the backing store.
716  *
717  *	The object and page must be locked.
718  *	The underlying pmap entry (if any) is NOT removed here.
719  *	This routine may not block.
720  */
721 void
722 vm_page_remove(vm_page_t m)
723 {
724 	vm_object_t object;
725 	vm_page_t root;
726 
727 	if ((object = m->object) == NULL)
728 		return;
729 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
730 	if (m->oflags & VPO_BUSY) {
731 		m->oflags &= ~VPO_BUSY;
732 		vm_page_flash(m);
733 	}
734 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
735 
736 	/*
737 	 * Now remove from the object's list of backed pages.
738 	 */
739 	if (m != object->root)
740 		vm_page_splay(m->pindex, object->root);
741 	if (m->left == NULL)
742 		root = m->right;
743 	else {
744 		root = vm_page_splay(m->pindex, m->left);
745 		root->right = m->right;
746 	}
747 	object->root = root;
748 	TAILQ_REMOVE(&object->memq, m, listq);
749 
750 	/*
751 	 * And show that the object has one fewer resident page.
752 	 */
753 	object->resident_page_count--;
754 	object->generation++;
755 	/*
756 	 * The vnode may now be recycled.
757 	 */
758 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
759 		vdrop((struct vnode *)object->handle);
760 
761 	m->object = NULL;
762 }
763 
764 /*
765  *	vm_page_lookup:
766  *
767  *	Returns the page associated with the object/offset
768  *	pair specified; if none is found, NULL is returned.
769  *
770  *	The object must be locked.
771  *	This routine may not block.
772  *	This is a critical path routine
773  */
774 vm_page_t
775 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
776 {
777 	vm_page_t m;
778 
779 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
780 	if ((m = object->root) != NULL && m->pindex != pindex) {
781 		m = vm_page_splay(pindex, m);
782 		if ((object->root = m)->pindex != pindex)
783 			m = NULL;
784 	}
785 	return (m);
786 }
787 
788 /*
789  *	vm_page_rename:
790  *
791  *	Move the given memory entry from its
792  *	current object to the specified target object/offset.
793  *
794  *	The object must be locked.
795  *	This routine may not block.
796  *
797  *	Note: swap associated with the page must be invalidated by the move.  We
798  *	      have to do this for several reasons:  (1) we aren't freeing the
799  *	      page, (2) we are dirtying the page, (3) the VM system is probably
800  *	      moving the page from object A to B, and will then later move
801  *	      the backing store from A to B and we can't have a conflict.
802  *
803  *	Note: we *always* dirty the page.  It is necessary both for the
804  *	      fact that we moved it, and because we may be invalidating
805  *	      swap.  If the page is on the cache, we have to deactivate it
806  *	      or vm_page_dirty() will panic.  Dirty pages are not allowed
807  *	      on the cache.
808  */
809 void
810 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
811 {
812 
813 	vm_page_remove(m);
814 	vm_page_insert(m, new_object, new_pindex);
815 	vm_page_dirty(m);
816 }
817 
818 /*
819  *	Convert all of the given object's cached pages that have a
820  *	pindex within the given range into free pages.  If the value
821  *	zero is given for "end", then the range's upper bound is
822  *	infinity.  If the given object is backed by a vnode and it
823  *	transitions from having one or more cached pages to none, the
824  *	vnode's hold count is reduced.
825  */
826 void
827 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
828 {
829 	vm_page_t m, m_next;
830 	boolean_t empty;
831 
832 	mtx_lock(&vm_page_queue_free_mtx);
833 	if (__predict_false(object->cache == NULL)) {
834 		mtx_unlock(&vm_page_queue_free_mtx);
835 		return;
836 	}
837 	m = object->cache = vm_page_splay(start, object->cache);
838 	if (m->pindex < start) {
839 		if (m->right == NULL)
840 			m = NULL;
841 		else {
842 			m_next = vm_page_splay(start, m->right);
843 			m_next->left = m;
844 			m->right = NULL;
845 			m = object->cache = m_next;
846 		}
847 	}
848 
849 	/*
850 	 * At this point, "m" is either (1) a reference to the page
851 	 * with the least pindex that is greater than or equal to
852 	 * "start" or (2) NULL.
853 	 */
854 	for (; m != NULL && (m->pindex < end || end == 0); m = m_next) {
855 		/*
856 		 * Find "m"'s successor and remove "m" from the
857 		 * object's cache.
858 		 */
859 		if (m->right == NULL) {
860 			object->cache = m->left;
861 			m_next = NULL;
862 		} else {
863 			m_next = vm_page_splay(start, m->right);
864 			m_next->left = m->left;
865 			object->cache = m_next;
866 		}
867 		/* Convert "m" to a free page. */
868 		m->object = NULL;
869 		m->valid = 0;
870 		/* Clear PG_CACHED and set PG_FREE. */
871 		m->flags ^= PG_CACHED | PG_FREE;
872 		KASSERT((m->flags & (PG_CACHED | PG_FREE)) == PG_FREE,
873 		    ("vm_page_cache_free: page %p has inconsistent flags", m));
874 		cnt.v_cache_count--;
875 		cnt.v_free_count++;
876 	}
877 	empty = object->cache == NULL;
878 	mtx_unlock(&vm_page_queue_free_mtx);
879 	if (object->type == OBJT_VNODE && empty)
880 		vdrop(object->handle);
881 }
882 
883 /*
884  *	Returns the cached page that is associated with the given
885  *	object and offset.  If, however, none exists, returns NULL.
886  *
887  *	The free page queue must be locked.
888  */
889 static inline vm_page_t
890 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
891 {
892 	vm_page_t m;
893 
894 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
895 	if ((m = object->cache) != NULL && m->pindex != pindex) {
896 		m = vm_page_splay(pindex, m);
897 		if ((object->cache = m)->pindex != pindex)
898 			m = NULL;
899 	}
900 	return (m);
901 }
902 
903 /*
904  *	Remove the given cached page from its containing object's
905  *	collection of cached pages.
906  *
907  *	The free page queue must be locked.
908  */
909 void
910 vm_page_cache_remove(vm_page_t m)
911 {
912 	vm_object_t object;
913 	vm_page_t root;
914 
915 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
916 	KASSERT((m->flags & PG_CACHED) != 0,
917 	    ("vm_page_cache_remove: page %p is not cached", m));
918 	object = m->object;
919 	if (m != object->cache) {
920 		root = vm_page_splay(m->pindex, object->cache);
921 		KASSERT(root == m,
922 		    ("vm_page_cache_remove: page %p is not cached in object %p",
923 		    m, object));
924 	}
925 	if (m->left == NULL)
926 		root = m->right;
927 	else if (m->right == NULL)
928 		root = m->left;
929 	else {
930 		root = vm_page_splay(m->pindex, m->left);
931 		root->right = m->right;
932 	}
933 	object->cache = root;
934 	m->object = NULL;
935 	cnt.v_cache_count--;
936 }
937 
938 /*
939  *	Transfer all of the cached pages with offset greater than or
940  *	equal to 'offidxstart' from the original object's cache to the
941  *	new object's cache.  However, any cached pages with offset
942  *	greater than or equal to the new object's size are kept in the
943  *	original object.  Initially, the new object's cache must be
944  *	empty.  Offset 'offidxstart' in the original object must
945  *	correspond to offset zero in the new object.
946  *
947  *	The new object must be locked.
948  */
949 void
950 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
951     vm_object_t new_object)
952 {
953 	vm_page_t m, m_next;
954 
955 	/*
956 	 * Insertion into an object's collection of cached pages
957 	 * requires the object to be locked.  In contrast, removal does
958 	 * not.
959 	 */
960 	VM_OBJECT_LOCK_ASSERT(new_object, MA_OWNED);
961 	KASSERT(new_object->cache == NULL,
962 	    ("vm_page_cache_transfer: object %p has cached pages",
963 	    new_object));
964 	mtx_lock(&vm_page_queue_free_mtx);
965 	if ((m = orig_object->cache) != NULL) {
966 		/*
967 		 * Transfer all of the pages with offset greater than or
968 		 * equal to 'offidxstart' from the original object's
969 		 * cache to the new object's cache.
970 		 */
971 		m = vm_page_splay(offidxstart, m);
972 		if (m->pindex < offidxstart) {
973 			orig_object->cache = m;
974 			new_object->cache = m->right;
975 			m->right = NULL;
976 		} else {
977 			orig_object->cache = m->left;
978 			new_object->cache = m;
979 			m->left = NULL;
980 		}
981 		while ((m = new_object->cache) != NULL) {
982 			if ((m->pindex - offidxstart) >= new_object->size) {
983 				/*
984 				 * Return all of the cached pages with
985 				 * offset greater than or equal to the
986 				 * new object's size to the original
987 				 * object's cache.
988 				 */
989 				new_object->cache = m->left;
990 				m->left = orig_object->cache;
991 				orig_object->cache = m;
992 				break;
993 			}
994 			m_next = vm_page_splay(m->pindex, m->right);
995 			/* Update the page's object and offset. */
996 			m->object = new_object;
997 			m->pindex -= offidxstart;
998 			if (m_next == NULL)
999 				break;
1000 			m->right = NULL;
1001 			m_next->left = m;
1002 			new_object->cache = m_next;
1003 		}
1004 		KASSERT(new_object->cache == NULL ||
1005 		    new_object->type == OBJT_SWAP,
1006 		    ("vm_page_cache_transfer: object %p's type is incompatible"
1007 		    " with cached pages", new_object));
1008 	}
1009 	mtx_unlock(&vm_page_queue_free_mtx);
1010 }
1011 
1012 /*
1013  *	vm_page_alloc:
1014  *
1015  *	Allocate and return a memory cell associated
1016  *	with this VM object/offset pair.
1017  *
1018  *	page_req classes:
1019  *	VM_ALLOC_NORMAL		normal process request
1020  *	VM_ALLOC_SYSTEM		system *really* needs a page
1021  *	VM_ALLOC_INTERRUPT	interrupt time request
1022  *	VM_ALLOC_ZERO		zero page
1023  *
1024  *	This routine may not block.
1025  */
1026 vm_page_t
1027 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1028 {
1029 	struct vnode *vp = NULL;
1030 	vm_object_t m_object;
1031 	vm_page_t m;
1032 	int flags, page_req;
1033 
1034 	page_req = req & VM_ALLOC_CLASS_MASK;
1035 	KASSERT(curthread->td_intr_nesting_level == 0 ||
1036 	    page_req == VM_ALLOC_INTERRUPT,
1037 	    ("vm_page_alloc(NORMAL|SYSTEM) in interrupt context"));
1038 
1039 	if ((req & VM_ALLOC_NOOBJ) == 0) {
1040 		KASSERT(object != NULL,
1041 		    ("vm_page_alloc: NULL object."));
1042 		VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1043 	}
1044 
1045 	/*
1046 	 * The pager is allowed to eat deeper into the free page list.
1047 	 */
1048 	if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) {
1049 		page_req = VM_ALLOC_SYSTEM;
1050 	};
1051 
1052 	mtx_lock(&vm_page_queue_free_mtx);
1053 	if (cnt.v_free_count + cnt.v_cache_count > cnt.v_free_reserved ||
1054 	    (page_req == VM_ALLOC_SYSTEM &&
1055 	    cnt.v_free_count + cnt.v_cache_count > cnt.v_interrupt_free_min) ||
1056 	    (page_req == VM_ALLOC_INTERRUPT &&
1057 	    cnt.v_free_count + cnt.v_cache_count > 0)) {
1058 		/*
1059 		 * Allocate from the free queue if the number of free pages
1060 		 * exceeds the minimum for the request class.
1061 		 */
1062 		if (object != NULL &&
1063 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1064 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1065 				mtx_unlock(&vm_page_queue_free_mtx);
1066 				return (NULL);
1067 			}
1068 			if (vm_phys_unfree_page(m))
1069 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1070 #if VM_NRESERVLEVEL > 0
1071 			else if (!vm_reserv_reactivate_page(m))
1072 #else
1073 			else
1074 #endif
1075 				panic("vm_page_alloc: cache page %p is missing"
1076 				    " from the free queue", m);
1077 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1078 			mtx_unlock(&vm_page_queue_free_mtx);
1079 			return (NULL);
1080 #if VM_NRESERVLEVEL > 0
1081 		} else if (object == NULL || object->type == OBJT_DEVICE ||
1082 		    (object->flags & OBJ_COLORED) == 0 ||
1083 		    (m = vm_reserv_alloc_page(object, pindex)) == NULL) {
1084 #else
1085 		} else {
1086 #endif
1087 			m = vm_phys_alloc_pages(object != NULL ?
1088 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1089 #if VM_NRESERVLEVEL > 0
1090 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1091 				m = vm_phys_alloc_pages(object != NULL ?
1092 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1093 				    0);
1094 			}
1095 #endif
1096 		}
1097 	} else {
1098 		/*
1099 		 * Not allocatable, give up.
1100 		 */
1101 		mtx_unlock(&vm_page_queue_free_mtx);
1102 		atomic_add_int(&vm_pageout_deficit, 1);
1103 		pagedaemon_wakeup();
1104 		return (NULL);
1105 	}
1106 
1107 	/*
1108 	 *  At this point we had better have found a good page.
1109 	 */
1110 
1111 	KASSERT(
1112 	    m != NULL,
1113 	    ("vm_page_alloc(): missing page on free queue")
1114 	);
1115 	if ((m->flags & PG_CACHED) != 0) {
1116 		KASSERT(m->valid != 0,
1117 		    ("vm_page_alloc: cached page %p is invalid", m));
1118 		if (m->object == object && m->pindex == pindex)
1119 	  		cnt.v_reactivated++;
1120 		else
1121 			m->valid = 0;
1122 		m_object = m->object;
1123 		vm_page_cache_remove(m);
1124 		if (m_object->type == OBJT_VNODE && m_object->cache == NULL)
1125 			vp = m_object->handle;
1126 	} else {
1127 		KASSERT(VM_PAGE_IS_FREE(m),
1128 		    ("vm_page_alloc: page %p is not free", m));
1129 		KASSERT(m->valid == 0,
1130 		    ("vm_page_alloc: free page %p is valid", m));
1131 		cnt.v_free_count--;
1132 	}
1133 
1134 	/*
1135 	 * Initialize structure.  Only the PG_ZERO flag is inherited.
1136 	 */
1137 	flags = 0;
1138 	if (m->flags & PG_ZERO) {
1139 		vm_page_zero_count--;
1140 		if (req & VM_ALLOC_ZERO)
1141 			flags = PG_ZERO;
1142 	}
1143 	if (object == NULL || object->type == OBJT_PHYS)
1144 		flags |= PG_UNMANAGED;
1145 	m->flags = flags;
1146 	if (req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ))
1147 		m->oflags = 0;
1148 	else
1149 		m->oflags = VPO_BUSY;
1150 	if (req & VM_ALLOC_WIRED) {
1151 		atomic_add_int(&cnt.v_wire_count, 1);
1152 		m->wire_count = 1;
1153 	} else
1154 		m->wire_count = 0;
1155 	m->hold_count = 0;
1156 	m->act_count = 0;
1157 	m->busy = 0;
1158 	KASSERT(m->dirty == 0, ("vm_page_alloc: free/cache page %p was dirty", m));
1159 	mtx_unlock(&vm_page_queue_free_mtx);
1160 
1161 	if ((req & VM_ALLOC_NOOBJ) == 0)
1162 		vm_page_insert(m, object, pindex);
1163 	else
1164 		m->pindex = pindex;
1165 
1166 	/*
1167 	 * The following call to vdrop() must come after the above call
1168 	 * to vm_page_insert() in case both affect the same object and
1169 	 * vnode.  Otherwise, the affected vnode's hold count could
1170 	 * temporarily become zero.
1171 	 */
1172 	if (vp != NULL)
1173 		vdrop(vp);
1174 
1175 	/*
1176 	 * Don't wakeup too often - wakeup the pageout daemon when
1177 	 * we would be nearly out of memory.
1178 	 */
1179 	if (vm_paging_needed())
1180 		pagedaemon_wakeup();
1181 
1182 	return (m);
1183 }
1184 
1185 /*
1186  *	vm_wait:	(also see VM_WAIT macro)
1187  *
1188  *	Block until free pages are available for allocation
1189  *	- Called in various places before memory allocations.
1190  */
1191 void
1192 vm_wait(void)
1193 {
1194 
1195 	mtx_lock(&vm_page_queue_free_mtx);
1196 	if (curproc == pageproc) {
1197 		vm_pageout_pages_needed = 1;
1198 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
1199 		    PDROP | PSWP, "VMWait", 0);
1200 	} else {
1201 		if (!vm_pages_needed) {
1202 			vm_pages_needed = 1;
1203 			wakeup(&vm_pages_needed);
1204 		}
1205 		msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
1206 		    "vmwait", 0);
1207 	}
1208 }
1209 
1210 /*
1211  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
1212  *
1213  *	Block until free pages are available for allocation
1214  *	- Called only in vm_fault so that processes page faulting
1215  *	  can be easily tracked.
1216  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
1217  *	  processes will be able to grab memory first.  Do not change
1218  *	  this balance without careful testing first.
1219  */
1220 void
1221 vm_waitpfault(void)
1222 {
1223 
1224 	mtx_lock(&vm_page_queue_free_mtx);
1225 	if (!vm_pages_needed) {
1226 		vm_pages_needed = 1;
1227 		wakeup(&vm_pages_needed);
1228 	}
1229 	msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
1230 	    "pfault", 0);
1231 }
1232 
1233 /*
1234  *	vm_page_requeue:
1235  *
1236  *	If the given page is contained within a page queue, move it to the tail
1237  *	of that queue.
1238  *
1239  *	The page queues must be locked.
1240  */
1241 void
1242 vm_page_requeue(vm_page_t m)
1243 {
1244 	int queue = VM_PAGE_GETQUEUE(m);
1245 	struct vpgqueues *vpq;
1246 
1247 	if (queue != PQ_NONE) {
1248 		vpq = &vm_page_queues[queue];
1249 		TAILQ_REMOVE(&vpq->pl, m, pageq);
1250 		TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
1251 	}
1252 }
1253 
1254 /*
1255  *	vm_pageq_remove:
1256  *
1257  *	Remove a page from its queue.
1258  *
1259  *	The queue containing the given page must be locked.
1260  *	This routine may not block.
1261  */
1262 void
1263 vm_pageq_remove(vm_page_t m)
1264 {
1265 	int queue = VM_PAGE_GETQUEUE(m);
1266 	struct vpgqueues *pq;
1267 
1268 	if (queue != PQ_NONE) {
1269 		VM_PAGE_SETQUEUE2(m, PQ_NONE);
1270 		pq = &vm_page_queues[queue];
1271 		TAILQ_REMOVE(&pq->pl, m, pageq);
1272 		(*pq->cnt)--;
1273 	}
1274 }
1275 
1276 /*
1277  *	vm_page_enqueue:
1278  *
1279  *	Add the given page to the specified queue.
1280  *
1281  *	The page queues must be locked.
1282  */
1283 static void
1284 vm_page_enqueue(int queue, vm_page_t m)
1285 {
1286 	struct vpgqueues *vpq;
1287 
1288 	vpq = &vm_page_queues[queue];
1289 	VM_PAGE_SETQUEUE2(m, queue);
1290 	TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
1291 	++*vpq->cnt;
1292 }
1293 
1294 /*
1295  *	vm_page_activate:
1296  *
1297  *	Put the specified page on the active list (if appropriate).
1298  *	Ensure that act_count is at least ACT_INIT but do not otherwise
1299  *	mess with it.
1300  *
1301  *	The page queues must be locked.
1302  *	This routine may not block.
1303  */
1304 void
1305 vm_page_activate(vm_page_t m)
1306 {
1307 
1308 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1309 	if (VM_PAGE_GETKNOWNQUEUE2(m) != PQ_ACTIVE) {
1310 		vm_pageq_remove(m);
1311 		if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1312 			if (m->act_count < ACT_INIT)
1313 				m->act_count = ACT_INIT;
1314 			vm_page_enqueue(PQ_ACTIVE, m);
1315 		}
1316 	} else {
1317 		if (m->act_count < ACT_INIT)
1318 			m->act_count = ACT_INIT;
1319 	}
1320 }
1321 
1322 /*
1323  *	vm_page_free_wakeup:
1324  *
1325  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
1326  *	routine is called when a page has been added to the cache or free
1327  *	queues.
1328  *
1329  *	The page queues must be locked.
1330  *	This routine may not block.
1331  */
1332 static inline void
1333 vm_page_free_wakeup(void)
1334 {
1335 
1336 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1337 	/*
1338 	 * if pageout daemon needs pages, then tell it that there are
1339 	 * some free.
1340 	 */
1341 	if (vm_pageout_pages_needed &&
1342 	    cnt.v_cache_count + cnt.v_free_count >= cnt.v_pageout_free_min) {
1343 		wakeup(&vm_pageout_pages_needed);
1344 		vm_pageout_pages_needed = 0;
1345 	}
1346 	/*
1347 	 * wakeup processes that are waiting on memory if we hit a
1348 	 * high water mark. And wakeup scheduler process if we have
1349 	 * lots of memory. this process will swapin processes.
1350 	 */
1351 	if (vm_pages_needed && !vm_page_count_min()) {
1352 		vm_pages_needed = 0;
1353 		wakeup(&cnt.v_free_count);
1354 	}
1355 }
1356 
1357 /*
1358  *	vm_page_free_toq:
1359  *
1360  *	Returns the given page to the free list,
1361  *	disassociating it with any VM object.
1362  *
1363  *	Object and page must be locked prior to entry.
1364  *	This routine may not block.
1365  */
1366 
1367 void
1368 vm_page_free_toq(vm_page_t m)
1369 {
1370 
1371 	if (VM_PAGE_GETQUEUE(m) != PQ_NONE)
1372 		mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1373 	KASSERT(!pmap_page_is_mapped(m),
1374 	    ("vm_page_free_toq: freeing mapped page %p", m));
1375 	PCPU_INC(cnt.v_tfree);
1376 
1377 	if (m->busy || VM_PAGE_IS_FREE(m)) {
1378 		printf(
1379 		"vm_page_free: pindex(%lu), busy(%d), VPO_BUSY(%d), hold(%d)\n",
1380 		    (u_long)m->pindex, m->busy, (m->oflags & VPO_BUSY) ? 1 : 0,
1381 		    m->hold_count);
1382 		if (VM_PAGE_IS_FREE(m))
1383 			panic("vm_page_free: freeing free page");
1384 		else
1385 			panic("vm_page_free: freeing busy page");
1386 	}
1387 
1388 	/*
1389 	 * unqueue, then remove page.  Note that we cannot destroy
1390 	 * the page here because we do not want to call the pager's
1391 	 * callback routine until after we've put the page on the
1392 	 * appropriate free queue.
1393 	 */
1394 	vm_pageq_remove(m);
1395 	vm_page_remove(m);
1396 
1397 	/*
1398 	 * If fictitious remove object association and
1399 	 * return, otherwise delay object association removal.
1400 	 */
1401 	if ((m->flags & PG_FICTITIOUS) != 0) {
1402 		return;
1403 	}
1404 
1405 	m->valid = 0;
1406 	vm_page_undirty(m);
1407 
1408 	if (m->wire_count != 0) {
1409 		if (m->wire_count > 1) {
1410 			panic("vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1411 				m->wire_count, (long)m->pindex);
1412 		}
1413 		panic("vm_page_free: freeing wired page");
1414 	}
1415 	if (m->hold_count != 0) {
1416 		m->flags &= ~PG_ZERO;
1417 		vm_page_enqueue(PQ_HOLD, m);
1418 	} else {
1419 		mtx_lock(&vm_page_queue_free_mtx);
1420 		m->flags |= PG_FREE;
1421 		cnt.v_free_count++;
1422 #if VM_NRESERVLEVEL > 0
1423 		if (!vm_reserv_free_page(m))
1424 #else
1425 		if (TRUE)
1426 #endif
1427 			vm_phys_free_pages(m, 0);
1428 		if ((m->flags & PG_ZERO) != 0)
1429 			++vm_page_zero_count;
1430 		else
1431 			vm_page_zero_idle_wakeup();
1432 		vm_page_free_wakeup();
1433 		mtx_unlock(&vm_page_queue_free_mtx);
1434 	}
1435 }
1436 
1437 /*
1438  *	vm_page_wire:
1439  *
1440  *	Mark this page as wired down by yet
1441  *	another map, removing it from paging queues
1442  *	as necessary.
1443  *
1444  *	The page queues must be locked.
1445  *	This routine may not block.
1446  */
1447 void
1448 vm_page_wire(vm_page_t m)
1449 {
1450 
1451 	/*
1452 	 * Only bump the wire statistics if the page is not already wired,
1453 	 * and only unqueue the page if it is on some queue (if it is unmanaged
1454 	 * it is already off the queues).
1455 	 */
1456 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1457 	if (m->flags & PG_FICTITIOUS)
1458 		return;
1459 	if (m->wire_count == 0) {
1460 		if ((m->flags & PG_UNMANAGED) == 0)
1461 			vm_pageq_remove(m);
1462 		atomic_add_int(&cnt.v_wire_count, 1);
1463 	}
1464 	m->wire_count++;
1465 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
1466 }
1467 
1468 /*
1469  *	vm_page_unwire:
1470  *
1471  *	Release one wiring of this page, potentially
1472  *	enabling it to be paged again.
1473  *
1474  *	Many pages placed on the inactive queue should actually go
1475  *	into the cache, but it is difficult to figure out which.  What
1476  *	we do instead, if the inactive target is well met, is to put
1477  *	clean pages at the head of the inactive queue instead of the tail.
1478  *	This will cause them to be moved to the cache more quickly and
1479  *	if not actively re-referenced, freed more quickly.  If we just
1480  *	stick these pages at the end of the inactive queue, heavy filesystem
1481  *	meta-data accesses can cause an unnecessary paging load on memory bound
1482  *	processes.  This optimization causes one-time-use metadata to be
1483  *	reused more quickly.
1484  *
1485  *	BUT, if we are in a low-memory situation we have no choice but to
1486  *	put clean pages on the cache queue.
1487  *
1488  *	A number of routines use vm_page_unwire() to guarantee that the page
1489  *	will go into either the inactive or active queues, and will NEVER
1490  *	be placed in the cache - for example, just after dirtying a page.
1491  *	dirty pages in the cache are not allowed.
1492  *
1493  *	The page queues must be locked.
1494  *	This routine may not block.
1495  */
1496 void
1497 vm_page_unwire(vm_page_t m, int activate)
1498 {
1499 
1500 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1501 	if (m->flags & PG_FICTITIOUS)
1502 		return;
1503 	if (m->wire_count > 0) {
1504 		m->wire_count--;
1505 		if (m->wire_count == 0) {
1506 			atomic_subtract_int(&cnt.v_wire_count, 1);
1507 			if (m->flags & PG_UNMANAGED) {
1508 				;
1509 			} else if (activate)
1510 				vm_page_enqueue(PQ_ACTIVE, m);
1511 			else {
1512 				vm_page_flag_clear(m, PG_WINATCFLS);
1513 				vm_page_enqueue(PQ_INACTIVE, m);
1514 			}
1515 		}
1516 	} else {
1517 		panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
1518 	}
1519 }
1520 
1521 
1522 /*
1523  * Move the specified page to the inactive queue.  If the page has
1524  * any associated swap, the swap is deallocated.
1525  *
1526  * Normally athead is 0 resulting in LRU operation.  athead is set
1527  * to 1 if we want this page to be 'as if it were placed in the cache',
1528  * except without unmapping it from the process address space.
1529  *
1530  * This routine may not block.
1531  */
1532 static inline void
1533 _vm_page_deactivate(vm_page_t m, int athead)
1534 {
1535 
1536 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1537 
1538 	/*
1539 	 * Ignore if already inactive.
1540 	 */
1541 	if (VM_PAGE_INQUEUE2(m, PQ_INACTIVE))
1542 		return;
1543 	if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1544 		vm_page_flag_clear(m, PG_WINATCFLS);
1545 		vm_pageq_remove(m);
1546 		if (athead)
1547 			TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl, m, pageq);
1548 		else
1549 			TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl, m, pageq);
1550 		VM_PAGE_SETQUEUE2(m, PQ_INACTIVE);
1551 		cnt.v_inactive_count++;
1552 	}
1553 }
1554 
1555 void
1556 vm_page_deactivate(vm_page_t m)
1557 {
1558     _vm_page_deactivate(m, 0);
1559 }
1560 
1561 /*
1562  * vm_page_try_to_cache:
1563  *
1564  * Returns 0 on failure, 1 on success
1565  */
1566 int
1567 vm_page_try_to_cache(vm_page_t m)
1568 {
1569 
1570 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1571 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1572 	if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1573 	    (m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) {
1574 		return (0);
1575 	}
1576 	pmap_remove_all(m);
1577 	if (m->dirty)
1578 		return (0);
1579 	vm_page_cache(m);
1580 	return (1);
1581 }
1582 
1583 /*
1584  * vm_page_try_to_free()
1585  *
1586  *	Attempt to free the page.  If we cannot free it, we do nothing.
1587  *	1 is returned on success, 0 on failure.
1588  */
1589 int
1590 vm_page_try_to_free(vm_page_t m)
1591 {
1592 
1593 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1594 	if (m->object != NULL)
1595 		VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1596 	if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1597 	    (m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) {
1598 		return (0);
1599 	}
1600 	pmap_remove_all(m);
1601 	if (m->dirty)
1602 		return (0);
1603 	vm_page_free(m);
1604 	return (1);
1605 }
1606 
1607 /*
1608  * vm_page_cache
1609  *
1610  * Put the specified page onto the page cache queue (if appropriate).
1611  *
1612  * This routine may not block.
1613  */
1614 void
1615 vm_page_cache(vm_page_t m)
1616 {
1617 	vm_object_t object;
1618 	vm_page_t root;
1619 
1620 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1621 	object = m->object;
1622 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1623 	if ((m->flags & PG_UNMANAGED) || (m->oflags & VPO_BUSY) || m->busy ||
1624 	    m->hold_count || m->wire_count) {
1625 		panic("vm_page_cache: attempting to cache busy page");
1626 	}
1627 	pmap_remove_all(m);
1628 	if (m->dirty != 0)
1629 		panic("vm_page_cache: page %p is dirty", m);
1630 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
1631 	    (object->type == OBJT_SWAP &&
1632 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
1633 		/*
1634 		 * Hypothesis: A cache-elgible page belonging to a
1635 		 * default object or swap object but without a backing
1636 		 * store must be zero filled.
1637 		 */
1638 		vm_page_free(m);
1639 		return;
1640 	}
1641 	KASSERT((m->flags & PG_CACHED) == 0,
1642 	    ("vm_page_cache: page %p is already cached", m));
1643 	cnt.v_tcached++;
1644 
1645 	/*
1646 	 * Remove the page from the paging queues.
1647 	 */
1648 	vm_pageq_remove(m);
1649 
1650 	/*
1651 	 * Remove the page from the object's collection of resident
1652 	 * pages.
1653 	 */
1654 	if (m != object->root)
1655 		vm_page_splay(m->pindex, object->root);
1656 	if (m->left == NULL)
1657 		root = m->right;
1658 	else {
1659 		root = vm_page_splay(m->pindex, m->left);
1660 		root->right = m->right;
1661 	}
1662 	object->root = root;
1663 	TAILQ_REMOVE(&object->memq, m, listq);
1664 	object->resident_page_count--;
1665 	object->generation++;
1666 
1667 	/*
1668 	 * Insert the page into the object's collection of cached pages
1669 	 * and the physical memory allocator's cache/free page queues.
1670 	 */
1671 	vm_page_flag_clear(m, PG_ZERO);
1672 	mtx_lock(&vm_page_queue_free_mtx);
1673 	m->flags |= PG_CACHED;
1674 	cnt.v_cache_count++;
1675 	root = object->cache;
1676 	if (root == NULL) {
1677 		m->left = NULL;
1678 		m->right = NULL;
1679 	} else {
1680 		root = vm_page_splay(m->pindex, root);
1681 		if (m->pindex < root->pindex) {
1682 			m->left = root->left;
1683 			m->right = root;
1684 			root->left = NULL;
1685 		} else if (__predict_false(m->pindex == root->pindex))
1686 			panic("vm_page_cache: offset already cached");
1687 		else {
1688 			m->right = root->right;
1689 			m->left = root;
1690 			root->right = NULL;
1691 		}
1692 	}
1693 	object->cache = m;
1694 #if VM_NRESERVLEVEL > 0
1695 	if (!vm_reserv_free_page(m)) {
1696 #else
1697 	if (TRUE) {
1698 #endif
1699 		vm_phys_set_pool(VM_FREEPOOL_CACHE, m, 0);
1700 		vm_phys_free_pages(m, 0);
1701 	}
1702 	vm_page_free_wakeup();
1703 	mtx_unlock(&vm_page_queue_free_mtx);
1704 
1705 	/*
1706 	 * Increment the vnode's hold count if this is the object's only
1707 	 * cached page.  Decrement the vnode's hold count if this was
1708 	 * the object's only resident page.
1709 	 */
1710 	if (object->type == OBJT_VNODE) {
1711 		if (root == NULL && object->resident_page_count != 0)
1712 			vhold(object->handle);
1713 		else if (root != NULL && object->resident_page_count == 0)
1714 			vdrop(object->handle);
1715 	}
1716 }
1717 
1718 /*
1719  * vm_page_dontneed
1720  *
1721  *	Cache, deactivate, or do nothing as appropriate.  This routine
1722  *	is typically used by madvise() MADV_DONTNEED.
1723  *
1724  *	Generally speaking we want to move the page into the cache so
1725  *	it gets reused quickly.  However, this can result in a silly syndrome
1726  *	due to the page recycling too quickly.  Small objects will not be
1727  *	fully cached.  On the otherhand, if we move the page to the inactive
1728  *	queue we wind up with a problem whereby very large objects
1729  *	unnecessarily blow away our inactive and cache queues.
1730  *
1731  *	The solution is to move the pages based on a fixed weighting.  We
1732  *	either leave them alone, deactivate them, or move them to the cache,
1733  *	where moving them to the cache has the highest weighting.
1734  *	By forcing some pages into other queues we eventually force the
1735  *	system to balance the queues, potentially recovering other unrelated
1736  *	space from active.  The idea is to not force this to happen too
1737  *	often.
1738  */
1739 void
1740 vm_page_dontneed(vm_page_t m)
1741 {
1742 	static int dnweight;
1743 	int dnw;
1744 	int head;
1745 
1746 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1747 	dnw = ++dnweight;
1748 
1749 	/*
1750 	 * occassionally leave the page alone
1751 	 */
1752 	if ((dnw & 0x01F0) == 0 ||
1753 	    VM_PAGE_INQUEUE2(m, PQ_INACTIVE)) {
1754 		if (m->act_count >= ACT_INIT)
1755 			--m->act_count;
1756 		return;
1757 	}
1758 
1759 	/*
1760 	 * Clear any references to the page.  Otherwise, the page daemon will
1761 	 * immediately reactivate the page.
1762 	 */
1763 	vm_page_flag_clear(m, PG_REFERENCED);
1764 	pmap_clear_reference(m);
1765 
1766 	if (m->dirty == 0 && pmap_is_modified(m))
1767 		vm_page_dirty(m);
1768 
1769 	if (m->dirty || (dnw & 0x0070) == 0) {
1770 		/*
1771 		 * Deactivate the page 3 times out of 32.
1772 		 */
1773 		head = 0;
1774 	} else {
1775 		/*
1776 		 * Cache the page 28 times out of every 32.  Note that
1777 		 * the page is deactivated instead of cached, but placed
1778 		 * at the head of the queue instead of the tail.
1779 		 */
1780 		head = 1;
1781 	}
1782 	_vm_page_deactivate(m, head);
1783 }
1784 
1785 /*
1786  * Grab a page, waiting until we are waken up due to the page
1787  * changing state.  We keep on waiting, if the page continues
1788  * to be in the object.  If the page doesn't exist, first allocate it
1789  * and then conditionally zero it.
1790  *
1791  * This routine may block.
1792  */
1793 vm_page_t
1794 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
1795 {
1796 	vm_page_t m;
1797 
1798 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1799 retrylookup:
1800 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
1801 		if (vm_page_sleep_if_busy(m, TRUE, "pgrbwt")) {
1802 			if ((allocflags & VM_ALLOC_RETRY) == 0)
1803 				return (NULL);
1804 			goto retrylookup;
1805 		} else {
1806 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
1807 				vm_page_lock_queues();
1808 				vm_page_wire(m);
1809 				vm_page_unlock_queues();
1810 			}
1811 			if ((allocflags & VM_ALLOC_NOBUSY) == 0)
1812 				vm_page_busy(m);
1813 			return (m);
1814 		}
1815 	}
1816 	m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
1817 	if (m == NULL) {
1818 		VM_OBJECT_UNLOCK(object);
1819 		VM_WAIT;
1820 		VM_OBJECT_LOCK(object);
1821 		if ((allocflags & VM_ALLOC_RETRY) == 0)
1822 			return (NULL);
1823 		goto retrylookup;
1824 	} else if (m->valid != 0)
1825 		return (m);
1826 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
1827 		pmap_zero_page(m);
1828 	return (m);
1829 }
1830 
1831 /*
1832  * Mapping function for valid bits or for dirty bits in
1833  * a page.  May not block.
1834  *
1835  * Inputs are required to range within a page.
1836  */
1837 int
1838 vm_page_bits(int base, int size)
1839 {
1840 	int first_bit;
1841 	int last_bit;
1842 
1843 	KASSERT(
1844 	    base + size <= PAGE_SIZE,
1845 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
1846 	);
1847 
1848 	if (size == 0)		/* handle degenerate case */
1849 		return (0);
1850 
1851 	first_bit = base >> DEV_BSHIFT;
1852 	last_bit = (base + size - 1) >> DEV_BSHIFT;
1853 
1854 	return ((2 << last_bit) - (1 << first_bit));
1855 }
1856 
1857 /*
1858  *	vm_page_set_valid:
1859  *
1860  *	Sets portions of a page valid.  The arguments are expected
1861  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1862  *	of any partial chunks touched by the range.  The invalid portion of
1863  *	such chunks will be zeroed.
1864  *
1865  *	(base + size) must be less then or equal to PAGE_SIZE.
1866  */
1867 void
1868 vm_page_set_valid(vm_page_t m, int base, int size)
1869 {
1870 	int endoff, frag;
1871 
1872 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1873 	if (size == 0)	/* handle degenerate case */
1874 		return;
1875 
1876 	/*
1877 	 * If the base is not DEV_BSIZE aligned and the valid
1878 	 * bit is clear, we have to zero out a portion of the
1879 	 * first block.
1880 	 */
1881 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1882 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
1883 		pmap_zero_page_area(m, frag, base - frag);
1884 
1885 	/*
1886 	 * If the ending offset is not DEV_BSIZE aligned and the
1887 	 * valid bit is clear, we have to zero out a portion of
1888 	 * the last block.
1889 	 */
1890 	endoff = base + size;
1891 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1892 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
1893 		pmap_zero_page_area(m, endoff,
1894 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
1895 
1896 	/*
1897 	 * Assert that no previously invalid block that is now being validated
1898 	 * is already dirty.
1899 	 */
1900 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
1901 	    ("vm_page_set_valid: page %p is dirty", m));
1902 
1903 	/*
1904 	 * Set valid bits inclusive of any overlap.
1905 	 */
1906 	m->valid |= vm_page_bits(base, size);
1907 }
1908 
1909 /*
1910  *	vm_page_set_validclean:
1911  *
1912  *	Sets portions of a page valid and clean.  The arguments are expected
1913  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1914  *	of any partial chunks touched by the range.  The invalid portion of
1915  *	such chunks will be zero'd.
1916  *
1917  *	This routine may not block.
1918  *
1919  *	(base + size) must be less then or equal to PAGE_SIZE.
1920  */
1921 void
1922 vm_page_set_validclean(vm_page_t m, int base, int size)
1923 {
1924 	int pagebits;
1925 	int frag;
1926 	int endoff;
1927 
1928 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1929 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1930 	if (size == 0)	/* handle degenerate case */
1931 		return;
1932 
1933 	/*
1934 	 * If the base is not DEV_BSIZE aligned and the valid
1935 	 * bit is clear, we have to zero out a portion of the
1936 	 * first block.
1937 	 */
1938 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1939 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
1940 		pmap_zero_page_area(m, frag, base - frag);
1941 
1942 	/*
1943 	 * If the ending offset is not DEV_BSIZE aligned and the
1944 	 * valid bit is clear, we have to zero out a portion of
1945 	 * the last block.
1946 	 */
1947 	endoff = base + size;
1948 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1949 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
1950 		pmap_zero_page_area(m, endoff,
1951 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
1952 
1953 	/*
1954 	 * Set valid, clear dirty bits.  If validating the entire
1955 	 * page we can safely clear the pmap modify bit.  We also
1956 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
1957 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
1958 	 * be set again.
1959 	 *
1960 	 * We set valid bits inclusive of any overlap, but we can only
1961 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
1962 	 * the range.
1963 	 */
1964 	pagebits = vm_page_bits(base, size);
1965 	m->valid |= pagebits;
1966 #if 0	/* NOT YET */
1967 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
1968 		frag = DEV_BSIZE - frag;
1969 		base += frag;
1970 		size -= frag;
1971 		if (size < 0)
1972 			size = 0;
1973 	}
1974 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
1975 #endif
1976 	m->dirty &= ~pagebits;
1977 	if (base == 0 && size == PAGE_SIZE) {
1978 		pmap_clear_modify(m);
1979 		m->oflags &= ~VPO_NOSYNC;
1980 	}
1981 }
1982 
1983 void
1984 vm_page_clear_dirty(vm_page_t m, int base, int size)
1985 {
1986 
1987 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1988 	m->dirty &= ~vm_page_bits(base, size);
1989 }
1990 
1991 /*
1992  *	vm_page_set_invalid:
1993  *
1994  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
1995  *	valid and dirty bits for the effected areas are cleared.
1996  *
1997  *	May not block.
1998  */
1999 void
2000 vm_page_set_invalid(vm_page_t m, int base, int size)
2001 {
2002 	int bits;
2003 
2004 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2005 	bits = vm_page_bits(base, size);
2006 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2007 	if (m->valid == VM_PAGE_BITS_ALL && bits != 0)
2008 		pmap_remove_all(m);
2009 	m->valid &= ~bits;
2010 	m->dirty &= ~bits;
2011 	m->object->generation++;
2012 }
2013 
2014 /*
2015  * vm_page_zero_invalid()
2016  *
2017  *	The kernel assumes that the invalid portions of a page contain
2018  *	garbage, but such pages can be mapped into memory by user code.
2019  *	When this occurs, we must zero out the non-valid portions of the
2020  *	page so user code sees what it expects.
2021  *
2022  *	Pages are most often semi-valid when the end of a file is mapped
2023  *	into memory and the file's size is not page aligned.
2024  */
2025 void
2026 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
2027 {
2028 	int b;
2029 	int i;
2030 
2031 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2032 	/*
2033 	 * Scan the valid bits looking for invalid sections that
2034 	 * must be zerod.  Invalid sub-DEV_BSIZE'd areas ( where the
2035 	 * valid bit may be set ) have already been zerod by
2036 	 * vm_page_set_validclean().
2037 	 */
2038 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
2039 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
2040 		    (m->valid & (1 << i))
2041 		) {
2042 			if (i > b) {
2043 				pmap_zero_page_area(m,
2044 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
2045 			}
2046 			b = i + 1;
2047 		}
2048 	}
2049 
2050 	/*
2051 	 * setvalid is TRUE when we can safely set the zero'd areas
2052 	 * as being valid.  We can do this if there are no cache consistancy
2053 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
2054 	 */
2055 	if (setvalid)
2056 		m->valid = VM_PAGE_BITS_ALL;
2057 }
2058 
2059 /*
2060  *	vm_page_is_valid:
2061  *
2062  *	Is (partial) page valid?  Note that the case where size == 0
2063  *	will return FALSE in the degenerate case where the page is
2064  *	entirely invalid, and TRUE otherwise.
2065  *
2066  *	May not block.
2067  */
2068 int
2069 vm_page_is_valid(vm_page_t m, int base, int size)
2070 {
2071 	int bits = vm_page_bits(base, size);
2072 
2073 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2074 	if (m->valid && ((m->valid & bits) == bits))
2075 		return 1;
2076 	else
2077 		return 0;
2078 }
2079 
2080 /*
2081  * update dirty bits from pmap/mmu.  May not block.
2082  */
2083 void
2084 vm_page_test_dirty(vm_page_t m)
2085 {
2086 	if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
2087 		vm_page_dirty(m);
2088 	}
2089 }
2090 
2091 int so_zerocp_fullpage = 0;
2092 
2093 /*
2094  *	Replace the given page with a copy.  The copied page assumes
2095  *	the portion of the given page's "wire_count" that is not the
2096  *	responsibility of this copy-on-write mechanism.
2097  *
2098  *	The object containing the given page must have a non-zero
2099  *	paging-in-progress count and be locked.
2100  */
2101 void
2102 vm_page_cowfault(vm_page_t m)
2103 {
2104 	vm_page_t mnew;
2105 	vm_object_t object;
2106 	vm_pindex_t pindex;
2107 
2108 	object = m->object;
2109 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2110 	KASSERT(object->paging_in_progress != 0,
2111 	    ("vm_page_cowfault: object %p's paging-in-progress count is zero.",
2112 	    object));
2113 	pindex = m->pindex;
2114 
2115  retry_alloc:
2116 	pmap_remove_all(m);
2117 	vm_page_remove(m);
2118 	mnew = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
2119 	if (mnew == NULL) {
2120 		vm_page_insert(m, object, pindex);
2121 		vm_page_unlock_queues();
2122 		VM_OBJECT_UNLOCK(object);
2123 		VM_WAIT;
2124 		VM_OBJECT_LOCK(object);
2125 		if (m == vm_page_lookup(object, pindex)) {
2126 			vm_page_lock_queues();
2127 			goto retry_alloc;
2128 		} else {
2129 			/*
2130 			 * Page disappeared during the wait.
2131 			 */
2132 			vm_page_lock_queues();
2133 			return;
2134 		}
2135 	}
2136 
2137 	if (m->cow == 0) {
2138 		/*
2139 		 * check to see if we raced with an xmit complete when
2140 		 * waiting to allocate a page.  If so, put things back
2141 		 * the way they were
2142 		 */
2143 		vm_page_free(mnew);
2144 		vm_page_insert(m, object, pindex);
2145 	} else { /* clear COW & copy page */
2146 		if (!so_zerocp_fullpage)
2147 			pmap_copy_page(m, mnew);
2148 		mnew->valid = VM_PAGE_BITS_ALL;
2149 		vm_page_dirty(mnew);
2150 		mnew->wire_count = m->wire_count - m->cow;
2151 		m->wire_count = m->cow;
2152 	}
2153 }
2154 
2155 void
2156 vm_page_cowclear(vm_page_t m)
2157 {
2158 
2159 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2160 	if (m->cow) {
2161 		m->cow--;
2162 		/*
2163 		 * let vm_fault add back write permission  lazily
2164 		 */
2165 	}
2166 	/*
2167 	 *  sf_buf_free() will free the page, so we needn't do it here
2168 	 */
2169 }
2170 
2171 int
2172 vm_page_cowsetup(vm_page_t m)
2173 {
2174 
2175 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2176 	if (m->cow == USHRT_MAX - 1)
2177 		return (EBUSY);
2178 	m->cow++;
2179 	pmap_remove_write(m);
2180 	return (0);
2181 }
2182 
2183 #include "opt_ddb.h"
2184 #ifdef DDB
2185 #include <sys/kernel.h>
2186 
2187 #include <ddb/ddb.h>
2188 
2189 DB_SHOW_COMMAND(page, vm_page_print_page_info)
2190 {
2191 	db_printf("cnt.v_free_count: %d\n", cnt.v_free_count);
2192 	db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count);
2193 	db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count);
2194 	db_printf("cnt.v_active_count: %d\n", cnt.v_active_count);
2195 	db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count);
2196 	db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved);
2197 	db_printf("cnt.v_free_min: %d\n", cnt.v_free_min);
2198 	db_printf("cnt.v_free_target: %d\n", cnt.v_free_target);
2199 	db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min);
2200 	db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target);
2201 }
2202 
2203 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
2204 {
2205 
2206 	db_printf("PQ_FREE:");
2207 	db_printf(" %d", cnt.v_free_count);
2208 	db_printf("\n");
2209 
2210 	db_printf("PQ_CACHE:");
2211 	db_printf(" %d", cnt.v_cache_count);
2212 	db_printf("\n");
2213 
2214 	db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n",
2215 		*vm_page_queues[PQ_ACTIVE].cnt,
2216 		*vm_page_queues[PQ_INACTIVE].cnt);
2217 }
2218 #endif /* DDB */
2219