xref: /freebsd/sys/vm/vm_page.c (revision d8b878873e7aa8df1972cc6a642804b17eb61087)
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 vpglocks vm_page_queue_lock;
139 struct vpglocks vm_page_queue_free_lock;
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  *	VM_ALLOC_WIRED		wire the allocated page
1024  *	VM_ALLOC_NOOBJ		page is not associated with a vm object
1025  *	VM_ALLOC_NOBUSY		do not set the page busy
1026  *	VM_ALLOC_IFNOTCACHED	return NULL, do not reactivate if the page
1027  *				is cached
1028  *
1029  *	This routine may not sleep.
1030  */
1031 vm_page_t
1032 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1033 {
1034 	struct vnode *vp = NULL;
1035 	vm_object_t m_object;
1036 	vm_page_t m;
1037 	int flags, page_req;
1038 
1039 	page_req = req & VM_ALLOC_CLASS_MASK;
1040 	KASSERT(curthread->td_intr_nesting_level == 0 ||
1041 	    page_req == VM_ALLOC_INTERRUPT,
1042 	    ("vm_page_alloc(NORMAL|SYSTEM) in interrupt context"));
1043 
1044 	if ((req & VM_ALLOC_NOOBJ) == 0) {
1045 		KASSERT(object != NULL,
1046 		    ("vm_page_alloc: NULL object."));
1047 		VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1048 	}
1049 
1050 	/*
1051 	 * The pager is allowed to eat deeper into the free page list.
1052 	 */
1053 	if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) {
1054 		page_req = VM_ALLOC_SYSTEM;
1055 	};
1056 
1057 	mtx_lock(&vm_page_queue_free_mtx);
1058 	if (cnt.v_free_count + cnt.v_cache_count > cnt.v_free_reserved ||
1059 	    (page_req == VM_ALLOC_SYSTEM &&
1060 	    cnt.v_free_count + cnt.v_cache_count > cnt.v_interrupt_free_min) ||
1061 	    (page_req == VM_ALLOC_INTERRUPT &&
1062 	    cnt.v_free_count + cnt.v_cache_count > 0)) {
1063 		/*
1064 		 * Allocate from the free queue if the number of free pages
1065 		 * exceeds the minimum for the request class.
1066 		 */
1067 		if (object != NULL &&
1068 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1069 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1070 				mtx_unlock(&vm_page_queue_free_mtx);
1071 				return (NULL);
1072 			}
1073 			if (vm_phys_unfree_page(m))
1074 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1075 #if VM_NRESERVLEVEL > 0
1076 			else if (!vm_reserv_reactivate_page(m))
1077 #else
1078 			else
1079 #endif
1080 				panic("vm_page_alloc: cache page %p is missing"
1081 				    " from the free queue", m);
1082 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1083 			mtx_unlock(&vm_page_queue_free_mtx);
1084 			return (NULL);
1085 #if VM_NRESERVLEVEL > 0
1086 		} else if (object == NULL || object->type == OBJT_DEVICE ||
1087 		    object->type == OBJT_SG ||
1088 		    (object->flags & OBJ_COLORED) == 0 ||
1089 		    (m = vm_reserv_alloc_page(object, pindex)) == NULL) {
1090 #else
1091 		} else {
1092 #endif
1093 			m = vm_phys_alloc_pages(object != NULL ?
1094 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1095 #if VM_NRESERVLEVEL > 0
1096 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1097 				m = vm_phys_alloc_pages(object != NULL ?
1098 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1099 				    0);
1100 			}
1101 #endif
1102 		}
1103 	} else {
1104 		/*
1105 		 * Not allocatable, give up.
1106 		 */
1107 		mtx_unlock(&vm_page_queue_free_mtx);
1108 		atomic_add_int(&vm_pageout_deficit, 1);
1109 		pagedaemon_wakeup();
1110 		return (NULL);
1111 	}
1112 
1113 	/*
1114 	 *  At this point we had better have found a good page.
1115 	 */
1116 
1117 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1118 	KASSERT(m->queue == PQ_NONE,
1119 	    ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
1120 	KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
1121 	KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
1122 	KASSERT(m->busy == 0, ("vm_page_alloc: page %p is busy", m));
1123 	KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
1124 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1125 	    ("vm_page_alloc: page %p has unexpected memattr %d", m,
1126 	    pmap_page_get_memattr(m)));
1127 	if ((m->flags & PG_CACHED) != 0) {
1128 		KASSERT(m->valid != 0,
1129 		    ("vm_page_alloc: cached page %p is invalid", m));
1130 		if (m->object == object && m->pindex == pindex)
1131 	  		cnt.v_reactivated++;
1132 		else
1133 			m->valid = 0;
1134 		m_object = m->object;
1135 		vm_page_cache_remove(m);
1136 		if (m_object->type == OBJT_VNODE && m_object->cache == NULL)
1137 			vp = m_object->handle;
1138 	} else {
1139 		KASSERT(VM_PAGE_IS_FREE(m),
1140 		    ("vm_page_alloc: page %p is not free", m));
1141 		KASSERT(m->valid == 0,
1142 		    ("vm_page_alloc: free page %p is valid", m));
1143 		cnt.v_free_count--;
1144 	}
1145 
1146 	/*
1147 	 * Initialize structure.  Only the PG_ZERO flag is inherited.
1148 	 */
1149 	flags = 0;
1150 	if (m->flags & PG_ZERO) {
1151 		vm_page_zero_count--;
1152 		if (req & VM_ALLOC_ZERO)
1153 			flags = PG_ZERO;
1154 	}
1155 	if (object == NULL || object->type == OBJT_PHYS)
1156 		flags |= PG_UNMANAGED;
1157 	m->flags = flags;
1158 	if (req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ))
1159 		m->oflags = 0;
1160 	else
1161 		m->oflags = VPO_BUSY;
1162 	if (req & VM_ALLOC_WIRED) {
1163 		atomic_add_int(&cnt.v_wire_count, 1);
1164 		m->wire_count = 1;
1165 	}
1166 	m->act_count = 0;
1167 	mtx_unlock(&vm_page_queue_free_mtx);
1168 
1169 	if (object != NULL) {
1170 		/* Ignore device objects; the pager sets "memattr" for them. */
1171 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1172 		    object->type != OBJT_DEVICE && object->type != OBJT_SG)
1173 			pmap_page_set_memattr(m, object->memattr);
1174 		vm_page_insert(m, object, pindex);
1175 	} else
1176 		m->pindex = pindex;
1177 
1178 	/*
1179 	 * The following call to vdrop() must come after the above call
1180 	 * to vm_page_insert() in case both affect the same object and
1181 	 * vnode.  Otherwise, the affected vnode's hold count could
1182 	 * temporarily become zero.
1183 	 */
1184 	if (vp != NULL)
1185 		vdrop(vp);
1186 
1187 	/*
1188 	 * Don't wakeup too often - wakeup the pageout daemon when
1189 	 * we would be nearly out of memory.
1190 	 */
1191 	if (vm_paging_needed())
1192 		pagedaemon_wakeup();
1193 
1194 	return (m);
1195 }
1196 
1197 /*
1198  *	vm_wait:	(also see VM_WAIT macro)
1199  *
1200  *	Block until free pages are available for allocation
1201  *	- Called in various places before memory allocations.
1202  */
1203 void
1204 vm_wait(void)
1205 {
1206 
1207 	mtx_lock(&vm_page_queue_free_mtx);
1208 	if (curproc == pageproc) {
1209 		vm_pageout_pages_needed = 1;
1210 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
1211 		    PDROP | PSWP, "VMWait", 0);
1212 	} else {
1213 		if (!vm_pages_needed) {
1214 			vm_pages_needed = 1;
1215 			wakeup(&vm_pages_needed);
1216 		}
1217 		msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
1218 		    "vmwait", 0);
1219 	}
1220 }
1221 
1222 /*
1223  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
1224  *
1225  *	Block until free pages are available for allocation
1226  *	- Called only in vm_fault so that processes page faulting
1227  *	  can be easily tracked.
1228  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
1229  *	  processes will be able to grab memory first.  Do not change
1230  *	  this balance without careful testing first.
1231  */
1232 void
1233 vm_waitpfault(void)
1234 {
1235 
1236 	mtx_lock(&vm_page_queue_free_mtx);
1237 	if (!vm_pages_needed) {
1238 		vm_pages_needed = 1;
1239 		wakeup(&vm_pages_needed);
1240 	}
1241 	msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
1242 	    "pfault", 0);
1243 }
1244 
1245 /*
1246  *	vm_page_requeue:
1247  *
1248  *	If the given page is contained within a page queue, move it to the tail
1249  *	of that queue.
1250  *
1251  *	The page queues must be locked.
1252  */
1253 void
1254 vm_page_requeue(vm_page_t m)
1255 {
1256 	int queue = VM_PAGE_GETQUEUE(m);
1257 	struct vpgqueues *vpq;
1258 
1259 	if (queue != PQ_NONE) {
1260 		vpq = &vm_page_queues[queue];
1261 		TAILQ_REMOVE(&vpq->pl, m, pageq);
1262 		TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
1263 	}
1264 }
1265 
1266 /*
1267  *	vm_pageq_remove:
1268  *
1269  *	Remove a page from its queue.
1270  *
1271  *	The queue containing the given page must be locked.
1272  *	This routine may not block.
1273  */
1274 void
1275 vm_pageq_remove(vm_page_t m)
1276 {
1277 	int queue = VM_PAGE_GETQUEUE(m);
1278 	struct vpgqueues *pq;
1279 
1280 	if (queue != PQ_NONE) {
1281 		VM_PAGE_SETQUEUE2(m, PQ_NONE);
1282 		pq = &vm_page_queues[queue];
1283 		TAILQ_REMOVE(&pq->pl, m, pageq);
1284 		(*pq->cnt)--;
1285 	}
1286 }
1287 
1288 /*
1289  *	vm_page_enqueue:
1290  *
1291  *	Add the given page to the specified queue.
1292  *
1293  *	The page queues must be locked.
1294  */
1295 static void
1296 vm_page_enqueue(int queue, vm_page_t m)
1297 {
1298 	struct vpgqueues *vpq;
1299 
1300 	vpq = &vm_page_queues[queue];
1301 	VM_PAGE_SETQUEUE2(m, queue);
1302 	TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
1303 	++*vpq->cnt;
1304 }
1305 
1306 /*
1307  *	vm_page_activate:
1308  *
1309  *	Put the specified page on the active list (if appropriate).
1310  *	Ensure that act_count is at least ACT_INIT but do not otherwise
1311  *	mess with it.
1312  *
1313  *	The page queues must be locked.
1314  *	This routine may not block.
1315  */
1316 void
1317 vm_page_activate(vm_page_t m)
1318 {
1319 
1320 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1321 	if (VM_PAGE_GETKNOWNQUEUE2(m) != PQ_ACTIVE) {
1322 		vm_pageq_remove(m);
1323 		if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1324 			if (m->act_count < ACT_INIT)
1325 				m->act_count = ACT_INIT;
1326 			vm_page_enqueue(PQ_ACTIVE, m);
1327 		}
1328 	} else {
1329 		if (m->act_count < ACT_INIT)
1330 			m->act_count = ACT_INIT;
1331 	}
1332 }
1333 
1334 /*
1335  *	vm_page_free_wakeup:
1336  *
1337  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
1338  *	routine is called when a page has been added to the cache or free
1339  *	queues.
1340  *
1341  *	The page queues must be locked.
1342  *	This routine may not block.
1343  */
1344 static inline void
1345 vm_page_free_wakeup(void)
1346 {
1347 
1348 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1349 	/*
1350 	 * if pageout daemon needs pages, then tell it that there are
1351 	 * some free.
1352 	 */
1353 	if (vm_pageout_pages_needed &&
1354 	    cnt.v_cache_count + cnt.v_free_count >= cnt.v_pageout_free_min) {
1355 		wakeup(&vm_pageout_pages_needed);
1356 		vm_pageout_pages_needed = 0;
1357 	}
1358 	/*
1359 	 * wakeup processes that are waiting on memory if we hit a
1360 	 * high water mark. And wakeup scheduler process if we have
1361 	 * lots of memory. this process will swapin processes.
1362 	 */
1363 	if (vm_pages_needed && !vm_page_count_min()) {
1364 		vm_pages_needed = 0;
1365 		wakeup(&cnt.v_free_count);
1366 	}
1367 }
1368 
1369 /*
1370  *	vm_page_free_toq:
1371  *
1372  *	Returns the given page to the free list,
1373  *	disassociating it with any VM object.
1374  *
1375  *	Object and page must be locked prior to entry.
1376  *	This routine may not block.
1377  */
1378 
1379 void
1380 vm_page_free_toq(vm_page_t m)
1381 {
1382 
1383 	if (VM_PAGE_GETQUEUE(m) != PQ_NONE)
1384 		mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1385 	KASSERT(!pmap_page_is_mapped(m),
1386 	    ("vm_page_free_toq: freeing mapped page %p", m));
1387 	PCPU_INC(cnt.v_tfree);
1388 
1389 	if (m->busy || VM_PAGE_IS_FREE(m)) {
1390 		printf(
1391 		"vm_page_free: pindex(%lu), busy(%d), VPO_BUSY(%d), hold(%d)\n",
1392 		    (u_long)m->pindex, m->busy, (m->oflags & VPO_BUSY) ? 1 : 0,
1393 		    m->hold_count);
1394 		if (VM_PAGE_IS_FREE(m))
1395 			panic("vm_page_free: freeing free page");
1396 		else
1397 			panic("vm_page_free: freeing busy page");
1398 	}
1399 
1400 	/*
1401 	 * unqueue, then remove page.  Note that we cannot destroy
1402 	 * the page here because we do not want to call the pager's
1403 	 * callback routine until after we've put the page on the
1404 	 * appropriate free queue.
1405 	 */
1406 	vm_pageq_remove(m);
1407 	vm_page_remove(m);
1408 
1409 	/*
1410 	 * If fictitious remove object association and
1411 	 * return, otherwise delay object association removal.
1412 	 */
1413 	if ((m->flags & PG_FICTITIOUS) != 0) {
1414 		return;
1415 	}
1416 
1417 	m->valid = 0;
1418 	vm_page_undirty(m);
1419 
1420 	if (m->wire_count != 0) {
1421 		if (m->wire_count > 1) {
1422 			panic("vm_page_free: invalid wire count (%d), pindex: 0x%lx",
1423 				m->wire_count, (long)m->pindex);
1424 		}
1425 		panic("vm_page_free: freeing wired page");
1426 	}
1427 	if (m->hold_count != 0) {
1428 		m->flags &= ~PG_ZERO;
1429 		vm_page_enqueue(PQ_HOLD, m);
1430 	} else {
1431 		/*
1432 		 * Restore the default memory attribute to the page.
1433 		 */
1434 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
1435 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
1436 
1437 		/*
1438 		 * Insert the page into the physical memory allocator's
1439 		 * cache/free page queues.
1440 		 */
1441 		mtx_lock(&vm_page_queue_free_mtx);
1442 		m->flags |= PG_FREE;
1443 		cnt.v_free_count++;
1444 #if VM_NRESERVLEVEL > 0
1445 		if (!vm_reserv_free_page(m))
1446 #else
1447 		if (TRUE)
1448 #endif
1449 			vm_phys_free_pages(m, 0);
1450 		if ((m->flags & PG_ZERO) != 0)
1451 			++vm_page_zero_count;
1452 		else
1453 			vm_page_zero_idle_wakeup();
1454 		vm_page_free_wakeup();
1455 		mtx_unlock(&vm_page_queue_free_mtx);
1456 	}
1457 }
1458 
1459 /*
1460  *	vm_page_wire:
1461  *
1462  *	Mark this page as wired down by yet
1463  *	another map, removing it from paging queues
1464  *	as necessary.
1465  *
1466  *	The page queues must be locked.
1467  *	This routine may not block.
1468  */
1469 void
1470 vm_page_wire(vm_page_t m)
1471 {
1472 
1473 	/*
1474 	 * Only bump the wire statistics if the page is not already wired,
1475 	 * and only unqueue the page if it is on some queue (if it is unmanaged
1476 	 * it is already off the queues).
1477 	 */
1478 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1479 	if (m->flags & PG_FICTITIOUS)
1480 		return;
1481 	if (m->wire_count == 0) {
1482 		if ((m->flags & PG_UNMANAGED) == 0)
1483 			vm_pageq_remove(m);
1484 		atomic_add_int(&cnt.v_wire_count, 1);
1485 	}
1486 	m->wire_count++;
1487 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
1488 }
1489 
1490 /*
1491  *	vm_page_unwire:
1492  *
1493  *	Release one wiring of this page, potentially
1494  *	enabling it to be paged again.
1495  *
1496  *	Many pages placed on the inactive queue should actually go
1497  *	into the cache, but it is difficult to figure out which.  What
1498  *	we do instead, if the inactive target is well met, is to put
1499  *	clean pages at the head of the inactive queue instead of the tail.
1500  *	This will cause them to be moved to the cache more quickly and
1501  *	if not actively re-referenced, freed more quickly.  If we just
1502  *	stick these pages at the end of the inactive queue, heavy filesystem
1503  *	meta-data accesses can cause an unnecessary paging load on memory bound
1504  *	processes.  This optimization causes one-time-use metadata to be
1505  *	reused more quickly.
1506  *
1507  *	BUT, if we are in a low-memory situation we have no choice but to
1508  *	put clean pages on the cache queue.
1509  *
1510  *	A number of routines use vm_page_unwire() to guarantee that the page
1511  *	will go into either the inactive or active queues, and will NEVER
1512  *	be placed in the cache - for example, just after dirtying a page.
1513  *	dirty pages in the cache are not allowed.
1514  *
1515  *	The page queues must be locked.
1516  *	This routine may not block.
1517  */
1518 void
1519 vm_page_unwire(vm_page_t m, int activate)
1520 {
1521 
1522 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1523 	if (m->flags & PG_FICTITIOUS)
1524 		return;
1525 	if (m->wire_count > 0) {
1526 		m->wire_count--;
1527 		if (m->wire_count == 0) {
1528 			atomic_subtract_int(&cnt.v_wire_count, 1);
1529 			if (m->flags & PG_UNMANAGED) {
1530 				;
1531 			} else if (activate)
1532 				vm_page_enqueue(PQ_ACTIVE, m);
1533 			else {
1534 				vm_page_flag_clear(m, PG_WINATCFLS);
1535 				vm_page_enqueue(PQ_INACTIVE, m);
1536 			}
1537 		}
1538 	} else {
1539 		panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
1540 	}
1541 }
1542 
1543 
1544 /*
1545  * Move the specified page to the inactive queue.  If the page has
1546  * any associated swap, the swap is deallocated.
1547  *
1548  * Normally athead is 0 resulting in LRU operation.  athead is set
1549  * to 1 if we want this page to be 'as if it were placed in the cache',
1550  * except without unmapping it from the process address space.
1551  *
1552  * This routine may not block.
1553  */
1554 static inline void
1555 _vm_page_deactivate(vm_page_t m, int athead)
1556 {
1557 
1558 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1559 
1560 	/*
1561 	 * Ignore if already inactive.
1562 	 */
1563 	if (VM_PAGE_INQUEUE2(m, PQ_INACTIVE))
1564 		return;
1565 	if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
1566 		vm_page_flag_clear(m, PG_WINATCFLS);
1567 		vm_pageq_remove(m);
1568 		if (athead)
1569 			TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl, m, pageq);
1570 		else
1571 			TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl, m, pageq);
1572 		VM_PAGE_SETQUEUE2(m, PQ_INACTIVE);
1573 		cnt.v_inactive_count++;
1574 	}
1575 }
1576 
1577 void
1578 vm_page_deactivate(vm_page_t m)
1579 {
1580     _vm_page_deactivate(m, 0);
1581 }
1582 
1583 /*
1584  * vm_page_try_to_cache:
1585  *
1586  * Returns 0 on failure, 1 on success
1587  */
1588 int
1589 vm_page_try_to_cache(vm_page_t m)
1590 {
1591 
1592 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1593 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1594 	if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1595 	    (m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) {
1596 		return (0);
1597 	}
1598 	pmap_remove_all(m);
1599 	if (m->dirty)
1600 		return (0);
1601 	vm_page_cache(m);
1602 	return (1);
1603 }
1604 
1605 /*
1606  * vm_page_try_to_free()
1607  *
1608  *	Attempt to free the page.  If we cannot free it, we do nothing.
1609  *	1 is returned on success, 0 on failure.
1610  */
1611 int
1612 vm_page_try_to_free(vm_page_t m)
1613 {
1614 
1615 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1616 	if (m->object != NULL)
1617 		VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1618 	if (m->dirty || m->hold_count || m->busy || m->wire_count ||
1619 	    (m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) {
1620 		return (0);
1621 	}
1622 	pmap_remove_all(m);
1623 	if (m->dirty)
1624 		return (0);
1625 	vm_page_free(m);
1626 	return (1);
1627 }
1628 
1629 /*
1630  * vm_page_cache
1631  *
1632  * Put the specified page onto the page cache queue (if appropriate).
1633  *
1634  * This routine may not block.
1635  */
1636 void
1637 vm_page_cache(vm_page_t m)
1638 {
1639 	vm_object_t object;
1640 	vm_page_t root;
1641 
1642 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1643 	object = m->object;
1644 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1645 	if ((m->flags & PG_UNMANAGED) || (m->oflags & VPO_BUSY) || m->busy ||
1646 	    m->hold_count || m->wire_count) {
1647 		panic("vm_page_cache: attempting to cache busy page");
1648 	}
1649 	pmap_remove_all(m);
1650 	if (m->dirty != 0)
1651 		panic("vm_page_cache: page %p is dirty", m);
1652 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
1653 	    (object->type == OBJT_SWAP &&
1654 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
1655 		/*
1656 		 * Hypothesis: A cache-elgible page belonging to a
1657 		 * default object or swap object but without a backing
1658 		 * store must be zero filled.
1659 		 */
1660 		vm_page_free(m);
1661 		return;
1662 	}
1663 	KASSERT((m->flags & PG_CACHED) == 0,
1664 	    ("vm_page_cache: page %p is already cached", m));
1665 	cnt.v_tcached++;
1666 
1667 	/*
1668 	 * Remove the page from the paging queues.
1669 	 */
1670 	vm_pageq_remove(m);
1671 
1672 	/*
1673 	 * Remove the page from the object's collection of resident
1674 	 * pages.
1675 	 */
1676 	if (m != object->root)
1677 		vm_page_splay(m->pindex, object->root);
1678 	if (m->left == NULL)
1679 		root = m->right;
1680 	else {
1681 		root = vm_page_splay(m->pindex, m->left);
1682 		root->right = m->right;
1683 	}
1684 	object->root = root;
1685 	TAILQ_REMOVE(&object->memq, m, listq);
1686 	object->resident_page_count--;
1687 	object->generation++;
1688 
1689 	/*
1690 	 * Restore the default memory attribute to the page.
1691 	 */
1692 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
1693 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
1694 
1695 	/*
1696 	 * Insert the page into the object's collection of cached pages
1697 	 * and the physical memory allocator's cache/free page queues.
1698 	 */
1699 	vm_page_flag_clear(m, PG_ZERO);
1700 	mtx_lock(&vm_page_queue_free_mtx);
1701 	m->flags |= PG_CACHED;
1702 	cnt.v_cache_count++;
1703 	root = object->cache;
1704 	if (root == NULL) {
1705 		m->left = NULL;
1706 		m->right = NULL;
1707 	} else {
1708 		root = vm_page_splay(m->pindex, root);
1709 		if (m->pindex < root->pindex) {
1710 			m->left = root->left;
1711 			m->right = root;
1712 			root->left = NULL;
1713 		} else if (__predict_false(m->pindex == root->pindex))
1714 			panic("vm_page_cache: offset already cached");
1715 		else {
1716 			m->right = root->right;
1717 			m->left = root;
1718 			root->right = NULL;
1719 		}
1720 	}
1721 	object->cache = m;
1722 #if VM_NRESERVLEVEL > 0
1723 	if (!vm_reserv_free_page(m)) {
1724 #else
1725 	if (TRUE) {
1726 #endif
1727 		vm_phys_set_pool(VM_FREEPOOL_CACHE, m, 0);
1728 		vm_phys_free_pages(m, 0);
1729 	}
1730 	vm_page_free_wakeup();
1731 	mtx_unlock(&vm_page_queue_free_mtx);
1732 
1733 	/*
1734 	 * Increment the vnode's hold count if this is the object's only
1735 	 * cached page.  Decrement the vnode's hold count if this was
1736 	 * the object's only resident page.
1737 	 */
1738 	if (object->type == OBJT_VNODE) {
1739 		if (root == NULL && object->resident_page_count != 0)
1740 			vhold(object->handle);
1741 		else if (root != NULL && object->resident_page_count == 0)
1742 			vdrop(object->handle);
1743 	}
1744 }
1745 
1746 /*
1747  * vm_page_dontneed
1748  *
1749  *	Cache, deactivate, or do nothing as appropriate.  This routine
1750  *	is typically used by madvise() MADV_DONTNEED.
1751  *
1752  *	Generally speaking we want to move the page into the cache so
1753  *	it gets reused quickly.  However, this can result in a silly syndrome
1754  *	due to the page recycling too quickly.  Small objects will not be
1755  *	fully cached.  On the otherhand, if we move the page to the inactive
1756  *	queue we wind up with a problem whereby very large objects
1757  *	unnecessarily blow away our inactive and cache queues.
1758  *
1759  *	The solution is to move the pages based on a fixed weighting.  We
1760  *	either leave them alone, deactivate them, or move them to the cache,
1761  *	where moving them to the cache has the highest weighting.
1762  *	By forcing some pages into other queues we eventually force the
1763  *	system to balance the queues, potentially recovering other unrelated
1764  *	space from active.  The idea is to not force this to happen too
1765  *	often.
1766  */
1767 void
1768 vm_page_dontneed(vm_page_t m)
1769 {
1770 	static int dnweight;
1771 	int dnw;
1772 	int head;
1773 
1774 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1775 	dnw = ++dnweight;
1776 
1777 	/*
1778 	 * occassionally leave the page alone
1779 	 */
1780 	if ((dnw & 0x01F0) == 0 ||
1781 	    VM_PAGE_INQUEUE2(m, PQ_INACTIVE)) {
1782 		if (m->act_count >= ACT_INIT)
1783 			--m->act_count;
1784 		return;
1785 	}
1786 
1787 	/*
1788 	 * Clear any references to the page.  Otherwise, the page daemon will
1789 	 * immediately reactivate the page.
1790 	 */
1791 	vm_page_flag_clear(m, PG_REFERENCED);
1792 	pmap_clear_reference(m);
1793 
1794 	if (m->dirty == 0 && pmap_is_modified(m))
1795 		vm_page_dirty(m);
1796 
1797 	if (m->dirty || (dnw & 0x0070) == 0) {
1798 		/*
1799 		 * Deactivate the page 3 times out of 32.
1800 		 */
1801 		head = 0;
1802 	} else {
1803 		/*
1804 		 * Cache the page 28 times out of every 32.  Note that
1805 		 * the page is deactivated instead of cached, but placed
1806 		 * at the head of the queue instead of the tail.
1807 		 */
1808 		head = 1;
1809 	}
1810 	_vm_page_deactivate(m, head);
1811 }
1812 
1813 /*
1814  * Grab a page, waiting until we are waken up due to the page
1815  * changing state.  We keep on waiting, if the page continues
1816  * to be in the object.  If the page doesn't exist, first allocate it
1817  * and then conditionally zero it.
1818  *
1819  * This routine may block.
1820  */
1821 vm_page_t
1822 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
1823 {
1824 	vm_page_t m;
1825 
1826 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1827 retrylookup:
1828 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
1829 		if (vm_page_sleep_if_busy(m, TRUE, "pgrbwt")) {
1830 			if ((allocflags & VM_ALLOC_RETRY) == 0)
1831 				return (NULL);
1832 			goto retrylookup;
1833 		} else {
1834 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
1835 				vm_page_lock_queues();
1836 				vm_page_wire(m);
1837 				vm_page_unlock_queues();
1838 			}
1839 			if ((allocflags & VM_ALLOC_NOBUSY) == 0)
1840 				vm_page_busy(m);
1841 			return (m);
1842 		}
1843 	}
1844 	m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
1845 	if (m == NULL) {
1846 		VM_OBJECT_UNLOCK(object);
1847 		VM_WAIT;
1848 		VM_OBJECT_LOCK(object);
1849 		if ((allocflags & VM_ALLOC_RETRY) == 0)
1850 			return (NULL);
1851 		goto retrylookup;
1852 	} else if (m->valid != 0)
1853 		return (m);
1854 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
1855 		pmap_zero_page(m);
1856 	return (m);
1857 }
1858 
1859 /*
1860  * Mapping function for valid bits or for dirty bits in
1861  * a page.  May not block.
1862  *
1863  * Inputs are required to range within a page.
1864  */
1865 int
1866 vm_page_bits(int base, int size)
1867 {
1868 	int first_bit;
1869 	int last_bit;
1870 
1871 	KASSERT(
1872 	    base + size <= PAGE_SIZE,
1873 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
1874 	);
1875 
1876 	if (size == 0)		/* handle degenerate case */
1877 		return (0);
1878 
1879 	first_bit = base >> DEV_BSHIFT;
1880 	last_bit = (base + size - 1) >> DEV_BSHIFT;
1881 
1882 	return ((2 << last_bit) - (1 << first_bit));
1883 }
1884 
1885 /*
1886  *	vm_page_set_valid:
1887  *
1888  *	Sets portions of a page valid.  The arguments are expected
1889  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1890  *	of any partial chunks touched by the range.  The invalid portion of
1891  *	such chunks will be zeroed.
1892  *
1893  *	(base + size) must be less then or equal to PAGE_SIZE.
1894  */
1895 void
1896 vm_page_set_valid(vm_page_t m, int base, int size)
1897 {
1898 	int endoff, frag;
1899 
1900 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1901 	if (size == 0)	/* handle degenerate case */
1902 		return;
1903 
1904 	/*
1905 	 * If the base is not DEV_BSIZE aligned and the valid
1906 	 * bit is clear, we have to zero out a portion of the
1907 	 * first block.
1908 	 */
1909 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1910 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
1911 		pmap_zero_page_area(m, frag, base - frag);
1912 
1913 	/*
1914 	 * If the ending offset is not DEV_BSIZE aligned and the
1915 	 * valid bit is clear, we have to zero out a portion of
1916 	 * the last block.
1917 	 */
1918 	endoff = base + size;
1919 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1920 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
1921 		pmap_zero_page_area(m, endoff,
1922 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
1923 
1924 	/*
1925 	 * Assert that no previously invalid block that is now being validated
1926 	 * is already dirty.
1927 	 */
1928 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
1929 	    ("vm_page_set_valid: page %p is dirty", m));
1930 
1931 	/*
1932 	 * Set valid bits inclusive of any overlap.
1933 	 */
1934 	m->valid |= vm_page_bits(base, size);
1935 }
1936 
1937 /*
1938  *	vm_page_set_validclean:
1939  *
1940  *	Sets portions of a page valid and clean.  The arguments are expected
1941  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
1942  *	of any partial chunks touched by the range.  The invalid portion of
1943  *	such chunks will be zero'd.
1944  *
1945  *	This routine may not block.
1946  *
1947  *	(base + size) must be less then or equal to PAGE_SIZE.
1948  */
1949 void
1950 vm_page_set_validclean(vm_page_t m, int base, int size)
1951 {
1952 	int pagebits;
1953 	int frag;
1954 	int endoff;
1955 
1956 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1957 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1958 	if (size == 0)	/* handle degenerate case */
1959 		return;
1960 
1961 	/*
1962 	 * If the base is not DEV_BSIZE aligned and the valid
1963 	 * bit is clear, we have to zero out a portion of the
1964 	 * first block.
1965 	 */
1966 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
1967 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
1968 		pmap_zero_page_area(m, frag, base - frag);
1969 
1970 	/*
1971 	 * If the ending offset is not DEV_BSIZE aligned and the
1972 	 * valid bit is clear, we have to zero out a portion of
1973 	 * the last block.
1974 	 */
1975 	endoff = base + size;
1976 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
1977 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
1978 		pmap_zero_page_area(m, endoff,
1979 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
1980 
1981 	/*
1982 	 * Set valid, clear dirty bits.  If validating the entire
1983 	 * page we can safely clear the pmap modify bit.  We also
1984 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
1985 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
1986 	 * be set again.
1987 	 *
1988 	 * We set valid bits inclusive of any overlap, but we can only
1989 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
1990 	 * the range.
1991 	 */
1992 	pagebits = vm_page_bits(base, size);
1993 	m->valid |= pagebits;
1994 #if 0	/* NOT YET */
1995 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
1996 		frag = DEV_BSIZE - frag;
1997 		base += frag;
1998 		size -= frag;
1999 		if (size < 0)
2000 			size = 0;
2001 	}
2002 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
2003 #endif
2004 	m->dirty &= ~pagebits;
2005 	if (base == 0 && size == PAGE_SIZE) {
2006 		pmap_clear_modify(m);
2007 		m->oflags &= ~VPO_NOSYNC;
2008 	}
2009 }
2010 
2011 void
2012 vm_page_clear_dirty(vm_page_t m, int base, int size)
2013 {
2014 
2015 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2016 	m->dirty &= ~vm_page_bits(base, size);
2017 }
2018 
2019 /*
2020  *	vm_page_set_invalid:
2021  *
2022  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
2023  *	valid and dirty bits for the effected areas are cleared.
2024  *
2025  *	May not block.
2026  */
2027 void
2028 vm_page_set_invalid(vm_page_t m, int base, int size)
2029 {
2030 	int bits;
2031 
2032 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2033 	bits = vm_page_bits(base, size);
2034 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2035 	if (m->valid == VM_PAGE_BITS_ALL && bits != 0)
2036 		pmap_remove_all(m);
2037 	m->valid &= ~bits;
2038 	m->dirty &= ~bits;
2039 	m->object->generation++;
2040 }
2041 
2042 /*
2043  * vm_page_zero_invalid()
2044  *
2045  *	The kernel assumes that the invalid portions of a page contain
2046  *	garbage, but such pages can be mapped into memory by user code.
2047  *	When this occurs, we must zero out the non-valid portions of the
2048  *	page so user code sees what it expects.
2049  *
2050  *	Pages are most often semi-valid when the end of a file is mapped
2051  *	into memory and the file's size is not page aligned.
2052  */
2053 void
2054 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
2055 {
2056 	int b;
2057 	int i;
2058 
2059 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2060 	/*
2061 	 * Scan the valid bits looking for invalid sections that
2062 	 * must be zerod.  Invalid sub-DEV_BSIZE'd areas ( where the
2063 	 * valid bit may be set ) have already been zerod by
2064 	 * vm_page_set_validclean().
2065 	 */
2066 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
2067 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
2068 		    (m->valid & (1 << i))
2069 		) {
2070 			if (i > b) {
2071 				pmap_zero_page_area(m,
2072 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
2073 			}
2074 			b = i + 1;
2075 		}
2076 	}
2077 
2078 	/*
2079 	 * setvalid is TRUE when we can safely set the zero'd areas
2080 	 * as being valid.  We can do this if there are no cache consistancy
2081 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
2082 	 */
2083 	if (setvalid)
2084 		m->valid = VM_PAGE_BITS_ALL;
2085 }
2086 
2087 /*
2088  *	vm_page_is_valid:
2089  *
2090  *	Is (partial) page valid?  Note that the case where size == 0
2091  *	will return FALSE in the degenerate case where the page is
2092  *	entirely invalid, and TRUE otherwise.
2093  *
2094  *	May not block.
2095  */
2096 int
2097 vm_page_is_valid(vm_page_t m, int base, int size)
2098 {
2099 	int bits = vm_page_bits(base, size);
2100 
2101 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2102 	if (m->valid && ((m->valid & bits) == bits))
2103 		return 1;
2104 	else
2105 		return 0;
2106 }
2107 
2108 /*
2109  * update dirty bits from pmap/mmu.  May not block.
2110  */
2111 void
2112 vm_page_test_dirty(vm_page_t m)
2113 {
2114 	if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
2115 		vm_page_dirty(m);
2116 	}
2117 }
2118 
2119 int so_zerocp_fullpage = 0;
2120 
2121 /*
2122  *	Replace the given page with a copy.  The copied page assumes
2123  *	the portion of the given page's "wire_count" that is not the
2124  *	responsibility of this copy-on-write mechanism.
2125  *
2126  *	The object containing the given page must have a non-zero
2127  *	paging-in-progress count and be locked.
2128  */
2129 void
2130 vm_page_cowfault(vm_page_t m)
2131 {
2132 	vm_page_t mnew;
2133 	vm_object_t object;
2134 	vm_pindex_t pindex;
2135 
2136 	object = m->object;
2137 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2138 	KASSERT(object->paging_in_progress != 0,
2139 	    ("vm_page_cowfault: object %p's paging-in-progress count is zero.",
2140 	    object));
2141 	pindex = m->pindex;
2142 
2143  retry_alloc:
2144 	pmap_remove_all(m);
2145 	vm_page_remove(m);
2146 	mnew = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
2147 	if (mnew == NULL) {
2148 		vm_page_insert(m, object, pindex);
2149 		vm_page_unlock_queues();
2150 		VM_OBJECT_UNLOCK(object);
2151 		VM_WAIT;
2152 		VM_OBJECT_LOCK(object);
2153 		if (m == vm_page_lookup(object, pindex)) {
2154 			vm_page_lock_queues();
2155 			goto retry_alloc;
2156 		} else {
2157 			/*
2158 			 * Page disappeared during the wait.
2159 			 */
2160 			vm_page_lock_queues();
2161 			return;
2162 		}
2163 	}
2164 
2165 	if (m->cow == 0) {
2166 		/*
2167 		 * check to see if we raced with an xmit complete when
2168 		 * waiting to allocate a page.  If so, put things back
2169 		 * the way they were
2170 		 */
2171 		vm_page_free(mnew);
2172 		vm_page_insert(m, object, pindex);
2173 	} else { /* clear COW & copy page */
2174 		if (!so_zerocp_fullpage)
2175 			pmap_copy_page(m, mnew);
2176 		mnew->valid = VM_PAGE_BITS_ALL;
2177 		vm_page_dirty(mnew);
2178 		mnew->wire_count = m->wire_count - m->cow;
2179 		m->wire_count = m->cow;
2180 	}
2181 }
2182 
2183 void
2184 vm_page_cowclear(vm_page_t m)
2185 {
2186 
2187 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2188 	if (m->cow) {
2189 		m->cow--;
2190 		/*
2191 		 * let vm_fault add back write permission  lazily
2192 		 */
2193 	}
2194 	/*
2195 	 *  sf_buf_free() will free the page, so we needn't do it here
2196 	 */
2197 }
2198 
2199 int
2200 vm_page_cowsetup(vm_page_t m)
2201 {
2202 
2203 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2204 	if (m->cow == USHRT_MAX - 1)
2205 		return (EBUSY);
2206 	m->cow++;
2207 	pmap_remove_write(m);
2208 	return (0);
2209 }
2210 
2211 #include "opt_ddb.h"
2212 #ifdef DDB
2213 #include <sys/kernel.h>
2214 
2215 #include <ddb/ddb.h>
2216 
2217 DB_SHOW_COMMAND(page, vm_page_print_page_info)
2218 {
2219 	db_printf("cnt.v_free_count: %d\n", cnt.v_free_count);
2220 	db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count);
2221 	db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count);
2222 	db_printf("cnt.v_active_count: %d\n", cnt.v_active_count);
2223 	db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count);
2224 	db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved);
2225 	db_printf("cnt.v_free_min: %d\n", cnt.v_free_min);
2226 	db_printf("cnt.v_free_target: %d\n", cnt.v_free_target);
2227 	db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min);
2228 	db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target);
2229 }
2230 
2231 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
2232 {
2233 
2234 	db_printf("PQ_FREE:");
2235 	db_printf(" %d", cnt.v_free_count);
2236 	db_printf("\n");
2237 
2238 	db_printf("PQ_CACHE:");
2239 	db_printf(" %d", cnt.v_cache_count);
2240 	db_printf("\n");
2241 
2242 	db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n",
2243 		*vm_page_queues[PQ_ACTIVE].cnt,
2244 		*vm_page_queues[PQ_INACTIVE].cnt);
2245 }
2246 #endif /* DDB */
2247