xref: /freebsd/sys/vm/vm_page.c (revision d940bfec8c329dd82d8d54efebd81c8aa420503b)
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 page queue lock is required when adding or removing a page from a
67  *	  page queue regardless of other locks or the busy state of a page.
68  *
69  *		* In general, no thread besides the page daemon can acquire or
70  *		  hold more than one page queue lock at a time.
71  *
72  *		* The page daemon can acquire and hold any pair of page queue
73  *		  locks in any order.
74  *
75  *	- The object lock is required when inserting or removing
76  *	  pages from an object (vm_page_insert() or vm_page_remove()).
77  *
78  */
79 
80 /*
81  *	Resident memory management module.
82  */
83 
84 #include <sys/cdefs.h>
85 __FBSDID("$FreeBSD$");
86 
87 #include "opt_vm.h"
88 
89 #include <sys/param.h>
90 #include <sys/systm.h>
91 #include <sys/lock.h>
92 #include <sys/kernel.h>
93 #include <sys/limits.h>
94 #include <sys/malloc.h>
95 #include <sys/mman.h>
96 #include <sys/msgbuf.h>
97 #include <sys/mutex.h>
98 #include <sys/proc.h>
99 #include <sys/rwlock.h>
100 #include <sys/sysctl.h>
101 #include <sys/vmmeter.h>
102 #include <sys/vnode.h>
103 
104 #include <vm/vm.h>
105 #include <vm/pmap.h>
106 #include <vm/vm_param.h>
107 #include <vm/vm_kern.h>
108 #include <vm/vm_object.h>
109 #include <vm/vm_page.h>
110 #include <vm/vm_pageout.h>
111 #include <vm/vm_pager.h>
112 #include <vm/vm_phys.h>
113 #include <vm/vm_radix.h>
114 #include <vm/vm_reserv.h>
115 #include <vm/vm_extern.h>
116 #include <vm/uma.h>
117 #include <vm/uma_int.h>
118 
119 #include <machine/md_var.h>
120 
121 /*
122  *	Associated with page of user-allocatable memory is a
123  *	page structure.
124  */
125 
126 struct vm_domain vm_dom[MAXMEMDOM];
127 struct mtx_padalign vm_page_queue_free_mtx;
128 
129 struct mtx_padalign pa_lock[PA_LOCK_COUNT];
130 
131 vm_page_t vm_page_array;
132 long vm_page_array_size;
133 long first_page;
134 int vm_page_zero_count;
135 
136 static int boot_pages = UMA_BOOT_PAGES;
137 TUNABLE_INT("vm.boot_pages", &boot_pages);
138 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RD, &boot_pages, 0,
139 	"number of pages allocated for bootstrapping the VM system");
140 
141 static int pa_tryrelock_restart;
142 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
143     &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
144 
145 static uma_zone_t fakepg_zone;
146 
147 static struct vnode *vm_page_alloc_init(vm_page_t m);
148 static void vm_page_cache_turn_free(vm_page_t m);
149 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
150 static void vm_page_enqueue(int queue, vm_page_t m);
151 static void vm_page_init_fakepg(void *dummy);
152 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
153     vm_pindex_t pindex, vm_page_t mpred);
154 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
155     vm_page_t mpred);
156 
157 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL);
158 
159 static void
160 vm_page_init_fakepg(void *dummy)
161 {
162 
163 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
164 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
165 }
166 
167 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
168 #if PAGE_SIZE == 32768
169 #ifdef CTASSERT
170 CTASSERT(sizeof(u_long) >= 8);
171 #endif
172 #endif
173 
174 /*
175  * Try to acquire a physical address lock while a pmap is locked.  If we
176  * fail to trylock we unlock and lock the pmap directly and cache the
177  * locked pa in *locked.  The caller should then restart their loop in case
178  * the virtual to physical mapping has changed.
179  */
180 int
181 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
182 {
183 	vm_paddr_t lockpa;
184 
185 	lockpa = *locked;
186 	*locked = pa;
187 	if (lockpa) {
188 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
189 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
190 			return (0);
191 		PA_UNLOCK(lockpa);
192 	}
193 	if (PA_TRYLOCK(pa))
194 		return (0);
195 	PMAP_UNLOCK(pmap);
196 	atomic_add_int(&pa_tryrelock_restart, 1);
197 	PA_LOCK(pa);
198 	PMAP_LOCK(pmap);
199 	return (EAGAIN);
200 }
201 
202 /*
203  *	vm_set_page_size:
204  *
205  *	Sets the page size, perhaps based upon the memory
206  *	size.  Must be called before any use of page-size
207  *	dependent functions.
208  */
209 void
210 vm_set_page_size(void)
211 {
212 	if (cnt.v_page_size == 0)
213 		cnt.v_page_size = PAGE_SIZE;
214 	if (((cnt.v_page_size - 1) & cnt.v_page_size) != 0)
215 		panic("vm_set_page_size: page size not a power of two");
216 }
217 
218 /*
219  *	vm_page_blacklist_lookup:
220  *
221  *	See if a physical address in this page has been listed
222  *	in the blacklist tunable.  Entries in the tunable are
223  *	separated by spaces or commas.  If an invalid integer is
224  *	encountered then the rest of the string is skipped.
225  */
226 static int
227 vm_page_blacklist_lookup(char *list, vm_paddr_t pa)
228 {
229 	vm_paddr_t bad;
230 	char *cp, *pos;
231 
232 	for (pos = list; *pos != '\0'; pos = cp) {
233 		bad = strtoq(pos, &cp, 0);
234 		if (*cp != '\0') {
235 			if (*cp == ' ' || *cp == ',') {
236 				cp++;
237 				if (cp == pos)
238 					continue;
239 			} else
240 				break;
241 		}
242 		if (pa == trunc_page(bad))
243 			return (1);
244 	}
245 	return (0);
246 }
247 
248 static void
249 vm_page_domain_init(struct vm_domain *vmd)
250 {
251 	struct vm_pagequeue *pq;
252 	int i;
253 
254 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
255 	    "vm inactive pagequeue";
256 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
257 	    &cnt.v_inactive_count;
258 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
259 	    "vm active pagequeue";
260 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
261 	    &cnt.v_active_count;
262 	vmd->vmd_page_count = 0;
263 	vmd->vmd_free_count = 0;
264 	vmd->vmd_segs = 0;
265 	vmd->vmd_oom = FALSE;
266 	vmd->vmd_pass = 0;
267 	for (i = 0; i < PQ_COUNT; i++) {
268 		pq = &vmd->vmd_pagequeues[i];
269 		TAILQ_INIT(&pq->pq_pl);
270 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
271 		    MTX_DEF | MTX_DUPOK);
272 	}
273 }
274 
275 /*
276  *	vm_page_startup:
277  *
278  *	Initializes the resident memory module.
279  *
280  *	Allocates memory for the page cells, and
281  *	for the object/offset-to-page hash table headers.
282  *	Each page cell is initialized and placed on the free list.
283  */
284 vm_offset_t
285 vm_page_startup(vm_offset_t vaddr)
286 {
287 	vm_offset_t mapped;
288 	vm_paddr_t page_range;
289 	vm_paddr_t new_end;
290 	int i;
291 	vm_paddr_t pa;
292 	vm_paddr_t last_pa;
293 	char *list;
294 
295 	/* the biggest memory array is the second group of pages */
296 	vm_paddr_t end;
297 	vm_paddr_t biggestsize;
298 	vm_paddr_t low_water, high_water;
299 	int biggestone;
300 
301 	biggestsize = 0;
302 	biggestone = 0;
303 	vaddr = round_page(vaddr);
304 
305 	for (i = 0; phys_avail[i + 1]; i += 2) {
306 		phys_avail[i] = round_page(phys_avail[i]);
307 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
308 	}
309 
310 	low_water = phys_avail[0];
311 	high_water = phys_avail[1];
312 
313 	for (i = 0; phys_avail[i + 1]; i += 2) {
314 		vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
315 
316 		if (size > biggestsize) {
317 			biggestone = i;
318 			biggestsize = size;
319 		}
320 		if (phys_avail[i] < low_water)
321 			low_water = phys_avail[i];
322 		if (phys_avail[i + 1] > high_water)
323 			high_water = phys_avail[i + 1];
324 	}
325 
326 #ifdef XEN
327 	low_water = 0;
328 #endif
329 
330 	end = phys_avail[biggestone+1];
331 
332 	/*
333 	 * Initialize the page and queue locks.
334 	 */
335 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
336 	for (i = 0; i < PA_LOCK_COUNT; i++)
337 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
338 	for (i = 0; i < vm_ndomains; i++)
339 		vm_page_domain_init(&vm_dom[i]);
340 
341 	/*
342 	 * Allocate memory for use when boot strapping the kernel memory
343 	 * allocator.
344 	 */
345 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
346 	new_end = trunc_page(new_end);
347 	mapped = pmap_map(&vaddr, new_end, end,
348 	    VM_PROT_READ | VM_PROT_WRITE);
349 	bzero((void *)mapped, end - new_end);
350 	uma_startup((void *)mapped, boot_pages);
351 
352 #if defined(__amd64__) || defined(__i386__) || defined(__arm__) || \
353     defined(__mips__)
354 	/*
355 	 * Allocate a bitmap to indicate that a random physical page
356 	 * needs to be included in a minidump.
357 	 *
358 	 * The amd64 port needs this to indicate which direct map pages
359 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
360 	 *
361 	 * However, i386 still needs this workspace internally within the
362 	 * minidump code.  In theory, they are not needed on i386, but are
363 	 * included should the sf_buf code decide to use them.
364 	 */
365 	last_pa = 0;
366 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
367 		if (dump_avail[i + 1] > last_pa)
368 			last_pa = dump_avail[i + 1];
369 	page_range = last_pa / PAGE_SIZE;
370 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
371 	new_end -= vm_page_dump_size;
372 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
373 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
374 	bzero((void *)vm_page_dump, vm_page_dump_size);
375 #endif
376 #ifdef __amd64__
377 	/*
378 	 * Request that the physical pages underlying the message buffer be
379 	 * included in a crash dump.  Since the message buffer is accessed
380 	 * through the direct map, they are not automatically included.
381 	 */
382 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
383 	last_pa = pa + round_page(msgbufsize);
384 	while (pa < last_pa) {
385 		dump_add_page(pa);
386 		pa += PAGE_SIZE;
387 	}
388 #endif
389 	/*
390 	 * Compute the number of pages of memory that will be available for
391 	 * use (taking into account the overhead of a page structure per
392 	 * page).
393 	 */
394 	first_page = low_water / PAGE_SIZE;
395 #ifdef VM_PHYSSEG_SPARSE
396 	page_range = 0;
397 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
398 		page_range += atop(phys_avail[i + 1] - phys_avail[i]);
399 #elif defined(VM_PHYSSEG_DENSE)
400 	page_range = high_water / PAGE_SIZE - first_page;
401 #else
402 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
403 #endif
404 	end = new_end;
405 
406 	/*
407 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
408 	 */
409 	vaddr += PAGE_SIZE;
410 
411 	/*
412 	 * Initialize the mem entry structures now, and put them in the free
413 	 * queue.
414 	 */
415 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
416 	mapped = pmap_map(&vaddr, new_end, end,
417 	    VM_PROT_READ | VM_PROT_WRITE);
418 	vm_page_array = (vm_page_t) mapped;
419 #if VM_NRESERVLEVEL > 0
420 	/*
421 	 * Allocate memory for the reservation management system's data
422 	 * structures.
423 	 */
424 	new_end = vm_reserv_startup(&vaddr, new_end, high_water);
425 #endif
426 #if defined(__amd64__) || defined(__mips__)
427 	/*
428 	 * pmap_map on amd64 and mips can come out of the direct-map, not kvm
429 	 * like i386, so the pages must be tracked for a crashdump to include
430 	 * this data.  This includes the vm_page_array and the early UMA
431 	 * bootstrap pages.
432 	 */
433 	for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
434 		dump_add_page(pa);
435 #endif
436 	phys_avail[biggestone + 1] = new_end;
437 
438 	/*
439 	 * Clear all of the page structures
440 	 */
441 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
442 	for (i = 0; i < page_range; i++)
443 		vm_page_array[i].order = VM_NFREEORDER;
444 	vm_page_array_size = page_range;
445 
446 	/*
447 	 * Initialize the physical memory allocator.
448 	 */
449 	vm_phys_init();
450 
451 	/*
452 	 * Add every available physical page that is not blacklisted to
453 	 * the free lists.
454 	 */
455 	cnt.v_page_count = 0;
456 	cnt.v_free_count = 0;
457 	list = getenv("vm.blacklist");
458 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
459 		pa = phys_avail[i];
460 		last_pa = phys_avail[i + 1];
461 		while (pa < last_pa) {
462 			if (list != NULL &&
463 			    vm_page_blacklist_lookup(list, pa))
464 				printf("Skipping page with pa 0x%jx\n",
465 				    (uintmax_t)pa);
466 			else
467 				vm_phys_add_page(pa);
468 			pa += PAGE_SIZE;
469 		}
470 	}
471 	freeenv(list);
472 #if VM_NRESERVLEVEL > 0
473 	/*
474 	 * Initialize the reservation management system.
475 	 */
476 	vm_reserv_init();
477 #endif
478 	return (vaddr);
479 }
480 
481 void
482 vm_page_reference(vm_page_t m)
483 {
484 
485 	vm_page_aflag_set(m, PGA_REFERENCED);
486 }
487 
488 /*
489  *	vm_page_busy_downgrade:
490  *
491  *	Downgrade an exclusive busy page into a single shared busy page.
492  */
493 void
494 vm_page_busy_downgrade(vm_page_t m)
495 {
496 	u_int x;
497 
498 	vm_page_assert_xbusied(m);
499 
500 	for (;;) {
501 		x = m->busy_lock;
502 		x &= VPB_BIT_WAITERS;
503 		if (atomic_cmpset_rel_int(&m->busy_lock,
504 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x))
505 			break;
506 	}
507 }
508 
509 /*
510  *	vm_page_sbusied:
511  *
512  *	Return a positive value if the page is shared busied, 0 otherwise.
513  */
514 int
515 vm_page_sbusied(vm_page_t m)
516 {
517 	u_int x;
518 
519 	x = m->busy_lock;
520 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
521 }
522 
523 /*
524  *	vm_page_sunbusy:
525  *
526  *	Shared unbusy a page.
527  */
528 void
529 vm_page_sunbusy(vm_page_t m)
530 {
531 	u_int x;
532 
533 	vm_page_assert_sbusied(m);
534 
535 	for (;;) {
536 		x = m->busy_lock;
537 		if (VPB_SHARERS(x) > 1) {
538 			if (atomic_cmpset_int(&m->busy_lock, x,
539 			    x - VPB_ONE_SHARER))
540 				break;
541 			continue;
542 		}
543 		if ((x & VPB_BIT_WAITERS) == 0) {
544 			KASSERT(x == VPB_SHARERS_WORD(1),
545 			    ("vm_page_sunbusy: invalid lock state"));
546 			if (atomic_cmpset_int(&m->busy_lock,
547 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
548 				break;
549 			continue;
550 		}
551 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
552 		    ("vm_page_sunbusy: invalid lock state for waiters"));
553 
554 		vm_page_lock(m);
555 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
556 			vm_page_unlock(m);
557 			continue;
558 		}
559 		wakeup(m);
560 		vm_page_unlock(m);
561 		break;
562 	}
563 }
564 
565 /*
566  *	vm_page_busy_sleep:
567  *
568  *	Sleep and release the page lock, using the page pointer as wchan.
569  *	This is used to implement the hard-path of busying mechanism.
570  *
571  *	The given page must be locked.
572  */
573 void
574 vm_page_busy_sleep(vm_page_t m, const char *wmesg)
575 {
576 	u_int x;
577 
578 	vm_page_lock_assert(m, MA_OWNED);
579 
580 	x = m->busy_lock;
581 	if (x == VPB_UNBUSIED) {
582 		vm_page_unlock(m);
583 		return;
584 	}
585 	if ((x & VPB_BIT_WAITERS) == 0 &&
586 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) {
587 		vm_page_unlock(m);
588 		return;
589 	}
590 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
591 }
592 
593 /*
594  *	vm_page_trysbusy:
595  *
596  *	Try to shared busy a page.
597  *	If the operation succeeds 1 is returned otherwise 0.
598  *	The operation never sleeps.
599  */
600 int
601 vm_page_trysbusy(vm_page_t m)
602 {
603 	u_int x;
604 
605 	x = m->busy_lock;
606 	return ((x & VPB_BIT_SHARED) != 0 &&
607 	    atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER));
608 }
609 
610 /*
611  *	vm_page_xunbusy_hard:
612  *
613  *	Called after the first try the exclusive unbusy of a page failed.
614  *	It is assumed that the waiters bit is on.
615  */
616 void
617 vm_page_xunbusy_hard(vm_page_t m)
618 {
619 
620 	vm_page_assert_xbusied(m);
621 
622 	vm_page_lock(m);
623 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
624 	wakeup(m);
625 	vm_page_unlock(m);
626 }
627 
628 /*
629  *	vm_page_flash:
630  *
631  *	Wakeup anyone waiting for the page.
632  *	The ownership bits do not change.
633  *
634  *	The given page must be locked.
635  */
636 void
637 vm_page_flash(vm_page_t m)
638 {
639 	u_int x;
640 
641 	vm_page_lock_assert(m, MA_OWNED);
642 
643 	for (;;) {
644 		x = m->busy_lock;
645 		if ((x & VPB_BIT_WAITERS) == 0)
646 			return;
647 		if (atomic_cmpset_int(&m->busy_lock, x,
648 		    x & (~VPB_BIT_WAITERS)))
649 			break;
650 	}
651 	wakeup(m);
652 }
653 
654 /*
655  * Keep page from being freed by the page daemon
656  * much of the same effect as wiring, except much lower
657  * overhead and should be used only for *very* temporary
658  * holding ("wiring").
659  */
660 void
661 vm_page_hold(vm_page_t mem)
662 {
663 
664 	vm_page_lock_assert(mem, MA_OWNED);
665         mem->hold_count++;
666 }
667 
668 void
669 vm_page_unhold(vm_page_t mem)
670 {
671 
672 	vm_page_lock_assert(mem, MA_OWNED);
673 	--mem->hold_count;
674 	KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
675 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
676 		vm_page_free_toq(mem);
677 }
678 
679 /*
680  *	vm_page_unhold_pages:
681  *
682  *	Unhold each of the pages that is referenced by the given array.
683  */
684 void
685 vm_page_unhold_pages(vm_page_t *ma, int count)
686 {
687 	struct mtx *mtx, *new_mtx;
688 
689 	mtx = NULL;
690 	for (; count != 0; count--) {
691 		/*
692 		 * Avoid releasing and reacquiring the same page lock.
693 		 */
694 		new_mtx = vm_page_lockptr(*ma);
695 		if (mtx != new_mtx) {
696 			if (mtx != NULL)
697 				mtx_unlock(mtx);
698 			mtx = new_mtx;
699 			mtx_lock(mtx);
700 		}
701 		vm_page_unhold(*ma);
702 		ma++;
703 	}
704 	if (mtx != NULL)
705 		mtx_unlock(mtx);
706 }
707 
708 vm_page_t
709 PHYS_TO_VM_PAGE(vm_paddr_t pa)
710 {
711 	vm_page_t m;
712 
713 #ifdef VM_PHYSSEG_SPARSE
714 	m = vm_phys_paddr_to_vm_page(pa);
715 	if (m == NULL)
716 		m = vm_phys_fictitious_to_vm_page(pa);
717 	return (m);
718 #elif defined(VM_PHYSSEG_DENSE)
719 	long pi;
720 
721 	pi = atop(pa);
722 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
723 		m = &vm_page_array[pi - first_page];
724 		return (m);
725 	}
726 	return (vm_phys_fictitious_to_vm_page(pa));
727 #else
728 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
729 #endif
730 }
731 
732 /*
733  *	vm_page_getfake:
734  *
735  *	Create a fictitious page with the specified physical address and
736  *	memory attribute.  The memory attribute is the only the machine-
737  *	dependent aspect of a fictitious page that must be initialized.
738  */
739 vm_page_t
740 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
741 {
742 	vm_page_t m;
743 
744 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
745 	vm_page_initfake(m, paddr, memattr);
746 	return (m);
747 }
748 
749 void
750 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
751 {
752 
753 	if ((m->flags & PG_FICTITIOUS) != 0) {
754 		/*
755 		 * The page's memattr might have changed since the
756 		 * previous initialization.  Update the pmap to the
757 		 * new memattr.
758 		 */
759 		goto memattr;
760 	}
761 	m->phys_addr = paddr;
762 	m->queue = PQ_NONE;
763 	/* Fictitious pages don't use "segind". */
764 	m->flags = PG_FICTITIOUS;
765 	/* Fictitious pages don't use "order" or "pool". */
766 	m->oflags = VPO_UNMANAGED;
767 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
768 	m->wire_count = 1;
769 	pmap_page_init(m);
770 memattr:
771 	pmap_page_set_memattr(m, memattr);
772 }
773 
774 /*
775  *	vm_page_putfake:
776  *
777  *	Release a fictitious page.
778  */
779 void
780 vm_page_putfake(vm_page_t m)
781 {
782 
783 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
784 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
785 	    ("vm_page_putfake: bad page %p", m));
786 	uma_zfree(fakepg_zone, m);
787 }
788 
789 /*
790  *	vm_page_updatefake:
791  *
792  *	Update the given fictitious page to the specified physical address and
793  *	memory attribute.
794  */
795 void
796 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
797 {
798 
799 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
800 	    ("vm_page_updatefake: bad page %p", m));
801 	m->phys_addr = paddr;
802 	pmap_page_set_memattr(m, memattr);
803 }
804 
805 /*
806  *	vm_page_free:
807  *
808  *	Free a page.
809  */
810 void
811 vm_page_free(vm_page_t m)
812 {
813 
814 	m->flags &= ~PG_ZERO;
815 	vm_page_free_toq(m);
816 }
817 
818 /*
819  *	vm_page_free_zero:
820  *
821  *	Free a page to the zerod-pages queue
822  */
823 void
824 vm_page_free_zero(vm_page_t m)
825 {
826 
827 	m->flags |= PG_ZERO;
828 	vm_page_free_toq(m);
829 }
830 
831 /*
832  * Unbusy and handle the page queueing for a page from the VOP_GETPAGES()
833  * array which is not the request page.
834  */
835 void
836 vm_page_readahead_finish(vm_page_t m)
837 {
838 
839 	if (m->valid != 0) {
840 		/*
841 		 * Since the page is not the requested page, whether
842 		 * it should be activated or deactivated is not
843 		 * obvious.  Empirical results have shown that
844 		 * deactivating the page is usually the best choice,
845 		 * unless the page is wanted by another thread.
846 		 */
847 		vm_page_lock(m);
848 		if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
849 			vm_page_activate(m);
850 		else
851 			vm_page_deactivate(m);
852 		vm_page_unlock(m);
853 		vm_page_xunbusy(m);
854 	} else {
855 		/*
856 		 * Free the completely invalid page.  Such page state
857 		 * occurs due to the short read operation which did
858 		 * not covered our page at all, or in case when a read
859 		 * error happens.
860 		 */
861 		vm_page_lock(m);
862 		vm_page_free(m);
863 		vm_page_unlock(m);
864 	}
865 }
866 
867 /*
868  *	vm_page_sleep_if_busy:
869  *
870  *	Sleep and release the page queues lock if the page is busied.
871  *	Returns TRUE if the thread slept.
872  *
873  *	The given page must be unlocked and object containing it must
874  *	be locked.
875  */
876 int
877 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
878 {
879 	vm_object_t obj;
880 
881 	vm_page_lock_assert(m, MA_NOTOWNED);
882 	VM_OBJECT_ASSERT_WLOCKED(m->object);
883 
884 	if (vm_page_busied(m)) {
885 		/*
886 		 * The page-specific object must be cached because page
887 		 * identity can change during the sleep, causing the
888 		 * re-lock of a different object.
889 		 * It is assumed that a reference to the object is already
890 		 * held by the callers.
891 		 */
892 		obj = m->object;
893 		vm_page_lock(m);
894 		VM_OBJECT_WUNLOCK(obj);
895 		vm_page_busy_sleep(m, msg);
896 		VM_OBJECT_WLOCK(obj);
897 		return (TRUE);
898 	}
899 	return (FALSE);
900 }
901 
902 /*
903  *	vm_page_dirty_KBI:		[ internal use only ]
904  *
905  *	Set all bits in the page's dirty field.
906  *
907  *	The object containing the specified page must be locked if the
908  *	call is made from the machine-independent layer.
909  *
910  *	See vm_page_clear_dirty_mask().
911  *
912  *	This function should only be called by vm_page_dirty().
913  */
914 void
915 vm_page_dirty_KBI(vm_page_t m)
916 {
917 
918 	/* These assertions refer to this operation by its public name. */
919 	KASSERT((m->flags & PG_CACHED) == 0,
920 	    ("vm_page_dirty: page in cache!"));
921 	KASSERT(!VM_PAGE_IS_FREE(m),
922 	    ("vm_page_dirty: page is free!"));
923 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
924 	    ("vm_page_dirty: page is invalid!"));
925 	m->dirty = VM_PAGE_BITS_ALL;
926 }
927 
928 /*
929  *	vm_page_insert:		[ internal use only ]
930  *
931  *	Inserts the given mem entry into the object and object list.
932  *
933  *	The object must be locked.
934  */
935 int
936 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
937 {
938 	vm_page_t mpred;
939 
940 	VM_OBJECT_ASSERT_WLOCKED(object);
941 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
942 	return (vm_page_insert_after(m, object, pindex, mpred));
943 }
944 
945 /*
946  *	vm_page_insert_after:
947  *
948  *	Inserts the page "m" into the specified object at offset "pindex".
949  *
950  *	The page "mpred" must immediately precede the offset "pindex" within
951  *	the specified object.
952  *
953  *	The object must be locked.
954  */
955 static int
956 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
957     vm_page_t mpred)
958 {
959 	vm_pindex_t sidx;
960 	vm_object_t sobj;
961 	vm_page_t msucc;
962 
963 	VM_OBJECT_ASSERT_WLOCKED(object);
964 	KASSERT(m->object == NULL,
965 	    ("vm_page_insert_after: page already inserted"));
966 	if (mpred != NULL) {
967 		KASSERT(mpred->object == object ||
968 		    (mpred->flags & PG_SLAB) != 0,
969 		    ("vm_page_insert_after: object doesn't contain mpred"));
970 		KASSERT(mpred->pindex < pindex,
971 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
972 		msucc = TAILQ_NEXT(mpred, listq);
973 	} else
974 		msucc = TAILQ_FIRST(&object->memq);
975 	if (msucc != NULL)
976 		KASSERT(msucc->pindex > pindex,
977 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
978 
979 	/*
980 	 * Record the object/offset pair in this page
981 	 */
982 	sobj = m->object;
983 	sidx = m->pindex;
984 	m->object = object;
985 	m->pindex = pindex;
986 
987 	/*
988 	 * Now link into the object's ordered list of backed pages.
989 	 */
990 	if (vm_radix_insert(&object->rtree, m)) {
991 		m->object = sobj;
992 		m->pindex = sidx;
993 		return (1);
994 	}
995 	vm_page_insert_radixdone(m, object, mpred);
996 	return (0);
997 }
998 
999 /*
1000  *	vm_page_insert_radixdone:
1001  *
1002  *	Complete page "m" insertion into the specified object after the
1003  *	radix trie hooking.
1004  *
1005  *	The page "mpred" must precede the offset "m->pindex" within the
1006  *	specified object.
1007  *
1008  *	The object must be locked.
1009  */
1010 static void
1011 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1012 {
1013 
1014 	VM_OBJECT_ASSERT_WLOCKED(object);
1015 	KASSERT(object != NULL && m->object == object,
1016 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1017 	if (mpred != NULL) {
1018 		KASSERT(mpred->object == object ||
1019 		    (mpred->flags & PG_SLAB) != 0,
1020 		    ("vm_page_insert_after: object doesn't contain mpred"));
1021 		KASSERT(mpred->pindex < m->pindex,
1022 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1023 	}
1024 
1025 	if (mpred != NULL)
1026 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1027 	else
1028 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1029 
1030 	/*
1031 	 * Show that the object has one more resident page.
1032 	 */
1033 	object->resident_page_count++;
1034 
1035 	/*
1036 	 * Hold the vnode until the last page is released.
1037 	 */
1038 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1039 		vhold(object->handle);
1040 
1041 	/*
1042 	 * Since we are inserting a new and possibly dirty page,
1043 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1044 	 */
1045 	if (pmap_page_is_write_mapped(m))
1046 		vm_object_set_writeable_dirty(object);
1047 }
1048 
1049 /*
1050  *	vm_page_remove:
1051  *
1052  *	Removes the given mem entry from the object/offset-page
1053  *	table and the object page list, but do not invalidate/terminate
1054  *	the backing store.
1055  *
1056  *	The object must be locked.  The page must be locked if it is managed.
1057  */
1058 void
1059 vm_page_remove(vm_page_t m)
1060 {
1061 	vm_object_t object;
1062 	boolean_t lockacq;
1063 
1064 	if ((m->oflags & VPO_UNMANAGED) == 0)
1065 		vm_page_lock_assert(m, MA_OWNED);
1066 	if ((object = m->object) == NULL)
1067 		return;
1068 	VM_OBJECT_ASSERT_WLOCKED(object);
1069 	if (vm_page_xbusied(m)) {
1070 		lockacq = FALSE;
1071 		if ((m->oflags & VPO_UNMANAGED) != 0 &&
1072 		    !mtx_owned(vm_page_lockptr(m))) {
1073 			lockacq = TRUE;
1074 			vm_page_lock(m);
1075 		}
1076 		vm_page_flash(m);
1077 		atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
1078 		if (lockacq)
1079 			vm_page_unlock(m);
1080 	}
1081 
1082 	/*
1083 	 * Now remove from the object's list of backed pages.
1084 	 */
1085 	vm_radix_remove(&object->rtree, m->pindex);
1086 	TAILQ_REMOVE(&object->memq, m, listq);
1087 
1088 	/*
1089 	 * And show that the object has one fewer resident page.
1090 	 */
1091 	object->resident_page_count--;
1092 
1093 	/*
1094 	 * The vnode may now be recycled.
1095 	 */
1096 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1097 		vdrop(object->handle);
1098 
1099 	m->object = NULL;
1100 }
1101 
1102 /*
1103  *	vm_page_lookup:
1104  *
1105  *	Returns the page associated with the object/offset
1106  *	pair specified; if none is found, NULL is returned.
1107  *
1108  *	The object must be locked.
1109  */
1110 vm_page_t
1111 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1112 {
1113 
1114 	VM_OBJECT_ASSERT_LOCKED(object);
1115 	return (vm_radix_lookup(&object->rtree, pindex));
1116 }
1117 
1118 /*
1119  *	vm_page_find_least:
1120  *
1121  *	Returns the page associated with the object with least pindex
1122  *	greater than or equal to the parameter pindex, or NULL.
1123  *
1124  *	The object must be locked.
1125  */
1126 vm_page_t
1127 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1128 {
1129 	vm_page_t m;
1130 
1131 	VM_OBJECT_ASSERT_LOCKED(object);
1132 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1133 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1134 	return (m);
1135 }
1136 
1137 /*
1138  * Returns the given page's successor (by pindex) within the object if it is
1139  * resident; if none is found, NULL is returned.
1140  *
1141  * The object must be locked.
1142  */
1143 vm_page_t
1144 vm_page_next(vm_page_t m)
1145 {
1146 	vm_page_t next;
1147 
1148 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1149 	if ((next = TAILQ_NEXT(m, listq)) != NULL &&
1150 	    next->pindex != m->pindex + 1)
1151 		next = NULL;
1152 	return (next);
1153 }
1154 
1155 /*
1156  * Returns the given page's predecessor (by pindex) within the object if it is
1157  * resident; if none is found, NULL is returned.
1158  *
1159  * The object must be locked.
1160  */
1161 vm_page_t
1162 vm_page_prev(vm_page_t m)
1163 {
1164 	vm_page_t prev;
1165 
1166 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1167 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL &&
1168 	    prev->pindex != m->pindex - 1)
1169 		prev = NULL;
1170 	return (prev);
1171 }
1172 
1173 /*
1174  * Uses the page mnew as a replacement for an existing page at index
1175  * pindex which must be already present in the object.
1176  *
1177  * The existing page must not be on a paging queue.
1178  */
1179 vm_page_t
1180 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1181 {
1182 	vm_page_t mold, mpred;
1183 
1184 	VM_OBJECT_ASSERT_WLOCKED(object);
1185 
1186 	/*
1187 	 * This function mostly follows vm_page_insert() and
1188 	 * vm_page_remove() without the radix, object count and vnode
1189 	 * dance.  Double check such functions for more comments.
1190 	 */
1191 	mpred = vm_radix_lookup(&object->rtree, pindex);
1192 	KASSERT(mpred != NULL,
1193 	    ("vm_page_replace: replacing page not present with pindex"));
1194 	mpred = TAILQ_PREV(mpred, respgs, listq);
1195 	if (mpred != NULL)
1196 		KASSERT(mpred->pindex < pindex,
1197 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1198 
1199 	mnew->object = object;
1200 	mnew->pindex = pindex;
1201 	mold = vm_radix_replace(&object->rtree, mnew, pindex);
1202 	KASSERT(mold->queue == PQ_NONE,
1203 	    ("vm_page_replace: mold is on a paging queue"));
1204 
1205 	/* Detach the old page from the resident tailq. */
1206 	TAILQ_REMOVE(&object->memq, mold, listq);
1207 
1208 	mold->object = NULL;
1209 	vm_page_xunbusy(mold);
1210 
1211 	/* Insert the new page in the resident tailq. */
1212 	if (mpred != NULL)
1213 		TAILQ_INSERT_AFTER(&object->memq, mpred, mnew, listq);
1214 	else
1215 		TAILQ_INSERT_HEAD(&object->memq, mnew, listq);
1216 	if (pmap_page_is_write_mapped(mnew))
1217 		vm_object_set_writeable_dirty(object);
1218 	return (mold);
1219 }
1220 
1221 /*
1222  *	vm_page_rename:
1223  *
1224  *	Move the given memory entry from its
1225  *	current object to the specified target object/offset.
1226  *
1227  *	Note: swap associated with the page must be invalidated by the move.  We
1228  *	      have to do this for several reasons:  (1) we aren't freeing the
1229  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1230  *	      moving the page from object A to B, and will then later move
1231  *	      the backing store from A to B and we can't have a conflict.
1232  *
1233  *	Note: we *always* dirty the page.  It is necessary both for the
1234  *	      fact that we moved it, and because we may be invalidating
1235  *	      swap.  If the page is on the cache, we have to deactivate it
1236  *	      or vm_page_dirty() will panic.  Dirty pages are not allowed
1237  *	      on the cache.
1238  *
1239  *	The objects must be locked.
1240  */
1241 int
1242 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1243 {
1244 	vm_page_t mpred;
1245 	vm_pindex_t opidx;
1246 
1247 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1248 
1249 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1250 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1251 	    ("vm_page_rename: pindex already renamed"));
1252 
1253 	/*
1254 	 * Create a custom version of vm_page_insert() which does not depend
1255 	 * by m_prev and can cheat on the implementation aspects of the
1256 	 * function.
1257 	 */
1258 	opidx = m->pindex;
1259 	m->pindex = new_pindex;
1260 	if (vm_radix_insert(&new_object->rtree, m)) {
1261 		m->pindex = opidx;
1262 		return (1);
1263 	}
1264 
1265 	/*
1266 	 * The operation cannot fail anymore.  The removal must happen before
1267 	 * the listq iterator is tainted.
1268 	 */
1269 	m->pindex = opidx;
1270 	vm_page_lock(m);
1271 	vm_page_remove(m);
1272 
1273 	/* Return back to the new pindex to complete vm_page_insert(). */
1274 	m->pindex = new_pindex;
1275 	m->object = new_object;
1276 	vm_page_unlock(m);
1277 	vm_page_insert_radixdone(m, new_object, mpred);
1278 	vm_page_dirty(m);
1279 	return (0);
1280 }
1281 
1282 /*
1283  *	Convert all of the given object's cached pages that have a
1284  *	pindex within the given range into free pages.  If the value
1285  *	zero is given for "end", then the range's upper bound is
1286  *	infinity.  If the given object is backed by a vnode and it
1287  *	transitions from having one or more cached pages to none, the
1288  *	vnode's hold count is reduced.
1289  */
1290 void
1291 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1292 {
1293 	vm_page_t m;
1294 	boolean_t empty;
1295 
1296 	mtx_lock(&vm_page_queue_free_mtx);
1297 	if (__predict_false(vm_radix_is_empty(&object->cache))) {
1298 		mtx_unlock(&vm_page_queue_free_mtx);
1299 		return;
1300 	}
1301 	while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) {
1302 		if (end != 0 && m->pindex >= end)
1303 			break;
1304 		vm_radix_remove(&object->cache, m->pindex);
1305 		vm_page_cache_turn_free(m);
1306 	}
1307 	empty = vm_radix_is_empty(&object->cache);
1308 	mtx_unlock(&vm_page_queue_free_mtx);
1309 	if (object->type == OBJT_VNODE && empty)
1310 		vdrop(object->handle);
1311 }
1312 
1313 /*
1314  *	Returns the cached page that is associated with the given
1315  *	object and offset.  If, however, none exists, returns NULL.
1316  *
1317  *	The free page queue must be locked.
1318  */
1319 static inline vm_page_t
1320 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
1321 {
1322 
1323 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1324 	return (vm_radix_lookup(&object->cache, pindex));
1325 }
1326 
1327 /*
1328  *	Remove the given cached page from its containing object's
1329  *	collection of cached pages.
1330  *
1331  *	The free page queue must be locked.
1332  */
1333 static void
1334 vm_page_cache_remove(vm_page_t m)
1335 {
1336 
1337 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1338 	KASSERT((m->flags & PG_CACHED) != 0,
1339 	    ("vm_page_cache_remove: page %p is not cached", m));
1340 	vm_radix_remove(&m->object->cache, m->pindex);
1341 	m->object = NULL;
1342 	cnt.v_cache_count--;
1343 }
1344 
1345 /*
1346  *	Transfer all of the cached pages with offset greater than or
1347  *	equal to 'offidxstart' from the original object's cache to the
1348  *	new object's cache.  However, any cached pages with offset
1349  *	greater than or equal to the new object's size are kept in the
1350  *	original object.  Initially, the new object's cache must be
1351  *	empty.  Offset 'offidxstart' in the original object must
1352  *	correspond to offset zero in the new object.
1353  *
1354  *	The new object must be locked.
1355  */
1356 void
1357 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
1358     vm_object_t new_object)
1359 {
1360 	vm_page_t m;
1361 
1362 	/*
1363 	 * Insertion into an object's collection of cached pages
1364 	 * requires the object to be locked.  In contrast, removal does
1365 	 * not.
1366 	 */
1367 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1368 	KASSERT(vm_radix_is_empty(&new_object->cache),
1369 	    ("vm_page_cache_transfer: object %p has cached pages",
1370 	    new_object));
1371 	mtx_lock(&vm_page_queue_free_mtx);
1372 	while ((m = vm_radix_lookup_ge(&orig_object->cache,
1373 	    offidxstart)) != NULL) {
1374 		/*
1375 		 * Transfer all of the pages with offset greater than or
1376 		 * equal to 'offidxstart' from the original object's
1377 		 * cache to the new object's cache.
1378 		 */
1379 		if ((m->pindex - offidxstart) >= new_object->size)
1380 			break;
1381 		vm_radix_remove(&orig_object->cache, m->pindex);
1382 		/* Update the page's object and offset. */
1383 		m->object = new_object;
1384 		m->pindex -= offidxstart;
1385 		if (vm_radix_insert(&new_object->cache, m))
1386 			vm_page_cache_turn_free(m);
1387 	}
1388 	mtx_unlock(&vm_page_queue_free_mtx);
1389 }
1390 
1391 /*
1392  *	Returns TRUE if a cached page is associated with the given object and
1393  *	offset, and FALSE otherwise.
1394  *
1395  *	The object must be locked.
1396  */
1397 boolean_t
1398 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex)
1399 {
1400 	vm_page_t m;
1401 
1402 	/*
1403 	 * Insertion into an object's collection of cached pages requires the
1404 	 * object to be locked.  Therefore, if the object is locked and the
1405 	 * object's collection is empty, there is no need to acquire the free
1406 	 * page queues lock in order to prove that the specified page doesn't
1407 	 * exist.
1408 	 */
1409 	VM_OBJECT_ASSERT_WLOCKED(object);
1410 	if (__predict_true(vm_object_cache_is_empty(object)))
1411 		return (FALSE);
1412 	mtx_lock(&vm_page_queue_free_mtx);
1413 	m = vm_page_cache_lookup(object, pindex);
1414 	mtx_unlock(&vm_page_queue_free_mtx);
1415 	return (m != NULL);
1416 }
1417 
1418 /*
1419  *	vm_page_alloc:
1420  *
1421  *	Allocate and return a page that is associated with the specified
1422  *	object and offset pair.  By default, this page is exclusive busied.
1423  *
1424  *	The caller must always specify an allocation class.
1425  *
1426  *	allocation classes:
1427  *	VM_ALLOC_NORMAL		normal process request
1428  *	VM_ALLOC_SYSTEM		system *really* needs a page
1429  *	VM_ALLOC_INTERRUPT	interrupt time request
1430  *
1431  *	optional allocation flags:
1432  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1433  *				intends to allocate
1434  *	VM_ALLOC_IFCACHED	return page only if it is cached
1435  *	VM_ALLOC_IFNOTCACHED	return NULL, do not reactivate if the page
1436  *				is cached
1437  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1438  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1439  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1440  *				should not be exclusive busy
1441  *	VM_ALLOC_SBUSY		shared busy the allocated page
1442  *	VM_ALLOC_WIRED		wire the allocated page
1443  *	VM_ALLOC_ZERO		prefer a zeroed page
1444  *
1445  *	This routine may not sleep.
1446  */
1447 vm_page_t
1448 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1449 {
1450 	struct vnode *vp = NULL;
1451 	vm_object_t m_object;
1452 	vm_page_t m, mpred;
1453 	int flags, req_class;
1454 
1455 	mpred = 0;	/* XXX: pacify gcc */
1456 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1457 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1458 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1459 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1460 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1461 	    req));
1462 	if (object != NULL)
1463 		VM_OBJECT_ASSERT_WLOCKED(object);
1464 
1465 	req_class = req & VM_ALLOC_CLASS_MASK;
1466 
1467 	/*
1468 	 * The page daemon is allowed to dig deeper into the free page list.
1469 	 */
1470 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1471 		req_class = VM_ALLOC_SYSTEM;
1472 
1473 	if (object != NULL) {
1474 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1475 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1476 		   ("vm_page_alloc: pindex already allocated"));
1477 	}
1478 
1479 	/*
1480 	 * The page allocation request can came from consumers which already
1481 	 * hold the free page queue mutex, like vm_page_insert() in
1482 	 * vm_page_cache().
1483 	 */
1484 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
1485 	if (cnt.v_free_count + cnt.v_cache_count > cnt.v_free_reserved ||
1486 	    (req_class == VM_ALLOC_SYSTEM &&
1487 	    cnt.v_free_count + cnt.v_cache_count > cnt.v_interrupt_free_min) ||
1488 	    (req_class == VM_ALLOC_INTERRUPT &&
1489 	    cnt.v_free_count + cnt.v_cache_count > 0)) {
1490 		/*
1491 		 * Allocate from the free queue if the number of free pages
1492 		 * exceeds the minimum for the request class.
1493 		 */
1494 		if (object != NULL &&
1495 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1496 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1497 				mtx_unlock(&vm_page_queue_free_mtx);
1498 				return (NULL);
1499 			}
1500 			if (vm_phys_unfree_page(m))
1501 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1502 #if VM_NRESERVLEVEL > 0
1503 			else if (!vm_reserv_reactivate_page(m))
1504 #else
1505 			else
1506 #endif
1507 				panic("vm_page_alloc: cache page %p is missing"
1508 				    " from the free queue", m);
1509 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1510 			mtx_unlock(&vm_page_queue_free_mtx);
1511 			return (NULL);
1512 #if VM_NRESERVLEVEL > 0
1513 		} else if (object == NULL || (object->flags & (OBJ_COLORED |
1514 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1515 		    vm_reserv_alloc_page(object, pindex, mpred)) == NULL) {
1516 #else
1517 		} else {
1518 #endif
1519 			m = vm_phys_alloc_pages(object != NULL ?
1520 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1521 #if VM_NRESERVLEVEL > 0
1522 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1523 				m = vm_phys_alloc_pages(object != NULL ?
1524 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1525 				    0);
1526 			}
1527 #endif
1528 		}
1529 	} else {
1530 		/*
1531 		 * Not allocatable, give up.
1532 		 */
1533 		mtx_unlock(&vm_page_queue_free_mtx);
1534 		atomic_add_int(&vm_pageout_deficit,
1535 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1536 		pagedaemon_wakeup();
1537 		return (NULL);
1538 	}
1539 
1540 	/*
1541 	 *  At this point we had better have found a good page.
1542 	 */
1543 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1544 	KASSERT(m->queue == PQ_NONE,
1545 	    ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
1546 	KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
1547 	KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
1548 	KASSERT(!vm_page_sbusied(m),
1549 	    ("vm_page_alloc: page %p is busy", m));
1550 	KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
1551 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1552 	    ("vm_page_alloc: page %p has unexpected memattr %d", m,
1553 	    pmap_page_get_memattr(m)));
1554 	if ((m->flags & PG_CACHED) != 0) {
1555 		KASSERT((m->flags & PG_ZERO) == 0,
1556 		    ("vm_page_alloc: cached page %p is PG_ZERO", m));
1557 		KASSERT(m->valid != 0,
1558 		    ("vm_page_alloc: cached page %p is invalid", m));
1559 		if (m->object == object && m->pindex == pindex)
1560 	  		cnt.v_reactivated++;
1561 		else
1562 			m->valid = 0;
1563 		m_object = m->object;
1564 		vm_page_cache_remove(m);
1565 		if (m_object->type == OBJT_VNODE &&
1566 		    vm_object_cache_is_empty(m_object))
1567 			vp = m_object->handle;
1568 	} else {
1569 		KASSERT(VM_PAGE_IS_FREE(m),
1570 		    ("vm_page_alloc: page %p is not free", m));
1571 		KASSERT(m->valid == 0,
1572 		    ("vm_page_alloc: free page %p is valid", m));
1573 		vm_phys_freecnt_adj(m, -1);
1574 	}
1575 
1576 	/*
1577 	 * Only the PG_ZERO flag is inherited.  The PG_CACHED or PG_FREE flag
1578 	 * must be cleared before the free page queues lock is released.
1579 	 */
1580 	flags = 0;
1581 	if (m->flags & PG_ZERO) {
1582 		vm_page_zero_count--;
1583 		if (req & VM_ALLOC_ZERO)
1584 			flags = PG_ZERO;
1585 	}
1586 	if (req & VM_ALLOC_NODUMP)
1587 		flags |= PG_NODUMP;
1588 	m->flags = flags;
1589 	mtx_unlock(&vm_page_queue_free_mtx);
1590 	m->aflags = 0;
1591 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1592 	    VPO_UNMANAGED : 0;
1593 	m->busy_lock = VPB_UNBUSIED;
1594 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1595 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1596 	if ((req & VM_ALLOC_SBUSY) != 0)
1597 		m->busy_lock = VPB_SHARERS_WORD(1);
1598 	if (req & VM_ALLOC_WIRED) {
1599 		/*
1600 		 * The page lock is not required for wiring a page until that
1601 		 * page is inserted into the object.
1602 		 */
1603 		atomic_add_int(&cnt.v_wire_count, 1);
1604 		m->wire_count = 1;
1605 	}
1606 	m->act_count = 0;
1607 
1608 	if (object != NULL) {
1609 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1610 			/* See the comment below about hold count. */
1611 			if (vp != NULL)
1612 				vdrop(vp);
1613 			pagedaemon_wakeup();
1614 			if (req & VM_ALLOC_WIRED) {
1615 				atomic_subtract_int(&cnt.v_wire_count, 1);
1616 				m->wire_count = 0;
1617 			}
1618 			m->object = NULL;
1619 			vm_page_free(m);
1620 			return (NULL);
1621 		}
1622 
1623 		/* Ignore device objects; the pager sets "memattr" for them. */
1624 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1625 		    (object->flags & OBJ_FICTITIOUS) == 0)
1626 			pmap_page_set_memattr(m, object->memattr);
1627 	} else
1628 		m->pindex = pindex;
1629 
1630 	/*
1631 	 * The following call to vdrop() must come after the above call
1632 	 * to vm_page_insert() in case both affect the same object and
1633 	 * vnode.  Otherwise, the affected vnode's hold count could
1634 	 * temporarily become zero.
1635 	 */
1636 	if (vp != NULL)
1637 		vdrop(vp);
1638 
1639 	/*
1640 	 * Don't wakeup too often - wakeup the pageout daemon when
1641 	 * we would be nearly out of memory.
1642 	 */
1643 	if (vm_paging_needed())
1644 		pagedaemon_wakeup();
1645 
1646 	return (m);
1647 }
1648 
1649 static void
1650 vm_page_alloc_contig_vdrop(struct spglist *lst)
1651 {
1652 
1653 	while (!SLIST_EMPTY(lst)) {
1654 		vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv);
1655 		SLIST_REMOVE_HEAD(lst, plinks.s.ss);
1656 	}
1657 }
1658 
1659 /*
1660  *	vm_page_alloc_contig:
1661  *
1662  *	Allocate a contiguous set of physical pages of the given size "npages"
1663  *	from the free lists.  All of the physical pages must be at or above
1664  *	the given physical address "low" and below the given physical address
1665  *	"high".  The given value "alignment" determines the alignment of the
1666  *	first physical page in the set.  If the given value "boundary" is
1667  *	non-zero, then the set of physical pages cannot cross any physical
1668  *	address boundary that is a multiple of that value.  Both "alignment"
1669  *	and "boundary" must be a power of two.
1670  *
1671  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1672  *	then the memory attribute setting for the physical pages is configured
1673  *	to the object's memory attribute setting.  Otherwise, the memory
1674  *	attribute setting for the physical pages is configured to "memattr",
1675  *	overriding the object's memory attribute setting.  However, if the
1676  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1677  *	memory attribute setting for the physical pages cannot be configured
1678  *	to VM_MEMATTR_DEFAULT.
1679  *
1680  *	The caller must always specify an allocation class.
1681  *
1682  *	allocation classes:
1683  *	VM_ALLOC_NORMAL		normal process request
1684  *	VM_ALLOC_SYSTEM		system *really* needs a page
1685  *	VM_ALLOC_INTERRUPT	interrupt time request
1686  *
1687  *	optional allocation flags:
1688  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1689  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1690  *				should not be exclusive busy
1691  *	VM_ALLOC_SBUSY		shared busy the allocated page
1692  *	VM_ALLOC_WIRED		wire the allocated page
1693  *	VM_ALLOC_ZERO		prefer a zeroed page
1694  *
1695  *	This routine may not sleep.
1696  */
1697 vm_page_t
1698 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1699     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1700     vm_paddr_t boundary, vm_memattr_t memattr)
1701 {
1702 	struct vnode *drop;
1703 	struct spglist deferred_vdrop_list;
1704 	vm_page_t m, m_tmp, m_ret;
1705 	u_int flags, oflags;
1706 	int req_class;
1707 
1708 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1709 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1710 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1711 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1712 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1713 	    req));
1714 	if (object != NULL) {
1715 		VM_OBJECT_ASSERT_WLOCKED(object);
1716 		KASSERT(object->type == OBJT_PHYS,
1717 		    ("vm_page_alloc_contig: object %p isn't OBJT_PHYS",
1718 		    object));
1719 	}
1720 	KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1721 	req_class = req & VM_ALLOC_CLASS_MASK;
1722 
1723 	/*
1724 	 * The page daemon is allowed to dig deeper into the free page list.
1725 	 */
1726 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1727 		req_class = VM_ALLOC_SYSTEM;
1728 
1729 	SLIST_INIT(&deferred_vdrop_list);
1730 	mtx_lock(&vm_page_queue_free_mtx);
1731 	if (cnt.v_free_count + cnt.v_cache_count >= npages +
1732 	    cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM &&
1733 	    cnt.v_free_count + cnt.v_cache_count >= npages +
1734 	    cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT &&
1735 	    cnt.v_free_count + cnt.v_cache_count >= npages)) {
1736 #if VM_NRESERVLEVEL > 0
1737 retry:
1738 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1739 		    (m_ret = vm_reserv_alloc_contig(object, pindex, npages,
1740 		    low, high, alignment, boundary)) == NULL)
1741 #endif
1742 			m_ret = vm_phys_alloc_contig(npages, low, high,
1743 			    alignment, boundary);
1744 	} else {
1745 		mtx_unlock(&vm_page_queue_free_mtx);
1746 		atomic_add_int(&vm_pageout_deficit, npages);
1747 		pagedaemon_wakeup();
1748 		return (NULL);
1749 	}
1750 	if (m_ret != NULL)
1751 		for (m = m_ret; m < &m_ret[npages]; m++) {
1752 			drop = vm_page_alloc_init(m);
1753 			if (drop != NULL) {
1754 				/*
1755 				 * Enqueue the vnode for deferred vdrop().
1756 				 */
1757 				m->plinks.s.pv = drop;
1758 				SLIST_INSERT_HEAD(&deferred_vdrop_list, m,
1759 				    plinks.s.ss);
1760 			}
1761 		}
1762 	else {
1763 #if VM_NRESERVLEVEL > 0
1764 		if (vm_reserv_reclaim_contig(npages, low, high, alignment,
1765 		    boundary))
1766 			goto retry;
1767 #endif
1768 	}
1769 	mtx_unlock(&vm_page_queue_free_mtx);
1770 	if (m_ret == NULL)
1771 		return (NULL);
1772 
1773 	/*
1774 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1775 	 */
1776 	flags = 0;
1777 	if ((req & VM_ALLOC_ZERO) != 0)
1778 		flags = PG_ZERO;
1779 	if ((req & VM_ALLOC_NODUMP) != 0)
1780 		flags |= PG_NODUMP;
1781 	if ((req & VM_ALLOC_WIRED) != 0)
1782 		atomic_add_int(&cnt.v_wire_count, npages);
1783 	oflags = VPO_UNMANAGED;
1784 	if (object != NULL) {
1785 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1786 		    memattr == VM_MEMATTR_DEFAULT)
1787 			memattr = object->memattr;
1788 	}
1789 	for (m = m_ret; m < &m_ret[npages]; m++) {
1790 		m->aflags = 0;
1791 		m->flags = (m->flags | PG_NODUMP) & flags;
1792 		m->busy_lock = VPB_UNBUSIED;
1793 		if (object != NULL) {
1794 			if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
1795 				m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1796 			if ((req & VM_ALLOC_SBUSY) != 0)
1797 				m->busy_lock = VPB_SHARERS_WORD(1);
1798 		}
1799 		if ((req & VM_ALLOC_WIRED) != 0)
1800 			m->wire_count = 1;
1801 		/* Unmanaged pages don't use "act_count". */
1802 		m->oflags = oflags;
1803 		if (object != NULL) {
1804 			if (vm_page_insert(m, object, pindex)) {
1805 				vm_page_alloc_contig_vdrop(
1806 				    &deferred_vdrop_list);
1807 				if (vm_paging_needed())
1808 					pagedaemon_wakeup();
1809 				if ((req & VM_ALLOC_WIRED) != 0)
1810 					atomic_subtract_int(&cnt.v_wire_count,
1811 					    npages);
1812 				for (m_tmp = m, m = m_ret;
1813 				    m < &m_ret[npages]; m++) {
1814 					if ((req & VM_ALLOC_WIRED) != 0)
1815 						m->wire_count = 0;
1816 					if (m >= m_tmp)
1817 						m->object = NULL;
1818 					vm_page_free(m);
1819 				}
1820 				return (NULL);
1821 			}
1822 		} else
1823 			m->pindex = pindex;
1824 		if (memattr != VM_MEMATTR_DEFAULT)
1825 			pmap_page_set_memattr(m, memattr);
1826 		pindex++;
1827 	}
1828 	vm_page_alloc_contig_vdrop(&deferred_vdrop_list);
1829 	if (vm_paging_needed())
1830 		pagedaemon_wakeup();
1831 	return (m_ret);
1832 }
1833 
1834 /*
1835  * Initialize a page that has been freshly dequeued from a freelist.
1836  * The caller has to drop the vnode returned, if it is not NULL.
1837  *
1838  * This function may only be used to initialize unmanaged pages.
1839  *
1840  * To be called with vm_page_queue_free_mtx held.
1841  */
1842 static struct vnode *
1843 vm_page_alloc_init(vm_page_t m)
1844 {
1845 	struct vnode *drop;
1846 	vm_object_t m_object;
1847 
1848 	KASSERT(m->queue == PQ_NONE,
1849 	    ("vm_page_alloc_init: page %p has unexpected queue %d",
1850 	    m, m->queue));
1851 	KASSERT(m->wire_count == 0,
1852 	    ("vm_page_alloc_init: page %p is wired", m));
1853 	KASSERT(m->hold_count == 0,
1854 	    ("vm_page_alloc_init: page %p is held", m));
1855 	KASSERT(!vm_page_sbusied(m),
1856 	    ("vm_page_alloc_init: page %p is busy", m));
1857 	KASSERT(m->dirty == 0,
1858 	    ("vm_page_alloc_init: page %p is dirty", m));
1859 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1860 	    ("vm_page_alloc_init: page %p has unexpected memattr %d",
1861 	    m, pmap_page_get_memattr(m)));
1862 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1863 	drop = NULL;
1864 	if ((m->flags & PG_CACHED) != 0) {
1865 		KASSERT((m->flags & PG_ZERO) == 0,
1866 		    ("vm_page_alloc_init: cached page %p is PG_ZERO", m));
1867 		m->valid = 0;
1868 		m_object = m->object;
1869 		vm_page_cache_remove(m);
1870 		if (m_object->type == OBJT_VNODE &&
1871 		    vm_object_cache_is_empty(m_object))
1872 			drop = m_object->handle;
1873 	} else {
1874 		KASSERT(VM_PAGE_IS_FREE(m),
1875 		    ("vm_page_alloc_init: page %p is not free", m));
1876 		KASSERT(m->valid == 0,
1877 		    ("vm_page_alloc_init: free page %p is valid", m));
1878 		vm_phys_freecnt_adj(m, -1);
1879 		if ((m->flags & PG_ZERO) != 0)
1880 			vm_page_zero_count--;
1881 	}
1882 	/* Don't clear the PG_ZERO flag; we'll need it later. */
1883 	m->flags &= PG_ZERO;
1884 	return (drop);
1885 }
1886 
1887 /*
1888  * 	vm_page_alloc_freelist:
1889  *
1890  *	Allocate a physical page from the specified free page list.
1891  *
1892  *	The caller must always specify an allocation class.
1893  *
1894  *	allocation classes:
1895  *	VM_ALLOC_NORMAL		normal process request
1896  *	VM_ALLOC_SYSTEM		system *really* needs a page
1897  *	VM_ALLOC_INTERRUPT	interrupt time request
1898  *
1899  *	optional allocation flags:
1900  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1901  *				intends to allocate
1902  *	VM_ALLOC_WIRED		wire the allocated page
1903  *	VM_ALLOC_ZERO		prefer a zeroed page
1904  *
1905  *	This routine may not sleep.
1906  */
1907 vm_page_t
1908 vm_page_alloc_freelist(int flind, int req)
1909 {
1910 	struct vnode *drop;
1911 	vm_page_t m;
1912 	u_int flags;
1913 	int req_class;
1914 
1915 	req_class = req & VM_ALLOC_CLASS_MASK;
1916 
1917 	/*
1918 	 * The page daemon is allowed to dig deeper into the free page list.
1919 	 */
1920 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1921 		req_class = VM_ALLOC_SYSTEM;
1922 
1923 	/*
1924 	 * Do not allocate reserved pages unless the req has asked for it.
1925 	 */
1926 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
1927 	if (cnt.v_free_count + cnt.v_cache_count > cnt.v_free_reserved ||
1928 	    (req_class == VM_ALLOC_SYSTEM &&
1929 	    cnt.v_free_count + cnt.v_cache_count > cnt.v_interrupt_free_min) ||
1930 	    (req_class == VM_ALLOC_INTERRUPT &&
1931 	    cnt.v_free_count + cnt.v_cache_count > 0))
1932 		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0);
1933 	else {
1934 		mtx_unlock(&vm_page_queue_free_mtx);
1935 		atomic_add_int(&vm_pageout_deficit,
1936 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1937 		pagedaemon_wakeup();
1938 		return (NULL);
1939 	}
1940 	if (m == NULL) {
1941 		mtx_unlock(&vm_page_queue_free_mtx);
1942 		return (NULL);
1943 	}
1944 	drop = vm_page_alloc_init(m);
1945 	mtx_unlock(&vm_page_queue_free_mtx);
1946 
1947 	/*
1948 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1949 	 */
1950 	m->aflags = 0;
1951 	flags = 0;
1952 	if ((req & VM_ALLOC_ZERO) != 0)
1953 		flags = PG_ZERO;
1954 	m->flags &= flags;
1955 	if ((req & VM_ALLOC_WIRED) != 0) {
1956 		/*
1957 		 * The page lock is not required for wiring a page that does
1958 		 * not belong to an object.
1959 		 */
1960 		atomic_add_int(&cnt.v_wire_count, 1);
1961 		m->wire_count = 1;
1962 	}
1963 	/* Unmanaged pages don't use "act_count". */
1964 	m->oflags = VPO_UNMANAGED;
1965 	if (drop != NULL)
1966 		vdrop(drop);
1967 	if (vm_paging_needed())
1968 		pagedaemon_wakeup();
1969 	return (m);
1970 }
1971 
1972 /*
1973  *	vm_wait:	(also see VM_WAIT macro)
1974  *
1975  *	Sleep until free pages are available for allocation.
1976  *	- Called in various places before memory allocations.
1977  */
1978 void
1979 vm_wait(void)
1980 {
1981 
1982 	mtx_lock(&vm_page_queue_free_mtx);
1983 	if (curproc == pageproc) {
1984 		vm_pageout_pages_needed = 1;
1985 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
1986 		    PDROP | PSWP, "VMWait", 0);
1987 	} else {
1988 		if (!vm_pages_needed) {
1989 			vm_pages_needed = 1;
1990 			wakeup(&vm_pages_needed);
1991 		}
1992 		msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
1993 		    "vmwait", 0);
1994 	}
1995 }
1996 
1997 /*
1998  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
1999  *
2000  *	Sleep until free pages are available for allocation.
2001  *	- Called only in vm_fault so that processes page faulting
2002  *	  can be easily tracked.
2003  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2004  *	  processes will be able to grab memory first.  Do not change
2005  *	  this balance without careful testing first.
2006  */
2007 void
2008 vm_waitpfault(void)
2009 {
2010 
2011 	mtx_lock(&vm_page_queue_free_mtx);
2012 	if (!vm_pages_needed) {
2013 		vm_pages_needed = 1;
2014 		wakeup(&vm_pages_needed);
2015 	}
2016 	msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
2017 	    "pfault", 0);
2018 }
2019 
2020 struct vm_pagequeue *
2021 vm_page_pagequeue(vm_page_t m)
2022 {
2023 
2024 	return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2025 }
2026 
2027 /*
2028  *	vm_page_dequeue:
2029  *
2030  *	Remove the given page from its current page queue.
2031  *
2032  *	The page must be locked.
2033  */
2034 void
2035 vm_page_dequeue(vm_page_t m)
2036 {
2037 	struct vm_pagequeue *pq;
2038 
2039 	vm_page_lock_assert(m, MA_OWNED);
2040 	KASSERT(m->queue != PQ_NONE,
2041 	    ("vm_page_dequeue: page %p is not queued", m));
2042 	pq = vm_page_pagequeue(m);
2043 	vm_pagequeue_lock(pq);
2044 	m->queue = PQ_NONE;
2045 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2046 	vm_pagequeue_cnt_dec(pq);
2047 	vm_pagequeue_unlock(pq);
2048 }
2049 
2050 /*
2051  *	vm_page_dequeue_locked:
2052  *
2053  *	Remove the given page from its current page queue.
2054  *
2055  *	The page and page queue must be locked.
2056  */
2057 void
2058 vm_page_dequeue_locked(vm_page_t m)
2059 {
2060 	struct vm_pagequeue *pq;
2061 
2062 	vm_page_lock_assert(m, MA_OWNED);
2063 	pq = vm_page_pagequeue(m);
2064 	vm_pagequeue_assert_locked(pq);
2065 	m->queue = PQ_NONE;
2066 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2067 	vm_pagequeue_cnt_dec(pq);
2068 }
2069 
2070 /*
2071  *	vm_page_enqueue:
2072  *
2073  *	Add the given page to the specified page queue.
2074  *
2075  *	The page must be locked.
2076  */
2077 static void
2078 vm_page_enqueue(int queue, vm_page_t m)
2079 {
2080 	struct vm_pagequeue *pq;
2081 
2082 	vm_page_lock_assert(m, MA_OWNED);
2083 	pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2084 	vm_pagequeue_lock(pq);
2085 	m->queue = queue;
2086 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2087 	vm_pagequeue_cnt_inc(pq);
2088 	vm_pagequeue_unlock(pq);
2089 }
2090 
2091 /*
2092  *	vm_page_requeue:
2093  *
2094  *	Move the given page to the tail of its current page queue.
2095  *
2096  *	The page must be locked.
2097  */
2098 void
2099 vm_page_requeue(vm_page_t m)
2100 {
2101 	struct vm_pagequeue *pq;
2102 
2103 	vm_page_lock_assert(m, MA_OWNED);
2104 	KASSERT(m->queue != PQ_NONE,
2105 	    ("vm_page_requeue: page %p is not queued", m));
2106 	pq = vm_page_pagequeue(m);
2107 	vm_pagequeue_lock(pq);
2108 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2109 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2110 	vm_pagequeue_unlock(pq);
2111 }
2112 
2113 /*
2114  *	vm_page_requeue_locked:
2115  *
2116  *	Move the given page to the tail of its current page queue.
2117  *
2118  *	The page queue must be locked.
2119  */
2120 void
2121 vm_page_requeue_locked(vm_page_t m)
2122 {
2123 	struct vm_pagequeue *pq;
2124 
2125 	KASSERT(m->queue != PQ_NONE,
2126 	    ("vm_page_requeue_locked: page %p is not queued", m));
2127 	pq = vm_page_pagequeue(m);
2128 	vm_pagequeue_assert_locked(pq);
2129 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2130 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2131 }
2132 
2133 /*
2134  *	vm_page_activate:
2135  *
2136  *	Put the specified page on the active list (if appropriate).
2137  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2138  *	mess with it.
2139  *
2140  *	The page must be locked.
2141  */
2142 void
2143 vm_page_activate(vm_page_t m)
2144 {
2145 	int queue;
2146 
2147 	vm_page_lock_assert(m, MA_OWNED);
2148 	if ((queue = m->queue) != PQ_ACTIVE) {
2149 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2150 			if (m->act_count < ACT_INIT)
2151 				m->act_count = ACT_INIT;
2152 			if (queue != PQ_NONE)
2153 				vm_page_dequeue(m);
2154 			vm_page_enqueue(PQ_ACTIVE, m);
2155 		} else
2156 			KASSERT(queue == PQ_NONE,
2157 			    ("vm_page_activate: wired page %p is queued", m));
2158 	} else {
2159 		if (m->act_count < ACT_INIT)
2160 			m->act_count = ACT_INIT;
2161 	}
2162 }
2163 
2164 /*
2165  *	vm_page_free_wakeup:
2166  *
2167  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
2168  *	routine is called when a page has been added to the cache or free
2169  *	queues.
2170  *
2171  *	The page queues must be locked.
2172  */
2173 static inline void
2174 vm_page_free_wakeup(void)
2175 {
2176 
2177 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2178 	/*
2179 	 * if pageout daemon needs pages, then tell it that there are
2180 	 * some free.
2181 	 */
2182 	if (vm_pageout_pages_needed &&
2183 	    cnt.v_cache_count + cnt.v_free_count >= cnt.v_pageout_free_min) {
2184 		wakeup(&vm_pageout_pages_needed);
2185 		vm_pageout_pages_needed = 0;
2186 	}
2187 	/*
2188 	 * wakeup processes that are waiting on memory if we hit a
2189 	 * high water mark. And wakeup scheduler process if we have
2190 	 * lots of memory. this process will swapin processes.
2191 	 */
2192 	if (vm_pages_needed && !vm_page_count_min()) {
2193 		vm_pages_needed = 0;
2194 		wakeup(&cnt.v_free_count);
2195 	}
2196 }
2197 
2198 /*
2199  *	Turn a cached page into a free page, by changing its attributes.
2200  *	Keep the statistics up-to-date.
2201  *
2202  *	The free page queue must be locked.
2203  */
2204 static void
2205 vm_page_cache_turn_free(vm_page_t m)
2206 {
2207 
2208 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2209 
2210 	m->object = NULL;
2211 	m->valid = 0;
2212 	/* Clear PG_CACHED and set PG_FREE. */
2213 	m->flags ^= PG_CACHED | PG_FREE;
2214 	KASSERT((m->flags & (PG_CACHED | PG_FREE)) == PG_FREE,
2215 	    ("vm_page_cache_free: page %p has inconsistent flags", m));
2216 	cnt.v_cache_count--;
2217 	vm_phys_freecnt_adj(m, 1);
2218 }
2219 
2220 /*
2221  *	vm_page_free_toq:
2222  *
2223  *	Returns the given page to the free list,
2224  *	disassociating it with any VM object.
2225  *
2226  *	The object must be locked.  The page must be locked if it is managed.
2227  */
2228 void
2229 vm_page_free_toq(vm_page_t m)
2230 {
2231 
2232 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2233 		vm_page_lock_assert(m, MA_OWNED);
2234 		KASSERT(!pmap_page_is_mapped(m),
2235 		    ("vm_page_free_toq: freeing mapped page %p", m));
2236 	} else
2237 		KASSERT(m->queue == PQ_NONE,
2238 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2239 	PCPU_INC(cnt.v_tfree);
2240 
2241 	if (VM_PAGE_IS_FREE(m))
2242 		panic("vm_page_free: freeing free page %p", m);
2243 	else if (vm_page_sbusied(m))
2244 		panic("vm_page_free: freeing busy page %p", m);
2245 
2246 	/*
2247 	 * Unqueue, then remove page.  Note that we cannot destroy
2248 	 * the page here because we do not want to call the pager's
2249 	 * callback routine until after we've put the page on the
2250 	 * appropriate free queue.
2251 	 */
2252 	vm_page_remque(m);
2253 	vm_page_remove(m);
2254 
2255 	/*
2256 	 * If fictitious remove object association and
2257 	 * return, otherwise delay object association removal.
2258 	 */
2259 	if ((m->flags & PG_FICTITIOUS) != 0) {
2260 		return;
2261 	}
2262 
2263 	m->valid = 0;
2264 	vm_page_undirty(m);
2265 
2266 	if (m->wire_count != 0)
2267 		panic("vm_page_free: freeing wired page %p", m);
2268 	if (m->hold_count != 0) {
2269 		m->flags &= ~PG_ZERO;
2270 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2271 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
2272 		m->flags |= PG_UNHOLDFREE;
2273 	} else {
2274 		/*
2275 		 * Restore the default memory attribute to the page.
2276 		 */
2277 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2278 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2279 
2280 		/*
2281 		 * Insert the page into the physical memory allocator's
2282 		 * cache/free page queues.
2283 		 */
2284 		mtx_lock(&vm_page_queue_free_mtx);
2285 		m->flags |= PG_FREE;
2286 		vm_phys_freecnt_adj(m, 1);
2287 #if VM_NRESERVLEVEL > 0
2288 		if (!vm_reserv_free_page(m))
2289 #else
2290 		if (TRUE)
2291 #endif
2292 			vm_phys_free_pages(m, 0);
2293 		if ((m->flags & PG_ZERO) != 0)
2294 			++vm_page_zero_count;
2295 		else
2296 			vm_page_zero_idle_wakeup();
2297 		vm_page_free_wakeup();
2298 		mtx_unlock(&vm_page_queue_free_mtx);
2299 	}
2300 }
2301 
2302 /*
2303  *	vm_page_wire:
2304  *
2305  *	Mark this page as wired down by yet
2306  *	another map, removing it from paging queues
2307  *	as necessary.
2308  *
2309  *	If the page is fictitious, then its wire count must remain one.
2310  *
2311  *	The page must be locked.
2312  */
2313 void
2314 vm_page_wire(vm_page_t m)
2315 {
2316 
2317 	/*
2318 	 * Only bump the wire statistics if the page is not already wired,
2319 	 * and only unqueue the page if it is on some queue (if it is unmanaged
2320 	 * it is already off the queues).
2321 	 */
2322 	vm_page_lock_assert(m, MA_OWNED);
2323 	if ((m->flags & PG_FICTITIOUS) != 0) {
2324 		KASSERT(m->wire_count == 1,
2325 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
2326 		    m));
2327 		return;
2328 	}
2329 	if (m->wire_count == 0) {
2330 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
2331 		    m->queue == PQ_NONE,
2332 		    ("vm_page_wire: unmanaged page %p is queued", m));
2333 		vm_page_remque(m);
2334 		atomic_add_int(&cnt.v_wire_count, 1);
2335 	}
2336 	m->wire_count++;
2337 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
2338 }
2339 
2340 /*
2341  * vm_page_unwire:
2342  *
2343  * Release one wiring of the specified page, potentially enabling it to be
2344  * paged again.  If paging is enabled, then the value of the parameter
2345  * "activate" determines to which queue the page is added.  If "activate" is
2346  * non-zero, then the page is added to the active queue.  Otherwise, it is
2347  * added to the inactive queue.
2348  *
2349  * However, unless the page belongs to an object, it is not enqueued because
2350  * it cannot be paged out.
2351  *
2352  * If a page is fictitious, then its wire count must always be one.
2353  *
2354  * A managed page must be locked.
2355  */
2356 void
2357 vm_page_unwire(vm_page_t m, int activate)
2358 {
2359 
2360 	if ((m->oflags & VPO_UNMANAGED) == 0)
2361 		vm_page_lock_assert(m, MA_OWNED);
2362 	if ((m->flags & PG_FICTITIOUS) != 0) {
2363 		KASSERT(m->wire_count == 1,
2364 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
2365 		return;
2366 	}
2367 	if (m->wire_count > 0) {
2368 		m->wire_count--;
2369 		if (m->wire_count == 0) {
2370 			atomic_subtract_int(&cnt.v_wire_count, 1);
2371 			if ((m->oflags & VPO_UNMANAGED) != 0 ||
2372 			    m->object == NULL)
2373 				return;
2374 			if (!activate)
2375 				m->flags &= ~PG_WINATCFLS;
2376 			vm_page_enqueue(activate ? PQ_ACTIVE : PQ_INACTIVE, m);
2377 		}
2378 	} else
2379 		panic("vm_page_unwire: page %p's wire count is zero", m);
2380 }
2381 
2382 /*
2383  * Move the specified page to the inactive queue.
2384  *
2385  * Many pages placed on the inactive queue should actually go
2386  * into the cache, but it is difficult to figure out which.  What
2387  * we do instead, if the inactive target is well met, is to put
2388  * clean pages at the head of the inactive queue instead of the tail.
2389  * This will cause them to be moved to the cache more quickly and
2390  * if not actively re-referenced, reclaimed more quickly.  If we just
2391  * stick these pages at the end of the inactive queue, heavy filesystem
2392  * meta-data accesses can cause an unnecessary paging load on memory bound
2393  * processes.  This optimization causes one-time-use metadata to be
2394  * reused more quickly.
2395  *
2396  * Normally athead is 0 resulting in LRU operation.  athead is set
2397  * to 1 if we want this page to be 'as if it were placed in the cache',
2398  * except without unmapping it from the process address space.
2399  *
2400  * The page must be locked.
2401  */
2402 static inline void
2403 _vm_page_deactivate(vm_page_t m, int athead)
2404 {
2405 	struct vm_pagequeue *pq;
2406 	int queue;
2407 
2408 	vm_page_lock_assert(m, MA_OWNED);
2409 
2410 	/*
2411 	 * Ignore if already inactive.
2412 	 */
2413 	if ((queue = m->queue) == PQ_INACTIVE)
2414 		return;
2415 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2416 		if (queue != PQ_NONE)
2417 			vm_page_dequeue(m);
2418 		m->flags &= ~PG_WINATCFLS;
2419 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
2420 		vm_pagequeue_lock(pq);
2421 		m->queue = PQ_INACTIVE;
2422 		if (athead)
2423 			TAILQ_INSERT_HEAD(&pq->pq_pl, m, plinks.q);
2424 		else
2425 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2426 		vm_pagequeue_cnt_inc(pq);
2427 		vm_pagequeue_unlock(pq);
2428 	}
2429 }
2430 
2431 /*
2432  * Move the specified page to the inactive queue.
2433  *
2434  * The page must be locked.
2435  */
2436 void
2437 vm_page_deactivate(vm_page_t m)
2438 {
2439 
2440 	_vm_page_deactivate(m, 0);
2441 }
2442 
2443 /*
2444  * vm_page_try_to_cache:
2445  *
2446  * Returns 0 on failure, 1 on success
2447  */
2448 int
2449 vm_page_try_to_cache(vm_page_t m)
2450 {
2451 
2452 	vm_page_lock_assert(m, MA_OWNED);
2453 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2454 	if (m->dirty || m->hold_count || m->wire_count ||
2455 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
2456 		return (0);
2457 	pmap_remove_all(m);
2458 	if (m->dirty)
2459 		return (0);
2460 	vm_page_cache(m);
2461 	return (1);
2462 }
2463 
2464 /*
2465  * vm_page_try_to_free()
2466  *
2467  *	Attempt to free the page.  If we cannot free it, we do nothing.
2468  *	1 is returned on success, 0 on failure.
2469  */
2470 int
2471 vm_page_try_to_free(vm_page_t m)
2472 {
2473 
2474 	vm_page_lock_assert(m, MA_OWNED);
2475 	if (m->object != NULL)
2476 		VM_OBJECT_ASSERT_WLOCKED(m->object);
2477 	if (m->dirty || m->hold_count || m->wire_count ||
2478 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
2479 		return (0);
2480 	pmap_remove_all(m);
2481 	if (m->dirty)
2482 		return (0);
2483 	vm_page_free(m);
2484 	return (1);
2485 }
2486 
2487 /*
2488  * vm_page_cache
2489  *
2490  * Put the specified page onto the page cache queue (if appropriate).
2491  *
2492  * The object and page must be locked.
2493  */
2494 void
2495 vm_page_cache(vm_page_t m)
2496 {
2497 	vm_object_t object;
2498 	boolean_t cache_was_empty;
2499 
2500 	vm_page_lock_assert(m, MA_OWNED);
2501 	object = m->object;
2502 	VM_OBJECT_ASSERT_WLOCKED(object);
2503 	if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) ||
2504 	    m->hold_count || m->wire_count)
2505 		panic("vm_page_cache: attempting to cache busy page");
2506 	KASSERT(!pmap_page_is_mapped(m),
2507 	    ("vm_page_cache: page %p is mapped", m));
2508 	KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m));
2509 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
2510 	    (object->type == OBJT_SWAP &&
2511 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
2512 		/*
2513 		 * Hypothesis: A cache-elgible page belonging to a
2514 		 * default object or swap object but without a backing
2515 		 * store must be zero filled.
2516 		 */
2517 		vm_page_free(m);
2518 		return;
2519 	}
2520 	KASSERT((m->flags & PG_CACHED) == 0,
2521 	    ("vm_page_cache: page %p is already cached", m));
2522 
2523 	/*
2524 	 * Remove the page from the paging queues.
2525 	 */
2526 	vm_page_remque(m);
2527 
2528 	/*
2529 	 * Remove the page from the object's collection of resident
2530 	 * pages.
2531 	 */
2532 	vm_radix_remove(&object->rtree, m->pindex);
2533 	TAILQ_REMOVE(&object->memq, m, listq);
2534 	object->resident_page_count--;
2535 
2536 	/*
2537 	 * Restore the default memory attribute to the page.
2538 	 */
2539 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2540 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2541 
2542 	/*
2543 	 * Insert the page into the object's collection of cached pages
2544 	 * and the physical memory allocator's cache/free page queues.
2545 	 */
2546 	m->flags &= ~PG_ZERO;
2547 	mtx_lock(&vm_page_queue_free_mtx);
2548 	cache_was_empty = vm_radix_is_empty(&object->cache);
2549 	if (vm_radix_insert(&object->cache, m)) {
2550 		mtx_unlock(&vm_page_queue_free_mtx);
2551 		if (object->resident_page_count == 0)
2552 			vdrop(object->handle);
2553 		m->object = NULL;
2554 		vm_page_free(m);
2555 		return;
2556 	}
2557 
2558 	/*
2559 	 * The above call to vm_radix_insert() could reclaim the one pre-
2560 	 * existing cached page from this object, resulting in a call to
2561 	 * vdrop().
2562 	 */
2563 	if (!cache_was_empty)
2564 		cache_was_empty = vm_radix_is_singleton(&object->cache);
2565 
2566 	m->flags |= PG_CACHED;
2567 	cnt.v_cache_count++;
2568 	PCPU_INC(cnt.v_tcached);
2569 #if VM_NRESERVLEVEL > 0
2570 	if (!vm_reserv_free_page(m)) {
2571 #else
2572 	if (TRUE) {
2573 #endif
2574 		vm_phys_set_pool(VM_FREEPOOL_CACHE, m, 0);
2575 		vm_phys_free_pages(m, 0);
2576 	}
2577 	vm_page_free_wakeup();
2578 	mtx_unlock(&vm_page_queue_free_mtx);
2579 
2580 	/*
2581 	 * Increment the vnode's hold count if this is the object's only
2582 	 * cached page.  Decrement the vnode's hold count if this was
2583 	 * the object's only resident page.
2584 	 */
2585 	if (object->type == OBJT_VNODE) {
2586 		if (cache_was_empty && object->resident_page_count != 0)
2587 			vhold(object->handle);
2588 		else if (!cache_was_empty && object->resident_page_count == 0)
2589 			vdrop(object->handle);
2590 	}
2591 }
2592 
2593 /*
2594  * vm_page_advise
2595  *
2596  *	Cache, deactivate, or do nothing as appropriate.  This routine
2597  *	is used by madvise().
2598  *
2599  *	Generally speaking we want to move the page into the cache so
2600  *	it gets reused quickly.  However, this can result in a silly syndrome
2601  *	due to the page recycling too quickly.  Small objects will not be
2602  *	fully cached.  On the other hand, if we move the page to the inactive
2603  *	queue we wind up with a problem whereby very large objects
2604  *	unnecessarily blow away our inactive and cache queues.
2605  *
2606  *	The solution is to move the pages based on a fixed weighting.  We
2607  *	either leave them alone, deactivate them, or move them to the cache,
2608  *	where moving them to the cache has the highest weighting.
2609  *	By forcing some pages into other queues we eventually force the
2610  *	system to balance the queues, potentially recovering other unrelated
2611  *	space from active.  The idea is to not force this to happen too
2612  *	often.
2613  *
2614  *	The object and page must be locked.
2615  */
2616 void
2617 vm_page_advise(vm_page_t m, int advice)
2618 {
2619 	int dnw, head;
2620 
2621 	vm_page_assert_locked(m);
2622 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2623 	if (advice == MADV_FREE) {
2624 		/*
2625 		 * Mark the page clean.  This will allow the page to be freed
2626 		 * up by the system.  However, such pages are often reused
2627 		 * quickly by malloc() so we do not do anything that would
2628 		 * cause a page fault if we can help it.
2629 		 *
2630 		 * Specifically, we do not try to actually free the page now
2631 		 * nor do we try to put it in the cache (which would cause a
2632 		 * page fault on reuse).
2633 		 *
2634 		 * But we do make the page is freeable as we can without
2635 		 * actually taking the step of unmapping it.
2636 		 */
2637 		m->dirty = 0;
2638 		m->act_count = 0;
2639 	} else if (advice != MADV_DONTNEED)
2640 		return;
2641 	dnw = PCPU_GET(dnweight);
2642 	PCPU_INC(dnweight);
2643 
2644 	/*
2645 	 * Occasionally leave the page alone.
2646 	 */
2647 	if ((dnw & 0x01F0) == 0 || m->queue == PQ_INACTIVE) {
2648 		if (m->act_count >= ACT_INIT)
2649 			--m->act_count;
2650 		return;
2651 	}
2652 
2653 	/*
2654 	 * Clear any references to the page.  Otherwise, the page daemon will
2655 	 * immediately reactivate the page.
2656 	 */
2657 	vm_page_aflag_clear(m, PGA_REFERENCED);
2658 
2659 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
2660 		vm_page_dirty(m);
2661 
2662 	if (m->dirty || (dnw & 0x0070) == 0) {
2663 		/*
2664 		 * Deactivate the page 3 times out of 32.
2665 		 */
2666 		head = 0;
2667 	} else {
2668 		/*
2669 		 * Cache the page 28 times out of every 32.  Note that
2670 		 * the page is deactivated instead of cached, but placed
2671 		 * at the head of the queue instead of the tail.
2672 		 */
2673 		head = 1;
2674 	}
2675 	_vm_page_deactivate(m, head);
2676 }
2677 
2678 /*
2679  * Grab a page, waiting until we are waken up due to the page
2680  * changing state.  We keep on waiting, if the page continues
2681  * to be in the object.  If the page doesn't exist, first allocate it
2682  * and then conditionally zero it.
2683  *
2684  * This routine may sleep.
2685  *
2686  * The object must be locked on entry.  The lock will, however, be released
2687  * and reacquired if the routine sleeps.
2688  */
2689 vm_page_t
2690 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
2691 {
2692 	vm_page_t m;
2693 	int sleep;
2694 
2695 	VM_OBJECT_ASSERT_WLOCKED(object);
2696 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
2697 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
2698 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
2699 retrylookup:
2700 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
2701 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
2702 		    vm_page_xbusied(m) : vm_page_busied(m);
2703 		if (sleep) {
2704 			/*
2705 			 * Reference the page before unlocking and
2706 			 * sleeping so that the page daemon is less
2707 			 * likely to reclaim it.
2708 			 */
2709 			vm_page_aflag_set(m, PGA_REFERENCED);
2710 			vm_page_lock(m);
2711 			VM_OBJECT_WUNLOCK(object);
2712 			vm_page_busy_sleep(m, "pgrbwt");
2713 			VM_OBJECT_WLOCK(object);
2714 			goto retrylookup;
2715 		} else {
2716 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
2717 				vm_page_lock(m);
2718 				vm_page_wire(m);
2719 				vm_page_unlock(m);
2720 			}
2721 			if ((allocflags &
2722 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
2723 				vm_page_xbusy(m);
2724 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
2725 				vm_page_sbusy(m);
2726 			return (m);
2727 		}
2728 	}
2729 	m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_IGN_SBUSY);
2730 	if (m == NULL) {
2731 		VM_OBJECT_WUNLOCK(object);
2732 		VM_WAIT;
2733 		VM_OBJECT_WLOCK(object);
2734 		goto retrylookup;
2735 	} else if (m->valid != 0)
2736 		return (m);
2737 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
2738 		pmap_zero_page(m);
2739 	return (m);
2740 }
2741 
2742 /*
2743  * Mapping function for valid or dirty bits in a page.
2744  *
2745  * Inputs are required to range within a page.
2746  */
2747 vm_page_bits_t
2748 vm_page_bits(int base, int size)
2749 {
2750 	int first_bit;
2751 	int last_bit;
2752 
2753 	KASSERT(
2754 	    base + size <= PAGE_SIZE,
2755 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
2756 	);
2757 
2758 	if (size == 0)		/* handle degenerate case */
2759 		return (0);
2760 
2761 	first_bit = base >> DEV_BSHIFT;
2762 	last_bit = (base + size - 1) >> DEV_BSHIFT;
2763 
2764 	return (((vm_page_bits_t)2 << last_bit) -
2765 	    ((vm_page_bits_t)1 << first_bit));
2766 }
2767 
2768 /*
2769  *	vm_page_set_valid_range:
2770  *
2771  *	Sets portions of a page valid.  The arguments are expected
2772  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
2773  *	of any partial chunks touched by the range.  The invalid portion of
2774  *	such chunks will be zeroed.
2775  *
2776  *	(base + size) must be less then or equal to PAGE_SIZE.
2777  */
2778 void
2779 vm_page_set_valid_range(vm_page_t m, int base, int size)
2780 {
2781 	int endoff, frag;
2782 
2783 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2784 	if (size == 0)	/* handle degenerate case */
2785 		return;
2786 
2787 	/*
2788 	 * If the base is not DEV_BSIZE aligned and the valid
2789 	 * bit is clear, we have to zero out a portion of the
2790 	 * first block.
2791 	 */
2792 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
2793 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
2794 		pmap_zero_page_area(m, frag, base - frag);
2795 
2796 	/*
2797 	 * If the ending offset is not DEV_BSIZE aligned and the
2798 	 * valid bit is clear, we have to zero out a portion of
2799 	 * the last block.
2800 	 */
2801 	endoff = base + size;
2802 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
2803 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
2804 		pmap_zero_page_area(m, endoff,
2805 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
2806 
2807 	/*
2808 	 * Assert that no previously invalid block that is now being validated
2809 	 * is already dirty.
2810 	 */
2811 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
2812 	    ("vm_page_set_valid_range: page %p is dirty", m));
2813 
2814 	/*
2815 	 * Set valid bits inclusive of any overlap.
2816 	 */
2817 	m->valid |= vm_page_bits(base, size);
2818 }
2819 
2820 /*
2821  * Clear the given bits from the specified page's dirty field.
2822  */
2823 static __inline void
2824 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
2825 {
2826 	uintptr_t addr;
2827 #if PAGE_SIZE < 16384
2828 	int shift;
2829 #endif
2830 
2831 	/*
2832 	 * If the object is locked and the page is neither exclusive busy nor
2833 	 * write mapped, then the page's dirty field cannot possibly be
2834 	 * set by a concurrent pmap operation.
2835 	 */
2836 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2837 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
2838 		m->dirty &= ~pagebits;
2839 	else {
2840 		/*
2841 		 * The pmap layer can call vm_page_dirty() without
2842 		 * holding a distinguished lock.  The combination of
2843 		 * the object's lock and an atomic operation suffice
2844 		 * to guarantee consistency of the page dirty field.
2845 		 *
2846 		 * For PAGE_SIZE == 32768 case, compiler already
2847 		 * properly aligns the dirty field, so no forcible
2848 		 * alignment is needed. Only require existence of
2849 		 * atomic_clear_64 when page size is 32768.
2850 		 */
2851 		addr = (uintptr_t)&m->dirty;
2852 #if PAGE_SIZE == 32768
2853 		atomic_clear_64((uint64_t *)addr, pagebits);
2854 #elif PAGE_SIZE == 16384
2855 		atomic_clear_32((uint32_t *)addr, pagebits);
2856 #else		/* PAGE_SIZE <= 8192 */
2857 		/*
2858 		 * Use a trick to perform a 32-bit atomic on the
2859 		 * containing aligned word, to not depend on the existence
2860 		 * of atomic_clear_{8, 16}.
2861 		 */
2862 		shift = addr & (sizeof(uint32_t) - 1);
2863 #if BYTE_ORDER == BIG_ENDIAN
2864 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
2865 #else
2866 		shift *= NBBY;
2867 #endif
2868 		addr &= ~(sizeof(uint32_t) - 1);
2869 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
2870 #endif		/* PAGE_SIZE */
2871 	}
2872 }
2873 
2874 /*
2875  *	vm_page_set_validclean:
2876  *
2877  *	Sets portions of a page valid and clean.  The arguments are expected
2878  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
2879  *	of any partial chunks touched by the range.  The invalid portion of
2880  *	such chunks will be zero'd.
2881  *
2882  *	(base + size) must be less then or equal to PAGE_SIZE.
2883  */
2884 void
2885 vm_page_set_validclean(vm_page_t m, int base, int size)
2886 {
2887 	vm_page_bits_t oldvalid, pagebits;
2888 	int endoff, frag;
2889 
2890 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2891 	if (size == 0)	/* handle degenerate case */
2892 		return;
2893 
2894 	/*
2895 	 * If the base is not DEV_BSIZE aligned and the valid
2896 	 * bit is clear, we have to zero out a portion of the
2897 	 * first block.
2898 	 */
2899 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
2900 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
2901 		pmap_zero_page_area(m, frag, base - frag);
2902 
2903 	/*
2904 	 * If the ending offset is not DEV_BSIZE aligned and the
2905 	 * valid bit is clear, we have to zero out a portion of
2906 	 * the last block.
2907 	 */
2908 	endoff = base + size;
2909 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
2910 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
2911 		pmap_zero_page_area(m, endoff,
2912 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
2913 
2914 	/*
2915 	 * Set valid, clear dirty bits.  If validating the entire
2916 	 * page we can safely clear the pmap modify bit.  We also
2917 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
2918 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
2919 	 * be set again.
2920 	 *
2921 	 * We set valid bits inclusive of any overlap, but we can only
2922 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
2923 	 * the range.
2924 	 */
2925 	oldvalid = m->valid;
2926 	pagebits = vm_page_bits(base, size);
2927 	m->valid |= pagebits;
2928 #if 0	/* NOT YET */
2929 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
2930 		frag = DEV_BSIZE - frag;
2931 		base += frag;
2932 		size -= frag;
2933 		if (size < 0)
2934 			size = 0;
2935 	}
2936 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
2937 #endif
2938 	if (base == 0 && size == PAGE_SIZE) {
2939 		/*
2940 		 * The page can only be modified within the pmap if it is
2941 		 * mapped, and it can only be mapped if it was previously
2942 		 * fully valid.
2943 		 */
2944 		if (oldvalid == VM_PAGE_BITS_ALL)
2945 			/*
2946 			 * Perform the pmap_clear_modify() first.  Otherwise,
2947 			 * a concurrent pmap operation, such as
2948 			 * pmap_protect(), could clear a modification in the
2949 			 * pmap and set the dirty field on the page before
2950 			 * pmap_clear_modify() had begun and after the dirty
2951 			 * field was cleared here.
2952 			 */
2953 			pmap_clear_modify(m);
2954 		m->dirty = 0;
2955 		m->oflags &= ~VPO_NOSYNC;
2956 	} else if (oldvalid != VM_PAGE_BITS_ALL)
2957 		m->dirty &= ~pagebits;
2958 	else
2959 		vm_page_clear_dirty_mask(m, pagebits);
2960 }
2961 
2962 void
2963 vm_page_clear_dirty(vm_page_t m, int base, int size)
2964 {
2965 
2966 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
2967 }
2968 
2969 /*
2970  *	vm_page_set_invalid:
2971  *
2972  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
2973  *	valid and dirty bits for the effected areas are cleared.
2974  */
2975 void
2976 vm_page_set_invalid(vm_page_t m, int base, int size)
2977 {
2978 	vm_page_bits_t bits;
2979 
2980 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2981 	bits = vm_page_bits(base, size);
2982 	if (m->valid == VM_PAGE_BITS_ALL && bits != 0)
2983 		pmap_remove_all(m);
2984 	KASSERT(!pmap_page_is_mapped(m),
2985 	    ("vm_page_set_invalid: page %p is mapped", m));
2986 	m->valid &= ~bits;
2987 	m->dirty &= ~bits;
2988 }
2989 
2990 /*
2991  * vm_page_zero_invalid()
2992  *
2993  *	The kernel assumes that the invalid portions of a page contain
2994  *	garbage, but such pages can be mapped into memory by user code.
2995  *	When this occurs, we must zero out the non-valid portions of the
2996  *	page so user code sees what it expects.
2997  *
2998  *	Pages are most often semi-valid when the end of a file is mapped
2999  *	into memory and the file's size is not page aligned.
3000  */
3001 void
3002 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3003 {
3004 	int b;
3005 	int i;
3006 
3007 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3008 	/*
3009 	 * Scan the valid bits looking for invalid sections that
3010 	 * must be zerod.  Invalid sub-DEV_BSIZE'd areas ( where the
3011 	 * valid bit may be set ) have already been zerod by
3012 	 * vm_page_set_validclean().
3013 	 */
3014 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3015 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3016 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3017 			if (i > b) {
3018 				pmap_zero_page_area(m,
3019 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3020 			}
3021 			b = i + 1;
3022 		}
3023 	}
3024 
3025 	/*
3026 	 * setvalid is TRUE when we can safely set the zero'd areas
3027 	 * as being valid.  We can do this if there are no cache consistancy
3028 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3029 	 */
3030 	if (setvalid)
3031 		m->valid = VM_PAGE_BITS_ALL;
3032 }
3033 
3034 /*
3035  *	vm_page_is_valid:
3036  *
3037  *	Is (partial) page valid?  Note that the case where size == 0
3038  *	will return FALSE in the degenerate case where the page is
3039  *	entirely invalid, and TRUE otherwise.
3040  */
3041 int
3042 vm_page_is_valid(vm_page_t m, int base, int size)
3043 {
3044 	vm_page_bits_t bits;
3045 
3046 	VM_OBJECT_ASSERT_LOCKED(m->object);
3047 	bits = vm_page_bits(base, size);
3048 	return (m->valid != 0 && (m->valid & bits) == bits);
3049 }
3050 
3051 /*
3052  * Set the page's dirty bits if the page is modified.
3053  */
3054 void
3055 vm_page_test_dirty(vm_page_t m)
3056 {
3057 
3058 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3059 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3060 		vm_page_dirty(m);
3061 }
3062 
3063 void
3064 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3065 {
3066 
3067 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3068 }
3069 
3070 void
3071 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3072 {
3073 
3074 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3075 }
3076 
3077 int
3078 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3079 {
3080 
3081 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3082 }
3083 
3084 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3085 void
3086 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3087 {
3088 
3089 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3090 }
3091 
3092 void
3093 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3094 {
3095 
3096 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3097 }
3098 #endif
3099 
3100 int so_zerocp_fullpage = 0;
3101 
3102 /*
3103  *	Replace the given page with a copy.  The copied page assumes
3104  *	the portion of the given page's "wire_count" that is not the
3105  *	responsibility of this copy-on-write mechanism.
3106  *
3107  *	The object containing the given page must have a non-zero
3108  *	paging-in-progress count and be locked.
3109  */
3110 void
3111 vm_page_cowfault(vm_page_t m)
3112 {
3113 	vm_page_t mnew;
3114 	vm_object_t object;
3115 	vm_pindex_t pindex;
3116 
3117 	vm_page_lock_assert(m, MA_OWNED);
3118 	object = m->object;
3119 	VM_OBJECT_ASSERT_WLOCKED(object);
3120 	KASSERT(object->paging_in_progress != 0,
3121 	    ("vm_page_cowfault: object %p's paging-in-progress count is zero.",
3122 	    object));
3123 	pindex = m->pindex;
3124 
3125  retry_alloc:
3126 	mnew = vm_page_alloc(NULL, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ);
3127 	if (mnew == NULL) {
3128 		vm_page_unlock(m);
3129 		VM_OBJECT_WUNLOCK(object);
3130 		VM_WAIT;
3131 		VM_OBJECT_WLOCK(object);
3132 		if (m == vm_page_lookup(object, pindex)) {
3133 			vm_page_lock(m);
3134 			goto retry_alloc;
3135 		} else {
3136 			/*
3137 			 * Page disappeared during the wait.
3138 			 */
3139 			return;
3140 		}
3141 	}
3142 
3143 	if (m->cow == 0) {
3144 		/*
3145 		 * check to see if we raced with an xmit complete when
3146 		 * waiting to allocate a page.  If so, put things back
3147 		 * the way they were
3148 		 */
3149 		vm_page_unlock(m);
3150 		vm_page_lock(mnew);
3151 		vm_page_free(mnew);
3152 		vm_page_unlock(mnew);
3153 	} else { /* clear COW & copy page */
3154 		pmap_remove_all(m);
3155 		mnew->object = object;
3156 		if (object->memattr != VM_MEMATTR_DEFAULT &&
3157 		    (object->flags & OBJ_FICTITIOUS) == 0)
3158 			pmap_page_set_memattr(mnew, object->memattr);
3159 		if (vm_page_replace(mnew, object, pindex) != m)
3160 			panic("vm_page_cowfault: invalid page replacement");
3161 		if (!so_zerocp_fullpage)
3162 			pmap_copy_page(m, mnew);
3163 		mnew->valid = VM_PAGE_BITS_ALL;
3164 		vm_page_dirty(mnew);
3165 		mnew->wire_count = m->wire_count - m->cow;
3166 		m->wire_count = m->cow;
3167 		vm_page_unlock(m);
3168 	}
3169 }
3170 
3171 void
3172 vm_page_cowclear(vm_page_t m)
3173 {
3174 
3175 	vm_page_lock_assert(m, MA_OWNED);
3176 	if (m->cow) {
3177 		m->cow--;
3178 		/*
3179 		 * let vm_fault add back write permission  lazily
3180 		 */
3181 	}
3182 	/*
3183 	 *  sf_buf_free() will free the page, so we needn't do it here
3184 	 */
3185 }
3186 
3187 int
3188 vm_page_cowsetup(vm_page_t m)
3189 {
3190 
3191 	vm_page_lock_assert(m, MA_OWNED);
3192 	if ((m->flags & PG_FICTITIOUS) != 0 ||
3193 	    (m->oflags & VPO_UNMANAGED) != 0 ||
3194 	    m->cow == USHRT_MAX - 1 || !VM_OBJECT_TRYWLOCK(m->object))
3195 		return (EBUSY);
3196 	m->cow++;
3197 	pmap_remove_write(m);
3198 	VM_OBJECT_WUNLOCK(m->object);
3199 	return (0);
3200 }
3201 
3202 #ifdef INVARIANTS
3203 void
3204 vm_page_object_lock_assert(vm_page_t m)
3205 {
3206 
3207 	/*
3208 	 * Certain of the page's fields may only be modified by the
3209 	 * holder of the containing object's lock or the exclusive busy.
3210 	 * holder.  Unfortunately, the holder of the write busy is
3211 	 * not recorded, and thus cannot be checked here.
3212 	 */
3213 	if (m->object != NULL && !vm_page_xbusied(m))
3214 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3215 }
3216 #endif
3217 
3218 #include "opt_ddb.h"
3219 #ifdef DDB
3220 #include <sys/kernel.h>
3221 
3222 #include <ddb/ddb.h>
3223 
3224 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3225 {
3226 	db_printf("cnt.v_free_count: %d\n", cnt.v_free_count);
3227 	db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count);
3228 	db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count);
3229 	db_printf("cnt.v_active_count: %d\n", cnt.v_active_count);
3230 	db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count);
3231 	db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved);
3232 	db_printf("cnt.v_free_min: %d\n", cnt.v_free_min);
3233 	db_printf("cnt.v_free_target: %d\n", cnt.v_free_target);
3234 	db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min);
3235 	db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target);
3236 }
3237 
3238 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3239 {
3240 	int dom;
3241 
3242 	db_printf("pq_free %d pq_cache %d\n",
3243 	    cnt.v_free_count, cnt.v_cache_count);
3244 	for (dom = 0; dom < vm_ndomains; dom++) {
3245 		db_printf(
3246 	"dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n",
3247 		    dom,
3248 		    vm_dom[dom].vmd_page_count,
3249 		    vm_dom[dom].vmd_free_count,
3250 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3251 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3252 		    vm_dom[dom].vmd_pass);
3253 	}
3254 }
3255 
3256 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3257 {
3258 	vm_page_t m;
3259 	boolean_t phys;
3260 
3261 	if (!have_addr) {
3262 		db_printf("show pginfo addr\n");
3263 		return;
3264 	}
3265 
3266 	phys = strchr(modif, 'p') != NULL;
3267 	if (phys)
3268 		m = PHYS_TO_VM_PAGE(addr);
3269 	else
3270 		m = (vm_page_t)addr;
3271 	db_printf(
3272     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3273     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3274 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3275 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3276 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3277 }
3278 #endif /* DDB */
3279