xref: /freebsd/sys/vm/vm_page.c (revision 273c26a3c3bea87a241d6879abd4f991db180bf0)
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/linker.h>
95 #include <sys/malloc.h>
96 #include <sys/mman.h>
97 #include <sys/msgbuf.h>
98 #include <sys/mutex.h>
99 #include <sys/proc.h>
100 #include <sys/rwlock.h>
101 #include <sys/sbuf.h>
102 #include <sys/smp.h>
103 #include <sys/sysctl.h>
104 #include <sys/vmmeter.h>
105 #include <sys/vnode.h>
106 
107 #include <vm/vm.h>
108 #include <vm/pmap.h>
109 #include <vm/vm_param.h>
110 #include <vm/vm_kern.h>
111 #include <vm/vm_object.h>
112 #include <vm/vm_page.h>
113 #include <vm/vm_pageout.h>
114 #include <vm/vm_pager.h>
115 #include <vm/vm_phys.h>
116 #include <vm/vm_radix.h>
117 #include <vm/vm_reserv.h>
118 #include <vm/vm_extern.h>
119 #include <vm/uma.h>
120 #include <vm/uma_int.h>
121 
122 #include <machine/md_var.h>
123 
124 /*
125  *	Associated with page of user-allocatable memory is a
126  *	page structure.
127  */
128 
129 struct vm_domain vm_dom[MAXMEMDOM];
130 struct mtx_padalign vm_page_queue_free_mtx;
131 
132 struct mtx_padalign pa_lock[PA_LOCK_COUNT];
133 
134 vm_page_t vm_page_array;
135 long vm_page_array_size;
136 long first_page;
137 
138 static int boot_pages = UMA_BOOT_PAGES;
139 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
140     &boot_pages, 0,
141     "number of pages allocated for bootstrapping the VM system");
142 
143 static int pa_tryrelock_restart;
144 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
145     &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
146 
147 static TAILQ_HEAD(, vm_page) blacklist_head;
148 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
149 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
150     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
151 
152 /* Is the page daemon waiting for free pages? */
153 static int vm_pageout_pages_needed;
154 
155 static uma_zone_t fakepg_zone;
156 
157 static struct vnode *vm_page_alloc_init(vm_page_t m);
158 static void vm_page_cache_turn_free(vm_page_t m);
159 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
160 static void vm_page_enqueue(uint8_t queue, vm_page_t m);
161 static void vm_page_free_wakeup(void);
162 static void vm_page_init_fakepg(void *dummy);
163 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
164     vm_pindex_t pindex, vm_page_t mpred);
165 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
166     vm_page_t mpred);
167 static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
168     vm_paddr_t high);
169 
170 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL);
171 
172 static void
173 vm_page_init_fakepg(void *dummy)
174 {
175 
176 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
177 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
178 }
179 
180 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
181 #if PAGE_SIZE == 32768
182 #ifdef CTASSERT
183 CTASSERT(sizeof(u_long) >= 8);
184 #endif
185 #endif
186 
187 /*
188  * Try to acquire a physical address lock while a pmap is locked.  If we
189  * fail to trylock we unlock and lock the pmap directly and cache the
190  * locked pa in *locked.  The caller should then restart their loop in case
191  * the virtual to physical mapping has changed.
192  */
193 int
194 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
195 {
196 	vm_paddr_t lockpa;
197 
198 	lockpa = *locked;
199 	*locked = pa;
200 	if (lockpa) {
201 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
202 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
203 			return (0);
204 		PA_UNLOCK(lockpa);
205 	}
206 	if (PA_TRYLOCK(pa))
207 		return (0);
208 	PMAP_UNLOCK(pmap);
209 	atomic_add_int(&pa_tryrelock_restart, 1);
210 	PA_LOCK(pa);
211 	PMAP_LOCK(pmap);
212 	return (EAGAIN);
213 }
214 
215 /*
216  *	vm_set_page_size:
217  *
218  *	Sets the page size, perhaps based upon the memory
219  *	size.  Must be called before any use of page-size
220  *	dependent functions.
221  */
222 void
223 vm_set_page_size(void)
224 {
225 	if (vm_cnt.v_page_size == 0)
226 		vm_cnt.v_page_size = PAGE_SIZE;
227 	if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
228 		panic("vm_set_page_size: page size not a power of two");
229 }
230 
231 /*
232  *	vm_page_blacklist_next:
233  *
234  *	Find the next entry in the provided string of blacklist
235  *	addresses.  Entries are separated by space, comma, or newline.
236  *	If an invalid integer is encountered then the rest of the
237  *	string is skipped.  Updates the list pointer to the next
238  *	character, or NULL if the string is exhausted or invalid.
239  */
240 static vm_paddr_t
241 vm_page_blacklist_next(char **list, char *end)
242 {
243 	vm_paddr_t bad;
244 	char *cp, *pos;
245 
246 	if (list == NULL || *list == NULL)
247 		return (0);
248 	if (**list =='\0') {
249 		*list = NULL;
250 		return (0);
251 	}
252 
253 	/*
254 	 * If there's no end pointer then the buffer is coming from
255 	 * the kenv and we know it's null-terminated.
256 	 */
257 	if (end == NULL)
258 		end = *list + strlen(*list);
259 
260 	/* Ensure that strtoq() won't walk off the end */
261 	if (*end != '\0') {
262 		if (*end == '\n' || *end == ' ' || *end  == ',')
263 			*end = '\0';
264 		else {
265 			printf("Blacklist not terminated, skipping\n");
266 			*list = NULL;
267 			return (0);
268 		}
269 	}
270 
271 	for (pos = *list; *pos != '\0'; pos = cp) {
272 		bad = strtoq(pos, &cp, 0);
273 		if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
274 			if (bad == 0) {
275 				if (++cp < end)
276 					continue;
277 				else
278 					break;
279 			}
280 		} else
281 			break;
282 		if (*cp == '\0' || ++cp >= end)
283 			*list = NULL;
284 		else
285 			*list = cp;
286 		return (trunc_page(bad));
287 	}
288 	printf("Garbage in RAM blacklist, skipping\n");
289 	*list = NULL;
290 	return (0);
291 }
292 
293 /*
294  *	vm_page_blacklist_check:
295  *
296  *	Iterate through the provided string of blacklist addresses, pulling
297  *	each entry out of the physical allocator free list and putting it
298  *	onto a list for reporting via the vm.page_blacklist sysctl.
299  */
300 static void
301 vm_page_blacklist_check(char *list, char *end)
302 {
303 	vm_paddr_t pa;
304 	vm_page_t m;
305 	char *next;
306 	int ret;
307 
308 	next = list;
309 	while (next != NULL) {
310 		if ((pa = vm_page_blacklist_next(&next, end)) == 0)
311 			continue;
312 		m = vm_phys_paddr_to_vm_page(pa);
313 		if (m == NULL)
314 			continue;
315 		mtx_lock(&vm_page_queue_free_mtx);
316 		ret = vm_phys_unfree_page(m);
317 		mtx_unlock(&vm_page_queue_free_mtx);
318 		if (ret == TRUE) {
319 			TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
320 			if (bootverbose)
321 				printf("Skipping page with pa 0x%jx\n",
322 				    (uintmax_t)pa);
323 		}
324 	}
325 }
326 
327 /*
328  *	vm_page_blacklist_load:
329  *
330  *	Search for a special module named "ram_blacklist".  It'll be a
331  *	plain text file provided by the user via the loader directive
332  *	of the same name.
333  */
334 static void
335 vm_page_blacklist_load(char **list, char **end)
336 {
337 	void *mod;
338 	u_char *ptr;
339 	u_int len;
340 
341 	mod = NULL;
342 	ptr = NULL;
343 
344 	mod = preload_search_by_type("ram_blacklist");
345 	if (mod != NULL) {
346 		ptr = preload_fetch_addr(mod);
347 		len = preload_fetch_size(mod);
348         }
349 	*list = ptr;
350 	if (ptr != NULL)
351 		*end = ptr + len;
352 	else
353 		*end = NULL;
354 	return;
355 }
356 
357 static int
358 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
359 {
360 	vm_page_t m;
361 	struct sbuf sbuf;
362 	int error, first;
363 
364 	first = 1;
365 	error = sysctl_wire_old_buffer(req, 0);
366 	if (error != 0)
367 		return (error);
368 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
369 	TAILQ_FOREACH(m, &blacklist_head, listq) {
370 		sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
371 		    (uintmax_t)m->phys_addr);
372 		first = 0;
373 	}
374 	error = sbuf_finish(&sbuf);
375 	sbuf_delete(&sbuf);
376 	return (error);
377 }
378 
379 static void
380 vm_page_domain_init(struct vm_domain *vmd)
381 {
382 	struct vm_pagequeue *pq;
383 	int i;
384 
385 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
386 	    "vm inactive pagequeue";
387 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
388 	    &vm_cnt.v_inactive_count;
389 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
390 	    "vm active pagequeue";
391 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
392 	    &vm_cnt.v_active_count;
393 	vmd->vmd_page_count = 0;
394 	vmd->vmd_free_count = 0;
395 	vmd->vmd_segs = 0;
396 	vmd->vmd_oom = FALSE;
397 	vmd->vmd_pass = 0;
398 	for (i = 0; i < PQ_COUNT; i++) {
399 		pq = &vmd->vmd_pagequeues[i];
400 		TAILQ_INIT(&pq->pq_pl);
401 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
402 		    MTX_DEF | MTX_DUPOK);
403 	}
404 }
405 
406 /*
407  *	vm_page_startup:
408  *
409  *	Initializes the resident memory module.
410  *
411  *	Allocates memory for the page cells, and
412  *	for the object/offset-to-page hash table headers.
413  *	Each page cell is initialized and placed on the free list.
414  */
415 vm_offset_t
416 vm_page_startup(vm_offset_t vaddr)
417 {
418 	vm_offset_t mapped;
419 	vm_paddr_t page_range;
420 	vm_paddr_t new_end;
421 	int i;
422 	vm_paddr_t pa;
423 	vm_paddr_t last_pa;
424 	char *list, *listend;
425 	vm_paddr_t end;
426 	vm_paddr_t biggestsize;
427 	vm_paddr_t low_water, high_water;
428 	int biggestone;
429 	int pages_per_zone;
430 
431 	biggestsize = 0;
432 	biggestone = 0;
433 	vaddr = round_page(vaddr);
434 
435 	for (i = 0; phys_avail[i + 1]; i += 2) {
436 		phys_avail[i] = round_page(phys_avail[i]);
437 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
438 	}
439 
440 	low_water = phys_avail[0];
441 	high_water = phys_avail[1];
442 
443 	for (i = 0; i < vm_phys_nsegs; i++) {
444 		if (vm_phys_segs[i].start < low_water)
445 			low_water = vm_phys_segs[i].start;
446 		if (vm_phys_segs[i].end > high_water)
447 			high_water = vm_phys_segs[i].end;
448 	}
449 	for (i = 0; phys_avail[i + 1]; i += 2) {
450 		vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
451 
452 		if (size > biggestsize) {
453 			biggestone = i;
454 			biggestsize = size;
455 		}
456 		if (phys_avail[i] < low_water)
457 			low_water = phys_avail[i];
458 		if (phys_avail[i + 1] > high_water)
459 			high_water = phys_avail[i + 1];
460 	}
461 
462 	end = phys_avail[biggestone+1];
463 
464 	/*
465 	 * Initialize the page and queue locks.
466 	 */
467 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
468 	for (i = 0; i < PA_LOCK_COUNT; i++)
469 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
470 	for (i = 0; i < vm_ndomains; i++)
471 		vm_page_domain_init(&vm_dom[i]);
472 
473 	/*
474 	 * Almost all of the pages needed for boot strapping UMA are used
475 	 * for zone structures, so if the number of CPUs results in those
476 	 * structures taking more than one page each, we set aside more pages
477 	 * in proportion to the zone structure size.
478 	 */
479 	pages_per_zone = howmany(sizeof(struct uma_zone) +
480 	    sizeof(struct uma_cache) * (mp_maxid + 1), UMA_SLAB_SIZE);
481 	if (pages_per_zone > 1) {
482 		/* Reserve more pages so that we don't run out. */
483 		boot_pages = UMA_BOOT_PAGES_ZONES * pages_per_zone;
484 	}
485 
486 	/*
487 	 * Allocate memory for use when boot strapping the kernel memory
488 	 * allocator.
489 	 *
490 	 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
491 	 * manually fetch the value.
492 	 */
493 	TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
494 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
495 	new_end = trunc_page(new_end);
496 	mapped = pmap_map(&vaddr, new_end, end,
497 	    VM_PROT_READ | VM_PROT_WRITE);
498 	bzero((void *)mapped, end - new_end);
499 	uma_startup((void *)mapped, boot_pages);
500 
501 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
502     defined(__i386__) || defined(__mips__)
503 	/*
504 	 * Allocate a bitmap to indicate that a random physical page
505 	 * needs to be included in a minidump.
506 	 *
507 	 * The amd64 port needs this to indicate which direct map pages
508 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
509 	 *
510 	 * However, i386 still needs this workspace internally within the
511 	 * minidump code.  In theory, they are not needed on i386, but are
512 	 * included should the sf_buf code decide to use them.
513 	 */
514 	last_pa = 0;
515 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
516 		if (dump_avail[i + 1] > last_pa)
517 			last_pa = dump_avail[i + 1];
518 	page_range = last_pa / PAGE_SIZE;
519 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
520 	new_end -= vm_page_dump_size;
521 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
522 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
523 	bzero((void *)vm_page_dump, vm_page_dump_size);
524 #endif
525 #ifdef __amd64__
526 	/*
527 	 * Request that the physical pages underlying the message buffer be
528 	 * included in a crash dump.  Since the message buffer is accessed
529 	 * through the direct map, they are not automatically included.
530 	 */
531 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
532 	last_pa = pa + round_page(msgbufsize);
533 	while (pa < last_pa) {
534 		dump_add_page(pa);
535 		pa += PAGE_SIZE;
536 	}
537 #endif
538 	/*
539 	 * Compute the number of pages of memory that will be available for
540 	 * use (taking into account the overhead of a page structure per
541 	 * page).
542 	 */
543 	first_page = low_water / PAGE_SIZE;
544 #ifdef VM_PHYSSEG_SPARSE
545 	page_range = 0;
546 	for (i = 0; i < vm_phys_nsegs; i++) {
547 		page_range += atop(vm_phys_segs[i].end -
548 		    vm_phys_segs[i].start);
549 	}
550 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
551 		page_range += atop(phys_avail[i + 1] - phys_avail[i]);
552 #elif defined(VM_PHYSSEG_DENSE)
553 	page_range = high_water / PAGE_SIZE - first_page;
554 #else
555 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
556 #endif
557 	end = new_end;
558 
559 	/*
560 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
561 	 */
562 	vaddr += PAGE_SIZE;
563 
564 	/*
565 	 * Initialize the mem entry structures now, and put them in the free
566 	 * queue.
567 	 */
568 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
569 	mapped = pmap_map(&vaddr, new_end, end,
570 	    VM_PROT_READ | VM_PROT_WRITE);
571 	vm_page_array = (vm_page_t) mapped;
572 #if VM_NRESERVLEVEL > 0
573 	/*
574 	 * Allocate memory for the reservation management system's data
575 	 * structures.
576 	 */
577 	new_end = vm_reserv_startup(&vaddr, new_end, high_water);
578 #endif
579 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
580 	/*
581 	 * pmap_map on arm64, amd64, and mips can come out of the direct-map,
582 	 * not kvm like i386, so the pages must be tracked for a crashdump to
583 	 * include this data.  This includes the vm_page_array and the early
584 	 * UMA bootstrap pages.
585 	 */
586 	for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
587 		dump_add_page(pa);
588 #endif
589 	phys_avail[biggestone + 1] = new_end;
590 
591 	/*
592 	 * Add physical memory segments corresponding to the available
593 	 * physical pages.
594 	 */
595 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
596 		vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
597 
598 	/*
599 	 * Clear all of the page structures
600 	 */
601 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
602 	for (i = 0; i < page_range; i++)
603 		vm_page_array[i].order = VM_NFREEORDER;
604 	vm_page_array_size = page_range;
605 
606 	/*
607 	 * Initialize the physical memory allocator.
608 	 */
609 	vm_phys_init();
610 
611 	/*
612 	 * Add every available physical page that is not blacklisted to
613 	 * the free lists.
614 	 */
615 	vm_cnt.v_page_count = 0;
616 	vm_cnt.v_free_count = 0;
617 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
618 		pa = phys_avail[i];
619 		last_pa = phys_avail[i + 1];
620 		while (pa < last_pa) {
621 			vm_phys_add_page(pa);
622 			pa += PAGE_SIZE;
623 		}
624 	}
625 
626 	TAILQ_INIT(&blacklist_head);
627 	vm_page_blacklist_load(&list, &listend);
628 	vm_page_blacklist_check(list, listend);
629 
630 	list = kern_getenv("vm.blacklist");
631 	vm_page_blacklist_check(list, NULL);
632 
633 	freeenv(list);
634 #if VM_NRESERVLEVEL > 0
635 	/*
636 	 * Initialize the reservation management system.
637 	 */
638 	vm_reserv_init();
639 #endif
640 	return (vaddr);
641 }
642 
643 void
644 vm_page_reference(vm_page_t m)
645 {
646 
647 	vm_page_aflag_set(m, PGA_REFERENCED);
648 }
649 
650 /*
651  *	vm_page_busy_downgrade:
652  *
653  *	Downgrade an exclusive busy page into a single shared busy page.
654  */
655 void
656 vm_page_busy_downgrade(vm_page_t m)
657 {
658 	u_int x;
659 
660 	vm_page_assert_xbusied(m);
661 
662 	for (;;) {
663 		x = m->busy_lock;
664 		x &= VPB_BIT_WAITERS;
665 		if (atomic_cmpset_rel_int(&m->busy_lock,
666 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x))
667 			break;
668 	}
669 }
670 
671 /*
672  *	vm_page_sbusied:
673  *
674  *	Return a positive value if the page is shared busied, 0 otherwise.
675  */
676 int
677 vm_page_sbusied(vm_page_t m)
678 {
679 	u_int x;
680 
681 	x = m->busy_lock;
682 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
683 }
684 
685 /*
686  *	vm_page_sunbusy:
687  *
688  *	Shared unbusy a page.
689  */
690 void
691 vm_page_sunbusy(vm_page_t m)
692 {
693 	u_int x;
694 
695 	vm_page_assert_sbusied(m);
696 
697 	for (;;) {
698 		x = m->busy_lock;
699 		if (VPB_SHARERS(x) > 1) {
700 			if (atomic_cmpset_int(&m->busy_lock, x,
701 			    x - VPB_ONE_SHARER))
702 				break;
703 			continue;
704 		}
705 		if ((x & VPB_BIT_WAITERS) == 0) {
706 			KASSERT(x == VPB_SHARERS_WORD(1),
707 			    ("vm_page_sunbusy: invalid lock state"));
708 			if (atomic_cmpset_int(&m->busy_lock,
709 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
710 				break;
711 			continue;
712 		}
713 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
714 		    ("vm_page_sunbusy: invalid lock state for waiters"));
715 
716 		vm_page_lock(m);
717 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
718 			vm_page_unlock(m);
719 			continue;
720 		}
721 		wakeup(m);
722 		vm_page_unlock(m);
723 		break;
724 	}
725 }
726 
727 /*
728  *	vm_page_busy_sleep:
729  *
730  *	Sleep and release the page lock, using the page pointer as wchan.
731  *	This is used to implement the hard-path of busying mechanism.
732  *
733  *	The given page must be locked.
734  */
735 void
736 vm_page_busy_sleep(vm_page_t m, const char *wmesg)
737 {
738 	u_int x;
739 
740 	vm_page_lock_assert(m, MA_OWNED);
741 
742 	x = m->busy_lock;
743 	if (x == VPB_UNBUSIED) {
744 		vm_page_unlock(m);
745 		return;
746 	}
747 	if ((x & VPB_BIT_WAITERS) == 0 &&
748 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) {
749 		vm_page_unlock(m);
750 		return;
751 	}
752 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
753 }
754 
755 /*
756  *	vm_page_trysbusy:
757  *
758  *	Try to shared busy a page.
759  *	If the operation succeeds 1 is returned otherwise 0.
760  *	The operation never sleeps.
761  */
762 int
763 vm_page_trysbusy(vm_page_t m)
764 {
765 	u_int x;
766 
767 	for (;;) {
768 		x = m->busy_lock;
769 		if ((x & VPB_BIT_SHARED) == 0)
770 			return (0);
771 		if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
772 			return (1);
773 	}
774 }
775 
776 static void
777 vm_page_xunbusy_locked(vm_page_t m)
778 {
779 
780 	vm_page_assert_xbusied(m);
781 	vm_page_assert_locked(m);
782 
783 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
784 	/* There is a waiter, do wakeup() instead of vm_page_flash(). */
785 	wakeup(m);
786 }
787 
788 static void
789 vm_page_xunbusy_maybelocked(vm_page_t m)
790 {
791 	bool lockacq;
792 
793 	vm_page_assert_xbusied(m);
794 
795 	/*
796 	 * Fast path for unbusy.  If it succeeds, we know that there
797 	 * are no waiters, so we do not need a wakeup.
798 	 */
799 	if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER,
800 	    VPB_UNBUSIED))
801 		return;
802 
803 	lockacq = !mtx_owned(vm_page_lockptr(m));
804 	if (lockacq)
805 		vm_page_lock(m);
806 	vm_page_xunbusy_locked(m);
807 	if (lockacq)
808 		vm_page_unlock(m);
809 }
810 
811 /*
812  *	vm_page_xunbusy_hard:
813  *
814  *	Called after the first try the exclusive unbusy of a page failed.
815  *	It is assumed that the waiters bit is on.
816  */
817 void
818 vm_page_xunbusy_hard(vm_page_t m)
819 {
820 
821 	vm_page_assert_xbusied(m);
822 
823 	vm_page_lock(m);
824 	vm_page_xunbusy_locked(m);
825 	vm_page_unlock(m);
826 }
827 
828 /*
829  *	vm_page_flash:
830  *
831  *	Wakeup anyone waiting for the page.
832  *	The ownership bits do not change.
833  *
834  *	The given page must be locked.
835  */
836 void
837 vm_page_flash(vm_page_t m)
838 {
839 	u_int x;
840 
841 	vm_page_lock_assert(m, MA_OWNED);
842 
843 	for (;;) {
844 		x = m->busy_lock;
845 		if ((x & VPB_BIT_WAITERS) == 0)
846 			return;
847 		if (atomic_cmpset_int(&m->busy_lock, x,
848 		    x & (~VPB_BIT_WAITERS)))
849 			break;
850 	}
851 	wakeup(m);
852 }
853 
854 /*
855  * Keep page from being freed by the page daemon
856  * much of the same effect as wiring, except much lower
857  * overhead and should be used only for *very* temporary
858  * holding ("wiring").
859  */
860 void
861 vm_page_hold(vm_page_t mem)
862 {
863 
864 	vm_page_lock_assert(mem, MA_OWNED);
865         mem->hold_count++;
866 }
867 
868 void
869 vm_page_unhold(vm_page_t mem)
870 {
871 
872 	vm_page_lock_assert(mem, MA_OWNED);
873 	KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
874 	--mem->hold_count;
875 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
876 		vm_page_free_toq(mem);
877 }
878 
879 /*
880  *	vm_page_unhold_pages:
881  *
882  *	Unhold each of the pages that is referenced by the given array.
883  */
884 void
885 vm_page_unhold_pages(vm_page_t *ma, int count)
886 {
887 	struct mtx *mtx, *new_mtx;
888 
889 	mtx = NULL;
890 	for (; count != 0; count--) {
891 		/*
892 		 * Avoid releasing and reacquiring the same page lock.
893 		 */
894 		new_mtx = vm_page_lockptr(*ma);
895 		if (mtx != new_mtx) {
896 			if (mtx != NULL)
897 				mtx_unlock(mtx);
898 			mtx = new_mtx;
899 			mtx_lock(mtx);
900 		}
901 		vm_page_unhold(*ma);
902 		ma++;
903 	}
904 	if (mtx != NULL)
905 		mtx_unlock(mtx);
906 }
907 
908 vm_page_t
909 PHYS_TO_VM_PAGE(vm_paddr_t pa)
910 {
911 	vm_page_t m;
912 
913 #ifdef VM_PHYSSEG_SPARSE
914 	m = vm_phys_paddr_to_vm_page(pa);
915 	if (m == NULL)
916 		m = vm_phys_fictitious_to_vm_page(pa);
917 	return (m);
918 #elif defined(VM_PHYSSEG_DENSE)
919 	long pi;
920 
921 	pi = atop(pa);
922 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
923 		m = &vm_page_array[pi - first_page];
924 		return (m);
925 	}
926 	return (vm_phys_fictitious_to_vm_page(pa));
927 #else
928 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
929 #endif
930 }
931 
932 /*
933  *	vm_page_getfake:
934  *
935  *	Create a fictitious page with the specified physical address and
936  *	memory attribute.  The memory attribute is the only the machine-
937  *	dependent aspect of a fictitious page that must be initialized.
938  */
939 vm_page_t
940 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
941 {
942 	vm_page_t m;
943 
944 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
945 	vm_page_initfake(m, paddr, memattr);
946 	return (m);
947 }
948 
949 void
950 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
951 {
952 
953 	if ((m->flags & PG_FICTITIOUS) != 0) {
954 		/*
955 		 * The page's memattr might have changed since the
956 		 * previous initialization.  Update the pmap to the
957 		 * new memattr.
958 		 */
959 		goto memattr;
960 	}
961 	m->phys_addr = paddr;
962 	m->queue = PQ_NONE;
963 	/* Fictitious pages don't use "segind". */
964 	m->flags = PG_FICTITIOUS;
965 	/* Fictitious pages don't use "order" or "pool". */
966 	m->oflags = VPO_UNMANAGED;
967 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
968 	m->wire_count = 1;
969 	pmap_page_init(m);
970 memattr:
971 	pmap_page_set_memattr(m, memattr);
972 }
973 
974 /*
975  *	vm_page_putfake:
976  *
977  *	Release a fictitious page.
978  */
979 void
980 vm_page_putfake(vm_page_t m)
981 {
982 
983 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
984 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
985 	    ("vm_page_putfake: bad page %p", m));
986 	uma_zfree(fakepg_zone, m);
987 }
988 
989 /*
990  *	vm_page_updatefake:
991  *
992  *	Update the given fictitious page to the specified physical address and
993  *	memory attribute.
994  */
995 void
996 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
997 {
998 
999 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
1000 	    ("vm_page_updatefake: bad page %p", m));
1001 	m->phys_addr = paddr;
1002 	pmap_page_set_memattr(m, memattr);
1003 }
1004 
1005 /*
1006  *	vm_page_free:
1007  *
1008  *	Free a page.
1009  */
1010 void
1011 vm_page_free(vm_page_t m)
1012 {
1013 
1014 	m->flags &= ~PG_ZERO;
1015 	vm_page_free_toq(m);
1016 }
1017 
1018 /*
1019  *	vm_page_free_zero:
1020  *
1021  *	Free a page to the zerod-pages queue
1022  */
1023 void
1024 vm_page_free_zero(vm_page_t m)
1025 {
1026 
1027 	m->flags |= PG_ZERO;
1028 	vm_page_free_toq(m);
1029 }
1030 
1031 /*
1032  * Unbusy and handle the page queueing for a page from a getpages request that
1033  * was optionally read ahead or behind.
1034  */
1035 void
1036 vm_page_readahead_finish(vm_page_t m)
1037 {
1038 
1039 	/* We shouldn't put invalid pages on queues. */
1040 	KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m));
1041 
1042 	/*
1043 	 * Since the page is not the actually needed one, whether it should
1044 	 * be activated or deactivated is not obvious.  Empirical results
1045 	 * have shown that deactivating the page is usually the best choice,
1046 	 * unless the page is wanted by another thread.
1047 	 */
1048 	vm_page_lock(m);
1049 	if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1050 		vm_page_activate(m);
1051 	else
1052 		vm_page_deactivate(m);
1053 	vm_page_unlock(m);
1054 	vm_page_xunbusy(m);
1055 }
1056 
1057 /*
1058  *	vm_page_sleep_if_busy:
1059  *
1060  *	Sleep and release the page queues lock if the page is busied.
1061  *	Returns TRUE if the thread slept.
1062  *
1063  *	The given page must be unlocked and object containing it must
1064  *	be locked.
1065  */
1066 int
1067 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1068 {
1069 	vm_object_t obj;
1070 
1071 	vm_page_lock_assert(m, MA_NOTOWNED);
1072 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1073 
1074 	if (vm_page_busied(m)) {
1075 		/*
1076 		 * The page-specific object must be cached because page
1077 		 * identity can change during the sleep, causing the
1078 		 * re-lock of a different object.
1079 		 * It is assumed that a reference to the object is already
1080 		 * held by the callers.
1081 		 */
1082 		obj = m->object;
1083 		vm_page_lock(m);
1084 		VM_OBJECT_WUNLOCK(obj);
1085 		vm_page_busy_sleep(m, msg);
1086 		VM_OBJECT_WLOCK(obj);
1087 		return (TRUE);
1088 	}
1089 	return (FALSE);
1090 }
1091 
1092 /*
1093  *	vm_page_dirty_KBI:		[ internal use only ]
1094  *
1095  *	Set all bits in the page's dirty field.
1096  *
1097  *	The object containing the specified page must be locked if the
1098  *	call is made from the machine-independent layer.
1099  *
1100  *	See vm_page_clear_dirty_mask().
1101  *
1102  *	This function should only be called by vm_page_dirty().
1103  */
1104 void
1105 vm_page_dirty_KBI(vm_page_t m)
1106 {
1107 
1108 	/* These assertions refer to this operation by its public name. */
1109 	KASSERT((m->flags & PG_CACHED) == 0,
1110 	    ("vm_page_dirty: page in cache!"));
1111 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
1112 	    ("vm_page_dirty: page is invalid!"));
1113 	m->dirty = VM_PAGE_BITS_ALL;
1114 }
1115 
1116 /*
1117  *	vm_page_insert:		[ internal use only ]
1118  *
1119  *	Inserts the given mem entry into the object and object list.
1120  *
1121  *	The object must be locked.
1122  */
1123 int
1124 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1125 {
1126 	vm_page_t mpred;
1127 
1128 	VM_OBJECT_ASSERT_WLOCKED(object);
1129 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
1130 	return (vm_page_insert_after(m, object, pindex, mpred));
1131 }
1132 
1133 /*
1134  *	vm_page_insert_after:
1135  *
1136  *	Inserts the page "m" into the specified object at offset "pindex".
1137  *
1138  *	The page "mpred" must immediately precede the offset "pindex" within
1139  *	the specified object.
1140  *
1141  *	The object must be locked.
1142  */
1143 static int
1144 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1145     vm_page_t mpred)
1146 {
1147 	vm_page_t msucc;
1148 
1149 	VM_OBJECT_ASSERT_WLOCKED(object);
1150 	KASSERT(m->object == NULL,
1151 	    ("vm_page_insert_after: page already inserted"));
1152 	if (mpred != NULL) {
1153 		KASSERT(mpred->object == object,
1154 		    ("vm_page_insert_after: object doesn't contain mpred"));
1155 		KASSERT(mpred->pindex < pindex,
1156 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1157 		msucc = TAILQ_NEXT(mpred, listq);
1158 	} else
1159 		msucc = TAILQ_FIRST(&object->memq);
1160 	if (msucc != NULL)
1161 		KASSERT(msucc->pindex > pindex,
1162 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
1163 
1164 	/*
1165 	 * Record the object/offset pair in this page
1166 	 */
1167 	m->object = object;
1168 	m->pindex = pindex;
1169 
1170 	/*
1171 	 * Now link into the object's ordered list of backed pages.
1172 	 */
1173 	if (vm_radix_insert(&object->rtree, m)) {
1174 		m->object = NULL;
1175 		m->pindex = 0;
1176 		return (1);
1177 	}
1178 	vm_page_insert_radixdone(m, object, mpred);
1179 	return (0);
1180 }
1181 
1182 /*
1183  *	vm_page_insert_radixdone:
1184  *
1185  *	Complete page "m" insertion into the specified object after the
1186  *	radix trie hooking.
1187  *
1188  *	The page "mpred" must precede the offset "m->pindex" within the
1189  *	specified object.
1190  *
1191  *	The object must be locked.
1192  */
1193 static void
1194 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1195 {
1196 
1197 	VM_OBJECT_ASSERT_WLOCKED(object);
1198 	KASSERT(object != NULL && m->object == object,
1199 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1200 	if (mpred != NULL) {
1201 		KASSERT(mpred->object == object,
1202 		    ("vm_page_insert_after: object doesn't contain mpred"));
1203 		KASSERT(mpred->pindex < m->pindex,
1204 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1205 	}
1206 
1207 	if (mpred != NULL)
1208 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1209 	else
1210 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1211 
1212 	/*
1213 	 * Show that the object has one more resident page.
1214 	 */
1215 	object->resident_page_count++;
1216 
1217 	/*
1218 	 * Hold the vnode until the last page is released.
1219 	 */
1220 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1221 		vhold(object->handle);
1222 
1223 	/*
1224 	 * Since we are inserting a new and possibly dirty page,
1225 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1226 	 */
1227 	if (pmap_page_is_write_mapped(m))
1228 		vm_object_set_writeable_dirty(object);
1229 }
1230 
1231 /*
1232  *	vm_page_remove:
1233  *
1234  *	Removes the given mem entry from the object/offset-page
1235  *	table and the object page list, but do not invalidate/terminate
1236  *	the backing store.
1237  *
1238  *	The object must be locked.  The page must be locked if it is managed.
1239  */
1240 void
1241 vm_page_remove(vm_page_t m)
1242 {
1243 	vm_object_t object;
1244 
1245 	if ((m->oflags & VPO_UNMANAGED) == 0)
1246 		vm_page_assert_locked(m);
1247 	if ((object = m->object) == NULL)
1248 		return;
1249 	VM_OBJECT_ASSERT_WLOCKED(object);
1250 	if (vm_page_xbusied(m))
1251 		vm_page_xunbusy_maybelocked(m);
1252 
1253 	/*
1254 	 * Now remove from the object's list of backed pages.
1255 	 */
1256 	vm_radix_remove(&object->rtree, m->pindex);
1257 	TAILQ_REMOVE(&object->memq, m, listq);
1258 
1259 	/*
1260 	 * And show that the object has one fewer resident page.
1261 	 */
1262 	object->resident_page_count--;
1263 
1264 	/*
1265 	 * The vnode may now be recycled.
1266 	 */
1267 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1268 		vdrop(object->handle);
1269 
1270 	m->object = NULL;
1271 }
1272 
1273 /*
1274  *	vm_page_lookup:
1275  *
1276  *	Returns the page associated with the object/offset
1277  *	pair specified; if none is found, NULL is returned.
1278  *
1279  *	The object must be locked.
1280  */
1281 vm_page_t
1282 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1283 {
1284 
1285 	VM_OBJECT_ASSERT_LOCKED(object);
1286 	return (vm_radix_lookup(&object->rtree, pindex));
1287 }
1288 
1289 /*
1290  *	vm_page_find_least:
1291  *
1292  *	Returns the page associated with the object with least pindex
1293  *	greater than or equal to the parameter pindex, or NULL.
1294  *
1295  *	The object must be locked.
1296  */
1297 vm_page_t
1298 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1299 {
1300 	vm_page_t m;
1301 
1302 	VM_OBJECT_ASSERT_LOCKED(object);
1303 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1304 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1305 	return (m);
1306 }
1307 
1308 /*
1309  * Returns the given page's successor (by pindex) within the object if it is
1310  * resident; if none is found, NULL is returned.
1311  *
1312  * The object must be locked.
1313  */
1314 vm_page_t
1315 vm_page_next(vm_page_t m)
1316 {
1317 	vm_page_t next;
1318 
1319 	VM_OBJECT_ASSERT_LOCKED(m->object);
1320 	if ((next = TAILQ_NEXT(m, listq)) != NULL &&
1321 	    next->pindex != m->pindex + 1)
1322 		next = NULL;
1323 	return (next);
1324 }
1325 
1326 /*
1327  * Returns the given page's predecessor (by pindex) within the object if it is
1328  * resident; if none is found, NULL is returned.
1329  *
1330  * The object must be locked.
1331  */
1332 vm_page_t
1333 vm_page_prev(vm_page_t m)
1334 {
1335 	vm_page_t prev;
1336 
1337 	VM_OBJECT_ASSERT_LOCKED(m->object);
1338 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL &&
1339 	    prev->pindex != m->pindex - 1)
1340 		prev = NULL;
1341 	return (prev);
1342 }
1343 
1344 /*
1345  * Uses the page mnew as a replacement for an existing page at index
1346  * pindex which must be already present in the object.
1347  *
1348  * The existing page must not be on a paging queue.
1349  */
1350 vm_page_t
1351 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1352 {
1353 	vm_page_t mold;
1354 
1355 	VM_OBJECT_ASSERT_WLOCKED(object);
1356 	KASSERT(mnew->object == NULL,
1357 	    ("vm_page_replace: page already in object"));
1358 
1359 	/*
1360 	 * This function mostly follows vm_page_insert() and
1361 	 * vm_page_remove() without the radix, object count and vnode
1362 	 * dance.  Double check such functions for more comments.
1363 	 */
1364 
1365 	mnew->object = object;
1366 	mnew->pindex = pindex;
1367 	mold = vm_radix_replace(&object->rtree, mnew);
1368 	KASSERT(mold->queue == PQ_NONE,
1369 	    ("vm_page_replace: mold is on a paging queue"));
1370 
1371 	/* Keep the resident page list in sorted order. */
1372 	TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1373 	TAILQ_REMOVE(&object->memq, mold, listq);
1374 
1375 	mold->object = NULL;
1376 	vm_page_xunbusy_maybelocked(mold);
1377 
1378 	/*
1379 	 * The object's resident_page_count does not change because we have
1380 	 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1381 	 */
1382 	if (pmap_page_is_write_mapped(mnew))
1383 		vm_object_set_writeable_dirty(object);
1384 	return (mold);
1385 }
1386 
1387 /*
1388  *	vm_page_rename:
1389  *
1390  *	Move the given memory entry from its
1391  *	current object to the specified target object/offset.
1392  *
1393  *	Note: swap associated with the page must be invalidated by the move.  We
1394  *	      have to do this for several reasons:  (1) we aren't freeing the
1395  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1396  *	      moving the page from object A to B, and will then later move
1397  *	      the backing store from A to B and we can't have a conflict.
1398  *
1399  *	Note: we *always* dirty the page.  It is necessary both for the
1400  *	      fact that we moved it, and because we may be invalidating
1401  *	      swap.  If the page is on the cache, we have to deactivate it
1402  *	      or vm_page_dirty() will panic.  Dirty pages are not allowed
1403  *	      on the cache.
1404  *
1405  *	The objects must be locked.
1406  */
1407 int
1408 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1409 {
1410 	vm_page_t mpred;
1411 	vm_pindex_t opidx;
1412 
1413 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1414 
1415 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1416 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1417 	    ("vm_page_rename: pindex already renamed"));
1418 
1419 	/*
1420 	 * Create a custom version of vm_page_insert() which does not depend
1421 	 * by m_prev and can cheat on the implementation aspects of the
1422 	 * function.
1423 	 */
1424 	opidx = m->pindex;
1425 	m->pindex = new_pindex;
1426 	if (vm_radix_insert(&new_object->rtree, m)) {
1427 		m->pindex = opidx;
1428 		return (1);
1429 	}
1430 
1431 	/*
1432 	 * The operation cannot fail anymore.  The removal must happen before
1433 	 * the listq iterator is tainted.
1434 	 */
1435 	m->pindex = opidx;
1436 	vm_page_lock(m);
1437 	vm_page_remove(m);
1438 
1439 	/* Return back to the new pindex to complete vm_page_insert(). */
1440 	m->pindex = new_pindex;
1441 	m->object = new_object;
1442 	vm_page_unlock(m);
1443 	vm_page_insert_radixdone(m, new_object, mpred);
1444 	vm_page_dirty(m);
1445 	return (0);
1446 }
1447 
1448 /*
1449  *	Convert all of the given object's cached pages that have a
1450  *	pindex within the given range into free pages.  If the value
1451  *	zero is given for "end", then the range's upper bound is
1452  *	infinity.  If the given object is backed by a vnode and it
1453  *	transitions from having one or more cached pages to none, the
1454  *	vnode's hold count is reduced.
1455  */
1456 void
1457 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1458 {
1459 	vm_page_t m;
1460 	boolean_t empty;
1461 
1462 	mtx_lock(&vm_page_queue_free_mtx);
1463 	if (__predict_false(vm_radix_is_empty(&object->cache))) {
1464 		mtx_unlock(&vm_page_queue_free_mtx);
1465 		return;
1466 	}
1467 	while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) {
1468 		if (end != 0 && m->pindex >= end)
1469 			break;
1470 		vm_radix_remove(&object->cache, m->pindex);
1471 		vm_page_cache_turn_free(m);
1472 	}
1473 	empty = vm_radix_is_empty(&object->cache);
1474 	mtx_unlock(&vm_page_queue_free_mtx);
1475 	if (object->type == OBJT_VNODE && empty)
1476 		vdrop(object->handle);
1477 }
1478 
1479 /*
1480  *	Returns the cached page that is associated with the given
1481  *	object and offset.  If, however, none exists, returns NULL.
1482  *
1483  *	The free page queue must be locked.
1484  */
1485 static inline vm_page_t
1486 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
1487 {
1488 
1489 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1490 	return (vm_radix_lookup(&object->cache, pindex));
1491 }
1492 
1493 /*
1494  *	Remove the given cached page from its containing object's
1495  *	collection of cached pages.
1496  *
1497  *	The free page queue must be locked.
1498  */
1499 static void
1500 vm_page_cache_remove(vm_page_t m)
1501 {
1502 
1503 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1504 	KASSERT((m->flags & PG_CACHED) != 0,
1505 	    ("vm_page_cache_remove: page %p is not cached", m));
1506 	vm_radix_remove(&m->object->cache, m->pindex);
1507 	m->object = NULL;
1508 	vm_cnt.v_cache_count--;
1509 }
1510 
1511 /*
1512  *	Transfer all of the cached pages with offset greater than or
1513  *	equal to 'offidxstart' from the original object's cache to the
1514  *	new object's cache.  However, any cached pages with offset
1515  *	greater than or equal to the new object's size are kept in the
1516  *	original object.  Initially, the new object's cache must be
1517  *	empty.  Offset 'offidxstart' in the original object must
1518  *	correspond to offset zero in the new object.
1519  *
1520  *	The new object must be locked.
1521  */
1522 void
1523 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
1524     vm_object_t new_object)
1525 {
1526 	vm_page_t m;
1527 
1528 	/*
1529 	 * Insertion into an object's collection of cached pages
1530 	 * requires the object to be locked.  In contrast, removal does
1531 	 * not.
1532 	 */
1533 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1534 	KASSERT(vm_radix_is_empty(&new_object->cache),
1535 	    ("vm_page_cache_transfer: object %p has cached pages",
1536 	    new_object));
1537 	mtx_lock(&vm_page_queue_free_mtx);
1538 	while ((m = vm_radix_lookup_ge(&orig_object->cache,
1539 	    offidxstart)) != NULL) {
1540 		/*
1541 		 * Transfer all of the pages with offset greater than or
1542 		 * equal to 'offidxstart' from the original object's
1543 		 * cache to the new object's cache.
1544 		 */
1545 		if ((m->pindex - offidxstart) >= new_object->size)
1546 			break;
1547 		vm_radix_remove(&orig_object->cache, m->pindex);
1548 		/* Update the page's object and offset. */
1549 		m->object = new_object;
1550 		m->pindex -= offidxstart;
1551 		if (vm_radix_insert(&new_object->cache, m))
1552 			vm_page_cache_turn_free(m);
1553 	}
1554 	mtx_unlock(&vm_page_queue_free_mtx);
1555 }
1556 
1557 /*
1558  *	Returns TRUE if a cached page is associated with the given object and
1559  *	offset, and FALSE otherwise.
1560  *
1561  *	The object must be locked.
1562  */
1563 boolean_t
1564 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex)
1565 {
1566 	vm_page_t m;
1567 
1568 	/*
1569 	 * Insertion into an object's collection of cached pages requires the
1570 	 * object to be locked.  Therefore, if the object is locked and the
1571 	 * object's collection is empty, there is no need to acquire the free
1572 	 * page queues lock in order to prove that the specified page doesn't
1573 	 * exist.
1574 	 */
1575 	VM_OBJECT_ASSERT_WLOCKED(object);
1576 	if (__predict_true(vm_object_cache_is_empty(object)))
1577 		return (FALSE);
1578 	mtx_lock(&vm_page_queue_free_mtx);
1579 	m = vm_page_cache_lookup(object, pindex);
1580 	mtx_unlock(&vm_page_queue_free_mtx);
1581 	return (m != NULL);
1582 }
1583 
1584 /*
1585  *	vm_page_alloc:
1586  *
1587  *	Allocate and return a page that is associated with the specified
1588  *	object and offset pair.  By default, this page is exclusive busied.
1589  *
1590  *	The caller must always specify an allocation class.
1591  *
1592  *	allocation classes:
1593  *	VM_ALLOC_NORMAL		normal process request
1594  *	VM_ALLOC_SYSTEM		system *really* needs a page
1595  *	VM_ALLOC_INTERRUPT	interrupt time request
1596  *
1597  *	optional allocation flags:
1598  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1599  *				intends to allocate
1600  *	VM_ALLOC_IFCACHED	return page only if it is cached
1601  *	VM_ALLOC_IFNOTCACHED	return NULL, do not reactivate if the page
1602  *				is cached
1603  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1604  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1605  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1606  *				should not be exclusive busy
1607  *	VM_ALLOC_SBUSY		shared busy the allocated page
1608  *	VM_ALLOC_WIRED		wire the allocated page
1609  *	VM_ALLOC_ZERO		prefer a zeroed page
1610  *
1611  *	This routine may not sleep.
1612  */
1613 vm_page_t
1614 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1615 {
1616 	struct vnode *vp = NULL;
1617 	vm_object_t m_object;
1618 	vm_page_t m, mpred;
1619 	int flags, req_class;
1620 
1621 	mpred = 0;	/* XXX: pacify gcc */
1622 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1623 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1624 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1625 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1626 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1627 	    req));
1628 	if (object != NULL)
1629 		VM_OBJECT_ASSERT_WLOCKED(object);
1630 
1631 	req_class = req & VM_ALLOC_CLASS_MASK;
1632 
1633 	/*
1634 	 * The page daemon is allowed to dig deeper into the free page list.
1635 	 */
1636 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1637 		req_class = VM_ALLOC_SYSTEM;
1638 
1639 	if (object != NULL) {
1640 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1641 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1642 		   ("vm_page_alloc: pindex already allocated"));
1643 	}
1644 
1645 	/*
1646 	 * The page allocation request can came from consumers which already
1647 	 * hold the free page queue mutex, like vm_page_insert() in
1648 	 * vm_page_cache().
1649 	 */
1650 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
1651 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
1652 	    (req_class == VM_ALLOC_SYSTEM &&
1653 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
1654 	    (req_class == VM_ALLOC_INTERRUPT &&
1655 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) {
1656 		/*
1657 		 * Allocate from the free queue if the number of free pages
1658 		 * exceeds the minimum for the request class.
1659 		 */
1660 		if (object != NULL &&
1661 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1662 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1663 				mtx_unlock(&vm_page_queue_free_mtx);
1664 				return (NULL);
1665 			}
1666 			if (vm_phys_unfree_page(m))
1667 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1668 #if VM_NRESERVLEVEL > 0
1669 			else if (!vm_reserv_reactivate_page(m))
1670 #else
1671 			else
1672 #endif
1673 				panic("vm_page_alloc: cache page %p is missing"
1674 				    " from the free queue", m);
1675 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1676 			mtx_unlock(&vm_page_queue_free_mtx);
1677 			return (NULL);
1678 #if VM_NRESERVLEVEL > 0
1679 		} else if (object == NULL || (object->flags & (OBJ_COLORED |
1680 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1681 		    vm_reserv_alloc_page(object, pindex, mpred)) == NULL) {
1682 #else
1683 		} else {
1684 #endif
1685 			m = vm_phys_alloc_pages(object != NULL ?
1686 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1687 #if VM_NRESERVLEVEL > 0
1688 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1689 				m = vm_phys_alloc_pages(object != NULL ?
1690 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1691 				    0);
1692 			}
1693 #endif
1694 		}
1695 	} else {
1696 		/*
1697 		 * Not allocatable, give up.
1698 		 */
1699 		mtx_unlock(&vm_page_queue_free_mtx);
1700 		atomic_add_int(&vm_pageout_deficit,
1701 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1702 		pagedaemon_wakeup();
1703 		return (NULL);
1704 	}
1705 
1706 	/*
1707 	 *  At this point we had better have found a good page.
1708 	 */
1709 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1710 	KASSERT(m->queue == PQ_NONE,
1711 	    ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
1712 	KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
1713 	KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
1714 	KASSERT(!vm_page_busied(m), ("vm_page_alloc: page %p is busy", m));
1715 	KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
1716 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1717 	    ("vm_page_alloc: page %p has unexpected memattr %d", m,
1718 	    pmap_page_get_memattr(m)));
1719 	if ((m->flags & PG_CACHED) != 0) {
1720 		KASSERT((m->flags & PG_ZERO) == 0,
1721 		    ("vm_page_alloc: cached page %p is PG_ZERO", m));
1722 		KASSERT(m->valid != 0,
1723 		    ("vm_page_alloc: cached page %p is invalid", m));
1724 		if (m->object == object && m->pindex == pindex)
1725 			vm_cnt.v_reactivated++;
1726 		else
1727 			m->valid = 0;
1728 		m_object = m->object;
1729 		vm_page_cache_remove(m);
1730 		if (m_object->type == OBJT_VNODE &&
1731 		    vm_object_cache_is_empty(m_object))
1732 			vp = m_object->handle;
1733 	} else {
1734 		KASSERT(m->valid == 0,
1735 		    ("vm_page_alloc: free page %p is valid", m));
1736 		vm_phys_freecnt_adj(m, -1);
1737 	}
1738 	mtx_unlock(&vm_page_queue_free_mtx);
1739 
1740 	/*
1741 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1742 	 */
1743 	flags = 0;
1744 	if ((req & VM_ALLOC_ZERO) != 0)
1745 		flags = PG_ZERO;
1746 	flags &= m->flags;
1747 	if ((req & VM_ALLOC_NODUMP) != 0)
1748 		flags |= PG_NODUMP;
1749 	m->flags = flags;
1750 	m->aflags = 0;
1751 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1752 	    VPO_UNMANAGED : 0;
1753 	m->busy_lock = VPB_UNBUSIED;
1754 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1755 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1756 	if ((req & VM_ALLOC_SBUSY) != 0)
1757 		m->busy_lock = VPB_SHARERS_WORD(1);
1758 	if (req & VM_ALLOC_WIRED) {
1759 		/*
1760 		 * The page lock is not required for wiring a page until that
1761 		 * page is inserted into the object.
1762 		 */
1763 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1764 		m->wire_count = 1;
1765 	}
1766 	m->act_count = 0;
1767 
1768 	if (object != NULL) {
1769 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1770 			/* See the comment below about hold count. */
1771 			if (vp != NULL)
1772 				vdrop(vp);
1773 			pagedaemon_wakeup();
1774 			if (req & VM_ALLOC_WIRED) {
1775 				atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1776 				m->wire_count = 0;
1777 			}
1778 			m->object = NULL;
1779 			m->oflags = VPO_UNMANAGED;
1780 			m->busy_lock = VPB_UNBUSIED;
1781 			vm_page_free(m);
1782 			return (NULL);
1783 		}
1784 
1785 		/* Ignore device objects; the pager sets "memattr" for them. */
1786 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1787 		    (object->flags & OBJ_FICTITIOUS) == 0)
1788 			pmap_page_set_memattr(m, object->memattr);
1789 	} else
1790 		m->pindex = pindex;
1791 
1792 	/*
1793 	 * The following call to vdrop() must come after the above call
1794 	 * to vm_page_insert() in case both affect the same object and
1795 	 * vnode.  Otherwise, the affected vnode's hold count could
1796 	 * temporarily become zero.
1797 	 */
1798 	if (vp != NULL)
1799 		vdrop(vp);
1800 
1801 	/*
1802 	 * Don't wakeup too often - wakeup the pageout daemon when
1803 	 * we would be nearly out of memory.
1804 	 */
1805 	if (vm_paging_needed())
1806 		pagedaemon_wakeup();
1807 
1808 	return (m);
1809 }
1810 
1811 static void
1812 vm_page_alloc_contig_vdrop(struct spglist *lst)
1813 {
1814 
1815 	while (!SLIST_EMPTY(lst)) {
1816 		vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv);
1817 		SLIST_REMOVE_HEAD(lst, plinks.s.ss);
1818 	}
1819 }
1820 
1821 /*
1822  *	vm_page_alloc_contig:
1823  *
1824  *	Allocate a contiguous set of physical pages of the given size "npages"
1825  *	from the free lists.  All of the physical pages must be at or above
1826  *	the given physical address "low" and below the given physical address
1827  *	"high".  The given value "alignment" determines the alignment of the
1828  *	first physical page in the set.  If the given value "boundary" is
1829  *	non-zero, then the set of physical pages cannot cross any physical
1830  *	address boundary that is a multiple of that value.  Both "alignment"
1831  *	and "boundary" must be a power of two.
1832  *
1833  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1834  *	then the memory attribute setting for the physical pages is configured
1835  *	to the object's memory attribute setting.  Otherwise, the memory
1836  *	attribute setting for the physical pages is configured to "memattr",
1837  *	overriding the object's memory attribute setting.  However, if the
1838  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1839  *	memory attribute setting for the physical pages cannot be configured
1840  *	to VM_MEMATTR_DEFAULT.
1841  *
1842  *	The caller must always specify an allocation class.
1843  *
1844  *	allocation classes:
1845  *	VM_ALLOC_NORMAL		normal process request
1846  *	VM_ALLOC_SYSTEM		system *really* needs a page
1847  *	VM_ALLOC_INTERRUPT	interrupt time request
1848  *
1849  *	optional allocation flags:
1850  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1851  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1852  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1853  *				should not be exclusive busy
1854  *	VM_ALLOC_SBUSY		shared busy the allocated page
1855  *	VM_ALLOC_WIRED		wire the allocated page
1856  *	VM_ALLOC_ZERO		prefer a zeroed page
1857  *
1858  *	This routine may not sleep.
1859  */
1860 vm_page_t
1861 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1862     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1863     vm_paddr_t boundary, vm_memattr_t memattr)
1864 {
1865 	struct vnode *drop;
1866 	struct spglist deferred_vdrop_list;
1867 	vm_page_t m, m_tmp, m_ret;
1868 	u_int flags;
1869 	int req_class;
1870 
1871 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1872 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1873 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1874 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1875 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1876 	    req));
1877 	if (object != NULL) {
1878 		VM_OBJECT_ASSERT_WLOCKED(object);
1879 		KASSERT(object->type == OBJT_PHYS,
1880 		    ("vm_page_alloc_contig: object %p isn't OBJT_PHYS",
1881 		    object));
1882 	}
1883 	KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1884 	req_class = req & VM_ALLOC_CLASS_MASK;
1885 
1886 	/*
1887 	 * The page daemon is allowed to dig deeper into the free page list.
1888 	 */
1889 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1890 		req_class = VM_ALLOC_SYSTEM;
1891 
1892 	SLIST_INIT(&deferred_vdrop_list);
1893 	mtx_lock(&vm_page_queue_free_mtx);
1894 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1895 	    vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM &&
1896 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1897 	    vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT &&
1898 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) {
1899 #if VM_NRESERVLEVEL > 0
1900 retry:
1901 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1902 		    (m_ret = vm_reserv_alloc_contig(object, pindex, npages,
1903 		    low, high, alignment, boundary)) == NULL)
1904 #endif
1905 			m_ret = vm_phys_alloc_contig(npages, low, high,
1906 			    alignment, boundary);
1907 	} else {
1908 		mtx_unlock(&vm_page_queue_free_mtx);
1909 		atomic_add_int(&vm_pageout_deficit, npages);
1910 		pagedaemon_wakeup();
1911 		return (NULL);
1912 	}
1913 	if (m_ret != NULL)
1914 		for (m = m_ret; m < &m_ret[npages]; m++) {
1915 			drop = vm_page_alloc_init(m);
1916 			if (drop != NULL) {
1917 				/*
1918 				 * Enqueue the vnode for deferred vdrop().
1919 				 */
1920 				m->plinks.s.pv = drop;
1921 				SLIST_INSERT_HEAD(&deferred_vdrop_list, m,
1922 				    plinks.s.ss);
1923 			}
1924 		}
1925 	else {
1926 #if VM_NRESERVLEVEL > 0
1927 		if (vm_reserv_reclaim_contig(npages, low, high, alignment,
1928 		    boundary))
1929 			goto retry;
1930 #endif
1931 	}
1932 	mtx_unlock(&vm_page_queue_free_mtx);
1933 	if (m_ret == NULL)
1934 		return (NULL);
1935 
1936 	/*
1937 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1938 	 */
1939 	flags = 0;
1940 	if ((req & VM_ALLOC_ZERO) != 0)
1941 		flags = PG_ZERO;
1942 	if ((req & VM_ALLOC_NODUMP) != 0)
1943 		flags |= PG_NODUMP;
1944 	if ((req & VM_ALLOC_WIRED) != 0)
1945 		atomic_add_int(&vm_cnt.v_wire_count, npages);
1946 	if (object != NULL) {
1947 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1948 		    memattr == VM_MEMATTR_DEFAULT)
1949 			memattr = object->memattr;
1950 	}
1951 	for (m = m_ret; m < &m_ret[npages]; m++) {
1952 		m->aflags = 0;
1953 		m->flags = (m->flags | PG_NODUMP) & flags;
1954 		m->busy_lock = VPB_UNBUSIED;
1955 		if (object != NULL) {
1956 			if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
1957 				m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1958 			if ((req & VM_ALLOC_SBUSY) != 0)
1959 				m->busy_lock = VPB_SHARERS_WORD(1);
1960 		}
1961 		if ((req & VM_ALLOC_WIRED) != 0)
1962 			m->wire_count = 1;
1963 		/* Unmanaged pages don't use "act_count". */
1964 		m->oflags = VPO_UNMANAGED;
1965 		if (object != NULL) {
1966 			if (vm_page_insert(m, object, pindex)) {
1967 				vm_page_alloc_contig_vdrop(
1968 				    &deferred_vdrop_list);
1969 				if (vm_paging_needed())
1970 					pagedaemon_wakeup();
1971 				if ((req & VM_ALLOC_WIRED) != 0)
1972 					atomic_subtract_int(&vm_cnt.v_wire_count,
1973 					    npages);
1974 				for (m_tmp = m, m = m_ret;
1975 				    m < &m_ret[npages]; m++) {
1976 					if ((req & VM_ALLOC_WIRED) != 0)
1977 						m->wire_count = 0;
1978 					if (m >= m_tmp) {
1979 						m->object = NULL;
1980 						m->oflags |= VPO_UNMANAGED;
1981 					}
1982 					m->busy_lock = VPB_UNBUSIED;
1983 					vm_page_free(m);
1984 				}
1985 				return (NULL);
1986 			}
1987 		} else
1988 			m->pindex = pindex;
1989 		if (memattr != VM_MEMATTR_DEFAULT)
1990 			pmap_page_set_memattr(m, memattr);
1991 		pindex++;
1992 	}
1993 	vm_page_alloc_contig_vdrop(&deferred_vdrop_list);
1994 	if (vm_paging_needed())
1995 		pagedaemon_wakeup();
1996 	return (m_ret);
1997 }
1998 
1999 /*
2000  * Initialize a page that has been freshly dequeued from a freelist.
2001  * The caller has to drop the vnode returned, if it is not NULL.
2002  *
2003  * This function may only be used to initialize unmanaged pages.
2004  *
2005  * To be called with vm_page_queue_free_mtx held.
2006  */
2007 static struct vnode *
2008 vm_page_alloc_init(vm_page_t m)
2009 {
2010 	struct vnode *drop;
2011 	vm_object_t m_object;
2012 
2013 	KASSERT(m->queue == PQ_NONE,
2014 	    ("vm_page_alloc_init: page %p has unexpected queue %d",
2015 	    m, m->queue));
2016 	KASSERT(m->wire_count == 0,
2017 	    ("vm_page_alloc_init: page %p is wired", m));
2018 	KASSERT(m->hold_count == 0,
2019 	    ("vm_page_alloc_init: page %p is held", m));
2020 	KASSERT(!vm_page_busied(m),
2021 	    ("vm_page_alloc_init: page %p is busy", m));
2022 	KASSERT(m->dirty == 0,
2023 	    ("vm_page_alloc_init: page %p is dirty", m));
2024 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
2025 	    ("vm_page_alloc_init: page %p has unexpected memattr %d",
2026 	    m, pmap_page_get_memattr(m)));
2027 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2028 	drop = NULL;
2029 	if ((m->flags & PG_CACHED) != 0) {
2030 		KASSERT((m->flags & PG_ZERO) == 0,
2031 		    ("vm_page_alloc_init: cached page %p is PG_ZERO", m));
2032 		m->valid = 0;
2033 		m_object = m->object;
2034 		vm_page_cache_remove(m);
2035 		if (m_object->type == OBJT_VNODE &&
2036 		    vm_object_cache_is_empty(m_object))
2037 			drop = m_object->handle;
2038 	} else {
2039 		KASSERT(m->valid == 0,
2040 		    ("vm_page_alloc_init: free page %p is valid", m));
2041 		vm_phys_freecnt_adj(m, -1);
2042 	}
2043 	return (drop);
2044 }
2045 
2046 /*
2047  * 	vm_page_alloc_freelist:
2048  *
2049  *	Allocate a physical page from the specified free page list.
2050  *
2051  *	The caller must always specify an allocation class.
2052  *
2053  *	allocation classes:
2054  *	VM_ALLOC_NORMAL		normal process request
2055  *	VM_ALLOC_SYSTEM		system *really* needs a page
2056  *	VM_ALLOC_INTERRUPT	interrupt time request
2057  *
2058  *	optional allocation flags:
2059  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
2060  *				intends to allocate
2061  *	VM_ALLOC_WIRED		wire the allocated page
2062  *	VM_ALLOC_ZERO		prefer a zeroed page
2063  *
2064  *	This routine may not sleep.
2065  */
2066 vm_page_t
2067 vm_page_alloc_freelist(int flind, int req)
2068 {
2069 	struct vnode *drop;
2070 	vm_page_t m;
2071 	u_int flags;
2072 	int req_class;
2073 
2074 	req_class = req & VM_ALLOC_CLASS_MASK;
2075 
2076 	/*
2077 	 * The page daemon is allowed to dig deeper into the free page list.
2078 	 */
2079 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2080 		req_class = VM_ALLOC_SYSTEM;
2081 
2082 	/*
2083 	 * Do not allocate reserved pages unless the req has asked for it.
2084 	 */
2085 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
2086 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
2087 	    (req_class == VM_ALLOC_SYSTEM &&
2088 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
2089 	    (req_class == VM_ALLOC_INTERRUPT &&
2090 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0))
2091 		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0);
2092 	else {
2093 		mtx_unlock(&vm_page_queue_free_mtx);
2094 		atomic_add_int(&vm_pageout_deficit,
2095 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
2096 		pagedaemon_wakeup();
2097 		return (NULL);
2098 	}
2099 	if (m == NULL) {
2100 		mtx_unlock(&vm_page_queue_free_mtx);
2101 		return (NULL);
2102 	}
2103 	drop = vm_page_alloc_init(m);
2104 	mtx_unlock(&vm_page_queue_free_mtx);
2105 
2106 	/*
2107 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
2108 	 */
2109 	m->aflags = 0;
2110 	flags = 0;
2111 	if ((req & VM_ALLOC_ZERO) != 0)
2112 		flags = PG_ZERO;
2113 	m->flags &= flags;
2114 	if ((req & VM_ALLOC_WIRED) != 0) {
2115 		/*
2116 		 * The page lock is not required for wiring a page that does
2117 		 * not belong to an object.
2118 		 */
2119 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2120 		m->wire_count = 1;
2121 	}
2122 	/* Unmanaged pages don't use "act_count". */
2123 	m->oflags = VPO_UNMANAGED;
2124 	if (drop != NULL)
2125 		vdrop(drop);
2126 	if (vm_paging_needed())
2127 		pagedaemon_wakeup();
2128 	return (m);
2129 }
2130 
2131 #define	VPSC_ANY	0	/* No restrictions. */
2132 #define	VPSC_NORESERV	1	/* Skip reservations; implies VPSC_NOSUPER. */
2133 #define	VPSC_NOSUPER	2	/* Skip superpages. */
2134 
2135 /*
2136  *	vm_page_scan_contig:
2137  *
2138  *	Scan vm_page_array[] between the specified entries "m_start" and
2139  *	"m_end" for a run of contiguous physical pages that satisfy the
2140  *	specified conditions, and return the lowest page in the run.  The
2141  *	specified "alignment" determines the alignment of the lowest physical
2142  *	page in the run.  If the specified "boundary" is non-zero, then the
2143  *	run of physical pages cannot span a physical address that is a
2144  *	multiple of "boundary".
2145  *
2146  *	"m_end" is never dereferenced, so it need not point to a vm_page
2147  *	structure within vm_page_array[].
2148  *
2149  *	"npages" must be greater than zero.  "m_start" and "m_end" must not
2150  *	span a hole (or discontiguity) in the physical address space.  Both
2151  *	"alignment" and "boundary" must be a power of two.
2152  */
2153 vm_page_t
2154 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
2155     u_long alignment, vm_paddr_t boundary, int options)
2156 {
2157 	struct mtx *m_mtx, *new_mtx;
2158 	vm_object_t object;
2159 	vm_paddr_t pa;
2160 	vm_page_t m, m_run;
2161 #if VM_NRESERVLEVEL > 0
2162 	int level;
2163 #endif
2164 	int m_inc, order, run_ext, run_len;
2165 
2166 	KASSERT(npages > 0, ("npages is 0"));
2167 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2168 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2169 	m_run = NULL;
2170 	run_len = 0;
2171 	m_mtx = NULL;
2172 	for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
2173 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2174 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2175 
2176 		/*
2177 		 * If the current page would be the start of a run, check its
2178 		 * physical address against the end, alignment, and boundary
2179 		 * conditions.  If it doesn't satisfy these conditions, either
2180 		 * terminate the scan or advance to the next page that
2181 		 * satisfies the failed condition.
2182 		 */
2183 		if (run_len == 0) {
2184 			KASSERT(m_run == NULL, ("m_run != NULL"));
2185 			if (m + npages > m_end)
2186 				break;
2187 			pa = VM_PAGE_TO_PHYS(m);
2188 			if ((pa & (alignment - 1)) != 0) {
2189 				m_inc = atop(roundup2(pa, alignment) - pa);
2190 				continue;
2191 			}
2192 			if (rounddown2(pa ^ (pa + ptoa(npages) - 1),
2193 			    boundary) != 0) {
2194 				m_inc = atop(roundup2(pa, boundary) - pa);
2195 				continue;
2196 			}
2197 		} else
2198 			KASSERT(m_run != NULL, ("m_run == NULL"));
2199 
2200 		/*
2201 		 * Avoid releasing and reacquiring the same page lock.
2202 		 */
2203 		new_mtx = vm_page_lockptr(m);
2204 		if (m_mtx != new_mtx) {
2205 			if (m_mtx != NULL)
2206 				mtx_unlock(m_mtx);
2207 			m_mtx = new_mtx;
2208 			mtx_lock(m_mtx);
2209 		}
2210 		m_inc = 1;
2211 retry:
2212 		if (m->wire_count != 0 || m->hold_count != 0)
2213 			run_ext = 0;
2214 #if VM_NRESERVLEVEL > 0
2215 		else if ((level = vm_reserv_level(m)) >= 0 &&
2216 		    (options & VPSC_NORESERV) != 0) {
2217 			run_ext = 0;
2218 			/* Advance to the end of the reservation. */
2219 			pa = VM_PAGE_TO_PHYS(m);
2220 			m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
2221 			    pa);
2222 		}
2223 #endif
2224 		else if ((object = m->object) != NULL) {
2225 			/*
2226 			 * The page is considered eligible for relocation if
2227 			 * and only if it could be laundered or reclaimed by
2228 			 * the page daemon.
2229 			 */
2230 			if (!VM_OBJECT_TRYRLOCK(object)) {
2231 				mtx_unlock(m_mtx);
2232 				VM_OBJECT_RLOCK(object);
2233 				mtx_lock(m_mtx);
2234 				if (m->object != object) {
2235 					/*
2236 					 * The page may have been freed.
2237 					 */
2238 					VM_OBJECT_RUNLOCK(object);
2239 					goto retry;
2240 				} else if (m->wire_count != 0 ||
2241 				    m->hold_count != 0) {
2242 					run_ext = 0;
2243 					goto unlock;
2244 				}
2245 			}
2246 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2247 			    ("page %p is PG_UNHOLDFREE", m));
2248 			/* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */
2249 			if (object->type != OBJT_DEFAULT &&
2250 			    object->type != OBJT_SWAP &&
2251 			    object->type != OBJT_VNODE)
2252 				run_ext = 0;
2253 			else if ((m->flags & PG_CACHED) != 0 ||
2254 			    m != vm_page_lookup(object, m->pindex)) {
2255 				/*
2256 				 * The page is cached or recently converted
2257 				 * from cached to free.
2258 				 */
2259 #if VM_NRESERVLEVEL > 0
2260 				if (level >= 0) {
2261 					/*
2262 					 * The page is reserved.  Extend the
2263 					 * current run by one page.
2264 					 */
2265 					run_ext = 1;
2266 				} else
2267 #endif
2268 				if ((order = m->order) < VM_NFREEORDER) {
2269 					/*
2270 					 * The page is enqueued in the
2271 					 * physical memory allocator's cache/
2272 					 * free page queues.  Moreover, it is
2273 					 * the first page in a power-of-two-
2274 					 * sized run of contiguous cache/free
2275 					 * pages.  Add these pages to the end
2276 					 * of the current run, and jump
2277 					 * ahead.
2278 					 */
2279 					run_ext = 1 << order;
2280 					m_inc = 1 << order;
2281 				} else
2282 					run_ext = 0;
2283 #if VM_NRESERVLEVEL > 0
2284 			} else if ((options & VPSC_NOSUPER) != 0 &&
2285 			    (level = vm_reserv_level_iffullpop(m)) >= 0) {
2286 				run_ext = 0;
2287 				/* Advance to the end of the superpage. */
2288 				pa = VM_PAGE_TO_PHYS(m);
2289 				m_inc = atop(roundup2(pa + 1,
2290 				    vm_reserv_size(level)) - pa);
2291 #endif
2292 			} else if (object->memattr == VM_MEMATTR_DEFAULT &&
2293 			    m->queue != PQ_NONE && !vm_page_busied(m)) {
2294 				/*
2295 				 * The page is allocated but eligible for
2296 				 * relocation.  Extend the current run by one
2297 				 * page.
2298 				 */
2299 				KASSERT(pmap_page_get_memattr(m) ==
2300 				    VM_MEMATTR_DEFAULT,
2301 				    ("page %p has an unexpected memattr", m));
2302 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2303 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2304 				    ("page %p has unexpected oflags", m));
2305 				/* Don't care: VPO_NOSYNC. */
2306 				run_ext = 1;
2307 			} else
2308 				run_ext = 0;
2309 unlock:
2310 			VM_OBJECT_RUNLOCK(object);
2311 #if VM_NRESERVLEVEL > 0
2312 		} else if (level >= 0) {
2313 			/*
2314 			 * The page is reserved but not yet allocated.  In
2315 			 * other words, it is still cached or free.  Extend
2316 			 * the current run by one page.
2317 			 */
2318 			run_ext = 1;
2319 #endif
2320 		} else if ((order = m->order) < VM_NFREEORDER) {
2321 			/*
2322 			 * The page is enqueued in the physical memory
2323 			 * allocator's cache/free page queues.  Moreover, it
2324 			 * is the first page in a power-of-two-sized run of
2325 			 * contiguous cache/free pages.  Add these pages to
2326 			 * the end of the current run, and jump ahead.
2327 			 */
2328 			run_ext = 1 << order;
2329 			m_inc = 1 << order;
2330 		} else {
2331 			/*
2332 			 * Skip the page for one of the following reasons: (1)
2333 			 * It is enqueued in the physical memory allocator's
2334 			 * cache/free page queues.  However, it is not the
2335 			 * first page in a run of contiguous cache/free pages.
2336 			 * (This case rarely occurs because the scan is
2337 			 * performed in ascending order.) (2) It is not
2338 			 * reserved, and it is transitioning from free to
2339 			 * allocated.  (Conversely, the transition from
2340 			 * allocated to free for managed pages is blocked by
2341 			 * the page lock.) (3) It is allocated but not
2342 			 * contained by an object and not wired, e.g.,
2343 			 * allocated by Xen's balloon driver.
2344 			 */
2345 			run_ext = 0;
2346 		}
2347 
2348 		/*
2349 		 * Extend or reset the current run of pages.
2350 		 */
2351 		if (run_ext > 0) {
2352 			if (run_len == 0)
2353 				m_run = m;
2354 			run_len += run_ext;
2355 		} else {
2356 			if (run_len > 0) {
2357 				m_run = NULL;
2358 				run_len = 0;
2359 			}
2360 		}
2361 	}
2362 	if (m_mtx != NULL)
2363 		mtx_unlock(m_mtx);
2364 	if (run_len >= npages)
2365 		return (m_run);
2366 	return (NULL);
2367 }
2368 
2369 /*
2370  *	vm_page_reclaim_run:
2371  *
2372  *	Try to relocate each of the allocated virtual pages within the
2373  *	specified run of physical pages to a new physical address.  Free the
2374  *	physical pages underlying the relocated virtual pages.  A virtual page
2375  *	is relocatable if and only if it could be laundered or reclaimed by
2376  *	the page daemon.  Whenever possible, a virtual page is relocated to a
2377  *	physical address above "high".
2378  *
2379  *	Returns 0 if every physical page within the run was already free or
2380  *	just freed by a successful relocation.  Otherwise, returns a non-zero
2381  *	value indicating why the last attempt to relocate a virtual page was
2382  *	unsuccessful.
2383  *
2384  *	"req_class" must be an allocation class.
2385  */
2386 static int
2387 vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
2388     vm_paddr_t high)
2389 {
2390 	struct mtx *m_mtx, *new_mtx;
2391 	struct spglist free;
2392 	vm_object_t object;
2393 	vm_paddr_t pa;
2394 	vm_page_t m, m_end, m_new;
2395 	int error, order, req;
2396 
2397 	KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
2398 	    ("req_class is not an allocation class"));
2399 	SLIST_INIT(&free);
2400 	error = 0;
2401 	m = m_run;
2402 	m_end = m_run + npages;
2403 	m_mtx = NULL;
2404 	for (; error == 0 && m < m_end; m++) {
2405 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2406 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2407 
2408 		/*
2409 		 * Avoid releasing and reacquiring the same page lock.
2410 		 */
2411 		new_mtx = vm_page_lockptr(m);
2412 		if (m_mtx != new_mtx) {
2413 			if (m_mtx != NULL)
2414 				mtx_unlock(m_mtx);
2415 			m_mtx = new_mtx;
2416 			mtx_lock(m_mtx);
2417 		}
2418 retry:
2419 		if (m->wire_count != 0 || m->hold_count != 0)
2420 			error = EBUSY;
2421 		else if ((object = m->object) != NULL) {
2422 			/*
2423 			 * The page is relocated if and only if it could be
2424 			 * laundered or reclaimed by the page daemon.
2425 			 */
2426 			if (!VM_OBJECT_TRYWLOCK(object)) {
2427 				mtx_unlock(m_mtx);
2428 				VM_OBJECT_WLOCK(object);
2429 				mtx_lock(m_mtx);
2430 				if (m->object != object) {
2431 					/*
2432 					 * The page may have been freed.
2433 					 */
2434 					VM_OBJECT_WUNLOCK(object);
2435 					goto retry;
2436 				} else if (m->wire_count != 0 ||
2437 				    m->hold_count != 0) {
2438 					error = EBUSY;
2439 					goto unlock;
2440 				}
2441 			}
2442 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2443 			    ("page %p is PG_UNHOLDFREE", m));
2444 			/* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */
2445 			if (object->type != OBJT_DEFAULT &&
2446 			    object->type != OBJT_SWAP &&
2447 			    object->type != OBJT_VNODE)
2448 				error = EINVAL;
2449 			else if ((m->flags & PG_CACHED) != 0 ||
2450 			    m != vm_page_lookup(object, m->pindex)) {
2451 				/*
2452 				 * The page is cached or recently converted
2453 				 * from cached to free.
2454 				 */
2455 				VM_OBJECT_WUNLOCK(object);
2456 				goto cached;
2457 			} else if (object->memattr != VM_MEMATTR_DEFAULT)
2458 				error = EINVAL;
2459 			else if (m->queue != PQ_NONE && !vm_page_busied(m)) {
2460 				KASSERT(pmap_page_get_memattr(m) ==
2461 				    VM_MEMATTR_DEFAULT,
2462 				    ("page %p has an unexpected memattr", m));
2463 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2464 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2465 				    ("page %p has unexpected oflags", m));
2466 				/* Don't care: VPO_NOSYNC. */
2467 				if (m->valid != 0) {
2468 					/*
2469 					 * First, try to allocate a new page
2470 					 * that is above "high".  Failing
2471 					 * that, try to allocate a new page
2472 					 * that is below "m_run".  Allocate
2473 					 * the new page between the end of
2474 					 * "m_run" and "high" only as a last
2475 					 * resort.
2476 					 */
2477 					req = req_class | VM_ALLOC_NOOBJ;
2478 					if ((m->flags & PG_NODUMP) != 0)
2479 						req |= VM_ALLOC_NODUMP;
2480 					if (trunc_page(high) !=
2481 					    ~(vm_paddr_t)PAGE_MASK) {
2482 						m_new = vm_page_alloc_contig(
2483 						    NULL, 0, req, 1,
2484 						    round_page(high),
2485 						    ~(vm_paddr_t)0,
2486 						    PAGE_SIZE, 0,
2487 						    VM_MEMATTR_DEFAULT);
2488 					} else
2489 						m_new = NULL;
2490 					if (m_new == NULL) {
2491 						pa = VM_PAGE_TO_PHYS(m_run);
2492 						m_new = vm_page_alloc_contig(
2493 						    NULL, 0, req, 1,
2494 						    0, pa - 1, PAGE_SIZE, 0,
2495 						    VM_MEMATTR_DEFAULT);
2496 					}
2497 					if (m_new == NULL) {
2498 						pa += ptoa(npages);
2499 						m_new = vm_page_alloc_contig(
2500 						    NULL, 0, req, 1,
2501 						    pa, high, PAGE_SIZE, 0,
2502 						    VM_MEMATTR_DEFAULT);
2503 					}
2504 					if (m_new == NULL) {
2505 						error = ENOMEM;
2506 						goto unlock;
2507 					}
2508 					KASSERT(m_new->wire_count == 0,
2509 					    ("page %p is wired", m));
2510 
2511 					/*
2512 					 * Replace "m" with the new page.  For
2513 					 * vm_page_replace(), "m" must be busy
2514 					 * and dequeued.  Finally, change "m"
2515 					 * as if vm_page_free() was called.
2516 					 */
2517 					if (object->ref_count != 0)
2518 						pmap_remove_all(m);
2519 					m_new->aflags = m->aflags;
2520 					KASSERT(m_new->oflags == VPO_UNMANAGED,
2521 					    ("page %p is managed", m));
2522 					m_new->oflags = m->oflags & VPO_NOSYNC;
2523 					pmap_copy_page(m, m_new);
2524 					m_new->valid = m->valid;
2525 					m_new->dirty = m->dirty;
2526 					m->flags &= ~PG_ZERO;
2527 					vm_page_xbusy(m);
2528 					vm_page_remque(m);
2529 					vm_page_replace_checked(m_new, object,
2530 					    m->pindex, m);
2531 					m->valid = 0;
2532 					vm_page_undirty(m);
2533 
2534 					/*
2535 					 * The new page must be deactivated
2536 					 * before the object is unlocked.
2537 					 */
2538 					new_mtx = vm_page_lockptr(m_new);
2539 					if (m_mtx != new_mtx) {
2540 						mtx_unlock(m_mtx);
2541 						m_mtx = new_mtx;
2542 						mtx_lock(m_mtx);
2543 					}
2544 					vm_page_deactivate(m_new);
2545 				} else {
2546 					m->flags &= ~PG_ZERO;
2547 					vm_page_remque(m);
2548 					vm_page_remove(m);
2549 					KASSERT(m->dirty == 0,
2550 					    ("page %p is dirty", m));
2551 				}
2552 				SLIST_INSERT_HEAD(&free, m, plinks.s.ss);
2553 			} else
2554 				error = EBUSY;
2555 unlock:
2556 			VM_OBJECT_WUNLOCK(object);
2557 		} else {
2558 cached:
2559 			mtx_lock(&vm_page_queue_free_mtx);
2560 			order = m->order;
2561 			if (order < VM_NFREEORDER) {
2562 				/*
2563 				 * The page is enqueued in the physical memory
2564 				 * allocator's cache/free page queues.
2565 				 * Moreover, it is the first page in a power-
2566 				 * of-two-sized run of contiguous cache/free
2567 				 * pages.  Jump ahead to the last page within
2568 				 * that run, and continue from there.
2569 				 */
2570 				m += (1 << order) - 1;
2571 			}
2572 #if VM_NRESERVLEVEL > 0
2573 			else if (vm_reserv_is_page_free(m))
2574 				order = 0;
2575 #endif
2576 			mtx_unlock(&vm_page_queue_free_mtx);
2577 			if (order == VM_NFREEORDER)
2578 				error = EINVAL;
2579 		}
2580 	}
2581 	if (m_mtx != NULL)
2582 		mtx_unlock(m_mtx);
2583 	if ((m = SLIST_FIRST(&free)) != NULL) {
2584 		mtx_lock(&vm_page_queue_free_mtx);
2585 		do {
2586 			SLIST_REMOVE_HEAD(&free, plinks.s.ss);
2587 			vm_phys_freecnt_adj(m, 1);
2588 #if VM_NRESERVLEVEL > 0
2589 			if (!vm_reserv_free_page(m))
2590 #else
2591 			if (true)
2592 #endif
2593 				vm_phys_free_pages(m, 0);
2594 		} while ((m = SLIST_FIRST(&free)) != NULL);
2595 		vm_page_free_wakeup();
2596 		mtx_unlock(&vm_page_queue_free_mtx);
2597 	}
2598 	return (error);
2599 }
2600 
2601 #define	NRUNS	16
2602 
2603 CTASSERT(powerof2(NRUNS));
2604 
2605 #define	RUN_INDEX(count)	((count) & (NRUNS - 1))
2606 
2607 #define	MIN_RECLAIM	8
2608 
2609 /*
2610  *	vm_page_reclaim_contig:
2611  *
2612  *	Reclaim allocated, contiguous physical memory satisfying the specified
2613  *	conditions by relocating the virtual pages using that physical memory.
2614  *	Returns true if reclamation is successful and false otherwise.  Since
2615  *	relocation requires the allocation of physical pages, reclamation may
2616  *	fail due to a shortage of cache/free pages.  When reclamation fails,
2617  *	callers are expected to perform VM_WAIT before retrying a failed
2618  *	allocation operation, e.g., vm_page_alloc_contig().
2619  *
2620  *	The caller must always specify an allocation class through "req".
2621  *
2622  *	allocation classes:
2623  *	VM_ALLOC_NORMAL		normal process request
2624  *	VM_ALLOC_SYSTEM		system *really* needs a page
2625  *	VM_ALLOC_INTERRUPT	interrupt time request
2626  *
2627  *	The optional allocation flags are ignored.
2628  *
2629  *	"npages" must be greater than zero.  Both "alignment" and "boundary"
2630  *	must be a power of two.
2631  */
2632 bool
2633 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
2634     u_long alignment, vm_paddr_t boundary)
2635 {
2636 	vm_paddr_t curr_low;
2637 	vm_page_t m_run, m_runs[NRUNS];
2638 	u_long count, reclaimed;
2639 	int error, i, options, req_class;
2640 
2641 	KASSERT(npages > 0, ("npages is 0"));
2642 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2643 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2644 	req_class = req & VM_ALLOC_CLASS_MASK;
2645 
2646 	/*
2647 	 * The page daemon is allowed to dig deeper into the free page list.
2648 	 */
2649 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2650 		req_class = VM_ALLOC_SYSTEM;
2651 
2652 	/*
2653 	 * Return if the number of cached and free pages cannot satisfy the
2654 	 * requested allocation.
2655 	 */
2656 	count = vm_cnt.v_free_count + vm_cnt.v_cache_count;
2657 	if (count < npages + vm_cnt.v_free_reserved || (count < npages +
2658 	    vm_cnt.v_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
2659 	    (count < npages && req_class == VM_ALLOC_INTERRUPT))
2660 		return (false);
2661 
2662 	/*
2663 	 * Scan up to three times, relaxing the restrictions ("options") on
2664 	 * the reclamation of reservations and superpages each time.
2665 	 */
2666 	for (options = VPSC_NORESERV;;) {
2667 		/*
2668 		 * Find the highest runs that satisfy the given constraints
2669 		 * and restrictions, and record them in "m_runs".
2670 		 */
2671 		curr_low = low;
2672 		count = 0;
2673 		for (;;) {
2674 			m_run = vm_phys_scan_contig(npages, curr_low, high,
2675 			    alignment, boundary, options);
2676 			if (m_run == NULL)
2677 				break;
2678 			curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages);
2679 			m_runs[RUN_INDEX(count)] = m_run;
2680 			count++;
2681 		}
2682 
2683 		/*
2684 		 * Reclaim the highest runs in LIFO (descending) order until
2685 		 * the number of reclaimed pages, "reclaimed", is at least
2686 		 * MIN_RECLAIM.  Reset "reclaimed" each time because each
2687 		 * reclamation is idempotent, and runs will (likely) recur
2688 		 * from one scan to the next as restrictions are relaxed.
2689 		 */
2690 		reclaimed = 0;
2691 		for (i = 0; count > 0 && i < NRUNS; i++) {
2692 			count--;
2693 			m_run = m_runs[RUN_INDEX(count)];
2694 			error = vm_page_reclaim_run(req_class, npages, m_run,
2695 			    high);
2696 			if (error == 0) {
2697 				reclaimed += npages;
2698 				if (reclaimed >= MIN_RECLAIM)
2699 					return (true);
2700 			}
2701 		}
2702 
2703 		/*
2704 		 * Either relax the restrictions on the next scan or return if
2705 		 * the last scan had no restrictions.
2706 		 */
2707 		if (options == VPSC_NORESERV)
2708 			options = VPSC_NOSUPER;
2709 		else if (options == VPSC_NOSUPER)
2710 			options = VPSC_ANY;
2711 		else if (options == VPSC_ANY)
2712 			return (reclaimed != 0);
2713 	}
2714 }
2715 
2716 /*
2717  *	vm_wait:	(also see VM_WAIT macro)
2718  *
2719  *	Sleep until free pages are available for allocation.
2720  *	- Called in various places before memory allocations.
2721  */
2722 void
2723 vm_wait(void)
2724 {
2725 
2726 	mtx_lock(&vm_page_queue_free_mtx);
2727 	if (curproc == pageproc) {
2728 		vm_pageout_pages_needed = 1;
2729 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
2730 		    PDROP | PSWP, "VMWait", 0);
2731 	} else {
2732 		if (!vm_pageout_wanted) {
2733 			vm_pageout_wanted = true;
2734 			wakeup(&vm_pageout_wanted);
2735 		}
2736 		vm_pages_needed = true;
2737 		msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
2738 		    "vmwait", 0);
2739 	}
2740 }
2741 
2742 /*
2743  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
2744  *
2745  *	Sleep until free pages are available for allocation.
2746  *	- Called only in vm_fault so that processes page faulting
2747  *	  can be easily tracked.
2748  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2749  *	  processes will be able to grab memory first.  Do not change
2750  *	  this balance without careful testing first.
2751  */
2752 void
2753 vm_waitpfault(void)
2754 {
2755 
2756 	mtx_lock(&vm_page_queue_free_mtx);
2757 	if (!vm_pageout_wanted) {
2758 		vm_pageout_wanted = true;
2759 		wakeup(&vm_pageout_wanted);
2760 	}
2761 	vm_pages_needed = true;
2762 	msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
2763 	    "pfault", 0);
2764 }
2765 
2766 struct vm_pagequeue *
2767 vm_page_pagequeue(vm_page_t m)
2768 {
2769 
2770 	return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2771 }
2772 
2773 /*
2774  *	vm_page_dequeue:
2775  *
2776  *	Remove the given page from its current page queue.
2777  *
2778  *	The page must be locked.
2779  */
2780 void
2781 vm_page_dequeue(vm_page_t m)
2782 {
2783 	struct vm_pagequeue *pq;
2784 
2785 	vm_page_assert_locked(m);
2786 	KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2787 	    m));
2788 	pq = vm_page_pagequeue(m);
2789 	vm_pagequeue_lock(pq);
2790 	m->queue = PQ_NONE;
2791 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2792 	vm_pagequeue_cnt_dec(pq);
2793 	vm_pagequeue_unlock(pq);
2794 }
2795 
2796 /*
2797  *	vm_page_dequeue_locked:
2798  *
2799  *	Remove the given page from its current page queue.
2800  *
2801  *	The page and page queue must be locked.
2802  */
2803 void
2804 vm_page_dequeue_locked(vm_page_t m)
2805 {
2806 	struct vm_pagequeue *pq;
2807 
2808 	vm_page_lock_assert(m, MA_OWNED);
2809 	pq = vm_page_pagequeue(m);
2810 	vm_pagequeue_assert_locked(pq);
2811 	m->queue = PQ_NONE;
2812 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2813 	vm_pagequeue_cnt_dec(pq);
2814 }
2815 
2816 /*
2817  *	vm_page_enqueue:
2818  *
2819  *	Add the given page to the specified page queue.
2820  *
2821  *	The page must be locked.
2822  */
2823 static void
2824 vm_page_enqueue(uint8_t queue, vm_page_t m)
2825 {
2826 	struct vm_pagequeue *pq;
2827 
2828 	vm_page_lock_assert(m, MA_OWNED);
2829 	KASSERT(queue < PQ_COUNT,
2830 	    ("vm_page_enqueue: invalid queue %u request for page %p",
2831 	    queue, m));
2832 	pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2833 	vm_pagequeue_lock(pq);
2834 	m->queue = queue;
2835 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2836 	vm_pagequeue_cnt_inc(pq);
2837 	vm_pagequeue_unlock(pq);
2838 }
2839 
2840 /*
2841  *	vm_page_requeue:
2842  *
2843  *	Move the given page to the tail of its current page queue.
2844  *
2845  *	The page must be locked.
2846  */
2847 void
2848 vm_page_requeue(vm_page_t m)
2849 {
2850 	struct vm_pagequeue *pq;
2851 
2852 	vm_page_lock_assert(m, MA_OWNED);
2853 	KASSERT(m->queue != PQ_NONE,
2854 	    ("vm_page_requeue: page %p is not queued", m));
2855 	pq = vm_page_pagequeue(m);
2856 	vm_pagequeue_lock(pq);
2857 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2858 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2859 	vm_pagequeue_unlock(pq);
2860 }
2861 
2862 /*
2863  *	vm_page_requeue_locked:
2864  *
2865  *	Move the given page to the tail of its current page queue.
2866  *
2867  *	The page queue must be locked.
2868  */
2869 void
2870 vm_page_requeue_locked(vm_page_t m)
2871 {
2872 	struct vm_pagequeue *pq;
2873 
2874 	KASSERT(m->queue != PQ_NONE,
2875 	    ("vm_page_requeue_locked: page %p is not queued", m));
2876 	pq = vm_page_pagequeue(m);
2877 	vm_pagequeue_assert_locked(pq);
2878 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2879 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2880 }
2881 
2882 /*
2883  *	vm_page_activate:
2884  *
2885  *	Put the specified page on the active list (if appropriate).
2886  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2887  *	mess with it.
2888  *
2889  *	The page must be locked.
2890  */
2891 void
2892 vm_page_activate(vm_page_t m)
2893 {
2894 	int queue;
2895 
2896 	vm_page_lock_assert(m, MA_OWNED);
2897 	if ((queue = m->queue) != PQ_ACTIVE) {
2898 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2899 			if (m->act_count < ACT_INIT)
2900 				m->act_count = ACT_INIT;
2901 			if (queue != PQ_NONE)
2902 				vm_page_dequeue(m);
2903 			vm_page_enqueue(PQ_ACTIVE, m);
2904 		} else
2905 			KASSERT(queue == PQ_NONE,
2906 			    ("vm_page_activate: wired page %p is queued", m));
2907 	} else {
2908 		if (m->act_count < ACT_INIT)
2909 			m->act_count = ACT_INIT;
2910 	}
2911 }
2912 
2913 /*
2914  *	vm_page_free_wakeup:
2915  *
2916  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
2917  *	routine is called when a page has been added to the cache or free
2918  *	queues.
2919  *
2920  *	The page queues must be locked.
2921  */
2922 static inline void
2923 vm_page_free_wakeup(void)
2924 {
2925 
2926 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2927 	/*
2928 	 * if pageout daemon needs pages, then tell it that there are
2929 	 * some free.
2930 	 */
2931 	if (vm_pageout_pages_needed &&
2932 	    vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) {
2933 		wakeup(&vm_pageout_pages_needed);
2934 		vm_pageout_pages_needed = 0;
2935 	}
2936 	/*
2937 	 * wakeup processes that are waiting on memory if we hit a
2938 	 * high water mark. And wakeup scheduler process if we have
2939 	 * lots of memory. this process will swapin processes.
2940 	 */
2941 	if (vm_pages_needed && !vm_page_count_min()) {
2942 		vm_pages_needed = false;
2943 		wakeup(&vm_cnt.v_free_count);
2944 	}
2945 }
2946 
2947 /*
2948  *	Turn a cached page into a free page, by changing its attributes.
2949  *	Keep the statistics up-to-date.
2950  *
2951  *	The free page queue must be locked.
2952  */
2953 static void
2954 vm_page_cache_turn_free(vm_page_t m)
2955 {
2956 
2957 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2958 
2959 	m->object = NULL;
2960 	m->valid = 0;
2961 	KASSERT((m->flags & PG_CACHED) != 0,
2962 	    ("vm_page_cache_turn_free: page %p is not cached", m));
2963 	m->flags &= ~PG_CACHED;
2964 	vm_cnt.v_cache_count--;
2965 	vm_phys_freecnt_adj(m, 1);
2966 }
2967 
2968 /*
2969  *	vm_page_free_toq:
2970  *
2971  *	Returns the given page to the free list,
2972  *	disassociating it with any VM object.
2973  *
2974  *	The object must be locked.  The page must be locked if it is managed.
2975  */
2976 void
2977 vm_page_free_toq(vm_page_t m)
2978 {
2979 
2980 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2981 		vm_page_lock_assert(m, MA_OWNED);
2982 		KASSERT(!pmap_page_is_mapped(m),
2983 		    ("vm_page_free_toq: freeing mapped page %p", m));
2984 	} else
2985 		KASSERT(m->queue == PQ_NONE,
2986 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2987 	PCPU_INC(cnt.v_tfree);
2988 
2989 	if (vm_page_sbusied(m))
2990 		panic("vm_page_free: freeing busy page %p", m);
2991 
2992 	/*
2993 	 * Unqueue, then remove page.  Note that we cannot destroy
2994 	 * the page here because we do not want to call the pager's
2995 	 * callback routine until after we've put the page on the
2996 	 * appropriate free queue.
2997 	 */
2998 	vm_page_remque(m);
2999 	vm_page_remove(m);
3000 
3001 	/*
3002 	 * If fictitious remove object association and
3003 	 * return, otherwise delay object association removal.
3004 	 */
3005 	if ((m->flags & PG_FICTITIOUS) != 0) {
3006 		return;
3007 	}
3008 
3009 	m->valid = 0;
3010 	vm_page_undirty(m);
3011 
3012 	if (m->wire_count != 0)
3013 		panic("vm_page_free: freeing wired page %p", m);
3014 	if (m->hold_count != 0) {
3015 		m->flags &= ~PG_ZERO;
3016 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
3017 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
3018 		m->flags |= PG_UNHOLDFREE;
3019 	} else {
3020 		/*
3021 		 * Restore the default memory attribute to the page.
3022 		 */
3023 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
3024 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
3025 
3026 		/*
3027 		 * Insert the page into the physical memory allocator's
3028 		 * cache/free page queues.
3029 		 */
3030 		mtx_lock(&vm_page_queue_free_mtx);
3031 		vm_phys_freecnt_adj(m, 1);
3032 #if VM_NRESERVLEVEL > 0
3033 		if (!vm_reserv_free_page(m))
3034 #else
3035 		if (TRUE)
3036 #endif
3037 			vm_phys_free_pages(m, 0);
3038 		vm_page_free_wakeup();
3039 		mtx_unlock(&vm_page_queue_free_mtx);
3040 	}
3041 }
3042 
3043 /*
3044  *	vm_page_wire:
3045  *
3046  *	Mark this page as wired down by yet
3047  *	another map, removing it from paging queues
3048  *	as necessary.
3049  *
3050  *	If the page is fictitious, then its wire count must remain one.
3051  *
3052  *	The page must be locked.
3053  */
3054 void
3055 vm_page_wire(vm_page_t m)
3056 {
3057 
3058 	/*
3059 	 * Only bump the wire statistics if the page is not already wired,
3060 	 * and only unqueue the page if it is on some queue (if it is unmanaged
3061 	 * it is already off the queues).
3062 	 */
3063 	vm_page_lock_assert(m, MA_OWNED);
3064 	if ((m->flags & PG_FICTITIOUS) != 0) {
3065 		KASSERT(m->wire_count == 1,
3066 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
3067 		    m));
3068 		return;
3069 	}
3070 	if (m->wire_count == 0) {
3071 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
3072 		    m->queue == PQ_NONE,
3073 		    ("vm_page_wire: unmanaged page %p is queued", m));
3074 		vm_page_remque(m);
3075 		atomic_add_int(&vm_cnt.v_wire_count, 1);
3076 	}
3077 	m->wire_count++;
3078 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
3079 }
3080 
3081 /*
3082  * vm_page_unwire:
3083  *
3084  * Release one wiring of the specified page, potentially allowing it to be
3085  * paged out.  Returns TRUE if the number of wirings transitions to zero and
3086  * FALSE otherwise.
3087  *
3088  * Only managed pages belonging to an object can be paged out.  If the number
3089  * of wirings transitions to zero and the page is eligible for page out, then
3090  * the page is added to the specified paging queue (unless PQ_NONE is
3091  * specified).
3092  *
3093  * If a page is fictitious, then its wire count must always be one.
3094  *
3095  * A managed page must be locked.
3096  */
3097 boolean_t
3098 vm_page_unwire(vm_page_t m, uint8_t queue)
3099 {
3100 
3101 	KASSERT(queue < PQ_COUNT || queue == PQ_NONE,
3102 	    ("vm_page_unwire: invalid queue %u request for page %p",
3103 	    queue, m));
3104 	if ((m->oflags & VPO_UNMANAGED) == 0)
3105 		vm_page_assert_locked(m);
3106 	if ((m->flags & PG_FICTITIOUS) != 0) {
3107 		KASSERT(m->wire_count == 1,
3108 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
3109 		return (FALSE);
3110 	}
3111 	if (m->wire_count > 0) {
3112 		m->wire_count--;
3113 		if (m->wire_count == 0) {
3114 			atomic_subtract_int(&vm_cnt.v_wire_count, 1);
3115 			if ((m->oflags & VPO_UNMANAGED) == 0 &&
3116 			    m->object != NULL && queue != PQ_NONE) {
3117 				if (queue == PQ_INACTIVE)
3118 					m->flags &= ~PG_WINATCFLS;
3119 				vm_page_enqueue(queue, m);
3120 			}
3121 			return (TRUE);
3122 		} else
3123 			return (FALSE);
3124 	} else
3125 		panic("vm_page_unwire: page %p's wire count is zero", m);
3126 }
3127 
3128 /*
3129  * Move the specified page to the inactive queue.
3130  *
3131  * Many pages placed on the inactive queue should actually go
3132  * into the cache, but it is difficult to figure out which.  What
3133  * we do instead, if the inactive target is well met, is to put
3134  * clean pages at the head of the inactive queue instead of the tail.
3135  * This will cause them to be moved to the cache more quickly and
3136  * if not actively re-referenced, reclaimed more quickly.  If we just
3137  * stick these pages at the end of the inactive queue, heavy filesystem
3138  * meta-data accesses can cause an unnecessary paging load on memory bound
3139  * processes.  This optimization causes one-time-use metadata to be
3140  * reused more quickly.
3141  *
3142  * Normally noreuse is FALSE, resulting in LRU operation.  noreuse is set
3143  * to TRUE if we want this page to be 'as if it were placed in the cache',
3144  * except without unmapping it from the process address space.  In
3145  * practice this is implemented by inserting the page at the head of the
3146  * queue, using a marker page to guide FIFO insertion ordering.
3147  *
3148  * The page must be locked.
3149  */
3150 static inline void
3151 _vm_page_deactivate(vm_page_t m, boolean_t noreuse)
3152 {
3153 	struct vm_pagequeue *pq;
3154 	int queue;
3155 
3156 	vm_page_assert_locked(m);
3157 
3158 	/*
3159 	 * Ignore if the page is already inactive, unless it is unlikely to be
3160 	 * reactivated.
3161 	 */
3162 	if ((queue = m->queue) == PQ_INACTIVE && !noreuse)
3163 		return;
3164 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
3165 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
3166 		/* Avoid multiple acquisitions of the inactive queue lock. */
3167 		if (queue == PQ_INACTIVE) {
3168 			vm_pagequeue_lock(pq);
3169 			vm_page_dequeue_locked(m);
3170 		} else {
3171 			if (queue != PQ_NONE)
3172 				vm_page_dequeue(m);
3173 			m->flags &= ~PG_WINATCFLS;
3174 			vm_pagequeue_lock(pq);
3175 		}
3176 		m->queue = PQ_INACTIVE;
3177 		if (noreuse)
3178 			TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead,
3179 			    m, plinks.q);
3180 		else
3181 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3182 		vm_pagequeue_cnt_inc(pq);
3183 		vm_pagequeue_unlock(pq);
3184 	}
3185 }
3186 
3187 /*
3188  * Move the specified page to the inactive queue.
3189  *
3190  * The page must be locked.
3191  */
3192 void
3193 vm_page_deactivate(vm_page_t m)
3194 {
3195 
3196 	_vm_page_deactivate(m, FALSE);
3197 }
3198 
3199 /*
3200  * Move the specified page to the inactive queue with the expectation
3201  * that it is unlikely to be reused.
3202  *
3203  * The page must be locked.
3204  */
3205 void
3206 vm_page_deactivate_noreuse(vm_page_t m)
3207 {
3208 
3209 	_vm_page_deactivate(m, TRUE);
3210 }
3211 
3212 /*
3213  * vm_page_try_to_cache:
3214  *
3215  * Returns 0 on failure, 1 on success
3216  */
3217 int
3218 vm_page_try_to_cache(vm_page_t m)
3219 {
3220 
3221 	vm_page_lock_assert(m, MA_OWNED);
3222 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3223 	if (m->dirty || m->hold_count || m->wire_count ||
3224 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
3225 		return (0);
3226 	pmap_remove_all(m);
3227 	if (m->dirty)
3228 		return (0);
3229 	vm_page_cache(m);
3230 	return (1);
3231 }
3232 
3233 /*
3234  * vm_page_try_to_free()
3235  *
3236  *	Attempt to free the page.  If we cannot free it, we do nothing.
3237  *	1 is returned on success, 0 on failure.
3238  */
3239 int
3240 vm_page_try_to_free(vm_page_t m)
3241 {
3242 
3243 	vm_page_lock_assert(m, MA_OWNED);
3244 	if (m->object != NULL)
3245 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3246 	if (m->dirty || m->hold_count || m->wire_count ||
3247 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
3248 		return (0);
3249 	pmap_remove_all(m);
3250 	if (m->dirty)
3251 		return (0);
3252 	vm_page_free(m);
3253 	return (1);
3254 }
3255 
3256 /*
3257  * vm_page_cache
3258  *
3259  * Put the specified page onto the page cache queue (if appropriate).
3260  *
3261  * The object and page must be locked.
3262  */
3263 void
3264 vm_page_cache(vm_page_t m)
3265 {
3266 	vm_object_t object;
3267 	boolean_t cache_was_empty;
3268 
3269 	vm_page_lock_assert(m, MA_OWNED);
3270 	object = m->object;
3271 	VM_OBJECT_ASSERT_WLOCKED(object);
3272 	if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) ||
3273 	    m->hold_count || m->wire_count)
3274 		panic("vm_page_cache: attempting to cache busy page");
3275 	KASSERT(!pmap_page_is_mapped(m),
3276 	    ("vm_page_cache: page %p is mapped", m));
3277 	KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m));
3278 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
3279 	    (object->type == OBJT_SWAP &&
3280 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
3281 		/*
3282 		 * Hypothesis: A cache-eligible page belonging to a
3283 		 * default object or swap object but without a backing
3284 		 * store must be zero filled.
3285 		 */
3286 		vm_page_free(m);
3287 		return;
3288 	}
3289 	KASSERT((m->flags & PG_CACHED) == 0,
3290 	    ("vm_page_cache: page %p is already cached", m));
3291 
3292 	/*
3293 	 * Remove the page from the paging queues.
3294 	 */
3295 	vm_page_remque(m);
3296 
3297 	/*
3298 	 * Remove the page from the object's collection of resident
3299 	 * pages.
3300 	 */
3301 	vm_radix_remove(&object->rtree, m->pindex);
3302 	TAILQ_REMOVE(&object->memq, m, listq);
3303 	object->resident_page_count--;
3304 
3305 	/*
3306 	 * Restore the default memory attribute to the page.
3307 	 */
3308 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
3309 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
3310 
3311 	/*
3312 	 * Insert the page into the object's collection of cached pages
3313 	 * and the physical memory allocator's cache/free page queues.
3314 	 */
3315 	m->flags &= ~PG_ZERO;
3316 	mtx_lock(&vm_page_queue_free_mtx);
3317 	cache_was_empty = vm_radix_is_empty(&object->cache);
3318 	if (vm_radix_insert(&object->cache, m)) {
3319 		mtx_unlock(&vm_page_queue_free_mtx);
3320 		if (object->type == OBJT_VNODE &&
3321 		    object->resident_page_count == 0)
3322 			vdrop(object->handle);
3323 		m->object = NULL;
3324 		vm_page_free(m);
3325 		return;
3326 	}
3327 
3328 	/*
3329 	 * The above call to vm_radix_insert() could reclaim the one pre-
3330 	 * existing cached page from this object, resulting in a call to
3331 	 * vdrop().
3332 	 */
3333 	if (!cache_was_empty)
3334 		cache_was_empty = vm_radix_is_singleton(&object->cache);
3335 
3336 	m->flags |= PG_CACHED;
3337 	vm_cnt.v_cache_count++;
3338 	PCPU_INC(cnt.v_tcached);
3339 #if VM_NRESERVLEVEL > 0
3340 	if (!vm_reserv_free_page(m)) {
3341 #else
3342 	if (TRUE) {
3343 #endif
3344 		vm_phys_free_pages(m, 0);
3345 	}
3346 	vm_page_free_wakeup();
3347 	mtx_unlock(&vm_page_queue_free_mtx);
3348 
3349 	/*
3350 	 * Increment the vnode's hold count if this is the object's only
3351 	 * cached page.  Decrement the vnode's hold count if this was
3352 	 * the object's only resident page.
3353 	 */
3354 	if (object->type == OBJT_VNODE) {
3355 		if (cache_was_empty && object->resident_page_count != 0)
3356 			vhold(object->handle);
3357 		else if (!cache_was_empty && object->resident_page_count == 0)
3358 			vdrop(object->handle);
3359 	}
3360 }
3361 
3362 /*
3363  * vm_page_advise
3364  *
3365  * 	Deactivate or do nothing, as appropriate.
3366  *
3367  *	The object and page must be locked.
3368  */
3369 void
3370 vm_page_advise(vm_page_t m, int advice)
3371 {
3372 
3373 	vm_page_assert_locked(m);
3374 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3375 	if (advice == MADV_FREE)
3376 		/*
3377 		 * Mark the page clean.  This will allow the page to be freed
3378 		 * up by the system.  However, such pages are often reused
3379 		 * quickly by malloc() so we do not do anything that would
3380 		 * cause a page fault if we can help it.
3381 		 *
3382 		 * Specifically, we do not try to actually free the page now
3383 		 * nor do we try to put it in the cache (which would cause a
3384 		 * page fault on reuse).
3385 		 *
3386 		 * But we do make the page as freeable as we can without
3387 		 * actually taking the step of unmapping it.
3388 		 */
3389 		vm_page_undirty(m);
3390 	else if (advice != MADV_DONTNEED)
3391 		return;
3392 
3393 	/*
3394 	 * Clear any references to the page.  Otherwise, the page daemon will
3395 	 * immediately reactivate the page.
3396 	 */
3397 	vm_page_aflag_clear(m, PGA_REFERENCED);
3398 
3399 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
3400 		vm_page_dirty(m);
3401 
3402 	/*
3403 	 * Place clean pages near the head of the inactive queue rather than
3404 	 * the tail, thus defeating the queue's LRU operation and ensuring that
3405 	 * the page will be reused quickly.  Dirty pages are given a chance to
3406 	 * cycle once through the inactive queue before becoming eligible for
3407 	 * laundering.
3408 	 */
3409 	_vm_page_deactivate(m, m->dirty == 0);
3410 }
3411 
3412 /*
3413  * Grab a page, waiting until we are waken up due to the page
3414  * changing state.  We keep on waiting, if the page continues
3415  * to be in the object.  If the page doesn't exist, first allocate it
3416  * and then conditionally zero it.
3417  *
3418  * This routine may sleep.
3419  *
3420  * The object must be locked on entry.  The lock will, however, be released
3421  * and reacquired if the routine sleeps.
3422  */
3423 vm_page_t
3424 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
3425 {
3426 	vm_page_t m;
3427 	int sleep;
3428 
3429 	VM_OBJECT_ASSERT_WLOCKED(object);
3430 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3431 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3432 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
3433 retrylookup:
3434 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
3435 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3436 		    vm_page_xbusied(m) : vm_page_busied(m);
3437 		if (sleep) {
3438 			if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3439 				return (NULL);
3440 			/*
3441 			 * Reference the page before unlocking and
3442 			 * sleeping so that the page daemon is less
3443 			 * likely to reclaim it.
3444 			 */
3445 			vm_page_aflag_set(m, PGA_REFERENCED);
3446 			vm_page_lock(m);
3447 			VM_OBJECT_WUNLOCK(object);
3448 			vm_page_busy_sleep(m, "pgrbwt");
3449 			VM_OBJECT_WLOCK(object);
3450 			goto retrylookup;
3451 		} else {
3452 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
3453 				vm_page_lock(m);
3454 				vm_page_wire(m);
3455 				vm_page_unlock(m);
3456 			}
3457 			if ((allocflags &
3458 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
3459 				vm_page_xbusy(m);
3460 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
3461 				vm_page_sbusy(m);
3462 			return (m);
3463 		}
3464 	}
3465 	m = vm_page_alloc(object, pindex, allocflags);
3466 	if (m == NULL) {
3467 		if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3468 			return (NULL);
3469 		VM_OBJECT_WUNLOCK(object);
3470 		VM_WAIT;
3471 		VM_OBJECT_WLOCK(object);
3472 		goto retrylookup;
3473 	} else if (m->valid != 0)
3474 		return (m);
3475 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
3476 		pmap_zero_page(m);
3477 	return (m);
3478 }
3479 
3480 /*
3481  * Mapping function for valid or dirty bits in a page.
3482  *
3483  * Inputs are required to range within a page.
3484  */
3485 vm_page_bits_t
3486 vm_page_bits(int base, int size)
3487 {
3488 	int first_bit;
3489 	int last_bit;
3490 
3491 	KASSERT(
3492 	    base + size <= PAGE_SIZE,
3493 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
3494 	);
3495 
3496 	if (size == 0)		/* handle degenerate case */
3497 		return (0);
3498 
3499 	first_bit = base >> DEV_BSHIFT;
3500 	last_bit = (base + size - 1) >> DEV_BSHIFT;
3501 
3502 	return (((vm_page_bits_t)2 << last_bit) -
3503 	    ((vm_page_bits_t)1 << first_bit));
3504 }
3505 
3506 /*
3507  *	vm_page_set_valid_range:
3508  *
3509  *	Sets portions of a page valid.  The arguments are expected
3510  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3511  *	of any partial chunks touched by the range.  The invalid portion of
3512  *	such chunks will be zeroed.
3513  *
3514  *	(base + size) must be less then or equal to PAGE_SIZE.
3515  */
3516 void
3517 vm_page_set_valid_range(vm_page_t m, int base, int size)
3518 {
3519 	int endoff, frag;
3520 
3521 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3522 	if (size == 0)	/* handle degenerate case */
3523 		return;
3524 
3525 	/*
3526 	 * If the base is not DEV_BSIZE aligned and the valid
3527 	 * bit is clear, we have to zero out a portion of the
3528 	 * first block.
3529 	 */
3530 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3531 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
3532 		pmap_zero_page_area(m, frag, base - frag);
3533 
3534 	/*
3535 	 * If the ending offset is not DEV_BSIZE aligned and the
3536 	 * valid bit is clear, we have to zero out a portion of
3537 	 * the last block.
3538 	 */
3539 	endoff = base + size;
3540 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3541 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
3542 		pmap_zero_page_area(m, endoff,
3543 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3544 
3545 	/*
3546 	 * Assert that no previously invalid block that is now being validated
3547 	 * is already dirty.
3548 	 */
3549 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
3550 	    ("vm_page_set_valid_range: page %p is dirty", m));
3551 
3552 	/*
3553 	 * Set valid bits inclusive of any overlap.
3554 	 */
3555 	m->valid |= vm_page_bits(base, size);
3556 }
3557 
3558 /*
3559  * Clear the given bits from the specified page's dirty field.
3560  */
3561 static __inline void
3562 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
3563 {
3564 	uintptr_t addr;
3565 #if PAGE_SIZE < 16384
3566 	int shift;
3567 #endif
3568 
3569 	/*
3570 	 * If the object is locked and the page is neither exclusive busy nor
3571 	 * write mapped, then the page's dirty field cannot possibly be
3572 	 * set by a concurrent pmap operation.
3573 	 */
3574 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3575 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
3576 		m->dirty &= ~pagebits;
3577 	else {
3578 		/*
3579 		 * The pmap layer can call vm_page_dirty() without
3580 		 * holding a distinguished lock.  The combination of
3581 		 * the object's lock and an atomic operation suffice
3582 		 * to guarantee consistency of the page dirty field.
3583 		 *
3584 		 * For PAGE_SIZE == 32768 case, compiler already
3585 		 * properly aligns the dirty field, so no forcible
3586 		 * alignment is needed. Only require existence of
3587 		 * atomic_clear_64 when page size is 32768.
3588 		 */
3589 		addr = (uintptr_t)&m->dirty;
3590 #if PAGE_SIZE == 32768
3591 		atomic_clear_64((uint64_t *)addr, pagebits);
3592 #elif PAGE_SIZE == 16384
3593 		atomic_clear_32((uint32_t *)addr, pagebits);
3594 #else		/* PAGE_SIZE <= 8192 */
3595 		/*
3596 		 * Use a trick to perform a 32-bit atomic on the
3597 		 * containing aligned word, to not depend on the existence
3598 		 * of atomic_clear_{8, 16}.
3599 		 */
3600 		shift = addr & (sizeof(uint32_t) - 1);
3601 #if BYTE_ORDER == BIG_ENDIAN
3602 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
3603 #else
3604 		shift *= NBBY;
3605 #endif
3606 		addr &= ~(sizeof(uint32_t) - 1);
3607 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
3608 #endif		/* PAGE_SIZE */
3609 	}
3610 }
3611 
3612 /*
3613  *	vm_page_set_validclean:
3614  *
3615  *	Sets portions of a page valid and clean.  The arguments are expected
3616  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3617  *	of any partial chunks touched by the range.  The invalid portion of
3618  *	such chunks will be zero'd.
3619  *
3620  *	(base + size) must be less then or equal to PAGE_SIZE.
3621  */
3622 void
3623 vm_page_set_validclean(vm_page_t m, int base, int size)
3624 {
3625 	vm_page_bits_t oldvalid, pagebits;
3626 	int endoff, frag;
3627 
3628 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3629 	if (size == 0)	/* handle degenerate case */
3630 		return;
3631 
3632 	/*
3633 	 * If the base is not DEV_BSIZE aligned and the valid
3634 	 * bit is clear, we have to zero out a portion of the
3635 	 * first block.
3636 	 */
3637 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3638 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3639 		pmap_zero_page_area(m, frag, base - frag);
3640 
3641 	/*
3642 	 * If the ending offset is not DEV_BSIZE aligned and the
3643 	 * valid bit is clear, we have to zero out a portion of
3644 	 * the last block.
3645 	 */
3646 	endoff = base + size;
3647 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3648 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3649 		pmap_zero_page_area(m, endoff,
3650 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3651 
3652 	/*
3653 	 * Set valid, clear dirty bits.  If validating the entire
3654 	 * page we can safely clear the pmap modify bit.  We also
3655 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
3656 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
3657 	 * be set again.
3658 	 *
3659 	 * We set valid bits inclusive of any overlap, but we can only
3660 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
3661 	 * the range.
3662 	 */
3663 	oldvalid = m->valid;
3664 	pagebits = vm_page_bits(base, size);
3665 	m->valid |= pagebits;
3666 #if 0	/* NOT YET */
3667 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3668 		frag = DEV_BSIZE - frag;
3669 		base += frag;
3670 		size -= frag;
3671 		if (size < 0)
3672 			size = 0;
3673 	}
3674 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3675 #endif
3676 	if (base == 0 && size == PAGE_SIZE) {
3677 		/*
3678 		 * The page can only be modified within the pmap if it is
3679 		 * mapped, and it can only be mapped if it was previously
3680 		 * fully valid.
3681 		 */
3682 		if (oldvalid == VM_PAGE_BITS_ALL)
3683 			/*
3684 			 * Perform the pmap_clear_modify() first.  Otherwise,
3685 			 * a concurrent pmap operation, such as
3686 			 * pmap_protect(), could clear a modification in the
3687 			 * pmap and set the dirty field on the page before
3688 			 * pmap_clear_modify() had begun and after the dirty
3689 			 * field was cleared here.
3690 			 */
3691 			pmap_clear_modify(m);
3692 		m->dirty = 0;
3693 		m->oflags &= ~VPO_NOSYNC;
3694 	} else if (oldvalid != VM_PAGE_BITS_ALL)
3695 		m->dirty &= ~pagebits;
3696 	else
3697 		vm_page_clear_dirty_mask(m, pagebits);
3698 }
3699 
3700 void
3701 vm_page_clear_dirty(vm_page_t m, int base, int size)
3702 {
3703 
3704 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3705 }
3706 
3707 /*
3708  *	vm_page_set_invalid:
3709  *
3710  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
3711  *	valid and dirty bits for the effected areas are cleared.
3712  */
3713 void
3714 vm_page_set_invalid(vm_page_t m, int base, int size)
3715 {
3716 	vm_page_bits_t bits;
3717 	vm_object_t object;
3718 
3719 	object = m->object;
3720 	VM_OBJECT_ASSERT_WLOCKED(object);
3721 	if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3722 	    size >= object->un_pager.vnp.vnp_size)
3723 		bits = VM_PAGE_BITS_ALL;
3724 	else
3725 		bits = vm_page_bits(base, size);
3726 	if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
3727 	    bits != 0)
3728 		pmap_remove_all(m);
3729 	KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3730 	    !pmap_page_is_mapped(m),
3731 	    ("vm_page_set_invalid: page %p is mapped", m));
3732 	m->valid &= ~bits;
3733 	m->dirty &= ~bits;
3734 }
3735 
3736 /*
3737  * vm_page_zero_invalid()
3738  *
3739  *	The kernel assumes that the invalid portions of a page contain
3740  *	garbage, but such pages can be mapped into memory by user code.
3741  *	When this occurs, we must zero out the non-valid portions of the
3742  *	page so user code sees what it expects.
3743  *
3744  *	Pages are most often semi-valid when the end of a file is mapped
3745  *	into memory and the file's size is not page aligned.
3746  */
3747 void
3748 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3749 {
3750 	int b;
3751 	int i;
3752 
3753 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3754 	/*
3755 	 * Scan the valid bits looking for invalid sections that
3756 	 * must be zeroed.  Invalid sub-DEV_BSIZE'd areas ( where the
3757 	 * valid bit may be set ) have already been zeroed by
3758 	 * vm_page_set_validclean().
3759 	 */
3760 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3761 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3762 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3763 			if (i > b) {
3764 				pmap_zero_page_area(m,
3765 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3766 			}
3767 			b = i + 1;
3768 		}
3769 	}
3770 
3771 	/*
3772 	 * setvalid is TRUE when we can safely set the zero'd areas
3773 	 * as being valid.  We can do this if there are no cache consistancy
3774 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3775 	 */
3776 	if (setvalid)
3777 		m->valid = VM_PAGE_BITS_ALL;
3778 }
3779 
3780 /*
3781  *	vm_page_is_valid:
3782  *
3783  *	Is (partial) page valid?  Note that the case where size == 0
3784  *	will return FALSE in the degenerate case where the page is
3785  *	entirely invalid, and TRUE otherwise.
3786  */
3787 int
3788 vm_page_is_valid(vm_page_t m, int base, int size)
3789 {
3790 	vm_page_bits_t bits;
3791 
3792 	VM_OBJECT_ASSERT_LOCKED(m->object);
3793 	bits = vm_page_bits(base, size);
3794 	return (m->valid != 0 && (m->valid & bits) == bits);
3795 }
3796 
3797 /*
3798  *	vm_page_ps_is_valid:
3799  *
3800  *	Returns TRUE if the entire (super)page is valid and FALSE otherwise.
3801  */
3802 boolean_t
3803 vm_page_ps_is_valid(vm_page_t m)
3804 {
3805 	int i, npages;
3806 
3807 	VM_OBJECT_ASSERT_LOCKED(m->object);
3808 	npages = atop(pagesizes[m->psind]);
3809 
3810 	/*
3811 	 * The physically contiguous pages that make up a superpage, i.e., a
3812 	 * page with a page size index ("psind") greater than zero, will
3813 	 * occupy adjacent entries in vm_page_array[].
3814 	 */
3815 	for (i = 0; i < npages; i++) {
3816 		if (m[i].valid != VM_PAGE_BITS_ALL)
3817 			return (FALSE);
3818 	}
3819 	return (TRUE);
3820 }
3821 
3822 /*
3823  * Set the page's dirty bits if the page is modified.
3824  */
3825 void
3826 vm_page_test_dirty(vm_page_t m)
3827 {
3828 
3829 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3830 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3831 		vm_page_dirty(m);
3832 }
3833 
3834 void
3835 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3836 {
3837 
3838 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3839 }
3840 
3841 void
3842 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3843 {
3844 
3845 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3846 }
3847 
3848 int
3849 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3850 {
3851 
3852 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3853 }
3854 
3855 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3856 void
3857 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3858 {
3859 
3860 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3861 }
3862 
3863 void
3864 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3865 {
3866 
3867 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3868 }
3869 #endif
3870 
3871 #ifdef INVARIANTS
3872 void
3873 vm_page_object_lock_assert(vm_page_t m)
3874 {
3875 
3876 	/*
3877 	 * Certain of the page's fields may only be modified by the
3878 	 * holder of the containing object's lock or the exclusive busy.
3879 	 * holder.  Unfortunately, the holder of the write busy is
3880 	 * not recorded, and thus cannot be checked here.
3881 	 */
3882 	if (m->object != NULL && !vm_page_xbusied(m))
3883 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3884 }
3885 
3886 void
3887 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
3888 {
3889 
3890 	if ((bits & PGA_WRITEABLE) == 0)
3891 		return;
3892 
3893 	/*
3894 	 * The PGA_WRITEABLE flag can only be set if the page is
3895 	 * managed, is exclusively busied or the object is locked.
3896 	 * Currently, this flag is only set by pmap_enter().
3897 	 */
3898 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3899 	    ("PGA_WRITEABLE on unmanaged page"));
3900 	if (!vm_page_xbusied(m))
3901 		VM_OBJECT_ASSERT_LOCKED(m->object);
3902 }
3903 #endif
3904 
3905 #include "opt_ddb.h"
3906 #ifdef DDB
3907 #include <sys/kernel.h>
3908 
3909 #include <ddb/ddb.h>
3910 
3911 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3912 {
3913 	db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count);
3914 	db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count);
3915 	db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count);
3916 	db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count);
3917 	db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count);
3918 	db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
3919 	db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
3920 	db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
3921 	db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
3922 }
3923 
3924 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3925 {
3926 	int dom;
3927 
3928 	db_printf("pq_free %d pq_cache %d\n",
3929 	    vm_cnt.v_free_count, vm_cnt.v_cache_count);
3930 	for (dom = 0; dom < vm_ndomains; dom++) {
3931 		db_printf(
3932 	"dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n",
3933 		    dom,
3934 		    vm_dom[dom].vmd_page_count,
3935 		    vm_dom[dom].vmd_free_count,
3936 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3937 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3938 		    vm_dom[dom].vmd_pass);
3939 	}
3940 }
3941 
3942 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3943 {
3944 	vm_page_t m;
3945 	boolean_t phys;
3946 
3947 	if (!have_addr) {
3948 		db_printf("show pginfo addr\n");
3949 		return;
3950 	}
3951 
3952 	phys = strchr(modif, 'p') != NULL;
3953 	if (phys)
3954 		m = PHYS_TO_VM_PAGE(addr);
3955 	else
3956 		m = (vm_page_t)addr;
3957 	db_printf(
3958     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3959     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3960 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3961 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3962 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3963 }
3964 #endif /* DDB */
3965