xref: /freebsd/sys/vm/vm_page.c (revision 188e46ab0332a7887da6f1eb5494d92076e3d31e)
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  * 3. 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 /*
135  * bogus page -- for I/O to/from partially complete buffers,
136  * or for paging into sparsely invalid regions.
137  */
138 vm_page_t bogus_page;
139 
140 vm_page_t vm_page_array;
141 long vm_page_array_size;
142 long first_page;
143 
144 static int boot_pages = UMA_BOOT_PAGES;
145 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
146     &boot_pages, 0,
147     "number of pages allocated for bootstrapping the VM system");
148 
149 static int pa_tryrelock_restart;
150 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
151     &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
152 
153 static TAILQ_HEAD(, vm_page) blacklist_head;
154 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
155 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
156     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
157 
158 /* Is the page daemon waiting for free pages? */
159 static int vm_pageout_pages_needed;
160 
161 static uma_zone_t fakepg_zone;
162 
163 static void vm_page_alloc_check(vm_page_t m);
164 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
165 static void vm_page_enqueue(uint8_t queue, vm_page_t m);
166 static void vm_page_free_wakeup(void);
167 static void vm_page_init(void *dummy);
168 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
169     vm_pindex_t pindex, vm_page_t mpred);
170 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
171     vm_page_t mpred);
172 static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
173     vm_paddr_t high);
174 
175 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init, NULL);
176 
177 static void
178 vm_page_init(void *dummy)
179 {
180 
181 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
182 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
183 	bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ |
184 	    VM_ALLOC_NORMAL | VM_ALLOC_WIRED);
185 }
186 
187 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
188 #if PAGE_SIZE == 32768
189 #ifdef CTASSERT
190 CTASSERT(sizeof(u_long) >= 8);
191 #endif
192 #endif
193 
194 /*
195  * Try to acquire a physical address lock while a pmap is locked.  If we
196  * fail to trylock we unlock and lock the pmap directly and cache the
197  * locked pa in *locked.  The caller should then restart their loop in case
198  * the virtual to physical mapping has changed.
199  */
200 int
201 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
202 {
203 	vm_paddr_t lockpa;
204 
205 	lockpa = *locked;
206 	*locked = pa;
207 	if (lockpa) {
208 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
209 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
210 			return (0);
211 		PA_UNLOCK(lockpa);
212 	}
213 	if (PA_TRYLOCK(pa))
214 		return (0);
215 	PMAP_UNLOCK(pmap);
216 	atomic_add_int(&pa_tryrelock_restart, 1);
217 	PA_LOCK(pa);
218 	PMAP_LOCK(pmap);
219 	return (EAGAIN);
220 }
221 
222 /*
223  *	vm_set_page_size:
224  *
225  *	Sets the page size, perhaps based upon the memory
226  *	size.  Must be called before any use of page-size
227  *	dependent functions.
228  */
229 void
230 vm_set_page_size(void)
231 {
232 	if (vm_cnt.v_page_size == 0)
233 		vm_cnt.v_page_size = PAGE_SIZE;
234 	if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
235 		panic("vm_set_page_size: page size not a power of two");
236 }
237 
238 /*
239  *	vm_page_blacklist_next:
240  *
241  *	Find the next entry in the provided string of blacklist
242  *	addresses.  Entries are separated by space, comma, or newline.
243  *	If an invalid integer is encountered then the rest of the
244  *	string is skipped.  Updates the list pointer to the next
245  *	character, or NULL if the string is exhausted or invalid.
246  */
247 static vm_paddr_t
248 vm_page_blacklist_next(char **list, char *end)
249 {
250 	vm_paddr_t bad;
251 	char *cp, *pos;
252 
253 	if (list == NULL || *list == NULL)
254 		return (0);
255 	if (**list =='\0') {
256 		*list = NULL;
257 		return (0);
258 	}
259 
260 	/*
261 	 * If there's no end pointer then the buffer is coming from
262 	 * the kenv and we know it's null-terminated.
263 	 */
264 	if (end == NULL)
265 		end = *list + strlen(*list);
266 
267 	/* Ensure that strtoq() won't walk off the end */
268 	if (*end != '\0') {
269 		if (*end == '\n' || *end == ' ' || *end  == ',')
270 			*end = '\0';
271 		else {
272 			printf("Blacklist not terminated, skipping\n");
273 			*list = NULL;
274 			return (0);
275 		}
276 	}
277 
278 	for (pos = *list; *pos != '\0'; pos = cp) {
279 		bad = strtoq(pos, &cp, 0);
280 		if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
281 			if (bad == 0) {
282 				if (++cp < end)
283 					continue;
284 				else
285 					break;
286 			}
287 		} else
288 			break;
289 		if (*cp == '\0' || ++cp >= end)
290 			*list = NULL;
291 		else
292 			*list = cp;
293 		return (trunc_page(bad));
294 	}
295 	printf("Garbage in RAM blacklist, skipping\n");
296 	*list = NULL;
297 	return (0);
298 }
299 
300 /*
301  *	vm_page_blacklist_check:
302  *
303  *	Iterate through the provided string of blacklist addresses, pulling
304  *	each entry out of the physical allocator free list and putting it
305  *	onto a list for reporting via the vm.page_blacklist sysctl.
306  */
307 static void
308 vm_page_blacklist_check(char *list, char *end)
309 {
310 	vm_paddr_t pa;
311 	vm_page_t m;
312 	char *next;
313 	int ret;
314 
315 	next = list;
316 	while (next != NULL) {
317 		if ((pa = vm_page_blacklist_next(&next, end)) == 0)
318 			continue;
319 		m = vm_phys_paddr_to_vm_page(pa);
320 		if (m == NULL)
321 			continue;
322 		mtx_lock(&vm_page_queue_free_mtx);
323 		ret = vm_phys_unfree_page(m);
324 		mtx_unlock(&vm_page_queue_free_mtx);
325 		if (ret == TRUE) {
326 			TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
327 			if (bootverbose)
328 				printf("Skipping page with pa 0x%jx\n",
329 				    (uintmax_t)pa);
330 		}
331 	}
332 }
333 
334 /*
335  *	vm_page_blacklist_load:
336  *
337  *	Search for a special module named "ram_blacklist".  It'll be a
338  *	plain text file provided by the user via the loader directive
339  *	of the same name.
340  */
341 static void
342 vm_page_blacklist_load(char **list, char **end)
343 {
344 	void *mod;
345 	u_char *ptr;
346 	u_int len;
347 
348 	mod = NULL;
349 	ptr = NULL;
350 
351 	mod = preload_search_by_type("ram_blacklist");
352 	if (mod != NULL) {
353 		ptr = preload_fetch_addr(mod);
354 		len = preload_fetch_size(mod);
355         }
356 	*list = ptr;
357 	if (ptr != NULL)
358 		*end = ptr + len;
359 	else
360 		*end = NULL;
361 	return;
362 }
363 
364 static int
365 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
366 {
367 	vm_page_t m;
368 	struct sbuf sbuf;
369 	int error, first;
370 
371 	first = 1;
372 	error = sysctl_wire_old_buffer(req, 0);
373 	if (error != 0)
374 		return (error);
375 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
376 	TAILQ_FOREACH(m, &blacklist_head, listq) {
377 		sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
378 		    (uintmax_t)m->phys_addr);
379 		first = 0;
380 	}
381 	error = sbuf_finish(&sbuf);
382 	sbuf_delete(&sbuf);
383 	return (error);
384 }
385 
386 static void
387 vm_page_domain_init(struct vm_domain *vmd)
388 {
389 	struct vm_pagequeue *pq;
390 	int i;
391 
392 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
393 	    "vm inactive pagequeue";
394 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
395 	    &vm_cnt.v_inactive_count;
396 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
397 	    "vm active pagequeue";
398 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
399 	    &vm_cnt.v_active_count;
400 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_name) =
401 	    "vm laundry pagequeue";
402 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_vcnt) =
403 	    &vm_cnt.v_laundry_count;
404 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_name) =
405 	    "vm unswappable pagequeue";
406 	/* Unswappable dirty pages are counted as being in the laundry. */
407 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_vcnt) =
408 	    &vm_cnt.v_laundry_count;
409 	vmd->vmd_page_count = 0;
410 	vmd->vmd_free_count = 0;
411 	vmd->vmd_segs = 0;
412 	vmd->vmd_oom = FALSE;
413 	for (i = 0; i < PQ_COUNT; i++) {
414 		pq = &vmd->vmd_pagequeues[i];
415 		TAILQ_INIT(&pq->pq_pl);
416 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
417 		    MTX_DEF | MTX_DUPOK);
418 	}
419 }
420 
421 /*
422  *	vm_page_startup:
423  *
424  *	Initializes the resident memory module.  Allocates physical memory for
425  *	bootstrapping UMA and some data structures that are used to manage
426  *	physical pages.  Initializes these structures, and populates the free
427  *	page queues.
428  */
429 vm_offset_t
430 vm_page_startup(vm_offset_t vaddr)
431 {
432 	vm_offset_t mapped;
433 	vm_paddr_t high_avail, low_avail, page_range, size;
434 	vm_paddr_t new_end;
435 	int i;
436 	vm_paddr_t pa;
437 	vm_paddr_t last_pa;
438 	char *list, *listend;
439 	vm_paddr_t end;
440 	vm_paddr_t biggestsize;
441 	int biggestone;
442 	int pages_per_zone;
443 
444 	biggestsize = 0;
445 	biggestone = 0;
446 	vaddr = round_page(vaddr);
447 
448 	for (i = 0; phys_avail[i + 1]; i += 2) {
449 		phys_avail[i] = round_page(phys_avail[i]);
450 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
451 	}
452 	for (i = 0; phys_avail[i + 1]; i += 2) {
453 		size = phys_avail[i + 1] - phys_avail[i];
454 		if (size > biggestsize) {
455 			biggestone = i;
456 			biggestsize = size;
457 		}
458 	}
459 
460 	end = phys_avail[biggestone+1];
461 
462 	/*
463 	 * Initialize the page and queue locks.
464 	 */
465 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
466 	for (i = 0; i < PA_LOCK_COUNT; i++)
467 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
468 	for (i = 0; i < vm_ndomains; i++)
469 		vm_page_domain_init(&vm_dom[i]);
470 
471 	/*
472 	 * Almost all of the pages needed for bootstrapping UMA are used
473 	 * for zone structures, so if the number of CPUs results in those
474 	 * structures taking more than one page each, we set aside more pages
475 	 * in proportion to the zone structure size.
476 	 */
477 	pages_per_zone = howmany(sizeof(struct uma_zone) +
478 	    sizeof(struct uma_cache) * (mp_maxid + 1), UMA_SLAB_SIZE);
479 	if (pages_per_zone > 1) {
480 		/* Reserve more pages so that we don't run out. */
481 		boot_pages = UMA_BOOT_PAGES_ZONES * pages_per_zone;
482 	}
483 
484 	/*
485 	 * Allocate memory for use when boot strapping the kernel memory
486 	 * allocator.
487 	 *
488 	 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
489 	 * manually fetch the value.
490 	 */
491 	TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
492 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
493 	new_end = trunc_page(new_end);
494 	mapped = pmap_map(&vaddr, new_end, end,
495 	    VM_PROT_READ | VM_PROT_WRITE);
496 	bzero((void *)mapped, end - new_end);
497 	uma_startup((void *)mapped, boot_pages);
498 
499 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
500     defined(__i386__) || defined(__mips__)
501 	/*
502 	 * Allocate a bitmap to indicate that a random physical page
503 	 * needs to be included in a minidump.
504 	 *
505 	 * The amd64 port needs this to indicate which direct map pages
506 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
507 	 *
508 	 * However, i386 still needs this workspace internally within the
509 	 * minidump code.  In theory, they are not needed on i386, but are
510 	 * included should the sf_buf code decide to use them.
511 	 */
512 	last_pa = 0;
513 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
514 		if (dump_avail[i + 1] > last_pa)
515 			last_pa = dump_avail[i + 1];
516 	page_range = last_pa / PAGE_SIZE;
517 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
518 	new_end -= vm_page_dump_size;
519 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
520 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
521 	bzero((void *)vm_page_dump, vm_page_dump_size);
522 #endif
523 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
524 	/*
525 	 * Include the UMA bootstrap pages and vm_page_dump in a crash dump.
526 	 * When pmap_map() uses the direct map, they are not automatically
527 	 * included.
528 	 */
529 	for (pa = new_end; pa < end; pa += PAGE_SIZE)
530 		dump_add_page(pa);
531 #endif
532 	phys_avail[biggestone + 1] = new_end;
533 #ifdef __amd64__
534 	/*
535 	 * Request that the physical pages underlying the message buffer be
536 	 * included in a crash dump.  Since the message buffer is accessed
537 	 * through the direct map, they are not automatically included.
538 	 */
539 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
540 	last_pa = pa + round_page(msgbufsize);
541 	while (pa < last_pa) {
542 		dump_add_page(pa);
543 		pa += PAGE_SIZE;
544 	}
545 #endif
546 	/*
547 	 * Compute the number of pages of memory that will be available for
548 	 * use, taking into account the overhead of a page structure per page.
549 	 * In other words, solve
550 	 *	"available physical memory" - round_page(page_range *
551 	 *	    sizeof(struct vm_page)) = page_range * PAGE_SIZE
552 	 * for page_range.
553 	 */
554 	low_avail = phys_avail[0];
555 	high_avail = phys_avail[1];
556 	for (i = 0; i < vm_phys_nsegs; i++) {
557 		if (vm_phys_segs[i].start < low_avail)
558 			low_avail = vm_phys_segs[i].start;
559 		if (vm_phys_segs[i].end > high_avail)
560 			high_avail = vm_phys_segs[i].end;
561 	}
562 	/* Skip the first chunk.  It is already accounted for. */
563 	for (i = 2; phys_avail[i + 1] != 0; i += 2) {
564 		if (phys_avail[i] < low_avail)
565 			low_avail = phys_avail[i];
566 		if (phys_avail[i + 1] > high_avail)
567 			high_avail = phys_avail[i + 1];
568 	}
569 	first_page = low_avail / PAGE_SIZE;
570 #ifdef VM_PHYSSEG_SPARSE
571 	size = 0;
572 	for (i = 0; i < vm_phys_nsegs; i++)
573 		size += vm_phys_segs[i].end - vm_phys_segs[i].start;
574 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
575 		size += phys_avail[i + 1] - phys_avail[i];
576 #elif defined(VM_PHYSSEG_DENSE)
577 	size = high_avail - low_avail;
578 #else
579 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
580 #endif
581 
582 #ifdef VM_PHYSSEG_DENSE
583 	/*
584 	 * In the VM_PHYSSEG_DENSE case, the number of pages can account for
585 	 * the overhead of a page structure per page only if vm_page_array is
586 	 * allocated from the last physical memory chunk.  Otherwise, we must
587 	 * allocate page structures representing the physical memory
588 	 * underlying vm_page_array, even though they will not be used.
589 	 */
590 	if (new_end != high_avail)
591 		page_range = size / PAGE_SIZE;
592 	else
593 #endif
594 	{
595 		page_range = size / (PAGE_SIZE + sizeof(struct vm_page));
596 
597 		/*
598 		 * If the partial bytes remaining are large enough for
599 		 * a page (PAGE_SIZE) without a corresponding
600 		 * 'struct vm_page', then new_end will contain an
601 		 * extra page after subtracting the length of the VM
602 		 * page array.  Compensate by subtracting an extra
603 		 * page from new_end.
604 		 */
605 		if (size % (PAGE_SIZE + sizeof(struct vm_page)) >= PAGE_SIZE) {
606 			if (new_end == high_avail)
607 				high_avail -= PAGE_SIZE;
608 			new_end -= PAGE_SIZE;
609 		}
610 	}
611 	end = new_end;
612 
613 	/*
614 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
615 	 * However, because this page is allocated from KVM, out-of-bounds
616 	 * accesses using the direct map will not be trapped.
617 	 */
618 	vaddr += PAGE_SIZE;
619 
620 	/*
621 	 * Allocate physical memory for the page structures, and map it.
622 	 */
623 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
624 	mapped = pmap_map(&vaddr, new_end, end,
625 	    VM_PROT_READ | VM_PROT_WRITE);
626 	vm_page_array = (vm_page_t) mapped;
627 #if VM_NRESERVLEVEL > 0
628 	/*
629 	 * Allocate physical memory for the reservation management system's
630 	 * data structures, and map it.
631 	 */
632 	if (high_avail == end)
633 		high_avail = new_end;
634 	new_end = vm_reserv_startup(&vaddr, new_end, high_avail);
635 #endif
636 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
637 	/*
638 	 * Include vm_page_array and vm_reserv_array in a crash dump.
639 	 */
640 	for (pa = new_end; pa < end; pa += PAGE_SIZE)
641 		dump_add_page(pa);
642 #endif
643 	phys_avail[biggestone + 1] = new_end;
644 
645 	/*
646 	 * Add physical memory segments corresponding to the available
647 	 * physical pages.
648 	 */
649 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
650 		vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
651 
652 	/*
653 	 * Clear all of the page structures
654 	 */
655 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
656 	for (i = 0; i < page_range; i++)
657 		vm_page_array[i].order = VM_NFREEORDER;
658 	vm_page_array_size = page_range;
659 
660 	/*
661 	 * Initialize the physical memory allocator.
662 	 */
663 	vm_phys_init();
664 
665 	/*
666 	 * Add every available physical page that is not blacklisted to
667 	 * the free lists.
668 	 */
669 	vm_cnt.v_page_count = 0;
670 	vm_cnt.v_free_count = 0;
671 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
672 		pa = phys_avail[i];
673 		last_pa = phys_avail[i + 1];
674 		while (pa < last_pa) {
675 			vm_phys_add_page(pa);
676 			pa += PAGE_SIZE;
677 		}
678 	}
679 
680 	TAILQ_INIT(&blacklist_head);
681 	vm_page_blacklist_load(&list, &listend);
682 	vm_page_blacklist_check(list, listend);
683 
684 	list = kern_getenv("vm.blacklist");
685 	vm_page_blacklist_check(list, NULL);
686 
687 	freeenv(list);
688 #if VM_NRESERVLEVEL > 0
689 	/*
690 	 * Initialize the reservation management system.
691 	 */
692 	vm_reserv_init();
693 #endif
694 	return (vaddr);
695 }
696 
697 void
698 vm_page_reference(vm_page_t m)
699 {
700 
701 	vm_page_aflag_set(m, PGA_REFERENCED);
702 }
703 
704 /*
705  *	vm_page_busy_downgrade:
706  *
707  *	Downgrade an exclusive busy page into a single shared busy page.
708  */
709 void
710 vm_page_busy_downgrade(vm_page_t m)
711 {
712 	u_int x;
713 	bool locked;
714 
715 	vm_page_assert_xbusied(m);
716 	locked = mtx_owned(vm_page_lockptr(m));
717 
718 	for (;;) {
719 		x = m->busy_lock;
720 		x &= VPB_BIT_WAITERS;
721 		if (x != 0 && !locked)
722 			vm_page_lock(m);
723 		if (atomic_cmpset_rel_int(&m->busy_lock,
724 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1)))
725 			break;
726 		if (x != 0 && !locked)
727 			vm_page_unlock(m);
728 	}
729 	if (x != 0) {
730 		wakeup(m);
731 		if (!locked)
732 			vm_page_unlock(m);
733 	}
734 }
735 
736 /*
737  *	vm_page_sbusied:
738  *
739  *	Return a positive value if the page is shared busied, 0 otherwise.
740  */
741 int
742 vm_page_sbusied(vm_page_t m)
743 {
744 	u_int x;
745 
746 	x = m->busy_lock;
747 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
748 }
749 
750 /*
751  *	vm_page_sunbusy:
752  *
753  *	Shared unbusy a page.
754  */
755 void
756 vm_page_sunbusy(vm_page_t m)
757 {
758 	u_int x;
759 
760 	vm_page_lock_assert(m, MA_NOTOWNED);
761 	vm_page_assert_sbusied(m);
762 
763 	for (;;) {
764 		x = m->busy_lock;
765 		if (VPB_SHARERS(x) > 1) {
766 			if (atomic_cmpset_int(&m->busy_lock, x,
767 			    x - VPB_ONE_SHARER))
768 				break;
769 			continue;
770 		}
771 		if ((x & VPB_BIT_WAITERS) == 0) {
772 			KASSERT(x == VPB_SHARERS_WORD(1),
773 			    ("vm_page_sunbusy: invalid lock state"));
774 			if (atomic_cmpset_int(&m->busy_lock,
775 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
776 				break;
777 			continue;
778 		}
779 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
780 		    ("vm_page_sunbusy: invalid lock state for waiters"));
781 
782 		vm_page_lock(m);
783 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
784 			vm_page_unlock(m);
785 			continue;
786 		}
787 		wakeup(m);
788 		vm_page_unlock(m);
789 		break;
790 	}
791 }
792 
793 /*
794  *	vm_page_busy_sleep:
795  *
796  *	Sleep and release the page lock, using the page pointer as wchan.
797  *	This is used to implement the hard-path of busying mechanism.
798  *
799  *	The given page must be locked.
800  *
801  *	If nonshared is true, sleep only if the page is xbusy.
802  */
803 void
804 vm_page_busy_sleep(vm_page_t m, const char *wmesg, bool nonshared)
805 {
806 	u_int x;
807 
808 	vm_page_assert_locked(m);
809 
810 	x = m->busy_lock;
811 	if (x == VPB_UNBUSIED || (nonshared && (x & VPB_BIT_SHARED) != 0) ||
812 	    ((x & VPB_BIT_WAITERS) == 0 &&
813 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS))) {
814 		vm_page_unlock(m);
815 		return;
816 	}
817 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
818 }
819 
820 /*
821  *	vm_page_trysbusy:
822  *
823  *	Try to shared busy a page.
824  *	If the operation succeeds 1 is returned otherwise 0.
825  *	The operation never sleeps.
826  */
827 int
828 vm_page_trysbusy(vm_page_t m)
829 {
830 	u_int x;
831 
832 	for (;;) {
833 		x = m->busy_lock;
834 		if ((x & VPB_BIT_SHARED) == 0)
835 			return (0);
836 		if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
837 			return (1);
838 	}
839 }
840 
841 static void
842 vm_page_xunbusy_locked(vm_page_t m)
843 {
844 
845 	vm_page_assert_xbusied(m);
846 	vm_page_assert_locked(m);
847 
848 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
849 	/* There is a waiter, do wakeup() instead of vm_page_flash(). */
850 	wakeup(m);
851 }
852 
853 void
854 vm_page_xunbusy_maybelocked(vm_page_t m)
855 {
856 	bool lockacq;
857 
858 	vm_page_assert_xbusied(m);
859 
860 	/*
861 	 * Fast path for unbusy.  If it succeeds, we know that there
862 	 * are no waiters, so we do not need a wakeup.
863 	 */
864 	if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER,
865 	    VPB_UNBUSIED))
866 		return;
867 
868 	lockacq = !mtx_owned(vm_page_lockptr(m));
869 	if (lockacq)
870 		vm_page_lock(m);
871 	vm_page_xunbusy_locked(m);
872 	if (lockacq)
873 		vm_page_unlock(m);
874 }
875 
876 /*
877  *	vm_page_xunbusy_hard:
878  *
879  *	Called after the first try the exclusive unbusy of a page failed.
880  *	It is assumed that the waiters bit is on.
881  */
882 void
883 vm_page_xunbusy_hard(vm_page_t m)
884 {
885 
886 	vm_page_assert_xbusied(m);
887 
888 	vm_page_lock(m);
889 	vm_page_xunbusy_locked(m);
890 	vm_page_unlock(m);
891 }
892 
893 /*
894  *	vm_page_flash:
895  *
896  *	Wakeup anyone waiting for the page.
897  *	The ownership bits do not change.
898  *
899  *	The given page must be locked.
900  */
901 void
902 vm_page_flash(vm_page_t m)
903 {
904 	u_int x;
905 
906 	vm_page_lock_assert(m, MA_OWNED);
907 
908 	for (;;) {
909 		x = m->busy_lock;
910 		if ((x & VPB_BIT_WAITERS) == 0)
911 			return;
912 		if (atomic_cmpset_int(&m->busy_lock, x,
913 		    x & (~VPB_BIT_WAITERS)))
914 			break;
915 	}
916 	wakeup(m);
917 }
918 
919 /*
920  * Keep page from being freed by the page daemon
921  * much of the same effect as wiring, except much lower
922  * overhead and should be used only for *very* temporary
923  * holding ("wiring").
924  */
925 void
926 vm_page_hold(vm_page_t mem)
927 {
928 
929 	vm_page_lock_assert(mem, MA_OWNED);
930         mem->hold_count++;
931 }
932 
933 void
934 vm_page_unhold(vm_page_t mem)
935 {
936 
937 	vm_page_lock_assert(mem, MA_OWNED);
938 	KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
939 	--mem->hold_count;
940 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
941 		vm_page_free_toq(mem);
942 }
943 
944 /*
945  *	vm_page_unhold_pages:
946  *
947  *	Unhold each of the pages that is referenced by the given array.
948  */
949 void
950 vm_page_unhold_pages(vm_page_t *ma, int count)
951 {
952 	struct mtx *mtx, *new_mtx;
953 
954 	mtx = NULL;
955 	for (; count != 0; count--) {
956 		/*
957 		 * Avoid releasing and reacquiring the same page lock.
958 		 */
959 		new_mtx = vm_page_lockptr(*ma);
960 		if (mtx != new_mtx) {
961 			if (mtx != NULL)
962 				mtx_unlock(mtx);
963 			mtx = new_mtx;
964 			mtx_lock(mtx);
965 		}
966 		vm_page_unhold(*ma);
967 		ma++;
968 	}
969 	if (mtx != NULL)
970 		mtx_unlock(mtx);
971 }
972 
973 vm_page_t
974 PHYS_TO_VM_PAGE(vm_paddr_t pa)
975 {
976 	vm_page_t m;
977 
978 #ifdef VM_PHYSSEG_SPARSE
979 	m = vm_phys_paddr_to_vm_page(pa);
980 	if (m == NULL)
981 		m = vm_phys_fictitious_to_vm_page(pa);
982 	return (m);
983 #elif defined(VM_PHYSSEG_DENSE)
984 	long pi;
985 
986 	pi = atop(pa);
987 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
988 		m = &vm_page_array[pi - first_page];
989 		return (m);
990 	}
991 	return (vm_phys_fictitious_to_vm_page(pa));
992 #else
993 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
994 #endif
995 }
996 
997 /*
998  *	vm_page_getfake:
999  *
1000  *	Create a fictitious page with the specified physical address and
1001  *	memory attribute.  The memory attribute is the only the machine-
1002  *	dependent aspect of a fictitious page that must be initialized.
1003  */
1004 vm_page_t
1005 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
1006 {
1007 	vm_page_t m;
1008 
1009 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
1010 	vm_page_initfake(m, paddr, memattr);
1011 	return (m);
1012 }
1013 
1014 void
1015 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1016 {
1017 
1018 	if ((m->flags & PG_FICTITIOUS) != 0) {
1019 		/*
1020 		 * The page's memattr might have changed since the
1021 		 * previous initialization.  Update the pmap to the
1022 		 * new memattr.
1023 		 */
1024 		goto memattr;
1025 	}
1026 	m->phys_addr = paddr;
1027 	m->queue = PQ_NONE;
1028 	/* Fictitious pages don't use "segind". */
1029 	m->flags = PG_FICTITIOUS;
1030 	/* Fictitious pages don't use "order" or "pool". */
1031 	m->oflags = VPO_UNMANAGED;
1032 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1033 	m->wire_count = 1;
1034 	pmap_page_init(m);
1035 memattr:
1036 	pmap_page_set_memattr(m, memattr);
1037 }
1038 
1039 /*
1040  *	vm_page_putfake:
1041  *
1042  *	Release a fictitious page.
1043  */
1044 void
1045 vm_page_putfake(vm_page_t m)
1046 {
1047 
1048 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
1049 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
1050 	    ("vm_page_putfake: bad page %p", m));
1051 	uma_zfree(fakepg_zone, m);
1052 }
1053 
1054 /*
1055  *	vm_page_updatefake:
1056  *
1057  *	Update the given fictitious page to the specified physical address and
1058  *	memory attribute.
1059  */
1060 void
1061 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1062 {
1063 
1064 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
1065 	    ("vm_page_updatefake: bad page %p", m));
1066 	m->phys_addr = paddr;
1067 	pmap_page_set_memattr(m, memattr);
1068 }
1069 
1070 /*
1071  *	vm_page_free:
1072  *
1073  *	Free a page.
1074  */
1075 void
1076 vm_page_free(vm_page_t m)
1077 {
1078 
1079 	m->flags &= ~PG_ZERO;
1080 	vm_page_free_toq(m);
1081 }
1082 
1083 /*
1084  *	vm_page_free_zero:
1085  *
1086  *	Free a page to the zerod-pages queue
1087  */
1088 void
1089 vm_page_free_zero(vm_page_t m)
1090 {
1091 
1092 	m->flags |= PG_ZERO;
1093 	vm_page_free_toq(m);
1094 }
1095 
1096 /*
1097  * Unbusy and handle the page queueing for a page from a getpages request that
1098  * was optionally read ahead or behind.
1099  */
1100 void
1101 vm_page_readahead_finish(vm_page_t m)
1102 {
1103 
1104 	/* We shouldn't put invalid pages on queues. */
1105 	KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m));
1106 
1107 	/*
1108 	 * Since the page is not the actually needed one, whether it should
1109 	 * be activated or deactivated is not obvious.  Empirical results
1110 	 * have shown that deactivating the page is usually the best choice,
1111 	 * unless the page is wanted by another thread.
1112 	 */
1113 	vm_page_lock(m);
1114 	if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1115 		vm_page_activate(m);
1116 	else
1117 		vm_page_deactivate(m);
1118 	vm_page_unlock(m);
1119 	vm_page_xunbusy(m);
1120 }
1121 
1122 /*
1123  *	vm_page_sleep_if_busy:
1124  *
1125  *	Sleep and release the page queues lock if the page is busied.
1126  *	Returns TRUE if the thread slept.
1127  *
1128  *	The given page must be unlocked and object containing it must
1129  *	be locked.
1130  */
1131 int
1132 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1133 {
1134 	vm_object_t obj;
1135 
1136 	vm_page_lock_assert(m, MA_NOTOWNED);
1137 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1138 
1139 	if (vm_page_busied(m)) {
1140 		/*
1141 		 * The page-specific object must be cached because page
1142 		 * identity can change during the sleep, causing the
1143 		 * re-lock of a different object.
1144 		 * It is assumed that a reference to the object is already
1145 		 * held by the callers.
1146 		 */
1147 		obj = m->object;
1148 		vm_page_lock(m);
1149 		VM_OBJECT_WUNLOCK(obj);
1150 		vm_page_busy_sleep(m, msg, false);
1151 		VM_OBJECT_WLOCK(obj);
1152 		return (TRUE);
1153 	}
1154 	return (FALSE);
1155 }
1156 
1157 /*
1158  *	vm_page_dirty_KBI:		[ internal use only ]
1159  *
1160  *	Set all bits in the page's dirty field.
1161  *
1162  *	The object containing the specified page must be locked if the
1163  *	call is made from the machine-independent layer.
1164  *
1165  *	See vm_page_clear_dirty_mask().
1166  *
1167  *	This function should only be called by vm_page_dirty().
1168  */
1169 void
1170 vm_page_dirty_KBI(vm_page_t m)
1171 {
1172 
1173 	/* Refer to this operation by its public name. */
1174 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
1175 	    ("vm_page_dirty: page is invalid!"));
1176 	m->dirty = VM_PAGE_BITS_ALL;
1177 }
1178 
1179 /*
1180  *	vm_page_insert:		[ internal use only ]
1181  *
1182  *	Inserts the given mem entry into the object and object list.
1183  *
1184  *	The object must be locked.
1185  */
1186 int
1187 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1188 {
1189 	vm_page_t mpred;
1190 
1191 	VM_OBJECT_ASSERT_WLOCKED(object);
1192 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
1193 	return (vm_page_insert_after(m, object, pindex, mpred));
1194 }
1195 
1196 /*
1197  *	vm_page_insert_after:
1198  *
1199  *	Inserts the page "m" into the specified object at offset "pindex".
1200  *
1201  *	The page "mpred" must immediately precede the offset "pindex" within
1202  *	the specified object.
1203  *
1204  *	The object must be locked.
1205  */
1206 static int
1207 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1208     vm_page_t mpred)
1209 {
1210 	vm_page_t msucc;
1211 
1212 	VM_OBJECT_ASSERT_WLOCKED(object);
1213 	KASSERT(m->object == NULL,
1214 	    ("vm_page_insert_after: page already inserted"));
1215 	if (mpred != NULL) {
1216 		KASSERT(mpred->object == object,
1217 		    ("vm_page_insert_after: object doesn't contain mpred"));
1218 		KASSERT(mpred->pindex < pindex,
1219 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1220 		msucc = TAILQ_NEXT(mpred, listq);
1221 	} else
1222 		msucc = TAILQ_FIRST(&object->memq);
1223 	if (msucc != NULL)
1224 		KASSERT(msucc->pindex > pindex,
1225 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
1226 
1227 	/*
1228 	 * Record the object/offset pair in this page
1229 	 */
1230 	m->object = object;
1231 	m->pindex = pindex;
1232 
1233 	/*
1234 	 * Now link into the object's ordered list of backed pages.
1235 	 */
1236 	if (vm_radix_insert(&object->rtree, m)) {
1237 		m->object = NULL;
1238 		m->pindex = 0;
1239 		return (1);
1240 	}
1241 	vm_page_insert_radixdone(m, object, mpred);
1242 	return (0);
1243 }
1244 
1245 /*
1246  *	vm_page_insert_radixdone:
1247  *
1248  *	Complete page "m" insertion into the specified object after the
1249  *	radix trie hooking.
1250  *
1251  *	The page "mpred" must precede the offset "m->pindex" within the
1252  *	specified object.
1253  *
1254  *	The object must be locked.
1255  */
1256 static void
1257 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1258 {
1259 
1260 	VM_OBJECT_ASSERT_WLOCKED(object);
1261 	KASSERT(object != NULL && m->object == object,
1262 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1263 	if (mpred != NULL) {
1264 		KASSERT(mpred->object == object,
1265 		    ("vm_page_insert_after: object doesn't contain mpred"));
1266 		KASSERT(mpred->pindex < m->pindex,
1267 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1268 	}
1269 
1270 	if (mpred != NULL)
1271 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1272 	else
1273 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1274 
1275 	/*
1276 	 * Show that the object has one more resident page.
1277 	 */
1278 	object->resident_page_count++;
1279 
1280 	/*
1281 	 * Hold the vnode until the last page is released.
1282 	 */
1283 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1284 		vhold(object->handle);
1285 
1286 	/*
1287 	 * Since we are inserting a new and possibly dirty page,
1288 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1289 	 */
1290 	if (pmap_page_is_write_mapped(m))
1291 		vm_object_set_writeable_dirty(object);
1292 }
1293 
1294 /*
1295  *	vm_page_remove:
1296  *
1297  *	Removes the specified page from its containing object, but does not
1298  *	invalidate any backing storage.
1299  *
1300  *	The object must be locked.  The page must be locked if it is managed.
1301  */
1302 void
1303 vm_page_remove(vm_page_t m)
1304 {
1305 	vm_object_t object;
1306 	vm_page_t mrem;
1307 
1308 	if ((m->oflags & VPO_UNMANAGED) == 0)
1309 		vm_page_assert_locked(m);
1310 	if ((object = m->object) == NULL)
1311 		return;
1312 	VM_OBJECT_ASSERT_WLOCKED(object);
1313 	if (vm_page_xbusied(m))
1314 		vm_page_xunbusy_maybelocked(m);
1315 	mrem = vm_radix_remove(&object->rtree, m->pindex);
1316 	KASSERT(mrem == m, ("removed page %p, expected page %p", mrem, m));
1317 
1318 	/*
1319 	 * Now remove from the object's list of backed pages.
1320 	 */
1321 	TAILQ_REMOVE(&object->memq, m, listq);
1322 
1323 	/*
1324 	 * And show that the object has one fewer resident page.
1325 	 */
1326 	object->resident_page_count--;
1327 
1328 	/*
1329 	 * The vnode may now be recycled.
1330 	 */
1331 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1332 		vdrop(object->handle);
1333 
1334 	m->object = NULL;
1335 }
1336 
1337 /*
1338  *	vm_page_lookup:
1339  *
1340  *	Returns the page associated with the object/offset
1341  *	pair specified; if none is found, NULL is returned.
1342  *
1343  *	The object must be locked.
1344  */
1345 vm_page_t
1346 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1347 {
1348 
1349 	VM_OBJECT_ASSERT_LOCKED(object);
1350 	return (vm_radix_lookup(&object->rtree, pindex));
1351 }
1352 
1353 /*
1354  *	vm_page_find_least:
1355  *
1356  *	Returns the page associated with the object with least pindex
1357  *	greater than or equal to the parameter pindex, or NULL.
1358  *
1359  *	The object must be locked.
1360  */
1361 vm_page_t
1362 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1363 {
1364 	vm_page_t m;
1365 
1366 	VM_OBJECT_ASSERT_LOCKED(object);
1367 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1368 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1369 	return (m);
1370 }
1371 
1372 /*
1373  * Returns the given page's successor (by pindex) within the object if it is
1374  * resident; if none is found, NULL is returned.
1375  *
1376  * The object must be locked.
1377  */
1378 vm_page_t
1379 vm_page_next(vm_page_t m)
1380 {
1381 	vm_page_t next;
1382 
1383 	VM_OBJECT_ASSERT_LOCKED(m->object);
1384 	if ((next = TAILQ_NEXT(m, listq)) != NULL) {
1385 		MPASS(next->object == m->object);
1386 		if (next->pindex != m->pindex + 1)
1387 			next = NULL;
1388 	}
1389 	return (next);
1390 }
1391 
1392 /*
1393  * Returns the given page's predecessor (by pindex) within the object if it is
1394  * resident; if none is found, NULL is returned.
1395  *
1396  * The object must be locked.
1397  */
1398 vm_page_t
1399 vm_page_prev(vm_page_t m)
1400 {
1401 	vm_page_t prev;
1402 
1403 	VM_OBJECT_ASSERT_LOCKED(m->object);
1404 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL) {
1405 		MPASS(prev->object == m->object);
1406 		if (prev->pindex != m->pindex - 1)
1407 			prev = NULL;
1408 	}
1409 	return (prev);
1410 }
1411 
1412 /*
1413  * Uses the page mnew as a replacement for an existing page at index
1414  * pindex which must be already present in the object.
1415  *
1416  * The existing page must not be on a paging queue.
1417  */
1418 vm_page_t
1419 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1420 {
1421 	vm_page_t mold;
1422 
1423 	VM_OBJECT_ASSERT_WLOCKED(object);
1424 	KASSERT(mnew->object == NULL,
1425 	    ("vm_page_replace: page already in object"));
1426 
1427 	/*
1428 	 * This function mostly follows vm_page_insert() and
1429 	 * vm_page_remove() without the radix, object count and vnode
1430 	 * dance.  Double check such functions for more comments.
1431 	 */
1432 
1433 	mnew->object = object;
1434 	mnew->pindex = pindex;
1435 	mold = vm_radix_replace(&object->rtree, mnew);
1436 	KASSERT(mold->queue == PQ_NONE,
1437 	    ("vm_page_replace: mold is on a paging queue"));
1438 
1439 	/* Keep the resident page list in sorted order. */
1440 	TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1441 	TAILQ_REMOVE(&object->memq, mold, listq);
1442 
1443 	mold->object = NULL;
1444 	vm_page_xunbusy_maybelocked(mold);
1445 
1446 	/*
1447 	 * The object's resident_page_count does not change because we have
1448 	 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1449 	 */
1450 	if (pmap_page_is_write_mapped(mnew))
1451 		vm_object_set_writeable_dirty(object);
1452 	return (mold);
1453 }
1454 
1455 /*
1456  *	vm_page_rename:
1457  *
1458  *	Move the given memory entry from its
1459  *	current object to the specified target object/offset.
1460  *
1461  *	Note: swap associated with the page must be invalidated by the move.  We
1462  *	      have to do this for several reasons:  (1) we aren't freeing the
1463  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1464  *	      moving the page from object A to B, and will then later move
1465  *	      the backing store from A to B and we can't have a conflict.
1466  *
1467  *	Note: we *always* dirty the page.  It is necessary both for the
1468  *	      fact that we moved it, and because we may be invalidating
1469  *	      swap.
1470  *
1471  *	The objects must be locked.
1472  */
1473 int
1474 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1475 {
1476 	vm_page_t mpred;
1477 	vm_pindex_t opidx;
1478 
1479 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1480 
1481 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1482 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1483 	    ("vm_page_rename: pindex already renamed"));
1484 
1485 	/*
1486 	 * Create a custom version of vm_page_insert() which does not depend
1487 	 * by m_prev and can cheat on the implementation aspects of the
1488 	 * function.
1489 	 */
1490 	opidx = m->pindex;
1491 	m->pindex = new_pindex;
1492 	if (vm_radix_insert(&new_object->rtree, m)) {
1493 		m->pindex = opidx;
1494 		return (1);
1495 	}
1496 
1497 	/*
1498 	 * The operation cannot fail anymore.  The removal must happen before
1499 	 * the listq iterator is tainted.
1500 	 */
1501 	m->pindex = opidx;
1502 	vm_page_lock(m);
1503 	vm_page_remove(m);
1504 
1505 	/* Return back to the new pindex to complete vm_page_insert(). */
1506 	m->pindex = new_pindex;
1507 	m->object = new_object;
1508 	vm_page_unlock(m);
1509 	vm_page_insert_radixdone(m, new_object, mpred);
1510 	vm_page_dirty(m);
1511 	return (0);
1512 }
1513 
1514 /*
1515  *	vm_page_alloc:
1516  *
1517  *	Allocate and return a page that is associated with the specified
1518  *	object and offset pair.  By default, this page is exclusive busied.
1519  *
1520  *	The caller must always specify an allocation class.
1521  *
1522  *	allocation classes:
1523  *	VM_ALLOC_NORMAL		normal process request
1524  *	VM_ALLOC_SYSTEM		system *really* needs a page
1525  *	VM_ALLOC_INTERRUPT	interrupt time request
1526  *
1527  *	optional allocation flags:
1528  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1529  *				intends to allocate
1530  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1531  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1532  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1533  *				should not be exclusive busy
1534  *	VM_ALLOC_SBUSY		shared busy the allocated page
1535  *	VM_ALLOC_WIRED		wire the allocated page
1536  *	VM_ALLOC_ZERO		prefer a zeroed page
1537  *
1538  *	This routine may not sleep.
1539  */
1540 vm_page_t
1541 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1542 {
1543 	vm_page_t m, mpred;
1544 	int flags, req_class;
1545 
1546 	mpred = NULL;	/* XXX: pacify gcc */
1547 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1548 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1549 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1550 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1551 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", object, req));
1552 	if (object != NULL)
1553 		VM_OBJECT_ASSERT_WLOCKED(object);
1554 
1555 	req_class = req & VM_ALLOC_CLASS_MASK;
1556 
1557 	/*
1558 	 * The page daemon is allowed to dig deeper into the free page list.
1559 	 */
1560 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1561 		req_class = VM_ALLOC_SYSTEM;
1562 
1563 	if (object != NULL) {
1564 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1565 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1566 		   ("vm_page_alloc: pindex already allocated"));
1567 	}
1568 
1569 	/*
1570 	 * Allocate a page if the number of free pages exceeds the minimum
1571 	 * for the request class.
1572 	 */
1573 	mtx_lock(&vm_page_queue_free_mtx);
1574 	if (vm_cnt.v_free_count > vm_cnt.v_free_reserved ||
1575 	    (req_class == VM_ALLOC_SYSTEM &&
1576 	    vm_cnt.v_free_count > vm_cnt.v_interrupt_free_min) ||
1577 	    (req_class == VM_ALLOC_INTERRUPT &&
1578 	    vm_cnt.v_free_count > 0)) {
1579 		/*
1580 		 * Can we allocate the page from a reservation?
1581 		 */
1582 #if VM_NRESERVLEVEL > 0
1583 		if (object == NULL || (object->flags & (OBJ_COLORED |
1584 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1585 		    vm_reserv_alloc_page(object, pindex, mpred)) == NULL)
1586 #endif
1587 		{
1588 			/*
1589 			 * If not, allocate it from the free page queues.
1590 			 */
1591 			m = vm_phys_alloc_pages(object != NULL ?
1592 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1593 #if VM_NRESERVLEVEL > 0
1594 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1595 				m = vm_phys_alloc_pages(object != NULL ?
1596 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1597 				    0);
1598 			}
1599 #endif
1600 		}
1601 	} else {
1602 		/*
1603 		 * Not allocatable, give up.
1604 		 */
1605 		mtx_unlock(&vm_page_queue_free_mtx);
1606 		atomic_add_int(&vm_pageout_deficit,
1607 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1608 		pagedaemon_wakeup();
1609 		return (NULL);
1610 	}
1611 
1612 	/*
1613 	 *  At this point we had better have found a good page.
1614 	 */
1615 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1616 	vm_phys_freecnt_adj(m, -1);
1617 	mtx_unlock(&vm_page_queue_free_mtx);
1618 	vm_page_alloc_check(m);
1619 
1620 	/*
1621 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1622 	 */
1623 	flags = 0;
1624 	if ((req & VM_ALLOC_ZERO) != 0)
1625 		flags = PG_ZERO;
1626 	flags &= m->flags;
1627 	if ((req & VM_ALLOC_NODUMP) != 0)
1628 		flags |= PG_NODUMP;
1629 	m->flags = flags;
1630 	m->aflags = 0;
1631 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1632 	    VPO_UNMANAGED : 0;
1633 	m->busy_lock = VPB_UNBUSIED;
1634 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1635 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1636 	if ((req & VM_ALLOC_SBUSY) != 0)
1637 		m->busy_lock = VPB_SHARERS_WORD(1);
1638 	if (req & VM_ALLOC_WIRED) {
1639 		/*
1640 		 * The page lock is not required for wiring a page until that
1641 		 * page is inserted into the object.
1642 		 */
1643 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1644 		m->wire_count = 1;
1645 	}
1646 	m->act_count = 0;
1647 
1648 	if (object != NULL) {
1649 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1650 			pagedaemon_wakeup();
1651 			if (req & VM_ALLOC_WIRED) {
1652 				atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1653 				m->wire_count = 0;
1654 			}
1655 			KASSERT(m->object == NULL, ("page %p has object", m));
1656 			m->oflags = VPO_UNMANAGED;
1657 			m->busy_lock = VPB_UNBUSIED;
1658 			/* Don't change PG_ZERO. */
1659 			vm_page_free_toq(m);
1660 			return (NULL);
1661 		}
1662 
1663 		/* Ignore device objects; the pager sets "memattr" for them. */
1664 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1665 		    (object->flags & OBJ_FICTITIOUS) == 0)
1666 			pmap_page_set_memattr(m, object->memattr);
1667 	} else
1668 		m->pindex = pindex;
1669 
1670 	/*
1671 	 * Don't wakeup too often - wakeup the pageout daemon when
1672 	 * we would be nearly out of memory.
1673 	 */
1674 	if (vm_paging_needed())
1675 		pagedaemon_wakeup();
1676 
1677 	return (m);
1678 }
1679 
1680 /*
1681  *	vm_page_alloc_contig:
1682  *
1683  *	Allocate a contiguous set of physical pages of the given size "npages"
1684  *	from the free lists.  All of the physical pages must be at or above
1685  *	the given physical address "low" and below the given physical address
1686  *	"high".  The given value "alignment" determines the alignment of the
1687  *	first physical page in the set.  If the given value "boundary" is
1688  *	non-zero, then the set of physical pages cannot cross any physical
1689  *	address boundary that is a multiple of that value.  Both "alignment"
1690  *	and "boundary" must be a power of two.
1691  *
1692  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1693  *	then the memory attribute setting for the physical pages is configured
1694  *	to the object's memory attribute setting.  Otherwise, the memory
1695  *	attribute setting for the physical pages is configured to "memattr",
1696  *	overriding the object's memory attribute setting.  However, if the
1697  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1698  *	memory attribute setting for the physical pages cannot be configured
1699  *	to VM_MEMATTR_DEFAULT.
1700  *
1701  *	The specified object may not contain fictitious pages.
1702  *
1703  *	The caller must always specify an allocation class.
1704  *
1705  *	allocation classes:
1706  *	VM_ALLOC_NORMAL		normal process request
1707  *	VM_ALLOC_SYSTEM		system *really* needs a page
1708  *	VM_ALLOC_INTERRUPT	interrupt time request
1709  *
1710  *	optional allocation flags:
1711  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1712  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1713  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1714  *				should not be exclusive busy
1715  *	VM_ALLOC_SBUSY		shared busy the allocated page
1716  *	VM_ALLOC_WIRED		wire the allocated page
1717  *	VM_ALLOC_ZERO		prefer a zeroed page
1718  *
1719  *	This routine may not sleep.
1720  */
1721 vm_page_t
1722 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1723     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1724     vm_paddr_t boundary, vm_memattr_t memattr)
1725 {
1726 	vm_page_t m, m_ret, mpred;
1727 	u_int busy_lock, flags, oflags;
1728 	int req_class;
1729 
1730 	mpred = NULL;	/* XXX: pacify gcc */
1731 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1732 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1733 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1734 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1735 	    ("vm_page_alloc_contig: inconsistent object(%p)/req(%x)", object,
1736 	    req));
1737 	if (object != NULL) {
1738 		VM_OBJECT_ASSERT_WLOCKED(object);
1739 		KASSERT((object->flags & OBJ_FICTITIOUS) == 0,
1740 		    ("vm_page_alloc_contig: object %p has fictitious pages",
1741 		    object));
1742 	}
1743 	KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1744 	req_class = req & VM_ALLOC_CLASS_MASK;
1745 
1746 	/*
1747 	 * The page daemon is allowed to dig deeper into the free page list.
1748 	 */
1749 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1750 		req_class = VM_ALLOC_SYSTEM;
1751 
1752 	if (object != NULL) {
1753 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1754 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1755 		    ("vm_page_alloc_contig: pindex already allocated"));
1756 	}
1757 
1758 	/*
1759 	 * Can we allocate the pages without the number of free pages falling
1760 	 * below the lower bound for the allocation class?
1761 	 */
1762 	mtx_lock(&vm_page_queue_free_mtx);
1763 	if (vm_cnt.v_free_count >= npages + vm_cnt.v_free_reserved ||
1764 	    (req_class == VM_ALLOC_SYSTEM &&
1765 	    vm_cnt.v_free_count >= npages + vm_cnt.v_interrupt_free_min) ||
1766 	    (req_class == VM_ALLOC_INTERRUPT &&
1767 	    vm_cnt.v_free_count >= npages)) {
1768 		/*
1769 		 * Can we allocate the pages from a reservation?
1770 		 */
1771 #if VM_NRESERVLEVEL > 0
1772 retry:
1773 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1774 		    (m_ret = vm_reserv_alloc_contig(object, pindex, npages,
1775 		    low, high, alignment, boundary, mpred)) == NULL)
1776 #endif
1777 			/*
1778 			 * If not, allocate them from the free page queues.
1779 			 */
1780 			m_ret = vm_phys_alloc_contig(npages, low, high,
1781 			    alignment, boundary);
1782 	} else {
1783 		mtx_unlock(&vm_page_queue_free_mtx);
1784 		atomic_add_int(&vm_pageout_deficit, npages);
1785 		pagedaemon_wakeup();
1786 		return (NULL);
1787 	}
1788 	if (m_ret != NULL)
1789 		vm_phys_freecnt_adj(m_ret, -npages);
1790 	else {
1791 #if VM_NRESERVLEVEL > 0
1792 		if (vm_reserv_reclaim_contig(npages, low, high, alignment,
1793 		    boundary))
1794 			goto retry;
1795 #endif
1796 	}
1797 	mtx_unlock(&vm_page_queue_free_mtx);
1798 	if (m_ret == NULL)
1799 		return (NULL);
1800 	for (m = m_ret; m < &m_ret[npages]; m++)
1801 		vm_page_alloc_check(m);
1802 
1803 	/*
1804 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1805 	 */
1806 	flags = 0;
1807 	if ((req & VM_ALLOC_ZERO) != 0)
1808 		flags = PG_ZERO;
1809 	if ((req & VM_ALLOC_NODUMP) != 0)
1810 		flags |= PG_NODUMP;
1811 	oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1812 	    VPO_UNMANAGED : 0;
1813 	busy_lock = VPB_UNBUSIED;
1814 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1815 		busy_lock = VPB_SINGLE_EXCLUSIVER;
1816 	if ((req & VM_ALLOC_SBUSY) != 0)
1817 		busy_lock = VPB_SHARERS_WORD(1);
1818 	if ((req & VM_ALLOC_WIRED) != 0)
1819 		atomic_add_int(&vm_cnt.v_wire_count, npages);
1820 	if (object != NULL) {
1821 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1822 		    memattr == VM_MEMATTR_DEFAULT)
1823 			memattr = object->memattr;
1824 	}
1825 	for (m = m_ret; m < &m_ret[npages]; m++) {
1826 		m->aflags = 0;
1827 		m->flags = (m->flags | PG_NODUMP) & flags;
1828 		m->busy_lock = busy_lock;
1829 		if ((req & VM_ALLOC_WIRED) != 0)
1830 			m->wire_count = 1;
1831 		m->act_count = 0;
1832 		m->oflags = oflags;
1833 		if (object != NULL) {
1834 			if (vm_page_insert_after(m, object, pindex, mpred)) {
1835 				pagedaemon_wakeup();
1836 				if ((req & VM_ALLOC_WIRED) != 0)
1837 					atomic_subtract_int(
1838 					    &vm_cnt.v_wire_count, npages);
1839 				KASSERT(m->object == NULL,
1840 				    ("page %p has object", m));
1841 				mpred = m;
1842 				for (m = m_ret; m < &m_ret[npages]; m++) {
1843 					if (m <= mpred &&
1844 					    (req & VM_ALLOC_WIRED) != 0)
1845 						m->wire_count = 0;
1846 					m->oflags = VPO_UNMANAGED;
1847 					m->busy_lock = VPB_UNBUSIED;
1848 					/* Don't change PG_ZERO. */
1849 					vm_page_free_toq(m);
1850 				}
1851 				return (NULL);
1852 			}
1853 			mpred = m;
1854 		} else
1855 			m->pindex = pindex;
1856 		if (memattr != VM_MEMATTR_DEFAULT)
1857 			pmap_page_set_memattr(m, memattr);
1858 		pindex++;
1859 	}
1860 	if (vm_paging_needed())
1861 		pagedaemon_wakeup();
1862 	return (m_ret);
1863 }
1864 
1865 /*
1866  * Check a page that has been freshly dequeued from a freelist.
1867  */
1868 static void
1869 vm_page_alloc_check(vm_page_t m)
1870 {
1871 
1872 	KASSERT(m->object == NULL, ("page %p has object", m));
1873 	KASSERT(m->queue == PQ_NONE,
1874 	    ("page %p has unexpected queue %d", m, m->queue));
1875 	KASSERT(m->wire_count == 0, ("page %p is wired", m));
1876 	KASSERT(m->hold_count == 0, ("page %p is held", m));
1877 	KASSERT(!vm_page_busied(m), ("page %p is busy", m));
1878 	KASSERT(m->dirty == 0, ("page %p is dirty", m));
1879 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1880 	    ("page %p has unexpected memattr %d",
1881 	    m, pmap_page_get_memattr(m)));
1882 	KASSERT(m->valid == 0, ("free page %p is valid", m));
1883 }
1884 
1885 /*
1886  * 	vm_page_alloc_freelist:
1887  *
1888  *	Allocate a physical page from the specified free page list.
1889  *
1890  *	The caller must always specify an allocation class.
1891  *
1892  *	allocation classes:
1893  *	VM_ALLOC_NORMAL		normal process request
1894  *	VM_ALLOC_SYSTEM		system *really* needs a page
1895  *	VM_ALLOC_INTERRUPT	interrupt time request
1896  *
1897  *	optional allocation flags:
1898  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1899  *				intends to allocate
1900  *	VM_ALLOC_WIRED		wire the allocated page
1901  *	VM_ALLOC_ZERO		prefer a zeroed page
1902  *
1903  *	This routine may not sleep.
1904  */
1905 vm_page_t
1906 vm_page_alloc_freelist(int flind, int req)
1907 {
1908 	vm_page_t m;
1909 	u_int flags;
1910 	int req_class;
1911 
1912 	req_class = req & VM_ALLOC_CLASS_MASK;
1913 
1914 	/*
1915 	 * The page daemon is allowed to dig deeper into the free page list.
1916 	 */
1917 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1918 		req_class = VM_ALLOC_SYSTEM;
1919 
1920 	/*
1921 	 * Do not allocate reserved pages unless the req has asked for it.
1922 	 */
1923 	mtx_lock(&vm_page_queue_free_mtx);
1924 	if (vm_cnt.v_free_count > vm_cnt.v_free_reserved ||
1925 	    (req_class == VM_ALLOC_SYSTEM &&
1926 	    vm_cnt.v_free_count > vm_cnt.v_interrupt_free_min) ||
1927 	    (req_class == VM_ALLOC_INTERRUPT &&
1928 	    vm_cnt.v_free_count > 0))
1929 		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0);
1930 	else {
1931 		mtx_unlock(&vm_page_queue_free_mtx);
1932 		atomic_add_int(&vm_pageout_deficit,
1933 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1934 		pagedaemon_wakeup();
1935 		return (NULL);
1936 	}
1937 	if (m == NULL) {
1938 		mtx_unlock(&vm_page_queue_free_mtx);
1939 		return (NULL);
1940 	}
1941 	vm_phys_freecnt_adj(m, -1);
1942 	mtx_unlock(&vm_page_queue_free_mtx);
1943 	vm_page_alloc_check(m);
1944 
1945 	/*
1946 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1947 	 */
1948 	m->aflags = 0;
1949 	flags = 0;
1950 	if ((req & VM_ALLOC_ZERO) != 0)
1951 		flags = PG_ZERO;
1952 	m->flags &= flags;
1953 	if ((req & VM_ALLOC_WIRED) != 0) {
1954 		/*
1955 		 * The page lock is not required for wiring a page that does
1956 		 * not belong to an object.
1957 		 */
1958 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1959 		m->wire_count = 1;
1960 	}
1961 	/* Unmanaged pages don't use "act_count". */
1962 	m->oflags = VPO_UNMANAGED;
1963 	if (vm_paging_needed())
1964 		pagedaemon_wakeup();
1965 	return (m);
1966 }
1967 
1968 #define	VPSC_ANY	0	/* No restrictions. */
1969 #define	VPSC_NORESERV	1	/* Skip reservations; implies VPSC_NOSUPER. */
1970 #define	VPSC_NOSUPER	2	/* Skip superpages. */
1971 
1972 /*
1973  *	vm_page_scan_contig:
1974  *
1975  *	Scan vm_page_array[] between the specified entries "m_start" and
1976  *	"m_end" for a run of contiguous physical pages that satisfy the
1977  *	specified conditions, and return the lowest page in the run.  The
1978  *	specified "alignment" determines the alignment of the lowest physical
1979  *	page in the run.  If the specified "boundary" is non-zero, then the
1980  *	run of physical pages cannot span a physical address that is a
1981  *	multiple of "boundary".
1982  *
1983  *	"m_end" is never dereferenced, so it need not point to a vm_page
1984  *	structure within vm_page_array[].
1985  *
1986  *	"npages" must be greater than zero.  "m_start" and "m_end" must not
1987  *	span a hole (or discontiguity) in the physical address space.  Both
1988  *	"alignment" and "boundary" must be a power of two.
1989  */
1990 vm_page_t
1991 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
1992     u_long alignment, vm_paddr_t boundary, int options)
1993 {
1994 	struct mtx *m_mtx, *new_mtx;
1995 	vm_object_t object;
1996 	vm_paddr_t pa;
1997 	vm_page_t m, m_run;
1998 #if VM_NRESERVLEVEL > 0
1999 	int level;
2000 #endif
2001 	int m_inc, order, run_ext, run_len;
2002 
2003 	KASSERT(npages > 0, ("npages is 0"));
2004 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2005 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2006 	m_run = NULL;
2007 	run_len = 0;
2008 	m_mtx = NULL;
2009 	for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
2010 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2011 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2012 
2013 		/*
2014 		 * If the current page would be the start of a run, check its
2015 		 * physical address against the end, alignment, and boundary
2016 		 * conditions.  If it doesn't satisfy these conditions, either
2017 		 * terminate the scan or advance to the next page that
2018 		 * satisfies the failed condition.
2019 		 */
2020 		if (run_len == 0) {
2021 			KASSERT(m_run == NULL, ("m_run != NULL"));
2022 			if (m + npages > m_end)
2023 				break;
2024 			pa = VM_PAGE_TO_PHYS(m);
2025 			if ((pa & (alignment - 1)) != 0) {
2026 				m_inc = atop(roundup2(pa, alignment) - pa);
2027 				continue;
2028 			}
2029 			if (rounddown2(pa ^ (pa + ptoa(npages) - 1),
2030 			    boundary) != 0) {
2031 				m_inc = atop(roundup2(pa, boundary) - pa);
2032 				continue;
2033 			}
2034 		} else
2035 			KASSERT(m_run != NULL, ("m_run == NULL"));
2036 
2037 		/*
2038 		 * Avoid releasing and reacquiring the same page lock.
2039 		 */
2040 		new_mtx = vm_page_lockptr(m);
2041 		if (m_mtx != new_mtx) {
2042 			if (m_mtx != NULL)
2043 				mtx_unlock(m_mtx);
2044 			m_mtx = new_mtx;
2045 			mtx_lock(m_mtx);
2046 		}
2047 		m_inc = 1;
2048 retry:
2049 		if (m->wire_count != 0 || m->hold_count != 0)
2050 			run_ext = 0;
2051 #if VM_NRESERVLEVEL > 0
2052 		else if ((level = vm_reserv_level(m)) >= 0 &&
2053 		    (options & VPSC_NORESERV) != 0) {
2054 			run_ext = 0;
2055 			/* Advance to the end of the reservation. */
2056 			pa = VM_PAGE_TO_PHYS(m);
2057 			m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
2058 			    pa);
2059 		}
2060 #endif
2061 		else if ((object = m->object) != NULL) {
2062 			/*
2063 			 * The page is considered eligible for relocation if
2064 			 * and only if it could be laundered or reclaimed by
2065 			 * the page daemon.
2066 			 */
2067 			if (!VM_OBJECT_TRYRLOCK(object)) {
2068 				mtx_unlock(m_mtx);
2069 				VM_OBJECT_RLOCK(object);
2070 				mtx_lock(m_mtx);
2071 				if (m->object != object) {
2072 					/*
2073 					 * The page may have been freed.
2074 					 */
2075 					VM_OBJECT_RUNLOCK(object);
2076 					goto retry;
2077 				} else if (m->wire_count != 0 ||
2078 				    m->hold_count != 0) {
2079 					run_ext = 0;
2080 					goto unlock;
2081 				}
2082 			}
2083 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2084 			    ("page %p is PG_UNHOLDFREE", m));
2085 			/* Don't care: PG_NODUMP, PG_ZERO. */
2086 			if (object->type != OBJT_DEFAULT &&
2087 			    object->type != OBJT_SWAP &&
2088 			    object->type != OBJT_VNODE) {
2089 				run_ext = 0;
2090 #if VM_NRESERVLEVEL > 0
2091 			} else if ((options & VPSC_NOSUPER) != 0 &&
2092 			    (level = vm_reserv_level_iffullpop(m)) >= 0) {
2093 				run_ext = 0;
2094 				/* Advance to the end of the superpage. */
2095 				pa = VM_PAGE_TO_PHYS(m);
2096 				m_inc = atop(roundup2(pa + 1,
2097 				    vm_reserv_size(level)) - pa);
2098 #endif
2099 			} else if (object->memattr == VM_MEMATTR_DEFAULT &&
2100 			    m->queue != PQ_NONE && !vm_page_busied(m)) {
2101 				/*
2102 				 * The page is allocated but eligible for
2103 				 * relocation.  Extend the current run by one
2104 				 * page.
2105 				 */
2106 				KASSERT(pmap_page_get_memattr(m) ==
2107 				    VM_MEMATTR_DEFAULT,
2108 				    ("page %p has an unexpected memattr", m));
2109 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2110 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2111 				    ("page %p has unexpected oflags", m));
2112 				/* Don't care: VPO_NOSYNC. */
2113 				run_ext = 1;
2114 			} else
2115 				run_ext = 0;
2116 unlock:
2117 			VM_OBJECT_RUNLOCK(object);
2118 #if VM_NRESERVLEVEL > 0
2119 		} else if (level >= 0) {
2120 			/*
2121 			 * The page is reserved but not yet allocated.  In
2122 			 * other words, it is still free.  Extend the current
2123 			 * run by one page.
2124 			 */
2125 			run_ext = 1;
2126 #endif
2127 		} else if ((order = m->order) < VM_NFREEORDER) {
2128 			/*
2129 			 * The page is enqueued in the physical memory
2130 			 * allocator's free page queues.  Moreover, it is the
2131 			 * first page in a power-of-two-sized run of
2132 			 * contiguous free pages.  Add these pages to the end
2133 			 * of the current run, and jump ahead.
2134 			 */
2135 			run_ext = 1 << order;
2136 			m_inc = 1 << order;
2137 		} else {
2138 			/*
2139 			 * Skip the page for one of the following reasons: (1)
2140 			 * It is enqueued in the physical memory allocator's
2141 			 * free page queues.  However, it is not the first
2142 			 * page in a run of contiguous free pages.  (This case
2143 			 * rarely occurs because the scan is performed in
2144 			 * ascending order.) (2) It is not reserved, and it is
2145 			 * transitioning from free to allocated.  (Conversely,
2146 			 * the transition from allocated to free for managed
2147 			 * pages is blocked by the page lock.) (3) It is
2148 			 * allocated but not contained by an object and not
2149 			 * wired, e.g., allocated by Xen's balloon driver.
2150 			 */
2151 			run_ext = 0;
2152 		}
2153 
2154 		/*
2155 		 * Extend or reset the current run of pages.
2156 		 */
2157 		if (run_ext > 0) {
2158 			if (run_len == 0)
2159 				m_run = m;
2160 			run_len += run_ext;
2161 		} else {
2162 			if (run_len > 0) {
2163 				m_run = NULL;
2164 				run_len = 0;
2165 			}
2166 		}
2167 	}
2168 	if (m_mtx != NULL)
2169 		mtx_unlock(m_mtx);
2170 	if (run_len >= npages)
2171 		return (m_run);
2172 	return (NULL);
2173 }
2174 
2175 /*
2176  *	vm_page_reclaim_run:
2177  *
2178  *	Try to relocate each of the allocated virtual pages within the
2179  *	specified run of physical pages to a new physical address.  Free the
2180  *	physical pages underlying the relocated virtual pages.  A virtual page
2181  *	is relocatable if and only if it could be laundered or reclaimed by
2182  *	the page daemon.  Whenever possible, a virtual page is relocated to a
2183  *	physical address above "high".
2184  *
2185  *	Returns 0 if every physical page within the run was already free or
2186  *	just freed by a successful relocation.  Otherwise, returns a non-zero
2187  *	value indicating why the last attempt to relocate a virtual page was
2188  *	unsuccessful.
2189  *
2190  *	"req_class" must be an allocation class.
2191  */
2192 static int
2193 vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
2194     vm_paddr_t high)
2195 {
2196 	struct mtx *m_mtx, *new_mtx;
2197 	struct spglist free;
2198 	vm_object_t object;
2199 	vm_paddr_t pa;
2200 	vm_page_t m, m_end, m_new;
2201 	int error, order, req;
2202 
2203 	KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
2204 	    ("req_class is not an allocation class"));
2205 	SLIST_INIT(&free);
2206 	error = 0;
2207 	m = m_run;
2208 	m_end = m_run + npages;
2209 	m_mtx = NULL;
2210 	for (; error == 0 && m < m_end; m++) {
2211 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2212 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2213 
2214 		/*
2215 		 * Avoid releasing and reacquiring the same page lock.
2216 		 */
2217 		new_mtx = vm_page_lockptr(m);
2218 		if (m_mtx != new_mtx) {
2219 			if (m_mtx != NULL)
2220 				mtx_unlock(m_mtx);
2221 			m_mtx = new_mtx;
2222 			mtx_lock(m_mtx);
2223 		}
2224 retry:
2225 		if (m->wire_count != 0 || m->hold_count != 0)
2226 			error = EBUSY;
2227 		else if ((object = m->object) != NULL) {
2228 			/*
2229 			 * The page is relocated if and only if it could be
2230 			 * laundered or reclaimed by the page daemon.
2231 			 */
2232 			if (!VM_OBJECT_TRYWLOCK(object)) {
2233 				mtx_unlock(m_mtx);
2234 				VM_OBJECT_WLOCK(object);
2235 				mtx_lock(m_mtx);
2236 				if (m->object != object) {
2237 					/*
2238 					 * The page may have been freed.
2239 					 */
2240 					VM_OBJECT_WUNLOCK(object);
2241 					goto retry;
2242 				} else if (m->wire_count != 0 ||
2243 				    m->hold_count != 0) {
2244 					error = EBUSY;
2245 					goto unlock;
2246 				}
2247 			}
2248 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2249 			    ("page %p is PG_UNHOLDFREE", m));
2250 			/* Don't care: PG_NODUMP, PG_ZERO. */
2251 			if (object->type != OBJT_DEFAULT &&
2252 			    object->type != OBJT_SWAP &&
2253 			    object->type != OBJT_VNODE)
2254 				error = EINVAL;
2255 			else if (object->memattr != VM_MEMATTR_DEFAULT)
2256 				error = EINVAL;
2257 			else if (m->queue != PQ_NONE && !vm_page_busied(m)) {
2258 				KASSERT(pmap_page_get_memattr(m) ==
2259 				    VM_MEMATTR_DEFAULT,
2260 				    ("page %p has an unexpected memattr", m));
2261 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2262 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2263 				    ("page %p has unexpected oflags", m));
2264 				/* Don't care: VPO_NOSYNC. */
2265 				if (m->valid != 0) {
2266 					/*
2267 					 * First, try to allocate a new page
2268 					 * that is above "high".  Failing
2269 					 * that, try to allocate a new page
2270 					 * that is below "m_run".  Allocate
2271 					 * the new page between the end of
2272 					 * "m_run" and "high" only as a last
2273 					 * resort.
2274 					 */
2275 					req = req_class | VM_ALLOC_NOOBJ;
2276 					if ((m->flags & PG_NODUMP) != 0)
2277 						req |= VM_ALLOC_NODUMP;
2278 					if (trunc_page(high) !=
2279 					    ~(vm_paddr_t)PAGE_MASK) {
2280 						m_new = vm_page_alloc_contig(
2281 						    NULL, 0, req, 1,
2282 						    round_page(high),
2283 						    ~(vm_paddr_t)0,
2284 						    PAGE_SIZE, 0,
2285 						    VM_MEMATTR_DEFAULT);
2286 					} else
2287 						m_new = NULL;
2288 					if (m_new == NULL) {
2289 						pa = VM_PAGE_TO_PHYS(m_run);
2290 						m_new = vm_page_alloc_contig(
2291 						    NULL, 0, req, 1,
2292 						    0, pa - 1, PAGE_SIZE, 0,
2293 						    VM_MEMATTR_DEFAULT);
2294 					}
2295 					if (m_new == NULL) {
2296 						pa += ptoa(npages);
2297 						m_new = vm_page_alloc_contig(
2298 						    NULL, 0, req, 1,
2299 						    pa, high, PAGE_SIZE, 0,
2300 						    VM_MEMATTR_DEFAULT);
2301 					}
2302 					if (m_new == NULL) {
2303 						error = ENOMEM;
2304 						goto unlock;
2305 					}
2306 					KASSERT(m_new->wire_count == 0,
2307 					    ("page %p is wired", m));
2308 
2309 					/*
2310 					 * Replace "m" with the new page.  For
2311 					 * vm_page_replace(), "m" must be busy
2312 					 * and dequeued.  Finally, change "m"
2313 					 * as if vm_page_free() was called.
2314 					 */
2315 					if (object->ref_count != 0)
2316 						pmap_remove_all(m);
2317 					m_new->aflags = m->aflags;
2318 					KASSERT(m_new->oflags == VPO_UNMANAGED,
2319 					    ("page %p is managed", m));
2320 					m_new->oflags = m->oflags & VPO_NOSYNC;
2321 					pmap_copy_page(m, m_new);
2322 					m_new->valid = m->valid;
2323 					m_new->dirty = m->dirty;
2324 					m->flags &= ~PG_ZERO;
2325 					vm_page_xbusy(m);
2326 					vm_page_remque(m);
2327 					vm_page_replace_checked(m_new, object,
2328 					    m->pindex, m);
2329 					m->valid = 0;
2330 					vm_page_undirty(m);
2331 
2332 					/*
2333 					 * The new page must be deactivated
2334 					 * before the object is unlocked.
2335 					 */
2336 					new_mtx = vm_page_lockptr(m_new);
2337 					if (m_mtx != new_mtx) {
2338 						mtx_unlock(m_mtx);
2339 						m_mtx = new_mtx;
2340 						mtx_lock(m_mtx);
2341 					}
2342 					vm_page_deactivate(m_new);
2343 				} else {
2344 					m->flags &= ~PG_ZERO;
2345 					vm_page_remque(m);
2346 					vm_page_remove(m);
2347 					KASSERT(m->dirty == 0,
2348 					    ("page %p is dirty", m));
2349 				}
2350 				SLIST_INSERT_HEAD(&free, m, plinks.s.ss);
2351 			} else
2352 				error = EBUSY;
2353 unlock:
2354 			VM_OBJECT_WUNLOCK(object);
2355 		} else {
2356 			mtx_lock(&vm_page_queue_free_mtx);
2357 			order = m->order;
2358 			if (order < VM_NFREEORDER) {
2359 				/*
2360 				 * The page is enqueued in the physical memory
2361 				 * allocator's free page queues.  Moreover, it
2362 				 * is the first page in a power-of-two-sized
2363 				 * run of contiguous free pages.  Jump ahead
2364 				 * to the last page within that run, and
2365 				 * continue from there.
2366 				 */
2367 				m += (1 << order) - 1;
2368 			}
2369 #if VM_NRESERVLEVEL > 0
2370 			else if (vm_reserv_is_page_free(m))
2371 				order = 0;
2372 #endif
2373 			mtx_unlock(&vm_page_queue_free_mtx);
2374 			if (order == VM_NFREEORDER)
2375 				error = EINVAL;
2376 		}
2377 	}
2378 	if (m_mtx != NULL)
2379 		mtx_unlock(m_mtx);
2380 	if ((m = SLIST_FIRST(&free)) != NULL) {
2381 		mtx_lock(&vm_page_queue_free_mtx);
2382 		do {
2383 			SLIST_REMOVE_HEAD(&free, plinks.s.ss);
2384 			vm_phys_freecnt_adj(m, 1);
2385 #if VM_NRESERVLEVEL > 0
2386 			if (!vm_reserv_free_page(m))
2387 #else
2388 			if (true)
2389 #endif
2390 				vm_phys_free_pages(m, 0);
2391 		} while ((m = SLIST_FIRST(&free)) != NULL);
2392 		vm_page_free_wakeup();
2393 		mtx_unlock(&vm_page_queue_free_mtx);
2394 	}
2395 	return (error);
2396 }
2397 
2398 #define	NRUNS	16
2399 
2400 CTASSERT(powerof2(NRUNS));
2401 
2402 #define	RUN_INDEX(count)	((count) & (NRUNS - 1))
2403 
2404 #define	MIN_RECLAIM	8
2405 
2406 /*
2407  *	vm_page_reclaim_contig:
2408  *
2409  *	Reclaim allocated, contiguous physical memory satisfying the specified
2410  *	conditions by relocating the virtual pages using that physical memory.
2411  *	Returns true if reclamation is successful and false otherwise.  Since
2412  *	relocation requires the allocation of physical pages, reclamation may
2413  *	fail due to a shortage of free pages.  When reclamation fails, callers
2414  *	are expected to perform VM_WAIT before retrying a failed allocation
2415  *	operation, e.g., vm_page_alloc_contig().
2416  *
2417  *	The caller must always specify an allocation class through "req".
2418  *
2419  *	allocation classes:
2420  *	VM_ALLOC_NORMAL		normal process request
2421  *	VM_ALLOC_SYSTEM		system *really* needs a page
2422  *	VM_ALLOC_INTERRUPT	interrupt time request
2423  *
2424  *	The optional allocation flags are ignored.
2425  *
2426  *	"npages" must be greater than zero.  Both "alignment" and "boundary"
2427  *	must be a power of two.
2428  */
2429 bool
2430 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
2431     u_long alignment, vm_paddr_t boundary)
2432 {
2433 	vm_paddr_t curr_low;
2434 	vm_page_t m_run, m_runs[NRUNS];
2435 	u_long count, reclaimed;
2436 	int error, i, options, req_class;
2437 
2438 	KASSERT(npages > 0, ("npages is 0"));
2439 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2440 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2441 	req_class = req & VM_ALLOC_CLASS_MASK;
2442 
2443 	/*
2444 	 * The page daemon is allowed to dig deeper into the free page list.
2445 	 */
2446 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2447 		req_class = VM_ALLOC_SYSTEM;
2448 
2449 	/*
2450 	 * Return if the number of free pages cannot satisfy the requested
2451 	 * allocation.
2452 	 */
2453 	count = vm_cnt.v_free_count;
2454 	if (count < npages + vm_cnt.v_free_reserved || (count < npages +
2455 	    vm_cnt.v_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
2456 	    (count < npages && req_class == VM_ALLOC_INTERRUPT))
2457 		return (false);
2458 
2459 	/*
2460 	 * Scan up to three times, relaxing the restrictions ("options") on
2461 	 * the reclamation of reservations and superpages each time.
2462 	 */
2463 	for (options = VPSC_NORESERV;;) {
2464 		/*
2465 		 * Find the highest runs that satisfy the given constraints
2466 		 * and restrictions, and record them in "m_runs".
2467 		 */
2468 		curr_low = low;
2469 		count = 0;
2470 		for (;;) {
2471 			m_run = vm_phys_scan_contig(npages, curr_low, high,
2472 			    alignment, boundary, options);
2473 			if (m_run == NULL)
2474 				break;
2475 			curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages);
2476 			m_runs[RUN_INDEX(count)] = m_run;
2477 			count++;
2478 		}
2479 
2480 		/*
2481 		 * Reclaim the highest runs in LIFO (descending) order until
2482 		 * the number of reclaimed pages, "reclaimed", is at least
2483 		 * MIN_RECLAIM.  Reset "reclaimed" each time because each
2484 		 * reclamation is idempotent, and runs will (likely) recur
2485 		 * from one scan to the next as restrictions are relaxed.
2486 		 */
2487 		reclaimed = 0;
2488 		for (i = 0; count > 0 && i < NRUNS; i++) {
2489 			count--;
2490 			m_run = m_runs[RUN_INDEX(count)];
2491 			error = vm_page_reclaim_run(req_class, npages, m_run,
2492 			    high);
2493 			if (error == 0) {
2494 				reclaimed += npages;
2495 				if (reclaimed >= MIN_RECLAIM)
2496 					return (true);
2497 			}
2498 		}
2499 
2500 		/*
2501 		 * Either relax the restrictions on the next scan or return if
2502 		 * the last scan had no restrictions.
2503 		 */
2504 		if (options == VPSC_NORESERV)
2505 			options = VPSC_NOSUPER;
2506 		else if (options == VPSC_NOSUPER)
2507 			options = VPSC_ANY;
2508 		else if (options == VPSC_ANY)
2509 			return (reclaimed != 0);
2510 	}
2511 }
2512 
2513 /*
2514  *	vm_wait:	(also see VM_WAIT macro)
2515  *
2516  *	Sleep until free pages are available for allocation.
2517  *	- Called in various places before memory allocations.
2518  */
2519 void
2520 vm_wait(void)
2521 {
2522 
2523 	mtx_lock(&vm_page_queue_free_mtx);
2524 	if (curproc == pageproc) {
2525 		vm_pageout_pages_needed = 1;
2526 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
2527 		    PDROP | PSWP, "VMWait", 0);
2528 	} else {
2529 		if (__predict_false(pageproc == NULL))
2530 			panic("vm_wait in early boot");
2531 		if (!vm_pageout_wanted) {
2532 			vm_pageout_wanted = true;
2533 			wakeup(&vm_pageout_wanted);
2534 		}
2535 		vm_pages_needed = true;
2536 		msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
2537 		    "vmwait", 0);
2538 	}
2539 }
2540 
2541 /*
2542  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
2543  *
2544  *	Sleep until free pages are available for allocation.
2545  *	- Called only in vm_fault so that processes page faulting
2546  *	  can be easily tracked.
2547  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2548  *	  processes will be able to grab memory first.  Do not change
2549  *	  this balance without careful testing first.
2550  */
2551 void
2552 vm_waitpfault(void)
2553 {
2554 
2555 	mtx_lock(&vm_page_queue_free_mtx);
2556 	if (!vm_pageout_wanted) {
2557 		vm_pageout_wanted = true;
2558 		wakeup(&vm_pageout_wanted);
2559 	}
2560 	vm_pages_needed = true;
2561 	msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
2562 	    "pfault", 0);
2563 }
2564 
2565 struct vm_pagequeue *
2566 vm_page_pagequeue(vm_page_t m)
2567 {
2568 
2569 	if (vm_page_in_laundry(m))
2570 		return (&vm_dom[0].vmd_pagequeues[m->queue]);
2571 	else
2572 		return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2573 }
2574 
2575 /*
2576  *	vm_page_dequeue:
2577  *
2578  *	Remove the given page from its current page queue.
2579  *
2580  *	The page must be locked.
2581  */
2582 void
2583 vm_page_dequeue(vm_page_t m)
2584 {
2585 	struct vm_pagequeue *pq;
2586 
2587 	vm_page_assert_locked(m);
2588 	KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2589 	    m));
2590 	pq = vm_page_pagequeue(m);
2591 	vm_pagequeue_lock(pq);
2592 	m->queue = PQ_NONE;
2593 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2594 	vm_pagequeue_cnt_dec(pq);
2595 	vm_pagequeue_unlock(pq);
2596 }
2597 
2598 /*
2599  *	vm_page_dequeue_locked:
2600  *
2601  *	Remove the given page from its current page queue.
2602  *
2603  *	The page and page queue must be locked.
2604  */
2605 void
2606 vm_page_dequeue_locked(vm_page_t m)
2607 {
2608 	struct vm_pagequeue *pq;
2609 
2610 	vm_page_lock_assert(m, MA_OWNED);
2611 	pq = vm_page_pagequeue(m);
2612 	vm_pagequeue_assert_locked(pq);
2613 	m->queue = PQ_NONE;
2614 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2615 	vm_pagequeue_cnt_dec(pq);
2616 }
2617 
2618 /*
2619  *	vm_page_enqueue:
2620  *
2621  *	Add the given page to the specified page queue.
2622  *
2623  *	The page must be locked.
2624  */
2625 static void
2626 vm_page_enqueue(uint8_t queue, vm_page_t m)
2627 {
2628 	struct vm_pagequeue *pq;
2629 
2630 	vm_page_lock_assert(m, MA_OWNED);
2631 	KASSERT(queue < PQ_COUNT,
2632 	    ("vm_page_enqueue: invalid queue %u request for page %p",
2633 	    queue, m));
2634 	if (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE)
2635 		pq = &vm_dom[0].vmd_pagequeues[queue];
2636 	else
2637 		pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2638 	vm_pagequeue_lock(pq);
2639 	m->queue = queue;
2640 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2641 	vm_pagequeue_cnt_inc(pq);
2642 	vm_pagequeue_unlock(pq);
2643 }
2644 
2645 /*
2646  *	vm_page_requeue:
2647  *
2648  *	Move the given page to the tail of its current page queue.
2649  *
2650  *	The page must be locked.
2651  */
2652 void
2653 vm_page_requeue(vm_page_t m)
2654 {
2655 	struct vm_pagequeue *pq;
2656 
2657 	vm_page_lock_assert(m, MA_OWNED);
2658 	KASSERT(m->queue != PQ_NONE,
2659 	    ("vm_page_requeue: page %p is not queued", m));
2660 	pq = vm_page_pagequeue(m);
2661 	vm_pagequeue_lock(pq);
2662 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2663 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2664 	vm_pagequeue_unlock(pq);
2665 }
2666 
2667 /*
2668  *	vm_page_requeue_locked:
2669  *
2670  *	Move the given page to the tail of its current page queue.
2671  *
2672  *	The page queue must be locked.
2673  */
2674 void
2675 vm_page_requeue_locked(vm_page_t m)
2676 {
2677 	struct vm_pagequeue *pq;
2678 
2679 	KASSERT(m->queue != PQ_NONE,
2680 	    ("vm_page_requeue_locked: page %p is not queued", m));
2681 	pq = vm_page_pagequeue(m);
2682 	vm_pagequeue_assert_locked(pq);
2683 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2684 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2685 }
2686 
2687 /*
2688  *	vm_page_activate:
2689  *
2690  *	Put the specified page on the active list (if appropriate).
2691  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2692  *	mess with it.
2693  *
2694  *	The page must be locked.
2695  */
2696 void
2697 vm_page_activate(vm_page_t m)
2698 {
2699 	int queue;
2700 
2701 	vm_page_lock_assert(m, MA_OWNED);
2702 	if ((queue = m->queue) != PQ_ACTIVE) {
2703 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2704 			if (m->act_count < ACT_INIT)
2705 				m->act_count = ACT_INIT;
2706 			if (queue != PQ_NONE)
2707 				vm_page_dequeue(m);
2708 			vm_page_enqueue(PQ_ACTIVE, m);
2709 		} else
2710 			KASSERT(queue == PQ_NONE,
2711 			    ("vm_page_activate: wired page %p is queued", m));
2712 	} else {
2713 		if (m->act_count < ACT_INIT)
2714 			m->act_count = ACT_INIT;
2715 	}
2716 }
2717 
2718 /*
2719  *	vm_page_free_wakeup:
2720  *
2721  *	Helper routine for vm_page_free_toq().  This routine is called
2722  *	when a page is added to the free queues.
2723  *
2724  *	The page queues must be locked.
2725  */
2726 static inline void
2727 vm_page_free_wakeup(void)
2728 {
2729 
2730 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2731 	/*
2732 	 * if pageout daemon needs pages, then tell it that there are
2733 	 * some free.
2734 	 */
2735 	if (vm_pageout_pages_needed &&
2736 	    vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) {
2737 		wakeup(&vm_pageout_pages_needed);
2738 		vm_pageout_pages_needed = 0;
2739 	}
2740 	/*
2741 	 * wakeup processes that are waiting on memory if we hit a
2742 	 * high water mark. And wakeup scheduler process if we have
2743 	 * lots of memory. this process will swapin processes.
2744 	 */
2745 	if (vm_pages_needed && !vm_page_count_min()) {
2746 		vm_pages_needed = false;
2747 		wakeup(&vm_cnt.v_free_count);
2748 	}
2749 }
2750 
2751 /*
2752  *	vm_page_free_toq:
2753  *
2754  *	Returns the given page to the free list,
2755  *	disassociating it with any VM object.
2756  *
2757  *	The object must be locked.  The page must be locked if it is managed.
2758  */
2759 void
2760 vm_page_free_toq(vm_page_t m)
2761 {
2762 
2763 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2764 		vm_page_lock_assert(m, MA_OWNED);
2765 		KASSERT(!pmap_page_is_mapped(m),
2766 		    ("vm_page_free_toq: freeing mapped page %p", m));
2767 	} else
2768 		KASSERT(m->queue == PQ_NONE,
2769 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2770 	VM_CNT_INC(v_tfree);
2771 
2772 	if (vm_page_sbusied(m))
2773 		panic("vm_page_free: freeing busy page %p", m);
2774 
2775 	/*
2776 	 * Unqueue, then remove page.  Note that we cannot destroy
2777 	 * the page here because we do not want to call the pager's
2778 	 * callback routine until after we've put the page on the
2779 	 * appropriate free queue.
2780 	 */
2781 	vm_page_remque(m);
2782 	vm_page_remove(m);
2783 
2784 	/*
2785 	 * If fictitious remove object association and
2786 	 * return, otherwise delay object association removal.
2787 	 */
2788 	if ((m->flags & PG_FICTITIOUS) != 0) {
2789 		return;
2790 	}
2791 
2792 	m->valid = 0;
2793 	vm_page_undirty(m);
2794 
2795 	if (m->wire_count != 0)
2796 		panic("vm_page_free: freeing wired page %p", m);
2797 	if (m->hold_count != 0) {
2798 		m->flags &= ~PG_ZERO;
2799 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2800 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
2801 		m->flags |= PG_UNHOLDFREE;
2802 	} else {
2803 		/*
2804 		 * Restore the default memory attribute to the page.
2805 		 */
2806 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2807 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2808 
2809 		/*
2810 		 * Insert the page into the physical memory allocator's free
2811 		 * page queues.
2812 		 */
2813 		mtx_lock(&vm_page_queue_free_mtx);
2814 		vm_phys_freecnt_adj(m, 1);
2815 #if VM_NRESERVLEVEL > 0
2816 		if (!vm_reserv_free_page(m))
2817 #else
2818 		if (TRUE)
2819 #endif
2820 			vm_phys_free_pages(m, 0);
2821 		vm_page_free_wakeup();
2822 		mtx_unlock(&vm_page_queue_free_mtx);
2823 	}
2824 }
2825 
2826 /*
2827  *	vm_page_wire:
2828  *
2829  *	Mark this page as wired down by yet
2830  *	another map, removing it from paging queues
2831  *	as necessary.
2832  *
2833  *	If the page is fictitious, then its wire count must remain one.
2834  *
2835  *	The page must be locked.
2836  */
2837 void
2838 vm_page_wire(vm_page_t m)
2839 {
2840 
2841 	/*
2842 	 * Only bump the wire statistics if the page is not already wired,
2843 	 * and only unqueue the page if it is on some queue (if it is unmanaged
2844 	 * it is already off the queues).
2845 	 */
2846 	vm_page_lock_assert(m, MA_OWNED);
2847 	if ((m->flags & PG_FICTITIOUS) != 0) {
2848 		KASSERT(m->wire_count == 1,
2849 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
2850 		    m));
2851 		return;
2852 	}
2853 	if (m->wire_count == 0) {
2854 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
2855 		    m->queue == PQ_NONE,
2856 		    ("vm_page_wire: unmanaged page %p is queued", m));
2857 		vm_page_remque(m);
2858 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2859 	}
2860 	m->wire_count++;
2861 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
2862 }
2863 
2864 /*
2865  * vm_page_unwire:
2866  *
2867  * Release one wiring of the specified page, potentially allowing it to be
2868  * paged out.  Returns TRUE if the number of wirings transitions to zero and
2869  * FALSE otherwise.
2870  *
2871  * Only managed pages belonging to an object can be paged out.  If the number
2872  * of wirings transitions to zero and the page is eligible for page out, then
2873  * the page is added to the specified paging queue (unless PQ_NONE is
2874  * specified).
2875  *
2876  * If a page is fictitious, then its wire count must always be one.
2877  *
2878  * A managed page must be locked.
2879  */
2880 boolean_t
2881 vm_page_unwire(vm_page_t m, uint8_t queue)
2882 {
2883 
2884 	KASSERT(queue < PQ_COUNT || queue == PQ_NONE,
2885 	    ("vm_page_unwire: invalid queue %u request for page %p",
2886 	    queue, m));
2887 	if ((m->oflags & VPO_UNMANAGED) == 0)
2888 		vm_page_assert_locked(m);
2889 	if ((m->flags & PG_FICTITIOUS) != 0) {
2890 		KASSERT(m->wire_count == 1,
2891 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
2892 		return (FALSE);
2893 	}
2894 	if (m->wire_count > 0) {
2895 		m->wire_count--;
2896 		if (m->wire_count == 0) {
2897 			atomic_subtract_int(&vm_cnt.v_wire_count, 1);
2898 			if ((m->oflags & VPO_UNMANAGED) == 0 &&
2899 			    m->object != NULL && queue != PQ_NONE)
2900 				vm_page_enqueue(queue, m);
2901 			return (TRUE);
2902 		} else
2903 			return (FALSE);
2904 	} else
2905 		panic("vm_page_unwire: page %p's wire count is zero", m);
2906 }
2907 
2908 /*
2909  * Move the specified page to the inactive queue.
2910  *
2911  * Normally, "noreuse" is FALSE, resulting in LRU ordering of the inactive
2912  * queue.  However, setting "noreuse" to TRUE will accelerate the specified
2913  * page's reclamation, but it will not unmap the page from any address space.
2914  * This is implemented by inserting the page near the head of the inactive
2915  * queue, using a marker page to guide FIFO insertion ordering.
2916  *
2917  * The page must be locked.
2918  */
2919 static inline void
2920 _vm_page_deactivate(vm_page_t m, boolean_t noreuse)
2921 {
2922 	struct vm_pagequeue *pq;
2923 	int queue;
2924 
2925 	vm_page_assert_locked(m);
2926 
2927 	/*
2928 	 * Ignore if the page is already inactive, unless it is unlikely to be
2929 	 * reactivated.
2930 	 */
2931 	if ((queue = m->queue) == PQ_INACTIVE && !noreuse)
2932 		return;
2933 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2934 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
2935 		/* Avoid multiple acquisitions of the inactive queue lock. */
2936 		if (queue == PQ_INACTIVE) {
2937 			vm_pagequeue_lock(pq);
2938 			vm_page_dequeue_locked(m);
2939 		} else {
2940 			if (queue != PQ_NONE)
2941 				vm_page_dequeue(m);
2942 			vm_pagequeue_lock(pq);
2943 		}
2944 		m->queue = PQ_INACTIVE;
2945 		if (noreuse)
2946 			TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead,
2947 			    m, plinks.q);
2948 		else
2949 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2950 		vm_pagequeue_cnt_inc(pq);
2951 		vm_pagequeue_unlock(pq);
2952 	}
2953 }
2954 
2955 /*
2956  * Move the specified page to the inactive queue.
2957  *
2958  * The page must be locked.
2959  */
2960 void
2961 vm_page_deactivate(vm_page_t m)
2962 {
2963 
2964 	_vm_page_deactivate(m, FALSE);
2965 }
2966 
2967 /*
2968  * Move the specified page to the inactive queue with the expectation
2969  * that it is unlikely to be reused.
2970  *
2971  * The page must be locked.
2972  */
2973 void
2974 vm_page_deactivate_noreuse(vm_page_t m)
2975 {
2976 
2977 	_vm_page_deactivate(m, TRUE);
2978 }
2979 
2980 /*
2981  * vm_page_launder
2982  *
2983  * 	Put a page in the laundry.
2984  */
2985 void
2986 vm_page_launder(vm_page_t m)
2987 {
2988 	int queue;
2989 
2990 	vm_page_assert_locked(m);
2991 	if ((queue = m->queue) != PQ_LAUNDRY) {
2992 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2993 			if (queue != PQ_NONE)
2994 				vm_page_dequeue(m);
2995 			vm_page_enqueue(PQ_LAUNDRY, m);
2996 		} else
2997 			KASSERT(queue == PQ_NONE,
2998 			    ("wired page %p is queued", m));
2999 	}
3000 }
3001 
3002 /*
3003  * vm_page_unswappable
3004  *
3005  *	Put a page in the PQ_UNSWAPPABLE holding queue.
3006  */
3007 void
3008 vm_page_unswappable(vm_page_t m)
3009 {
3010 
3011 	vm_page_assert_locked(m);
3012 	KASSERT(m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0,
3013 	    ("page %p already unswappable", m));
3014 	if (m->queue != PQ_NONE)
3015 		vm_page_dequeue(m);
3016 	vm_page_enqueue(PQ_UNSWAPPABLE, m);
3017 }
3018 
3019 /*
3020  * vm_page_try_to_free()
3021  *
3022  *	Attempt to free the page.  If we cannot free it, we do nothing.
3023  *	1 is returned on success, 0 on failure.
3024  */
3025 int
3026 vm_page_try_to_free(vm_page_t m)
3027 {
3028 
3029 	vm_page_lock_assert(m, MA_OWNED);
3030 	if (m->object != NULL)
3031 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3032 	if (m->dirty || m->hold_count || m->wire_count ||
3033 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
3034 		return (0);
3035 	pmap_remove_all(m);
3036 	if (m->dirty)
3037 		return (0);
3038 	vm_page_free(m);
3039 	return (1);
3040 }
3041 
3042 /*
3043  * vm_page_advise
3044  *
3045  * 	Apply the specified advice to the given page.
3046  *
3047  *	The object and page must be locked.
3048  */
3049 void
3050 vm_page_advise(vm_page_t m, int advice)
3051 {
3052 
3053 	vm_page_assert_locked(m);
3054 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3055 	if (advice == MADV_FREE)
3056 		/*
3057 		 * Mark the page clean.  This will allow the page to be freed
3058 		 * without first paging it out.  MADV_FREE pages are often
3059 		 * quickly reused by malloc(3), so we do not do anything that
3060 		 * would result in a page fault on a later access.
3061 		 */
3062 		vm_page_undirty(m);
3063 	else if (advice != MADV_DONTNEED) {
3064 		if (advice == MADV_WILLNEED)
3065 			vm_page_activate(m);
3066 		return;
3067 	}
3068 
3069 	/*
3070 	 * Clear any references to the page.  Otherwise, the page daemon will
3071 	 * immediately reactivate the page.
3072 	 */
3073 	vm_page_aflag_clear(m, PGA_REFERENCED);
3074 
3075 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
3076 		vm_page_dirty(m);
3077 
3078 	/*
3079 	 * Place clean pages near the head of the inactive queue rather than
3080 	 * the tail, thus defeating the queue's LRU operation and ensuring that
3081 	 * the page will be reused quickly.  Dirty pages not already in the
3082 	 * laundry are moved there.
3083 	 */
3084 	if (m->dirty == 0)
3085 		vm_page_deactivate_noreuse(m);
3086 	else
3087 		vm_page_launder(m);
3088 }
3089 
3090 /*
3091  * Grab a page, waiting until we are waken up due to the page
3092  * changing state.  We keep on waiting, if the page continues
3093  * to be in the object.  If the page doesn't exist, first allocate it
3094  * and then conditionally zero it.
3095  *
3096  * This routine may sleep.
3097  *
3098  * The object must be locked on entry.  The lock will, however, be released
3099  * and reacquired if the routine sleeps.
3100  */
3101 vm_page_t
3102 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
3103 {
3104 	vm_page_t m;
3105 	int sleep;
3106 
3107 	VM_OBJECT_ASSERT_WLOCKED(object);
3108 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3109 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3110 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
3111 retrylookup:
3112 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
3113 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3114 		    vm_page_xbusied(m) : vm_page_busied(m);
3115 		if (sleep) {
3116 			if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3117 				return (NULL);
3118 			/*
3119 			 * Reference the page before unlocking and
3120 			 * sleeping so that the page daemon is less
3121 			 * likely to reclaim it.
3122 			 */
3123 			vm_page_aflag_set(m, PGA_REFERENCED);
3124 			vm_page_lock(m);
3125 			VM_OBJECT_WUNLOCK(object);
3126 			vm_page_busy_sleep(m, "pgrbwt", (allocflags &
3127 			    VM_ALLOC_IGN_SBUSY) != 0);
3128 			VM_OBJECT_WLOCK(object);
3129 			goto retrylookup;
3130 		} else {
3131 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
3132 				vm_page_lock(m);
3133 				vm_page_wire(m);
3134 				vm_page_unlock(m);
3135 			}
3136 			if ((allocflags &
3137 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
3138 				vm_page_xbusy(m);
3139 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
3140 				vm_page_sbusy(m);
3141 			return (m);
3142 		}
3143 	}
3144 	m = vm_page_alloc(object, pindex, allocflags);
3145 	if (m == NULL) {
3146 		if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3147 			return (NULL);
3148 		VM_OBJECT_WUNLOCK(object);
3149 		VM_WAIT;
3150 		VM_OBJECT_WLOCK(object);
3151 		goto retrylookup;
3152 	}
3153 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
3154 		pmap_zero_page(m);
3155 	return (m);
3156 }
3157 
3158 /*
3159  * Return the specified range of pages from the given object.  For each
3160  * page offset within the range, if a page already exists within the object
3161  * at that offset and it is busy, then wait for it to change state.  If,
3162  * instead, the page doesn't exist, then allocate it.
3163  *
3164  * The caller must always specify an allocation class.
3165  *
3166  * allocation classes:
3167  *	VM_ALLOC_NORMAL		normal process request
3168  *	VM_ALLOC_SYSTEM		system *really* needs the pages
3169  *
3170  * The caller must always specify that the pages are to be busied and/or
3171  * wired.
3172  *
3173  * optional allocation flags:
3174  *	VM_ALLOC_IGN_SBUSY	do not sleep on soft busy pages
3175  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
3176  *	VM_ALLOC_NOWAIT		do not sleep
3177  *	VM_ALLOC_SBUSY		set page to sbusy state
3178  *	VM_ALLOC_WIRED		wire the pages
3179  *	VM_ALLOC_ZERO		zero and validate any invalid pages
3180  *
3181  * If VM_ALLOC_NOWAIT is not specified, this routine may sleep.  Otherwise, it
3182  * may return a partial prefix of the requested range.
3183  */
3184 int
3185 vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
3186     vm_page_t *ma, int count)
3187 {
3188 	vm_page_t m;
3189 	int i;
3190 	bool sleep;
3191 
3192 	VM_OBJECT_ASSERT_WLOCKED(object);
3193 	KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0,
3194 	    ("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed"));
3195 	KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 ||
3196 	    (allocflags & VM_ALLOC_WIRED) != 0,
3197 	    ("vm_page_grab_pages: the pages must be busied or wired"));
3198 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3199 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3200 	    ("vm_page_grab_pages: VM_ALLOC_SBUSY/IGN_SBUSY mismatch"));
3201 	if (count == 0)
3202 		return (0);
3203 	i = 0;
3204 retrylookup:
3205 	m = vm_page_lookup(object, pindex + i);
3206 	for (; i < count; i++) {
3207 		if (m != NULL) {
3208 			sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3209 			    vm_page_xbusied(m) : vm_page_busied(m);
3210 			if (sleep) {
3211 				if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3212 					break;
3213 				/*
3214 				 * Reference the page before unlocking and
3215 				 * sleeping so that the page daemon is less
3216 				 * likely to reclaim it.
3217 				 */
3218 				vm_page_aflag_set(m, PGA_REFERENCED);
3219 				vm_page_lock(m);
3220 				VM_OBJECT_WUNLOCK(object);
3221 				vm_page_busy_sleep(m, "grbmaw", (allocflags &
3222 				    VM_ALLOC_IGN_SBUSY) != 0);
3223 				VM_OBJECT_WLOCK(object);
3224 				goto retrylookup;
3225 			}
3226 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
3227 				vm_page_lock(m);
3228 				vm_page_wire(m);
3229 				vm_page_unlock(m);
3230 			}
3231 			if ((allocflags & (VM_ALLOC_NOBUSY |
3232 			    VM_ALLOC_SBUSY)) == 0)
3233 				vm_page_xbusy(m);
3234 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
3235 				vm_page_sbusy(m);
3236 		} else {
3237 			m = vm_page_alloc(object, pindex + i, (allocflags &
3238 			    ~VM_ALLOC_IGN_SBUSY) | VM_ALLOC_COUNT(count - i));
3239 			if (m == NULL) {
3240 				if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3241 					break;
3242 				VM_OBJECT_WUNLOCK(object);
3243 				VM_WAIT;
3244 				VM_OBJECT_WLOCK(object);
3245 				goto retrylookup;
3246 			}
3247 		}
3248 		if (m->valid == 0 && (allocflags & VM_ALLOC_ZERO) != 0) {
3249 			if ((m->flags & PG_ZERO) == 0)
3250 				pmap_zero_page(m);
3251 			m->valid = VM_PAGE_BITS_ALL;
3252 		}
3253 		ma[i] = m;
3254 		m = vm_page_next(m);
3255 	}
3256 	return (i);
3257 }
3258 
3259 /*
3260  * Mapping function for valid or dirty bits in a page.
3261  *
3262  * Inputs are required to range within a page.
3263  */
3264 vm_page_bits_t
3265 vm_page_bits(int base, int size)
3266 {
3267 	int first_bit;
3268 	int last_bit;
3269 
3270 	KASSERT(
3271 	    base + size <= PAGE_SIZE,
3272 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
3273 	);
3274 
3275 	if (size == 0)		/* handle degenerate case */
3276 		return (0);
3277 
3278 	first_bit = base >> DEV_BSHIFT;
3279 	last_bit = (base + size - 1) >> DEV_BSHIFT;
3280 
3281 	return (((vm_page_bits_t)2 << last_bit) -
3282 	    ((vm_page_bits_t)1 << first_bit));
3283 }
3284 
3285 /*
3286  *	vm_page_set_valid_range:
3287  *
3288  *	Sets portions of a page valid.  The arguments are expected
3289  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3290  *	of any partial chunks touched by the range.  The invalid portion of
3291  *	such chunks will be zeroed.
3292  *
3293  *	(base + size) must be less then or equal to PAGE_SIZE.
3294  */
3295 void
3296 vm_page_set_valid_range(vm_page_t m, int base, int size)
3297 {
3298 	int endoff, frag;
3299 
3300 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3301 	if (size == 0)	/* handle degenerate case */
3302 		return;
3303 
3304 	/*
3305 	 * If the base is not DEV_BSIZE aligned and the valid
3306 	 * bit is clear, we have to zero out a portion of the
3307 	 * first block.
3308 	 */
3309 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3310 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
3311 		pmap_zero_page_area(m, frag, base - frag);
3312 
3313 	/*
3314 	 * If the ending offset is not DEV_BSIZE aligned and the
3315 	 * valid bit is clear, we have to zero out a portion of
3316 	 * the last block.
3317 	 */
3318 	endoff = base + size;
3319 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3320 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
3321 		pmap_zero_page_area(m, endoff,
3322 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3323 
3324 	/*
3325 	 * Assert that no previously invalid block that is now being validated
3326 	 * is already dirty.
3327 	 */
3328 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
3329 	    ("vm_page_set_valid_range: page %p is dirty", m));
3330 
3331 	/*
3332 	 * Set valid bits inclusive of any overlap.
3333 	 */
3334 	m->valid |= vm_page_bits(base, size);
3335 }
3336 
3337 /*
3338  * Clear the given bits from the specified page's dirty field.
3339  */
3340 static __inline void
3341 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
3342 {
3343 	uintptr_t addr;
3344 #if PAGE_SIZE < 16384
3345 	int shift;
3346 #endif
3347 
3348 	/*
3349 	 * If the object is locked and the page is neither exclusive busy nor
3350 	 * write mapped, then the page's dirty field cannot possibly be
3351 	 * set by a concurrent pmap operation.
3352 	 */
3353 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3354 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
3355 		m->dirty &= ~pagebits;
3356 	else {
3357 		/*
3358 		 * The pmap layer can call vm_page_dirty() without
3359 		 * holding a distinguished lock.  The combination of
3360 		 * the object's lock and an atomic operation suffice
3361 		 * to guarantee consistency of the page dirty field.
3362 		 *
3363 		 * For PAGE_SIZE == 32768 case, compiler already
3364 		 * properly aligns the dirty field, so no forcible
3365 		 * alignment is needed. Only require existence of
3366 		 * atomic_clear_64 when page size is 32768.
3367 		 */
3368 		addr = (uintptr_t)&m->dirty;
3369 #if PAGE_SIZE == 32768
3370 		atomic_clear_64((uint64_t *)addr, pagebits);
3371 #elif PAGE_SIZE == 16384
3372 		atomic_clear_32((uint32_t *)addr, pagebits);
3373 #else		/* PAGE_SIZE <= 8192 */
3374 		/*
3375 		 * Use a trick to perform a 32-bit atomic on the
3376 		 * containing aligned word, to not depend on the existence
3377 		 * of atomic_clear_{8, 16}.
3378 		 */
3379 		shift = addr & (sizeof(uint32_t) - 1);
3380 #if BYTE_ORDER == BIG_ENDIAN
3381 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
3382 #else
3383 		shift *= NBBY;
3384 #endif
3385 		addr &= ~(sizeof(uint32_t) - 1);
3386 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
3387 #endif		/* PAGE_SIZE */
3388 	}
3389 }
3390 
3391 /*
3392  *	vm_page_set_validclean:
3393  *
3394  *	Sets portions of a page valid and clean.  The arguments are expected
3395  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3396  *	of any partial chunks touched by the range.  The invalid portion of
3397  *	such chunks will be zero'd.
3398  *
3399  *	(base + size) must be less then or equal to PAGE_SIZE.
3400  */
3401 void
3402 vm_page_set_validclean(vm_page_t m, int base, int size)
3403 {
3404 	vm_page_bits_t oldvalid, pagebits;
3405 	int endoff, frag;
3406 
3407 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3408 	if (size == 0)	/* handle degenerate case */
3409 		return;
3410 
3411 	/*
3412 	 * If the base is not DEV_BSIZE aligned and the valid
3413 	 * bit is clear, we have to zero out a portion of the
3414 	 * first block.
3415 	 */
3416 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3417 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3418 		pmap_zero_page_area(m, frag, base - frag);
3419 
3420 	/*
3421 	 * If the ending offset is not DEV_BSIZE aligned and the
3422 	 * valid bit is clear, we have to zero out a portion of
3423 	 * the last block.
3424 	 */
3425 	endoff = base + size;
3426 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3427 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3428 		pmap_zero_page_area(m, endoff,
3429 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3430 
3431 	/*
3432 	 * Set valid, clear dirty bits.  If validating the entire
3433 	 * page we can safely clear the pmap modify bit.  We also
3434 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
3435 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
3436 	 * be set again.
3437 	 *
3438 	 * We set valid bits inclusive of any overlap, but we can only
3439 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
3440 	 * the range.
3441 	 */
3442 	oldvalid = m->valid;
3443 	pagebits = vm_page_bits(base, size);
3444 	m->valid |= pagebits;
3445 #if 0	/* NOT YET */
3446 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3447 		frag = DEV_BSIZE - frag;
3448 		base += frag;
3449 		size -= frag;
3450 		if (size < 0)
3451 			size = 0;
3452 	}
3453 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3454 #endif
3455 	if (base == 0 && size == PAGE_SIZE) {
3456 		/*
3457 		 * The page can only be modified within the pmap if it is
3458 		 * mapped, and it can only be mapped if it was previously
3459 		 * fully valid.
3460 		 */
3461 		if (oldvalid == VM_PAGE_BITS_ALL)
3462 			/*
3463 			 * Perform the pmap_clear_modify() first.  Otherwise,
3464 			 * a concurrent pmap operation, such as
3465 			 * pmap_protect(), could clear a modification in the
3466 			 * pmap and set the dirty field on the page before
3467 			 * pmap_clear_modify() had begun and after the dirty
3468 			 * field was cleared here.
3469 			 */
3470 			pmap_clear_modify(m);
3471 		m->dirty = 0;
3472 		m->oflags &= ~VPO_NOSYNC;
3473 	} else if (oldvalid != VM_PAGE_BITS_ALL)
3474 		m->dirty &= ~pagebits;
3475 	else
3476 		vm_page_clear_dirty_mask(m, pagebits);
3477 }
3478 
3479 void
3480 vm_page_clear_dirty(vm_page_t m, int base, int size)
3481 {
3482 
3483 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3484 }
3485 
3486 /*
3487  *	vm_page_set_invalid:
3488  *
3489  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
3490  *	valid and dirty bits for the effected areas are cleared.
3491  */
3492 void
3493 vm_page_set_invalid(vm_page_t m, int base, int size)
3494 {
3495 	vm_page_bits_t bits;
3496 	vm_object_t object;
3497 
3498 	object = m->object;
3499 	VM_OBJECT_ASSERT_WLOCKED(object);
3500 	if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3501 	    size >= object->un_pager.vnp.vnp_size)
3502 		bits = VM_PAGE_BITS_ALL;
3503 	else
3504 		bits = vm_page_bits(base, size);
3505 	if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
3506 	    bits != 0)
3507 		pmap_remove_all(m);
3508 	KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3509 	    !pmap_page_is_mapped(m),
3510 	    ("vm_page_set_invalid: page %p is mapped", m));
3511 	m->valid &= ~bits;
3512 	m->dirty &= ~bits;
3513 }
3514 
3515 /*
3516  * vm_page_zero_invalid()
3517  *
3518  *	The kernel assumes that the invalid portions of a page contain
3519  *	garbage, but such pages can be mapped into memory by user code.
3520  *	When this occurs, we must zero out the non-valid portions of the
3521  *	page so user code sees what it expects.
3522  *
3523  *	Pages are most often semi-valid when the end of a file is mapped
3524  *	into memory and the file's size is not page aligned.
3525  */
3526 void
3527 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3528 {
3529 	int b;
3530 	int i;
3531 
3532 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3533 	/*
3534 	 * Scan the valid bits looking for invalid sections that
3535 	 * must be zeroed.  Invalid sub-DEV_BSIZE'd areas ( where the
3536 	 * valid bit may be set ) have already been zeroed by
3537 	 * vm_page_set_validclean().
3538 	 */
3539 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3540 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3541 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3542 			if (i > b) {
3543 				pmap_zero_page_area(m,
3544 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3545 			}
3546 			b = i + 1;
3547 		}
3548 	}
3549 
3550 	/*
3551 	 * setvalid is TRUE when we can safely set the zero'd areas
3552 	 * as being valid.  We can do this if there are no cache consistancy
3553 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3554 	 */
3555 	if (setvalid)
3556 		m->valid = VM_PAGE_BITS_ALL;
3557 }
3558 
3559 /*
3560  *	vm_page_is_valid:
3561  *
3562  *	Is (partial) page valid?  Note that the case where size == 0
3563  *	will return FALSE in the degenerate case where the page is
3564  *	entirely invalid, and TRUE otherwise.
3565  */
3566 int
3567 vm_page_is_valid(vm_page_t m, int base, int size)
3568 {
3569 	vm_page_bits_t bits;
3570 
3571 	VM_OBJECT_ASSERT_LOCKED(m->object);
3572 	bits = vm_page_bits(base, size);
3573 	return (m->valid != 0 && (m->valid & bits) == bits);
3574 }
3575 
3576 /*
3577  * Returns true if all of the specified predicates are true for the entire
3578  * (super)page and false otherwise.
3579  */
3580 bool
3581 vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m)
3582 {
3583 	vm_object_t object;
3584 	int i, npages;
3585 
3586 	object = m->object;
3587 	VM_OBJECT_ASSERT_LOCKED(object);
3588 	npages = atop(pagesizes[m->psind]);
3589 
3590 	/*
3591 	 * The physically contiguous pages that make up a superpage, i.e., a
3592 	 * page with a page size index ("psind") greater than zero, will
3593 	 * occupy adjacent entries in vm_page_array[].
3594 	 */
3595 	for (i = 0; i < npages; i++) {
3596 		/* Always test object consistency, including "skip_m". */
3597 		if (m[i].object != object)
3598 			return (false);
3599 		if (&m[i] == skip_m)
3600 			continue;
3601 		if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i]))
3602 			return (false);
3603 		if ((flags & PS_ALL_DIRTY) != 0) {
3604 			/*
3605 			 * Calling vm_page_test_dirty() or pmap_is_modified()
3606 			 * might stop this case from spuriously returning
3607 			 * "false".  However, that would require a write lock
3608 			 * on the object containing "m[i]".
3609 			 */
3610 			if (m[i].dirty != VM_PAGE_BITS_ALL)
3611 				return (false);
3612 		}
3613 		if ((flags & PS_ALL_VALID) != 0 &&
3614 		    m[i].valid != VM_PAGE_BITS_ALL)
3615 			return (false);
3616 	}
3617 	return (true);
3618 }
3619 
3620 /*
3621  * Set the page's dirty bits if the page is modified.
3622  */
3623 void
3624 vm_page_test_dirty(vm_page_t m)
3625 {
3626 
3627 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3628 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3629 		vm_page_dirty(m);
3630 }
3631 
3632 void
3633 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3634 {
3635 
3636 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3637 }
3638 
3639 void
3640 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3641 {
3642 
3643 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3644 }
3645 
3646 int
3647 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3648 {
3649 
3650 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3651 }
3652 
3653 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3654 void
3655 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3656 {
3657 
3658 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3659 }
3660 
3661 void
3662 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3663 {
3664 
3665 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3666 }
3667 #endif
3668 
3669 #ifdef INVARIANTS
3670 void
3671 vm_page_object_lock_assert(vm_page_t m)
3672 {
3673 
3674 	/*
3675 	 * Certain of the page's fields may only be modified by the
3676 	 * holder of the containing object's lock or the exclusive busy.
3677 	 * holder.  Unfortunately, the holder of the write busy is
3678 	 * not recorded, and thus cannot be checked here.
3679 	 */
3680 	if (m->object != NULL && !vm_page_xbusied(m))
3681 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3682 }
3683 
3684 void
3685 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
3686 {
3687 
3688 	if ((bits & PGA_WRITEABLE) == 0)
3689 		return;
3690 
3691 	/*
3692 	 * The PGA_WRITEABLE flag can only be set if the page is
3693 	 * managed, is exclusively busied or the object is locked.
3694 	 * Currently, this flag is only set by pmap_enter().
3695 	 */
3696 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3697 	    ("PGA_WRITEABLE on unmanaged page"));
3698 	if (!vm_page_xbusied(m))
3699 		VM_OBJECT_ASSERT_LOCKED(m->object);
3700 }
3701 #endif
3702 
3703 #include "opt_ddb.h"
3704 #ifdef DDB
3705 #include <sys/kernel.h>
3706 
3707 #include <ddb/ddb.h>
3708 
3709 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3710 {
3711 
3712 	db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count);
3713 	db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count);
3714 	db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count);
3715 	db_printf("vm_cnt.v_laundry_count: %d\n", vm_cnt.v_laundry_count);
3716 	db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count);
3717 	db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
3718 	db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
3719 	db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
3720 	db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
3721 }
3722 
3723 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3724 {
3725 	int dom;
3726 
3727 	db_printf("pq_free %d\n", vm_cnt.v_free_count);
3728 	for (dom = 0; dom < vm_ndomains; dom++) {
3729 		db_printf(
3730     "dom %d page_cnt %d free %d pq_act %d pq_inact %d pq_laund %d pq_unsw %d\n",
3731 		    dom,
3732 		    vm_dom[dom].vmd_page_count,
3733 		    vm_dom[dom].vmd_free_count,
3734 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3735 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3736 		    vm_dom[dom].vmd_pagequeues[PQ_LAUNDRY].pq_cnt,
3737 		    vm_dom[dom].vmd_pagequeues[PQ_UNSWAPPABLE].pq_cnt);
3738 	}
3739 }
3740 
3741 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3742 {
3743 	vm_page_t m;
3744 	boolean_t phys;
3745 
3746 	if (!have_addr) {
3747 		db_printf("show pginfo addr\n");
3748 		return;
3749 	}
3750 
3751 	phys = strchr(modif, 'p') != NULL;
3752 	if (phys)
3753 		m = PHYS_TO_VM_PAGE(addr);
3754 	else
3755 		m = (vm_page_t)addr;
3756 	db_printf(
3757     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3758     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3759 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3760 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3761 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3762 }
3763 #endif /* DDB */
3764