xref: /freebsd/sys/vm/vm_page.c (revision a273973175e2a4cbbda3fd7b8cddd0bec0857082)
1 /*-
2  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3  *
4  * Copyright (c) 1991 Regents of the University of California.
5  * All rights reserved.
6  * Copyright (c) 1998 Matthew Dillon.  All Rights Reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * The Mach Operating System project at Carnegie-Mellon University.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	from: @(#)vm_page.c	7.4 (Berkeley) 5/7/91
36  */
37 
38 /*-
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  */
64 
65 /*
66  *			GENERAL RULES ON VM_PAGE MANIPULATION
67  *
68  *	- A page queue lock is required when adding or removing a page from a
69  *	  page queue regardless of other locks or the busy state of a page.
70  *
71  *		* In general, no thread besides the page daemon can acquire or
72  *		  hold more than one page queue lock at a time.
73  *
74  *		* The page daemon can acquire and hold any pair of page queue
75  *		  locks in any order.
76  *
77  *	- The object lock is required when inserting or removing
78  *	  pages from an object (vm_page_insert() or vm_page_remove()).
79  *
80  */
81 
82 /*
83  *	Resident memory management module.
84  */
85 
86 #include <sys/cdefs.h>
87 __FBSDID("$FreeBSD$");
88 
89 #include "opt_vm.h"
90 
91 #include <sys/param.h>
92 #include <sys/systm.h>
93 #include <sys/lock.h>
94 #include <sys/domainset.h>
95 #include <sys/kernel.h>
96 #include <sys/limits.h>
97 #include <sys/linker.h>
98 #include <sys/malloc.h>
99 #include <sys/mman.h>
100 #include <sys/msgbuf.h>
101 #include <sys/mutex.h>
102 #include <sys/proc.h>
103 #include <sys/rwlock.h>
104 #include <sys/sbuf.h>
105 #include <sys/smp.h>
106 #include <sys/sysctl.h>
107 #include <sys/vmmeter.h>
108 #include <sys/vnode.h>
109 
110 #include <vm/vm.h>
111 #include <vm/pmap.h>
112 #include <vm/vm_param.h>
113 #include <vm/vm_domainset.h>
114 #include <vm/vm_kern.h>
115 #include <vm/vm_object.h>
116 #include <vm/vm_page.h>
117 #include <vm/vm_pageout.h>
118 #include <vm/vm_pager.h>
119 #include <vm/vm_phys.h>
120 #include <vm/vm_radix.h>
121 #include <vm/vm_reserv.h>
122 #include <vm/vm_extern.h>
123 #include <vm/uma.h>
124 #include <vm/uma_int.h>
125 
126 #include <machine/md_var.h>
127 
128 extern int	uma_startup_count(int);
129 extern void	uma_startup(void *, int);
130 extern void	uma_startup1(void);
131 
132 /*
133  *	Associated with page of user-allocatable memory is a
134  *	page structure.
135  */
136 
137 struct vm_domain vm_dom[MAXMEMDOM];
138 struct mtx_padalign __exclusive_cache_line vm_page_queue_free_mtx;
139 
140 struct mtx_padalign __exclusive_cache_line pa_lock[PA_LOCK_COUNT];
141 
142 /*
143  * bogus page -- for I/O to/from partially complete buffers,
144  * or for paging into sparsely invalid regions.
145  */
146 vm_page_t bogus_page;
147 
148 vm_page_t vm_page_array;
149 long vm_page_array_size;
150 long first_page;
151 
152 static int boot_pages;
153 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
154     &boot_pages, 0,
155     "number of pages allocated for bootstrapping the VM system");
156 
157 static int pa_tryrelock_restart;
158 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
159     &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
160 
161 static TAILQ_HEAD(, vm_page) blacklist_head;
162 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
163 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
164     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
165 
166 /* Is the page daemon waiting for free pages? */
167 static int vm_pageout_pages_needed;
168 
169 static uma_zone_t fakepg_zone;
170 
171 static void vm_page_alloc_check(vm_page_t m);
172 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
173 static void vm_page_enqueue(uint8_t queue, vm_page_t m);
174 static void vm_page_free_phys(vm_page_t m);
175 static void vm_page_free_wakeup(void);
176 static void vm_page_init(void *dummy);
177 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
178     vm_pindex_t pindex, vm_page_t mpred);
179 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
180     vm_page_t mpred);
181 static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
182     vm_paddr_t high);
183 static int vm_page_alloc_fail(vm_object_t object, int req);
184 
185 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init, NULL);
186 
187 static void
188 vm_page_init(void *dummy)
189 {
190 
191 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
192 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
193 	bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ |
194 	    VM_ALLOC_NORMAL | VM_ALLOC_WIRED);
195 }
196 
197 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
198 #if PAGE_SIZE == 32768
199 #ifdef CTASSERT
200 CTASSERT(sizeof(u_long) >= 8);
201 #endif
202 #endif
203 
204 /*
205  * Try to acquire a physical address lock while a pmap is locked.  If we
206  * fail to trylock we unlock and lock the pmap directly and cache the
207  * locked pa in *locked.  The caller should then restart their loop in case
208  * the virtual to physical mapping has changed.
209  */
210 int
211 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
212 {
213 	vm_paddr_t lockpa;
214 
215 	lockpa = *locked;
216 	*locked = pa;
217 	if (lockpa) {
218 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
219 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
220 			return (0);
221 		PA_UNLOCK(lockpa);
222 	}
223 	if (PA_TRYLOCK(pa))
224 		return (0);
225 	PMAP_UNLOCK(pmap);
226 	atomic_add_int(&pa_tryrelock_restart, 1);
227 	PA_LOCK(pa);
228 	PMAP_LOCK(pmap);
229 	return (EAGAIN);
230 }
231 
232 /*
233  *	vm_set_page_size:
234  *
235  *	Sets the page size, perhaps based upon the memory
236  *	size.  Must be called before any use of page-size
237  *	dependent functions.
238  */
239 void
240 vm_set_page_size(void)
241 {
242 	if (vm_cnt.v_page_size == 0)
243 		vm_cnt.v_page_size = PAGE_SIZE;
244 	if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
245 		panic("vm_set_page_size: page size not a power of two");
246 }
247 
248 /*
249  *	vm_page_blacklist_next:
250  *
251  *	Find the next entry in the provided string of blacklist
252  *	addresses.  Entries are separated by space, comma, or newline.
253  *	If an invalid integer is encountered then the rest of the
254  *	string is skipped.  Updates the list pointer to the next
255  *	character, or NULL if the string is exhausted or invalid.
256  */
257 static vm_paddr_t
258 vm_page_blacklist_next(char **list, char *end)
259 {
260 	vm_paddr_t bad;
261 	char *cp, *pos;
262 
263 	if (list == NULL || *list == NULL)
264 		return (0);
265 	if (**list =='\0') {
266 		*list = NULL;
267 		return (0);
268 	}
269 
270 	/*
271 	 * If there's no end pointer then the buffer is coming from
272 	 * the kenv and we know it's null-terminated.
273 	 */
274 	if (end == NULL)
275 		end = *list + strlen(*list);
276 
277 	/* Ensure that strtoq() won't walk off the end */
278 	if (*end != '\0') {
279 		if (*end == '\n' || *end == ' ' || *end  == ',')
280 			*end = '\0';
281 		else {
282 			printf("Blacklist not terminated, skipping\n");
283 			*list = NULL;
284 			return (0);
285 		}
286 	}
287 
288 	for (pos = *list; *pos != '\0'; pos = cp) {
289 		bad = strtoq(pos, &cp, 0);
290 		if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
291 			if (bad == 0) {
292 				if (++cp < end)
293 					continue;
294 				else
295 					break;
296 			}
297 		} else
298 			break;
299 		if (*cp == '\0' || ++cp >= end)
300 			*list = NULL;
301 		else
302 			*list = cp;
303 		return (trunc_page(bad));
304 	}
305 	printf("Garbage in RAM blacklist, skipping\n");
306 	*list = NULL;
307 	return (0);
308 }
309 
310 /*
311  *	vm_page_blacklist_check:
312  *
313  *	Iterate through the provided string of blacklist addresses, pulling
314  *	each entry out of the physical allocator free list and putting it
315  *	onto a list for reporting via the vm.page_blacklist sysctl.
316  */
317 static void
318 vm_page_blacklist_check(char *list, char *end)
319 {
320 	vm_paddr_t pa;
321 	vm_page_t m;
322 	char *next;
323 	int ret;
324 
325 	next = list;
326 	while (next != NULL) {
327 		if ((pa = vm_page_blacklist_next(&next, end)) == 0)
328 			continue;
329 		m = vm_phys_paddr_to_vm_page(pa);
330 		if (m == NULL)
331 			continue;
332 		mtx_lock(&vm_page_queue_free_mtx);
333 		ret = vm_phys_unfree_page(m);
334 		mtx_unlock(&vm_page_queue_free_mtx);
335 		if (ret == TRUE) {
336 			TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
337 			if (bootverbose)
338 				printf("Skipping page with pa 0x%jx\n",
339 				    (uintmax_t)pa);
340 		}
341 	}
342 }
343 
344 /*
345  *	vm_page_blacklist_load:
346  *
347  *	Search for a special module named "ram_blacklist".  It'll be a
348  *	plain text file provided by the user via the loader directive
349  *	of the same name.
350  */
351 static void
352 vm_page_blacklist_load(char **list, char **end)
353 {
354 	void *mod;
355 	u_char *ptr;
356 	u_int len;
357 
358 	mod = NULL;
359 	ptr = NULL;
360 
361 	mod = preload_search_by_type("ram_blacklist");
362 	if (mod != NULL) {
363 		ptr = preload_fetch_addr(mod);
364 		len = preload_fetch_size(mod);
365         }
366 	*list = ptr;
367 	if (ptr != NULL)
368 		*end = ptr + len;
369 	else
370 		*end = NULL;
371 	return;
372 }
373 
374 static int
375 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
376 {
377 	vm_page_t m;
378 	struct sbuf sbuf;
379 	int error, first;
380 
381 	first = 1;
382 	error = sysctl_wire_old_buffer(req, 0);
383 	if (error != 0)
384 		return (error);
385 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
386 	TAILQ_FOREACH(m, &blacklist_head, listq) {
387 		sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
388 		    (uintmax_t)m->phys_addr);
389 		first = 0;
390 	}
391 	error = sbuf_finish(&sbuf);
392 	sbuf_delete(&sbuf);
393 	return (error);
394 }
395 
396 static void
397 vm_page_domain_init(struct vm_domain *vmd)
398 {
399 	struct vm_pagequeue *pq;
400 	int i;
401 
402 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
403 	    "vm inactive pagequeue";
404 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
405 	    &vm_cnt.v_inactive_count;
406 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
407 	    "vm active pagequeue";
408 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
409 	    &vm_cnt.v_active_count;
410 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_name) =
411 	    "vm laundry pagequeue";
412 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_vcnt) =
413 	    &vm_cnt.v_laundry_count;
414 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_name) =
415 	    "vm unswappable pagequeue";
416 	/* Unswappable dirty pages are counted as being in the laundry. */
417 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_vcnt) =
418 	    &vm_cnt.v_laundry_count;
419 	vmd->vmd_page_count = 0;
420 	vmd->vmd_free_count = 0;
421 	vmd->vmd_segs = 0;
422 	vmd->vmd_oom = FALSE;
423 	for (i = 0; i < PQ_COUNT; i++) {
424 		pq = &vmd->vmd_pagequeues[i];
425 		TAILQ_INIT(&pq->pq_pl);
426 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
427 		    MTX_DEF | MTX_DUPOK);
428 	}
429 }
430 
431 /*
432  * Initialize a physical page in preparation for adding it to the free
433  * lists.
434  */
435 static void
436 vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind)
437 {
438 
439 	m->object = NULL;
440 	m->wire_count = 0;
441 	m->busy_lock = VPB_UNBUSIED;
442 	m->hold_count = 0;
443 	m->flags = 0;
444 	m->phys_addr = pa;
445 	m->queue = PQ_NONE;
446 	m->psind = 0;
447 	m->segind = segind;
448 	m->order = VM_NFREEORDER;
449 	m->pool = VM_FREEPOOL_DEFAULT;
450 	m->valid = m->dirty = 0;
451 	pmap_page_init(m);
452 }
453 
454 /*
455  *	vm_page_startup:
456  *
457  *	Initializes the resident memory module.  Allocates physical memory for
458  *	bootstrapping UMA and some data structures that are used to manage
459  *	physical pages.  Initializes these structures, and populates the free
460  *	page queues.
461  */
462 vm_offset_t
463 vm_page_startup(vm_offset_t vaddr)
464 {
465 	struct vm_domain *vmd;
466 	struct vm_phys_seg *seg;
467 	vm_page_t m;
468 	char *list, *listend;
469 	vm_offset_t mapped;
470 	vm_paddr_t end, high_avail, low_avail, new_end, page_range, size;
471 	vm_paddr_t biggestsize, last_pa, pa;
472 	u_long pagecount;
473 	int biggestone, i, segind;
474 
475 	biggestsize = 0;
476 	biggestone = 0;
477 	vaddr = round_page(vaddr);
478 
479 	for (i = 0; phys_avail[i + 1]; i += 2) {
480 		phys_avail[i] = round_page(phys_avail[i]);
481 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
482 	}
483 	for (i = 0; phys_avail[i + 1]; i += 2) {
484 		size = phys_avail[i + 1] - phys_avail[i];
485 		if (size > biggestsize) {
486 			biggestone = i;
487 			biggestsize = size;
488 		}
489 	}
490 
491 	end = phys_avail[biggestone+1];
492 
493 	/*
494 	 * Initialize the page and queue locks.
495 	 */
496 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
497 	for (i = 0; i < PA_LOCK_COUNT; i++)
498 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
499 	for (i = 0; i < vm_ndomains; i++)
500 		vm_page_domain_init(&vm_dom[i]);
501 
502 	/*
503 	 * Allocate memory for use when boot strapping the kernel memory
504 	 * allocator.
505 	 *
506 	 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
507 	 * manually fetch the value.
508 	 */
509 	boot_pages = uma_startup_count(0);
510 	TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
511 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
512 	new_end = trunc_page(new_end);
513 	mapped = pmap_map(&vaddr, new_end, end,
514 	    VM_PROT_READ | VM_PROT_WRITE);
515 	bzero((void *)mapped, end - new_end);
516 	uma_startup((void *)mapped, boot_pages);
517 
518 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
519     defined(__i386__) || defined(__mips__)
520 	/*
521 	 * Allocate a bitmap to indicate that a random physical page
522 	 * needs to be included in a minidump.
523 	 *
524 	 * The amd64 port needs this to indicate which direct map pages
525 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
526 	 *
527 	 * However, i386 still needs this workspace internally within the
528 	 * minidump code.  In theory, they are not needed on i386, but are
529 	 * included should the sf_buf code decide to use them.
530 	 */
531 	last_pa = 0;
532 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
533 		if (dump_avail[i + 1] > last_pa)
534 			last_pa = dump_avail[i + 1];
535 	page_range = last_pa / PAGE_SIZE;
536 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
537 	new_end -= vm_page_dump_size;
538 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
539 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
540 	bzero((void *)vm_page_dump, vm_page_dump_size);
541 #else
542 	(void)last_pa;
543 #endif
544 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
545 	/*
546 	 * Include the UMA bootstrap pages and vm_page_dump in a crash dump.
547 	 * When pmap_map() uses the direct map, they are not automatically
548 	 * included.
549 	 */
550 	for (pa = new_end; pa < end; pa += PAGE_SIZE)
551 		dump_add_page(pa);
552 #endif
553 	phys_avail[biggestone + 1] = new_end;
554 #ifdef __amd64__
555 	/*
556 	 * Request that the physical pages underlying the message buffer be
557 	 * included in a crash dump.  Since the message buffer is accessed
558 	 * through the direct map, they are not automatically included.
559 	 */
560 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
561 	last_pa = pa + round_page(msgbufsize);
562 	while (pa < last_pa) {
563 		dump_add_page(pa);
564 		pa += PAGE_SIZE;
565 	}
566 #endif
567 	/*
568 	 * Compute the number of pages of memory that will be available for
569 	 * use, taking into account the overhead of a page structure per page.
570 	 * In other words, solve
571 	 *	"available physical memory" - round_page(page_range *
572 	 *	    sizeof(struct vm_page)) = page_range * PAGE_SIZE
573 	 * for page_range.
574 	 */
575 	low_avail = phys_avail[0];
576 	high_avail = phys_avail[1];
577 	for (i = 0; i < vm_phys_nsegs; i++) {
578 		if (vm_phys_segs[i].start < low_avail)
579 			low_avail = vm_phys_segs[i].start;
580 		if (vm_phys_segs[i].end > high_avail)
581 			high_avail = vm_phys_segs[i].end;
582 	}
583 	/* Skip the first chunk.  It is already accounted for. */
584 	for (i = 2; phys_avail[i + 1] != 0; i += 2) {
585 		if (phys_avail[i] < low_avail)
586 			low_avail = phys_avail[i];
587 		if (phys_avail[i + 1] > high_avail)
588 			high_avail = phys_avail[i + 1];
589 	}
590 	first_page = low_avail / PAGE_SIZE;
591 #ifdef VM_PHYSSEG_SPARSE
592 	size = 0;
593 	for (i = 0; i < vm_phys_nsegs; i++)
594 		size += vm_phys_segs[i].end - vm_phys_segs[i].start;
595 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
596 		size += phys_avail[i + 1] - phys_avail[i];
597 #elif defined(VM_PHYSSEG_DENSE)
598 	size = high_avail - low_avail;
599 #else
600 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
601 #endif
602 
603 #ifdef VM_PHYSSEG_DENSE
604 	/*
605 	 * In the VM_PHYSSEG_DENSE case, the number of pages can account for
606 	 * the overhead of a page structure per page only if vm_page_array is
607 	 * allocated from the last physical memory chunk.  Otherwise, we must
608 	 * allocate page structures representing the physical memory
609 	 * underlying vm_page_array, even though they will not be used.
610 	 */
611 	if (new_end != high_avail)
612 		page_range = size / PAGE_SIZE;
613 	else
614 #endif
615 	{
616 		page_range = size / (PAGE_SIZE + sizeof(struct vm_page));
617 
618 		/*
619 		 * If the partial bytes remaining are large enough for
620 		 * a page (PAGE_SIZE) without a corresponding
621 		 * 'struct vm_page', then new_end will contain an
622 		 * extra page after subtracting the length of the VM
623 		 * page array.  Compensate by subtracting an extra
624 		 * page from new_end.
625 		 */
626 		if (size % (PAGE_SIZE + sizeof(struct vm_page)) >= PAGE_SIZE) {
627 			if (new_end == high_avail)
628 				high_avail -= PAGE_SIZE;
629 			new_end -= PAGE_SIZE;
630 		}
631 	}
632 	end = new_end;
633 
634 	/*
635 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
636 	 * However, because this page is allocated from KVM, out-of-bounds
637 	 * accesses using the direct map will not be trapped.
638 	 */
639 	vaddr += PAGE_SIZE;
640 
641 	/*
642 	 * Allocate physical memory for the page structures, and map it.
643 	 */
644 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
645 	mapped = pmap_map(&vaddr, new_end, end,
646 	    VM_PROT_READ | VM_PROT_WRITE);
647 	vm_page_array = (vm_page_t)mapped;
648 	vm_page_array_size = page_range;
649 
650 #if VM_NRESERVLEVEL > 0
651 	/*
652 	 * Allocate physical memory for the reservation management system's
653 	 * data structures, and map it.
654 	 */
655 	if (high_avail == end)
656 		high_avail = new_end;
657 	new_end = vm_reserv_startup(&vaddr, new_end, high_avail);
658 #endif
659 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
660 	/*
661 	 * Include vm_page_array and vm_reserv_array in a crash dump.
662 	 */
663 	for (pa = new_end; pa < end; pa += PAGE_SIZE)
664 		dump_add_page(pa);
665 #endif
666 	phys_avail[biggestone + 1] = new_end;
667 
668 	/*
669 	 * Add physical memory segments corresponding to the available
670 	 * physical pages.
671 	 */
672 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
673 		vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
674 
675 	/*
676 	 * Initialize the physical memory allocator.
677 	 */
678 	vm_phys_init();
679 
680 	/*
681 	 * Initialize the page structures and add every available page to the
682 	 * physical memory allocator's free lists.
683 	 */
684 	vm_cnt.v_page_count = 0;
685 	vm_cnt.v_free_count = 0;
686 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
687 		seg = &vm_phys_segs[segind];
688 		for (m = seg->first_page, pa = seg->start; pa < seg->end;
689 		    m++, pa += PAGE_SIZE)
690 			vm_page_init_page(m, pa, segind);
691 
692 		/*
693 		 * Add the segment to the free lists only if it is covered by
694 		 * one of the ranges in phys_avail.  Because we've added the
695 		 * ranges to the vm_phys_segs array, we can assume that each
696 		 * segment is either entirely contained in one of the ranges,
697 		 * or doesn't overlap any of them.
698 		 */
699 		for (i = 0; phys_avail[i + 1] != 0; i += 2) {
700 			if (seg->start < phys_avail[i] ||
701 			    seg->end > phys_avail[i + 1])
702 				continue;
703 
704 			m = seg->first_page;
705 			pagecount = (u_long)atop(seg->end - seg->start);
706 
707 			mtx_lock(&vm_page_queue_free_mtx);
708 			vm_phys_free_contig(m, pagecount);
709 			vm_phys_freecnt_adj(m, (int)pagecount);
710 			mtx_unlock(&vm_page_queue_free_mtx);
711 			vm_cnt.v_page_count += (u_int)pagecount;
712 
713 			vmd = &vm_dom[seg->domain];
714 			vmd->vmd_page_count += (u_int)pagecount;
715 			vmd->vmd_segs |= 1UL << m->segind;
716 			break;
717 		}
718 	}
719 
720 	/*
721 	 * Remove blacklisted pages from the physical memory allocator.
722 	 */
723 	TAILQ_INIT(&blacklist_head);
724 	vm_page_blacklist_load(&list, &listend);
725 	vm_page_blacklist_check(list, listend);
726 
727 	list = kern_getenv("vm.blacklist");
728 	vm_page_blacklist_check(list, NULL);
729 
730 	freeenv(list);
731 #if VM_NRESERVLEVEL > 0
732 	/*
733 	 * Initialize the reservation management system.
734 	 */
735 	vm_reserv_init();
736 #endif
737 	/*
738 	 * Set an initial domain policy for thread0 so that allocations
739 	 * can work.
740 	 */
741 	domainset_zero();
742 
743 	/* Announce page availability to UMA. */
744 	uma_startup1();
745 
746 	return (vaddr);
747 }
748 
749 void
750 vm_page_reference(vm_page_t m)
751 {
752 
753 	vm_page_aflag_set(m, PGA_REFERENCED);
754 }
755 
756 /*
757  *	vm_page_busy_downgrade:
758  *
759  *	Downgrade an exclusive busy page into a single shared busy page.
760  */
761 void
762 vm_page_busy_downgrade(vm_page_t m)
763 {
764 	u_int x;
765 	bool locked;
766 
767 	vm_page_assert_xbusied(m);
768 	locked = mtx_owned(vm_page_lockptr(m));
769 
770 	for (;;) {
771 		x = m->busy_lock;
772 		x &= VPB_BIT_WAITERS;
773 		if (x != 0 && !locked)
774 			vm_page_lock(m);
775 		if (atomic_cmpset_rel_int(&m->busy_lock,
776 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1)))
777 			break;
778 		if (x != 0 && !locked)
779 			vm_page_unlock(m);
780 	}
781 	if (x != 0) {
782 		wakeup(m);
783 		if (!locked)
784 			vm_page_unlock(m);
785 	}
786 }
787 
788 /*
789  *	vm_page_sbusied:
790  *
791  *	Return a positive value if the page is shared busied, 0 otherwise.
792  */
793 int
794 vm_page_sbusied(vm_page_t m)
795 {
796 	u_int x;
797 
798 	x = m->busy_lock;
799 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
800 }
801 
802 /*
803  *	vm_page_sunbusy:
804  *
805  *	Shared unbusy a page.
806  */
807 void
808 vm_page_sunbusy(vm_page_t m)
809 {
810 	u_int x;
811 
812 	vm_page_lock_assert(m, MA_NOTOWNED);
813 	vm_page_assert_sbusied(m);
814 
815 	for (;;) {
816 		x = m->busy_lock;
817 		if (VPB_SHARERS(x) > 1) {
818 			if (atomic_cmpset_int(&m->busy_lock, x,
819 			    x - VPB_ONE_SHARER))
820 				break;
821 			continue;
822 		}
823 		if ((x & VPB_BIT_WAITERS) == 0) {
824 			KASSERT(x == VPB_SHARERS_WORD(1),
825 			    ("vm_page_sunbusy: invalid lock state"));
826 			if (atomic_cmpset_int(&m->busy_lock,
827 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
828 				break;
829 			continue;
830 		}
831 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
832 		    ("vm_page_sunbusy: invalid lock state for waiters"));
833 
834 		vm_page_lock(m);
835 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
836 			vm_page_unlock(m);
837 			continue;
838 		}
839 		wakeup(m);
840 		vm_page_unlock(m);
841 		break;
842 	}
843 }
844 
845 /*
846  *	vm_page_busy_sleep:
847  *
848  *	Sleep and release the page lock, using the page pointer as wchan.
849  *	This is used to implement the hard-path of busying mechanism.
850  *
851  *	The given page must be locked.
852  *
853  *	If nonshared is true, sleep only if the page is xbusy.
854  */
855 void
856 vm_page_busy_sleep(vm_page_t m, const char *wmesg, bool nonshared)
857 {
858 	u_int x;
859 
860 	vm_page_assert_locked(m);
861 
862 	x = m->busy_lock;
863 	if (x == VPB_UNBUSIED || (nonshared && (x & VPB_BIT_SHARED) != 0) ||
864 	    ((x & VPB_BIT_WAITERS) == 0 &&
865 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS))) {
866 		vm_page_unlock(m);
867 		return;
868 	}
869 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
870 }
871 
872 /*
873  *	vm_page_trysbusy:
874  *
875  *	Try to shared busy a page.
876  *	If the operation succeeds 1 is returned otherwise 0.
877  *	The operation never sleeps.
878  */
879 int
880 vm_page_trysbusy(vm_page_t m)
881 {
882 	u_int x;
883 
884 	for (;;) {
885 		x = m->busy_lock;
886 		if ((x & VPB_BIT_SHARED) == 0)
887 			return (0);
888 		if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
889 			return (1);
890 	}
891 }
892 
893 static void
894 vm_page_xunbusy_locked(vm_page_t m)
895 {
896 
897 	vm_page_assert_xbusied(m);
898 	vm_page_assert_locked(m);
899 
900 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
901 	/* There is a waiter, do wakeup() instead of vm_page_flash(). */
902 	wakeup(m);
903 }
904 
905 void
906 vm_page_xunbusy_maybelocked(vm_page_t m)
907 {
908 	bool lockacq;
909 
910 	vm_page_assert_xbusied(m);
911 
912 	/*
913 	 * Fast path for unbusy.  If it succeeds, we know that there
914 	 * are no waiters, so we do not need a wakeup.
915 	 */
916 	if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER,
917 	    VPB_UNBUSIED))
918 		return;
919 
920 	lockacq = !mtx_owned(vm_page_lockptr(m));
921 	if (lockacq)
922 		vm_page_lock(m);
923 	vm_page_xunbusy_locked(m);
924 	if (lockacq)
925 		vm_page_unlock(m);
926 }
927 
928 /*
929  *	vm_page_xunbusy_hard:
930  *
931  *	Called after the first try the exclusive unbusy of a page failed.
932  *	It is assumed that the waiters bit is on.
933  */
934 void
935 vm_page_xunbusy_hard(vm_page_t m)
936 {
937 
938 	vm_page_assert_xbusied(m);
939 
940 	vm_page_lock(m);
941 	vm_page_xunbusy_locked(m);
942 	vm_page_unlock(m);
943 }
944 
945 /*
946  *	vm_page_flash:
947  *
948  *	Wakeup anyone waiting for the page.
949  *	The ownership bits do not change.
950  *
951  *	The given page must be locked.
952  */
953 void
954 vm_page_flash(vm_page_t m)
955 {
956 	u_int x;
957 
958 	vm_page_lock_assert(m, MA_OWNED);
959 
960 	for (;;) {
961 		x = m->busy_lock;
962 		if ((x & VPB_BIT_WAITERS) == 0)
963 			return;
964 		if (atomic_cmpset_int(&m->busy_lock, x,
965 		    x & (~VPB_BIT_WAITERS)))
966 			break;
967 	}
968 	wakeup(m);
969 }
970 
971 /*
972  * Avoid releasing and reacquiring the same page lock.
973  */
974 void
975 vm_page_change_lock(vm_page_t m, struct mtx **mtx)
976 {
977 	struct mtx *mtx1;
978 
979 	mtx1 = vm_page_lockptr(m);
980 	if (*mtx == mtx1)
981 		return;
982 	if (*mtx != NULL)
983 		mtx_unlock(*mtx);
984 	*mtx = mtx1;
985 	mtx_lock(mtx1);
986 }
987 
988 /*
989  * Keep page from being freed by the page daemon
990  * much of the same effect as wiring, except much lower
991  * overhead and should be used only for *very* temporary
992  * holding ("wiring").
993  */
994 void
995 vm_page_hold(vm_page_t mem)
996 {
997 
998 	vm_page_lock_assert(mem, MA_OWNED);
999         mem->hold_count++;
1000 }
1001 
1002 void
1003 vm_page_unhold(vm_page_t mem)
1004 {
1005 
1006 	vm_page_lock_assert(mem, MA_OWNED);
1007 	KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
1008 	--mem->hold_count;
1009 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
1010 		vm_page_free_toq(mem);
1011 }
1012 
1013 /*
1014  *	vm_page_unhold_pages:
1015  *
1016  *	Unhold each of the pages that is referenced by the given array.
1017  */
1018 void
1019 vm_page_unhold_pages(vm_page_t *ma, int count)
1020 {
1021 	struct mtx *mtx;
1022 
1023 	mtx = NULL;
1024 	for (; count != 0; count--) {
1025 		vm_page_change_lock(*ma, &mtx);
1026 		vm_page_unhold(*ma);
1027 		ma++;
1028 	}
1029 	if (mtx != NULL)
1030 		mtx_unlock(mtx);
1031 }
1032 
1033 vm_page_t
1034 PHYS_TO_VM_PAGE(vm_paddr_t pa)
1035 {
1036 	vm_page_t m;
1037 
1038 #ifdef VM_PHYSSEG_SPARSE
1039 	m = vm_phys_paddr_to_vm_page(pa);
1040 	if (m == NULL)
1041 		m = vm_phys_fictitious_to_vm_page(pa);
1042 	return (m);
1043 #elif defined(VM_PHYSSEG_DENSE)
1044 	long pi;
1045 
1046 	pi = atop(pa);
1047 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
1048 		m = &vm_page_array[pi - first_page];
1049 		return (m);
1050 	}
1051 	return (vm_phys_fictitious_to_vm_page(pa));
1052 #else
1053 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
1054 #endif
1055 }
1056 
1057 /*
1058  *	vm_page_getfake:
1059  *
1060  *	Create a fictitious page with the specified physical address and
1061  *	memory attribute.  The memory attribute is the only the machine-
1062  *	dependent aspect of a fictitious page that must be initialized.
1063  */
1064 vm_page_t
1065 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
1066 {
1067 	vm_page_t m;
1068 
1069 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
1070 	vm_page_initfake(m, paddr, memattr);
1071 	return (m);
1072 }
1073 
1074 void
1075 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1076 {
1077 
1078 	if ((m->flags & PG_FICTITIOUS) != 0) {
1079 		/*
1080 		 * The page's memattr might have changed since the
1081 		 * previous initialization.  Update the pmap to the
1082 		 * new memattr.
1083 		 */
1084 		goto memattr;
1085 	}
1086 	m->phys_addr = paddr;
1087 	m->queue = PQ_NONE;
1088 	/* Fictitious pages don't use "segind". */
1089 	m->flags = PG_FICTITIOUS;
1090 	/* Fictitious pages don't use "order" or "pool". */
1091 	m->oflags = VPO_UNMANAGED;
1092 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1093 	m->wire_count = 1;
1094 	pmap_page_init(m);
1095 memattr:
1096 	pmap_page_set_memattr(m, memattr);
1097 }
1098 
1099 /*
1100  *	vm_page_putfake:
1101  *
1102  *	Release a fictitious page.
1103  */
1104 void
1105 vm_page_putfake(vm_page_t m)
1106 {
1107 
1108 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
1109 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
1110 	    ("vm_page_putfake: bad page %p", m));
1111 	uma_zfree(fakepg_zone, m);
1112 }
1113 
1114 /*
1115  *	vm_page_updatefake:
1116  *
1117  *	Update the given fictitious page to the specified physical address and
1118  *	memory attribute.
1119  */
1120 void
1121 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
1122 {
1123 
1124 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
1125 	    ("vm_page_updatefake: bad page %p", m));
1126 	m->phys_addr = paddr;
1127 	pmap_page_set_memattr(m, memattr);
1128 }
1129 
1130 /*
1131  *	vm_page_free:
1132  *
1133  *	Free a page.
1134  */
1135 void
1136 vm_page_free(vm_page_t m)
1137 {
1138 
1139 	m->flags &= ~PG_ZERO;
1140 	vm_page_free_toq(m);
1141 }
1142 
1143 /*
1144  *	vm_page_free_zero:
1145  *
1146  *	Free a page to the zerod-pages queue
1147  */
1148 void
1149 vm_page_free_zero(vm_page_t m)
1150 {
1151 
1152 	m->flags |= PG_ZERO;
1153 	vm_page_free_toq(m);
1154 }
1155 
1156 /*
1157  * Unbusy and handle the page queueing for a page from a getpages request that
1158  * was optionally read ahead or behind.
1159  */
1160 void
1161 vm_page_readahead_finish(vm_page_t m)
1162 {
1163 
1164 	/* We shouldn't put invalid pages on queues. */
1165 	KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m));
1166 
1167 	/*
1168 	 * Since the page is not the actually needed one, whether it should
1169 	 * be activated or deactivated is not obvious.  Empirical results
1170 	 * have shown that deactivating the page is usually the best choice,
1171 	 * unless the page is wanted by another thread.
1172 	 */
1173 	vm_page_lock(m);
1174 	if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1175 		vm_page_activate(m);
1176 	else
1177 		vm_page_deactivate(m);
1178 	vm_page_unlock(m);
1179 	vm_page_xunbusy(m);
1180 }
1181 
1182 /*
1183  *	vm_page_sleep_if_busy:
1184  *
1185  *	Sleep and release the page queues lock if the page is busied.
1186  *	Returns TRUE if the thread slept.
1187  *
1188  *	The given page must be unlocked and object containing it must
1189  *	be locked.
1190  */
1191 int
1192 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1193 {
1194 	vm_object_t obj;
1195 
1196 	vm_page_lock_assert(m, MA_NOTOWNED);
1197 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1198 
1199 	if (vm_page_busied(m)) {
1200 		/*
1201 		 * The page-specific object must be cached because page
1202 		 * identity can change during the sleep, causing the
1203 		 * re-lock of a different object.
1204 		 * It is assumed that a reference to the object is already
1205 		 * held by the callers.
1206 		 */
1207 		obj = m->object;
1208 		vm_page_lock(m);
1209 		VM_OBJECT_WUNLOCK(obj);
1210 		vm_page_busy_sleep(m, msg, false);
1211 		VM_OBJECT_WLOCK(obj);
1212 		return (TRUE);
1213 	}
1214 	return (FALSE);
1215 }
1216 
1217 /*
1218  *	vm_page_dirty_KBI:		[ internal use only ]
1219  *
1220  *	Set all bits in the page's dirty field.
1221  *
1222  *	The object containing the specified page must be locked if the
1223  *	call is made from the machine-independent layer.
1224  *
1225  *	See vm_page_clear_dirty_mask().
1226  *
1227  *	This function should only be called by vm_page_dirty().
1228  */
1229 void
1230 vm_page_dirty_KBI(vm_page_t m)
1231 {
1232 
1233 	/* Refer to this operation by its public name. */
1234 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
1235 	    ("vm_page_dirty: page is invalid!"));
1236 	m->dirty = VM_PAGE_BITS_ALL;
1237 }
1238 
1239 /*
1240  *	vm_page_insert:		[ internal use only ]
1241  *
1242  *	Inserts the given mem entry into the object and object list.
1243  *
1244  *	The object must be locked.
1245  */
1246 int
1247 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1248 {
1249 	vm_page_t mpred;
1250 
1251 	VM_OBJECT_ASSERT_WLOCKED(object);
1252 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
1253 	return (vm_page_insert_after(m, object, pindex, mpred));
1254 }
1255 
1256 /*
1257  *	vm_page_insert_after:
1258  *
1259  *	Inserts the page "m" into the specified object at offset "pindex".
1260  *
1261  *	The page "mpred" must immediately precede the offset "pindex" within
1262  *	the specified object.
1263  *
1264  *	The object must be locked.
1265  */
1266 static int
1267 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1268     vm_page_t mpred)
1269 {
1270 	vm_page_t msucc;
1271 
1272 	VM_OBJECT_ASSERT_WLOCKED(object);
1273 	KASSERT(m->object == NULL,
1274 	    ("vm_page_insert_after: page already inserted"));
1275 	if (mpred != NULL) {
1276 		KASSERT(mpred->object == object,
1277 		    ("vm_page_insert_after: object doesn't contain mpred"));
1278 		KASSERT(mpred->pindex < pindex,
1279 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1280 		msucc = TAILQ_NEXT(mpred, listq);
1281 	} else
1282 		msucc = TAILQ_FIRST(&object->memq);
1283 	if (msucc != NULL)
1284 		KASSERT(msucc->pindex > pindex,
1285 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
1286 
1287 	/*
1288 	 * Record the object/offset pair in this page
1289 	 */
1290 	m->object = object;
1291 	m->pindex = pindex;
1292 
1293 	/*
1294 	 * Now link into the object's ordered list of backed pages.
1295 	 */
1296 	if (vm_radix_insert(&object->rtree, m)) {
1297 		m->object = NULL;
1298 		m->pindex = 0;
1299 		return (1);
1300 	}
1301 	vm_page_insert_radixdone(m, object, mpred);
1302 	return (0);
1303 }
1304 
1305 /*
1306  *	vm_page_insert_radixdone:
1307  *
1308  *	Complete page "m" insertion into the specified object after the
1309  *	radix trie hooking.
1310  *
1311  *	The page "mpred" must precede the offset "m->pindex" within the
1312  *	specified object.
1313  *
1314  *	The object must be locked.
1315  */
1316 static void
1317 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1318 {
1319 
1320 	VM_OBJECT_ASSERT_WLOCKED(object);
1321 	KASSERT(object != NULL && m->object == object,
1322 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1323 	if (mpred != NULL) {
1324 		KASSERT(mpred->object == object,
1325 		    ("vm_page_insert_after: object doesn't contain mpred"));
1326 		KASSERT(mpred->pindex < m->pindex,
1327 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1328 	}
1329 
1330 	if (mpred != NULL)
1331 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1332 	else
1333 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1334 
1335 	/*
1336 	 * Show that the object has one more resident page.
1337 	 */
1338 	object->resident_page_count++;
1339 
1340 	/*
1341 	 * Hold the vnode until the last page is released.
1342 	 */
1343 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1344 		vhold(object->handle);
1345 
1346 	/*
1347 	 * Since we are inserting a new and possibly dirty page,
1348 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1349 	 */
1350 	if (pmap_page_is_write_mapped(m))
1351 		vm_object_set_writeable_dirty(object);
1352 }
1353 
1354 /*
1355  *	vm_page_remove:
1356  *
1357  *	Removes the specified page from its containing object, but does not
1358  *	invalidate any backing storage.
1359  *
1360  *	The object must be locked.  The page must be locked if it is managed.
1361  */
1362 void
1363 vm_page_remove(vm_page_t m)
1364 {
1365 	vm_object_t object;
1366 	vm_page_t mrem;
1367 
1368 	if ((m->oflags & VPO_UNMANAGED) == 0)
1369 		vm_page_assert_locked(m);
1370 	if ((object = m->object) == NULL)
1371 		return;
1372 	VM_OBJECT_ASSERT_WLOCKED(object);
1373 	if (vm_page_xbusied(m))
1374 		vm_page_xunbusy_maybelocked(m);
1375 	mrem = vm_radix_remove(&object->rtree, m->pindex);
1376 	KASSERT(mrem == m, ("removed page %p, expected page %p", mrem, m));
1377 
1378 	/*
1379 	 * Now remove from the object's list of backed pages.
1380 	 */
1381 	TAILQ_REMOVE(&object->memq, m, listq);
1382 
1383 	/*
1384 	 * And show that the object has one fewer resident page.
1385 	 */
1386 	object->resident_page_count--;
1387 
1388 	/*
1389 	 * The vnode may now be recycled.
1390 	 */
1391 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1392 		vdrop(object->handle);
1393 
1394 	m->object = NULL;
1395 }
1396 
1397 /*
1398  *	vm_page_lookup:
1399  *
1400  *	Returns the page associated with the object/offset
1401  *	pair specified; if none is found, NULL is returned.
1402  *
1403  *	The object must be locked.
1404  */
1405 vm_page_t
1406 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1407 {
1408 
1409 	VM_OBJECT_ASSERT_LOCKED(object);
1410 	return (vm_radix_lookup(&object->rtree, pindex));
1411 }
1412 
1413 /*
1414  *	vm_page_find_least:
1415  *
1416  *	Returns the page associated with the object with least pindex
1417  *	greater than or equal to the parameter pindex, or NULL.
1418  *
1419  *	The object must be locked.
1420  */
1421 vm_page_t
1422 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1423 {
1424 	vm_page_t m;
1425 
1426 	VM_OBJECT_ASSERT_LOCKED(object);
1427 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1428 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1429 	return (m);
1430 }
1431 
1432 /*
1433  * Returns the given page's successor (by pindex) within the object if it is
1434  * resident; if none is found, NULL is returned.
1435  *
1436  * The object must be locked.
1437  */
1438 vm_page_t
1439 vm_page_next(vm_page_t m)
1440 {
1441 	vm_page_t next;
1442 
1443 	VM_OBJECT_ASSERT_LOCKED(m->object);
1444 	if ((next = TAILQ_NEXT(m, listq)) != NULL) {
1445 		MPASS(next->object == m->object);
1446 		if (next->pindex != m->pindex + 1)
1447 			next = NULL;
1448 	}
1449 	return (next);
1450 }
1451 
1452 /*
1453  * Returns the given page's predecessor (by pindex) within the object if it is
1454  * resident; if none is found, NULL is returned.
1455  *
1456  * The object must be locked.
1457  */
1458 vm_page_t
1459 vm_page_prev(vm_page_t m)
1460 {
1461 	vm_page_t prev;
1462 
1463 	VM_OBJECT_ASSERT_LOCKED(m->object);
1464 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL) {
1465 		MPASS(prev->object == m->object);
1466 		if (prev->pindex != m->pindex - 1)
1467 			prev = NULL;
1468 	}
1469 	return (prev);
1470 }
1471 
1472 /*
1473  * Uses the page mnew as a replacement for an existing page at index
1474  * pindex which must be already present in the object.
1475  *
1476  * The existing page must not be on a paging queue.
1477  */
1478 vm_page_t
1479 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1480 {
1481 	vm_page_t mold;
1482 
1483 	VM_OBJECT_ASSERT_WLOCKED(object);
1484 	KASSERT(mnew->object == NULL,
1485 	    ("vm_page_replace: page already in object"));
1486 
1487 	/*
1488 	 * This function mostly follows vm_page_insert() and
1489 	 * vm_page_remove() without the radix, object count and vnode
1490 	 * dance.  Double check such functions for more comments.
1491 	 */
1492 
1493 	mnew->object = object;
1494 	mnew->pindex = pindex;
1495 	mold = vm_radix_replace(&object->rtree, mnew);
1496 	KASSERT(mold->queue == PQ_NONE,
1497 	    ("vm_page_replace: mold is on a paging queue"));
1498 
1499 	/* Keep the resident page list in sorted order. */
1500 	TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1501 	TAILQ_REMOVE(&object->memq, mold, listq);
1502 
1503 	mold->object = NULL;
1504 	vm_page_xunbusy_maybelocked(mold);
1505 
1506 	/*
1507 	 * The object's resident_page_count does not change because we have
1508 	 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1509 	 */
1510 	if (pmap_page_is_write_mapped(mnew))
1511 		vm_object_set_writeable_dirty(object);
1512 	return (mold);
1513 }
1514 
1515 /*
1516  *	vm_page_rename:
1517  *
1518  *	Move the given memory entry from its
1519  *	current object to the specified target object/offset.
1520  *
1521  *	Note: swap associated with the page must be invalidated by the move.  We
1522  *	      have to do this for several reasons:  (1) we aren't freeing the
1523  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1524  *	      moving the page from object A to B, and will then later move
1525  *	      the backing store from A to B and we can't have a conflict.
1526  *
1527  *	Note: we *always* dirty the page.  It is necessary both for the
1528  *	      fact that we moved it, and because we may be invalidating
1529  *	      swap.
1530  *
1531  *	The objects must be locked.
1532  */
1533 int
1534 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1535 {
1536 	vm_page_t mpred;
1537 	vm_pindex_t opidx;
1538 
1539 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1540 
1541 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1542 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1543 	    ("vm_page_rename: pindex already renamed"));
1544 
1545 	/*
1546 	 * Create a custom version of vm_page_insert() which does not depend
1547 	 * by m_prev and can cheat on the implementation aspects of the
1548 	 * function.
1549 	 */
1550 	opidx = m->pindex;
1551 	m->pindex = new_pindex;
1552 	if (vm_radix_insert(&new_object->rtree, m)) {
1553 		m->pindex = opidx;
1554 		return (1);
1555 	}
1556 
1557 	/*
1558 	 * The operation cannot fail anymore.  The removal must happen before
1559 	 * the listq iterator is tainted.
1560 	 */
1561 	m->pindex = opidx;
1562 	vm_page_lock(m);
1563 	vm_page_remove(m);
1564 
1565 	/* Return back to the new pindex to complete vm_page_insert(). */
1566 	m->pindex = new_pindex;
1567 	m->object = new_object;
1568 	vm_page_unlock(m);
1569 	vm_page_insert_radixdone(m, new_object, mpred);
1570 	vm_page_dirty(m);
1571 	return (0);
1572 }
1573 
1574 /*
1575  *	vm_page_alloc:
1576  *
1577  *	Allocate and return a page that is associated with the specified
1578  *	object and offset pair.  By default, this page is exclusive busied.
1579  *
1580  *	The caller must always specify an allocation class.
1581  *
1582  *	allocation classes:
1583  *	VM_ALLOC_NORMAL		normal process request
1584  *	VM_ALLOC_SYSTEM		system *really* needs a page
1585  *	VM_ALLOC_INTERRUPT	interrupt time request
1586  *
1587  *	optional allocation flags:
1588  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1589  *				intends to allocate
1590  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1591  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1592  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1593  *				should not be exclusive busy
1594  *	VM_ALLOC_SBUSY		shared busy the allocated page
1595  *	VM_ALLOC_WIRED		wire the allocated page
1596  *	VM_ALLOC_ZERO		prefer a zeroed page
1597  */
1598 vm_page_t
1599 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1600 {
1601 
1602 	return (vm_page_alloc_after(object, pindex, req, object != NULL ?
1603 	    vm_radix_lookup_le(&object->rtree, pindex) : NULL));
1604 }
1605 
1606 vm_page_t
1607 vm_page_alloc_domain(vm_object_t object, vm_pindex_t pindex, int domain,
1608     int req)
1609 {
1610 
1611 	return (vm_page_alloc_domain_after(object, pindex, domain, req,
1612 	    object != NULL ? vm_radix_lookup_le(&object->rtree, pindex) :
1613 	    NULL));
1614 }
1615 
1616 /*
1617  * Allocate a page in the specified object with the given page index.  To
1618  * optimize insertion of the page into the object, the caller must also specifiy
1619  * the resident page in the object with largest index smaller than the given
1620  * page index, or NULL if no such page exists.
1621  */
1622 vm_page_t
1623 vm_page_alloc_after(vm_object_t object, vm_pindex_t pindex,
1624     int req, vm_page_t mpred)
1625 {
1626 	struct vm_domainset_iter di;
1627 	vm_page_t m;
1628 	int domain;
1629 
1630 	vm_domainset_iter_page_init(&di, object, &domain, &req);
1631 	do {
1632 		m = vm_page_alloc_domain_after(object, pindex, domain, req,
1633 		    mpred);
1634 		if (m != NULL)
1635 			break;
1636 	} while (vm_domainset_iter_page(&di, &domain, &req) == 0);
1637 
1638 	return (m);
1639 }
1640 
1641 vm_page_t
1642 vm_page_alloc_domain_after(vm_object_t object, vm_pindex_t pindex, int domain,
1643     int req, vm_page_t mpred)
1644 {
1645 	vm_page_t m;
1646 	int flags, req_class;
1647 	u_int free_count;
1648 
1649 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1650 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1651 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1652 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1653 	    ("inconsistent object(%p)/req(%x)", object, req));
1654 	KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0,
1655 	    ("Can't sleep and retry object insertion."));
1656 	KASSERT(mpred == NULL || mpred->pindex < pindex,
1657 	    ("mpred %p doesn't precede pindex 0x%jx", mpred,
1658 	    (uintmax_t)pindex));
1659 	if (object != NULL)
1660 		VM_OBJECT_ASSERT_WLOCKED(object);
1661 
1662 	req_class = req & VM_ALLOC_CLASS_MASK;
1663 
1664 	/*
1665 	 * The page daemon is allowed to dig deeper into the free page list.
1666 	 */
1667 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1668 		req_class = VM_ALLOC_SYSTEM;
1669 
1670 	/*
1671 	 * Allocate a page if the number of free pages exceeds the minimum
1672 	 * for the request class.
1673 	 */
1674 again:
1675 	m = NULL;
1676 	mtx_lock(&vm_page_queue_free_mtx);
1677 	if (vm_cnt.v_free_count > vm_cnt.v_free_reserved ||
1678 	    (req_class == VM_ALLOC_SYSTEM &&
1679 	    vm_cnt.v_free_count > vm_cnt.v_interrupt_free_min) ||
1680 	    (req_class == VM_ALLOC_INTERRUPT &&
1681 	    vm_cnt.v_free_count > 0)) {
1682 		/*
1683 		 * Can we allocate the page from a reservation?
1684 		 */
1685 #if VM_NRESERVLEVEL > 0
1686 		if (object == NULL || (object->flags & (OBJ_COLORED |
1687 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1688 		    vm_reserv_alloc_page(object, pindex, domain,
1689 		    mpred)) == NULL)
1690 #endif
1691 		{
1692 			/*
1693 			 * If not, allocate it from the free page queues.
1694 			 */
1695 			m = vm_phys_alloc_pages(domain, object != NULL ?
1696 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1697 #if VM_NRESERVLEVEL > 0
1698 			if (m == NULL && vm_reserv_reclaim_inactive(domain)) {
1699 				m = vm_phys_alloc_pages(domain,
1700 				    object != NULL ?
1701 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1702 				    0);
1703 			}
1704 #endif
1705 		}
1706 	}
1707 	if (m == NULL) {
1708 		/*
1709 		 * Not allocatable, give up.
1710 		 */
1711 		if (vm_page_alloc_fail(object, req))
1712 			goto again;
1713 		return (NULL);
1714 	}
1715 
1716 	/*
1717 	 *  At this point we had better have found a good page.
1718 	 */
1719 	KASSERT(m != NULL, ("missing page"));
1720 	free_count = vm_phys_freecnt_adj(m, -1);
1721 	mtx_unlock(&vm_page_queue_free_mtx);
1722 	vm_page_alloc_check(m);
1723 
1724 	/*
1725 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1726 	 */
1727 	flags = 0;
1728 	if ((req & VM_ALLOC_ZERO) != 0)
1729 		flags = PG_ZERO;
1730 	flags &= m->flags;
1731 	if ((req & VM_ALLOC_NODUMP) != 0)
1732 		flags |= PG_NODUMP;
1733 	m->flags = flags;
1734 	m->aflags = 0;
1735 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1736 	    VPO_UNMANAGED : 0;
1737 	m->busy_lock = VPB_UNBUSIED;
1738 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1739 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1740 	if ((req & VM_ALLOC_SBUSY) != 0)
1741 		m->busy_lock = VPB_SHARERS_WORD(1);
1742 	if (req & VM_ALLOC_WIRED) {
1743 		/*
1744 		 * The page lock is not required for wiring a page until that
1745 		 * page is inserted into the object.
1746 		 */
1747 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1748 		m->wire_count = 1;
1749 	}
1750 	m->act_count = 0;
1751 
1752 	if (object != NULL) {
1753 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1754 			pagedaemon_wakeup();
1755 			if (req & VM_ALLOC_WIRED) {
1756 				atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1757 				m->wire_count = 0;
1758 			}
1759 			KASSERT(m->object == NULL, ("page %p has object", m));
1760 			m->oflags = VPO_UNMANAGED;
1761 			m->busy_lock = VPB_UNBUSIED;
1762 			/* Don't change PG_ZERO. */
1763 			vm_page_free_toq(m);
1764 			if (req & VM_ALLOC_WAITFAIL) {
1765 				VM_OBJECT_WUNLOCK(object);
1766 				vm_radix_wait();
1767 				VM_OBJECT_WLOCK(object);
1768 			}
1769 			return (NULL);
1770 		}
1771 
1772 		/* Ignore device objects; the pager sets "memattr" for them. */
1773 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1774 		    (object->flags & OBJ_FICTITIOUS) == 0)
1775 			pmap_page_set_memattr(m, object->memattr);
1776 	} else
1777 		m->pindex = pindex;
1778 
1779 	/*
1780 	 * Don't wakeup too often - wakeup the pageout daemon when
1781 	 * we would be nearly out of memory.
1782 	 */
1783 	if (vm_paging_needed(free_count))
1784 		pagedaemon_wakeup();
1785 
1786 	return (m);
1787 }
1788 
1789 /*
1790  *	vm_page_alloc_contig:
1791  *
1792  *	Allocate a contiguous set of physical pages of the given size "npages"
1793  *	from the free lists.  All of the physical pages must be at or above
1794  *	the given physical address "low" and below the given physical address
1795  *	"high".  The given value "alignment" determines the alignment of the
1796  *	first physical page in the set.  If the given value "boundary" is
1797  *	non-zero, then the set of physical pages cannot cross any physical
1798  *	address boundary that is a multiple of that value.  Both "alignment"
1799  *	and "boundary" must be a power of two.
1800  *
1801  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1802  *	then the memory attribute setting for the physical pages is configured
1803  *	to the object's memory attribute setting.  Otherwise, the memory
1804  *	attribute setting for the physical pages is configured to "memattr",
1805  *	overriding the object's memory attribute setting.  However, if the
1806  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1807  *	memory attribute setting for the physical pages cannot be configured
1808  *	to VM_MEMATTR_DEFAULT.
1809  *
1810  *	The specified object may not contain fictitious pages.
1811  *
1812  *	The caller must always specify an allocation class.
1813  *
1814  *	allocation classes:
1815  *	VM_ALLOC_NORMAL		normal process request
1816  *	VM_ALLOC_SYSTEM		system *really* needs a page
1817  *	VM_ALLOC_INTERRUPT	interrupt time request
1818  *
1819  *	optional allocation flags:
1820  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1821  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1822  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1823  *				should not be exclusive busy
1824  *	VM_ALLOC_SBUSY		shared busy the allocated page
1825  *	VM_ALLOC_WIRED		wire the allocated page
1826  *	VM_ALLOC_ZERO		prefer a zeroed page
1827  */
1828 vm_page_t
1829 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1830     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1831     vm_paddr_t boundary, vm_memattr_t memattr)
1832 {
1833 	struct vm_domainset_iter di;
1834 	vm_page_t m;
1835 	int domain;
1836 
1837 	vm_domainset_iter_page_init(&di, object, &domain, &req);
1838 	do {
1839 		m = vm_page_alloc_contig_domain(object, pindex, domain, req,
1840 		    npages, low, high, alignment, boundary, memattr);
1841 		if (m != NULL)
1842 			break;
1843 	} while (vm_domainset_iter_page(&di, &domain, &req) == 0);
1844 
1845 	return (m);
1846 }
1847 
1848 vm_page_t
1849 vm_page_alloc_contig_domain(vm_object_t object, vm_pindex_t pindex, int domain,
1850     int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1851     vm_paddr_t boundary, vm_memattr_t memattr)
1852 {
1853 	vm_page_t m, m_ret, mpred;
1854 	u_int busy_lock, flags, oflags;
1855 	int req_class;
1856 
1857 	mpred = NULL;	/* XXX: pacify gcc */
1858 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1859 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1860 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1861 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1862 	    ("vm_page_alloc_contig: inconsistent object(%p)/req(%x)", object,
1863 	    req));
1864 	KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0,
1865 	    ("Can't sleep and retry object insertion."));
1866 	if (object != NULL) {
1867 		VM_OBJECT_ASSERT_WLOCKED(object);
1868 		KASSERT((object->flags & OBJ_FICTITIOUS) == 0,
1869 		    ("vm_page_alloc_contig: object %p has fictitious pages",
1870 		    object));
1871 	}
1872 	KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1873 	req_class = req & VM_ALLOC_CLASS_MASK;
1874 
1875 	/*
1876 	 * The page daemon is allowed to dig deeper into the free page list.
1877 	 */
1878 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1879 		req_class = VM_ALLOC_SYSTEM;
1880 
1881 	if (object != NULL) {
1882 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1883 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1884 		    ("vm_page_alloc_contig: pindex already allocated"));
1885 	}
1886 
1887 	/*
1888 	 * Can we allocate the pages without the number of free pages falling
1889 	 * below the lower bound for the allocation class?
1890 	 */
1891 again:
1892 	m_ret = NULL;
1893 	mtx_lock(&vm_page_queue_free_mtx);
1894 	if (vm_cnt.v_free_count >= npages + vm_cnt.v_free_reserved ||
1895 	    (req_class == VM_ALLOC_SYSTEM &&
1896 	    vm_cnt.v_free_count >= npages + vm_cnt.v_interrupt_free_min) ||
1897 	    (req_class == VM_ALLOC_INTERRUPT &&
1898 	    vm_cnt.v_free_count >= npages)) {
1899 		/*
1900 		 * Can we allocate the pages from a reservation?
1901 		 */
1902 #if VM_NRESERVLEVEL > 0
1903 retry:
1904 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1905 		    (m_ret = vm_reserv_alloc_contig(object, pindex, domain,
1906 		    npages, low, high, alignment, boundary, mpred)) == NULL)
1907 #endif
1908 			/*
1909 			 * If not, allocate them from the free page queues.
1910 			 */
1911 			m_ret = vm_phys_alloc_contig(domain, npages, low, high,
1912 			    alignment, boundary);
1913 #if VM_NRESERVLEVEL > 0
1914 		if (m_ret == NULL && vm_reserv_reclaim_contig(
1915 		    domain, npages, low, high, alignment, boundary))
1916 			goto retry;
1917 #endif
1918 	}
1919 	if (m_ret == NULL) {
1920 		if (vm_page_alloc_fail(object, req))
1921 			goto again;
1922 		return (NULL);
1923 	}
1924 	vm_phys_freecnt_adj(m_ret, -npages);
1925 	mtx_unlock(&vm_page_queue_free_mtx);
1926 	for (m = m_ret; m < &m_ret[npages]; m++)
1927 		vm_page_alloc_check(m);
1928 
1929 	/*
1930 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1931 	 */
1932 	flags = 0;
1933 	if ((req & VM_ALLOC_ZERO) != 0)
1934 		flags = PG_ZERO;
1935 	if ((req & VM_ALLOC_NODUMP) != 0)
1936 		flags |= PG_NODUMP;
1937 	oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1938 	    VPO_UNMANAGED : 0;
1939 	busy_lock = VPB_UNBUSIED;
1940 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1941 		busy_lock = VPB_SINGLE_EXCLUSIVER;
1942 	if ((req & VM_ALLOC_SBUSY) != 0)
1943 		busy_lock = VPB_SHARERS_WORD(1);
1944 	if ((req & VM_ALLOC_WIRED) != 0)
1945 		atomic_add_int(&vm_cnt.v_wire_count, npages);
1946 	if (object != NULL) {
1947 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1948 		    memattr == VM_MEMATTR_DEFAULT)
1949 			memattr = object->memattr;
1950 	}
1951 	for (m = m_ret; m < &m_ret[npages]; m++) {
1952 		m->aflags = 0;
1953 		m->flags = (m->flags | PG_NODUMP) & flags;
1954 		m->busy_lock = busy_lock;
1955 		if ((req & VM_ALLOC_WIRED) != 0)
1956 			m->wire_count = 1;
1957 		m->act_count = 0;
1958 		m->oflags = oflags;
1959 		if (object != NULL) {
1960 			if (vm_page_insert_after(m, object, pindex, mpred)) {
1961 				pagedaemon_wakeup();
1962 				if ((req & VM_ALLOC_WIRED) != 0)
1963 					atomic_subtract_int(
1964 					    &vm_cnt.v_wire_count, npages);
1965 				KASSERT(m->object == NULL,
1966 				    ("page %p has object", m));
1967 				mpred = m;
1968 				for (m = m_ret; m < &m_ret[npages]; m++) {
1969 					if (m <= mpred &&
1970 					    (req & VM_ALLOC_WIRED) != 0)
1971 						m->wire_count = 0;
1972 					m->oflags = VPO_UNMANAGED;
1973 					m->busy_lock = VPB_UNBUSIED;
1974 					/* Don't change PG_ZERO. */
1975 					vm_page_free_toq(m);
1976 				}
1977 				if (req & VM_ALLOC_WAITFAIL) {
1978 					VM_OBJECT_WUNLOCK(object);
1979 					vm_radix_wait();
1980 					VM_OBJECT_WLOCK(object);
1981 				}
1982 				return (NULL);
1983 			}
1984 			mpred = m;
1985 		} else
1986 			m->pindex = pindex;
1987 		if (memattr != VM_MEMATTR_DEFAULT)
1988 			pmap_page_set_memattr(m, memattr);
1989 		pindex++;
1990 	}
1991 	if (vm_paging_needed(vm_cnt.v_free_count))
1992 		pagedaemon_wakeup();
1993 	return (m_ret);
1994 }
1995 
1996 /*
1997  * Check a page that has been freshly dequeued from a freelist.
1998  */
1999 static void
2000 vm_page_alloc_check(vm_page_t m)
2001 {
2002 
2003 	KASSERT(m->object == NULL, ("page %p has object", m));
2004 	KASSERT(m->queue == PQ_NONE,
2005 	    ("page %p has unexpected queue %d", m, m->queue));
2006 	KASSERT(m->wire_count == 0, ("page %p is wired", m));
2007 	KASSERT(m->hold_count == 0, ("page %p is held", m));
2008 	KASSERT(!vm_page_busied(m), ("page %p is busy", m));
2009 	KASSERT(m->dirty == 0, ("page %p is dirty", m));
2010 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
2011 	    ("page %p has unexpected memattr %d",
2012 	    m, pmap_page_get_memattr(m)));
2013 	KASSERT(m->valid == 0, ("free page %p is valid", m));
2014 }
2015 
2016 /*
2017  * 	vm_page_alloc_freelist:
2018  *
2019  *	Allocate a physical page from the specified free page list.
2020  *
2021  *	The caller must always specify an allocation class.
2022  *
2023  *	allocation classes:
2024  *	VM_ALLOC_NORMAL		normal process request
2025  *	VM_ALLOC_SYSTEM		system *really* needs a page
2026  *	VM_ALLOC_INTERRUPT	interrupt time request
2027  *
2028  *	optional allocation flags:
2029  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
2030  *				intends to allocate
2031  *	VM_ALLOC_WIRED		wire the allocated page
2032  *	VM_ALLOC_ZERO		prefer a zeroed page
2033  */
2034 vm_page_t
2035 vm_page_alloc_freelist(int freelist, int req)
2036 {
2037 	struct vm_domainset_iter di;
2038 	vm_page_t m;
2039 	int domain;
2040 
2041 	vm_domainset_iter_page_init(&di, kernel_object, &domain, &req);
2042 	do {
2043 		m = vm_page_alloc_freelist_domain(domain, freelist, req);
2044 		if (m != NULL)
2045 			break;
2046 	} while (vm_domainset_iter_page(&di, &domain, &req) == 0);
2047 
2048 	return (m);
2049 }
2050 
2051 vm_page_t
2052 vm_page_alloc_freelist_domain(int domain, int freelist, int req)
2053 {
2054 	vm_page_t m;
2055 	u_int flags, free_count;
2056 	int req_class;
2057 
2058 	req_class = req & VM_ALLOC_CLASS_MASK;
2059 
2060 	/*
2061 	 * The page daemon is allowed to dig deeper into the free page list.
2062 	 */
2063 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2064 		req_class = VM_ALLOC_SYSTEM;
2065 
2066 	/*
2067 	 * Do not allocate reserved pages unless the req has asked for it.
2068 	 */
2069 again:
2070 	mtx_lock(&vm_page_queue_free_mtx);
2071 	if (vm_cnt.v_free_count > vm_cnt.v_free_reserved ||
2072 	    (req_class == VM_ALLOC_SYSTEM &&
2073 	    vm_cnt.v_free_count > vm_cnt.v_interrupt_free_min) ||
2074 	    (req_class == VM_ALLOC_INTERRUPT &&
2075 	    vm_cnt.v_free_count > 0))
2076 		m = vm_phys_alloc_freelist_pages(domain, freelist,
2077 		    VM_FREEPOOL_DIRECT, 0);
2078 	if (m == NULL) {
2079 		if (vm_page_alloc_fail(NULL, req))
2080 			goto again;
2081 		return (NULL);
2082 	}
2083 	free_count = vm_phys_freecnt_adj(m, -1);
2084 	mtx_unlock(&vm_page_queue_free_mtx);
2085 	vm_page_alloc_check(m);
2086 
2087 	/*
2088 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
2089 	 */
2090 	m->aflags = 0;
2091 	flags = 0;
2092 	if ((req & VM_ALLOC_ZERO) != 0)
2093 		flags = PG_ZERO;
2094 	m->flags &= flags;
2095 	if ((req & VM_ALLOC_WIRED) != 0) {
2096 		/*
2097 		 * The page lock is not required for wiring a page that does
2098 		 * not belong to an object.
2099 		 */
2100 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2101 		m->wire_count = 1;
2102 	}
2103 	/* Unmanaged pages don't use "act_count". */
2104 	m->oflags = VPO_UNMANAGED;
2105 	if (vm_paging_needed(free_count))
2106 		pagedaemon_wakeup();
2107 	return (m);
2108 }
2109 
2110 #define	VPSC_ANY	0	/* No restrictions. */
2111 #define	VPSC_NORESERV	1	/* Skip reservations; implies VPSC_NOSUPER. */
2112 #define	VPSC_NOSUPER	2	/* Skip superpages. */
2113 
2114 /*
2115  *	vm_page_scan_contig:
2116  *
2117  *	Scan vm_page_array[] between the specified entries "m_start" and
2118  *	"m_end" for a run of contiguous physical pages that satisfy the
2119  *	specified conditions, and return the lowest page in the run.  The
2120  *	specified "alignment" determines the alignment of the lowest physical
2121  *	page in the run.  If the specified "boundary" is non-zero, then the
2122  *	run of physical pages cannot span a physical address that is a
2123  *	multiple of "boundary".
2124  *
2125  *	"m_end" is never dereferenced, so it need not point to a vm_page
2126  *	structure within vm_page_array[].
2127  *
2128  *	"npages" must be greater than zero.  "m_start" and "m_end" must not
2129  *	span a hole (or discontiguity) in the physical address space.  Both
2130  *	"alignment" and "boundary" must be a power of two.
2131  */
2132 vm_page_t
2133 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
2134     u_long alignment, vm_paddr_t boundary, int options)
2135 {
2136 	struct mtx *m_mtx;
2137 	vm_object_t object;
2138 	vm_paddr_t pa;
2139 	vm_page_t m, m_run;
2140 #if VM_NRESERVLEVEL > 0
2141 	int level;
2142 #endif
2143 	int m_inc, order, run_ext, run_len;
2144 
2145 	KASSERT(npages > 0, ("npages is 0"));
2146 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2147 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2148 	m_run = NULL;
2149 	run_len = 0;
2150 	m_mtx = NULL;
2151 	for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
2152 		KASSERT((m->flags & PG_MARKER) == 0,
2153 		    ("page %p is PG_MARKER", m));
2154 		KASSERT((m->flags & PG_FICTITIOUS) == 0 || m->wire_count == 1,
2155 		    ("fictitious page %p has invalid wire count", m));
2156 
2157 		/*
2158 		 * If the current page would be the start of a run, check its
2159 		 * physical address against the end, alignment, and boundary
2160 		 * conditions.  If it doesn't satisfy these conditions, either
2161 		 * terminate the scan or advance to the next page that
2162 		 * satisfies the failed condition.
2163 		 */
2164 		if (run_len == 0) {
2165 			KASSERT(m_run == NULL, ("m_run != NULL"));
2166 			if (m + npages > m_end)
2167 				break;
2168 			pa = VM_PAGE_TO_PHYS(m);
2169 			if ((pa & (alignment - 1)) != 0) {
2170 				m_inc = atop(roundup2(pa, alignment) - pa);
2171 				continue;
2172 			}
2173 			if (rounddown2(pa ^ (pa + ptoa(npages) - 1),
2174 			    boundary) != 0) {
2175 				m_inc = atop(roundup2(pa, boundary) - pa);
2176 				continue;
2177 			}
2178 		} else
2179 			KASSERT(m_run != NULL, ("m_run == NULL"));
2180 
2181 		vm_page_change_lock(m, &m_mtx);
2182 		m_inc = 1;
2183 retry:
2184 		if (m->wire_count != 0 || m->hold_count != 0)
2185 			run_ext = 0;
2186 #if VM_NRESERVLEVEL > 0
2187 		else if ((level = vm_reserv_level(m)) >= 0 &&
2188 		    (options & VPSC_NORESERV) != 0) {
2189 			run_ext = 0;
2190 			/* Advance to the end of the reservation. */
2191 			pa = VM_PAGE_TO_PHYS(m);
2192 			m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
2193 			    pa);
2194 		}
2195 #endif
2196 		else if ((object = m->object) != NULL) {
2197 			/*
2198 			 * The page is considered eligible for relocation if
2199 			 * and only if it could be laundered or reclaimed by
2200 			 * the page daemon.
2201 			 */
2202 			if (!VM_OBJECT_TRYRLOCK(object)) {
2203 				mtx_unlock(m_mtx);
2204 				VM_OBJECT_RLOCK(object);
2205 				mtx_lock(m_mtx);
2206 				if (m->object != object) {
2207 					/*
2208 					 * The page may have been freed.
2209 					 */
2210 					VM_OBJECT_RUNLOCK(object);
2211 					goto retry;
2212 				} else if (m->wire_count != 0 ||
2213 				    m->hold_count != 0) {
2214 					run_ext = 0;
2215 					goto unlock;
2216 				}
2217 			}
2218 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2219 			    ("page %p is PG_UNHOLDFREE", m));
2220 			/* Don't care: PG_NODUMP, PG_ZERO. */
2221 			if (object->type != OBJT_DEFAULT &&
2222 			    object->type != OBJT_SWAP &&
2223 			    object->type != OBJT_VNODE) {
2224 				run_ext = 0;
2225 #if VM_NRESERVLEVEL > 0
2226 			} else if ((options & VPSC_NOSUPER) != 0 &&
2227 			    (level = vm_reserv_level_iffullpop(m)) >= 0) {
2228 				run_ext = 0;
2229 				/* Advance to the end of the superpage. */
2230 				pa = VM_PAGE_TO_PHYS(m);
2231 				m_inc = atop(roundup2(pa + 1,
2232 				    vm_reserv_size(level)) - pa);
2233 #endif
2234 			} else if (object->memattr == VM_MEMATTR_DEFAULT &&
2235 			    m->queue != PQ_NONE && !vm_page_busied(m)) {
2236 				/*
2237 				 * The page is allocated but eligible for
2238 				 * relocation.  Extend the current run by one
2239 				 * page.
2240 				 */
2241 				KASSERT(pmap_page_get_memattr(m) ==
2242 				    VM_MEMATTR_DEFAULT,
2243 				    ("page %p has an unexpected memattr", m));
2244 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2245 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2246 				    ("page %p has unexpected oflags", m));
2247 				/* Don't care: VPO_NOSYNC. */
2248 				run_ext = 1;
2249 			} else
2250 				run_ext = 0;
2251 unlock:
2252 			VM_OBJECT_RUNLOCK(object);
2253 #if VM_NRESERVLEVEL > 0
2254 		} else if (level >= 0) {
2255 			/*
2256 			 * The page is reserved but not yet allocated.  In
2257 			 * other words, it is still free.  Extend the current
2258 			 * run by one page.
2259 			 */
2260 			run_ext = 1;
2261 #endif
2262 		} else if ((order = m->order) < VM_NFREEORDER) {
2263 			/*
2264 			 * The page is enqueued in the physical memory
2265 			 * allocator's free page queues.  Moreover, it is the
2266 			 * first page in a power-of-two-sized run of
2267 			 * contiguous free pages.  Add these pages to the end
2268 			 * of the current run, and jump ahead.
2269 			 */
2270 			run_ext = 1 << order;
2271 			m_inc = 1 << order;
2272 		} else {
2273 			/*
2274 			 * Skip the page for one of the following reasons: (1)
2275 			 * It is enqueued in the physical memory allocator's
2276 			 * free page queues.  However, it is not the first
2277 			 * page in a run of contiguous free pages.  (This case
2278 			 * rarely occurs because the scan is performed in
2279 			 * ascending order.) (2) It is not reserved, and it is
2280 			 * transitioning from free to allocated.  (Conversely,
2281 			 * the transition from allocated to free for managed
2282 			 * pages is blocked by the page lock.) (3) It is
2283 			 * allocated but not contained by an object and not
2284 			 * wired, e.g., allocated by Xen's balloon driver.
2285 			 */
2286 			run_ext = 0;
2287 		}
2288 
2289 		/*
2290 		 * Extend or reset the current run of pages.
2291 		 */
2292 		if (run_ext > 0) {
2293 			if (run_len == 0)
2294 				m_run = m;
2295 			run_len += run_ext;
2296 		} else {
2297 			if (run_len > 0) {
2298 				m_run = NULL;
2299 				run_len = 0;
2300 			}
2301 		}
2302 	}
2303 	if (m_mtx != NULL)
2304 		mtx_unlock(m_mtx);
2305 	if (run_len >= npages)
2306 		return (m_run);
2307 	return (NULL);
2308 }
2309 
2310 /*
2311  *	vm_page_reclaim_run:
2312  *
2313  *	Try to relocate each of the allocated virtual pages within the
2314  *	specified run of physical pages to a new physical address.  Free the
2315  *	physical pages underlying the relocated virtual pages.  A virtual page
2316  *	is relocatable if and only if it could be laundered or reclaimed by
2317  *	the page daemon.  Whenever possible, a virtual page is relocated to a
2318  *	physical address above "high".
2319  *
2320  *	Returns 0 if every physical page within the run was already free or
2321  *	just freed by a successful relocation.  Otherwise, returns a non-zero
2322  *	value indicating why the last attempt to relocate a virtual page was
2323  *	unsuccessful.
2324  *
2325  *	"req_class" must be an allocation class.
2326  */
2327 static int
2328 vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
2329     vm_paddr_t high)
2330 {
2331 	struct mtx *m_mtx;
2332 	struct spglist free;
2333 	vm_object_t object;
2334 	vm_paddr_t pa;
2335 	vm_page_t m, m_end, m_new;
2336 	int error, order, req;
2337 
2338 	KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
2339 	    ("req_class is not an allocation class"));
2340 	SLIST_INIT(&free);
2341 	error = 0;
2342 	m = m_run;
2343 	m_end = m_run + npages;
2344 	m_mtx = NULL;
2345 	for (; error == 0 && m < m_end; m++) {
2346 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2347 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2348 
2349 		/*
2350 		 * Avoid releasing and reacquiring the same page lock.
2351 		 */
2352 		vm_page_change_lock(m, &m_mtx);
2353 retry:
2354 		if (m->wire_count != 0 || m->hold_count != 0)
2355 			error = EBUSY;
2356 		else if ((object = m->object) != NULL) {
2357 			/*
2358 			 * The page is relocated if and only if it could be
2359 			 * laundered or reclaimed by the page daemon.
2360 			 */
2361 			if (!VM_OBJECT_TRYWLOCK(object)) {
2362 				mtx_unlock(m_mtx);
2363 				VM_OBJECT_WLOCK(object);
2364 				mtx_lock(m_mtx);
2365 				if (m->object != object) {
2366 					/*
2367 					 * The page may have been freed.
2368 					 */
2369 					VM_OBJECT_WUNLOCK(object);
2370 					goto retry;
2371 				} else if (m->wire_count != 0 ||
2372 				    m->hold_count != 0) {
2373 					error = EBUSY;
2374 					goto unlock;
2375 				}
2376 			}
2377 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2378 			    ("page %p is PG_UNHOLDFREE", m));
2379 			/* Don't care: PG_NODUMP, PG_ZERO. */
2380 			if (object->type != OBJT_DEFAULT &&
2381 			    object->type != OBJT_SWAP &&
2382 			    object->type != OBJT_VNODE)
2383 				error = EINVAL;
2384 			else if (object->memattr != VM_MEMATTR_DEFAULT)
2385 				error = EINVAL;
2386 			else if (m->queue != PQ_NONE && !vm_page_busied(m)) {
2387 				KASSERT(pmap_page_get_memattr(m) ==
2388 				    VM_MEMATTR_DEFAULT,
2389 				    ("page %p has an unexpected memattr", m));
2390 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2391 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2392 				    ("page %p has unexpected oflags", m));
2393 				/* Don't care: VPO_NOSYNC. */
2394 				if (m->valid != 0) {
2395 					/*
2396 					 * First, try to allocate a new page
2397 					 * that is above "high".  Failing
2398 					 * that, try to allocate a new page
2399 					 * that is below "m_run".  Allocate
2400 					 * the new page between the end of
2401 					 * "m_run" and "high" only as a last
2402 					 * resort.
2403 					 */
2404 					req = req_class | VM_ALLOC_NOOBJ;
2405 					if ((m->flags & PG_NODUMP) != 0)
2406 						req |= VM_ALLOC_NODUMP;
2407 					if (trunc_page(high) !=
2408 					    ~(vm_paddr_t)PAGE_MASK) {
2409 						m_new = vm_page_alloc_contig(
2410 						    NULL, 0, req, 1,
2411 						    round_page(high),
2412 						    ~(vm_paddr_t)0,
2413 						    PAGE_SIZE, 0,
2414 						    VM_MEMATTR_DEFAULT);
2415 					} else
2416 						m_new = NULL;
2417 					if (m_new == NULL) {
2418 						pa = VM_PAGE_TO_PHYS(m_run);
2419 						m_new = vm_page_alloc_contig(
2420 						    NULL, 0, req, 1,
2421 						    0, pa - 1, PAGE_SIZE, 0,
2422 						    VM_MEMATTR_DEFAULT);
2423 					}
2424 					if (m_new == NULL) {
2425 						pa += ptoa(npages);
2426 						m_new = vm_page_alloc_contig(
2427 						    NULL, 0, req, 1,
2428 						    pa, high, PAGE_SIZE, 0,
2429 						    VM_MEMATTR_DEFAULT);
2430 					}
2431 					if (m_new == NULL) {
2432 						error = ENOMEM;
2433 						goto unlock;
2434 					}
2435 					KASSERT(m_new->wire_count == 0,
2436 					    ("page %p is wired", m));
2437 
2438 					/*
2439 					 * Replace "m" with the new page.  For
2440 					 * vm_page_replace(), "m" must be busy
2441 					 * and dequeued.  Finally, change "m"
2442 					 * as if vm_page_free() was called.
2443 					 */
2444 					if (object->ref_count != 0)
2445 						pmap_remove_all(m);
2446 					m_new->aflags = m->aflags;
2447 					KASSERT(m_new->oflags == VPO_UNMANAGED,
2448 					    ("page %p is managed", m));
2449 					m_new->oflags = m->oflags & VPO_NOSYNC;
2450 					pmap_copy_page(m, m_new);
2451 					m_new->valid = m->valid;
2452 					m_new->dirty = m->dirty;
2453 					m->flags &= ~PG_ZERO;
2454 					vm_page_xbusy(m);
2455 					vm_page_remque(m);
2456 					vm_page_replace_checked(m_new, object,
2457 					    m->pindex, m);
2458 					m->valid = 0;
2459 					vm_page_undirty(m);
2460 
2461 					/*
2462 					 * The new page must be deactivated
2463 					 * before the object is unlocked.
2464 					 */
2465 					vm_page_change_lock(m_new, &m_mtx);
2466 					vm_page_deactivate(m_new);
2467 				} else {
2468 					m->flags &= ~PG_ZERO;
2469 					vm_page_remque(m);
2470 					vm_page_remove(m);
2471 					KASSERT(m->dirty == 0,
2472 					    ("page %p is dirty", m));
2473 				}
2474 				SLIST_INSERT_HEAD(&free, m, plinks.s.ss);
2475 			} else
2476 				error = EBUSY;
2477 unlock:
2478 			VM_OBJECT_WUNLOCK(object);
2479 		} else {
2480 			mtx_lock(&vm_page_queue_free_mtx);
2481 			order = m->order;
2482 			if (order < VM_NFREEORDER) {
2483 				/*
2484 				 * The page is enqueued in the physical memory
2485 				 * allocator's free page queues.  Moreover, it
2486 				 * is the first page in a power-of-two-sized
2487 				 * run of contiguous free pages.  Jump ahead
2488 				 * to the last page within that run, and
2489 				 * continue from there.
2490 				 */
2491 				m += (1 << order) - 1;
2492 			}
2493 #if VM_NRESERVLEVEL > 0
2494 			else if (vm_reserv_is_page_free(m))
2495 				order = 0;
2496 #endif
2497 			mtx_unlock(&vm_page_queue_free_mtx);
2498 			if (order == VM_NFREEORDER)
2499 				error = EINVAL;
2500 		}
2501 	}
2502 	if (m_mtx != NULL)
2503 		mtx_unlock(m_mtx);
2504 	if ((m = SLIST_FIRST(&free)) != NULL) {
2505 		mtx_lock(&vm_page_queue_free_mtx);
2506 		do {
2507 			SLIST_REMOVE_HEAD(&free, plinks.s.ss);
2508 			vm_page_free_phys(m);
2509 		} while ((m = SLIST_FIRST(&free)) != NULL);
2510 		vm_page_free_wakeup();
2511 		mtx_unlock(&vm_page_queue_free_mtx);
2512 	}
2513 	return (error);
2514 }
2515 
2516 #define	NRUNS	16
2517 
2518 CTASSERT(powerof2(NRUNS));
2519 
2520 #define	RUN_INDEX(count)	((count) & (NRUNS - 1))
2521 
2522 #define	MIN_RECLAIM	8
2523 
2524 /*
2525  *	vm_page_reclaim_contig:
2526  *
2527  *	Reclaim allocated, contiguous physical memory satisfying the specified
2528  *	conditions by relocating the virtual pages using that physical memory.
2529  *	Returns true if reclamation is successful and false otherwise.  Since
2530  *	relocation requires the allocation of physical pages, reclamation may
2531  *	fail due to a shortage of free pages.  When reclamation fails, callers
2532  *	are expected to perform VM_WAIT before retrying a failed allocation
2533  *	operation, e.g., vm_page_alloc_contig().
2534  *
2535  *	The caller must always specify an allocation class through "req".
2536  *
2537  *	allocation classes:
2538  *	VM_ALLOC_NORMAL		normal process request
2539  *	VM_ALLOC_SYSTEM		system *really* needs a page
2540  *	VM_ALLOC_INTERRUPT	interrupt time request
2541  *
2542  *	The optional allocation flags are ignored.
2543  *
2544  *	"npages" must be greater than zero.  Both "alignment" and "boundary"
2545  *	must be a power of two.
2546  */
2547 bool
2548 vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
2549     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
2550 {
2551 	vm_paddr_t curr_low;
2552 	vm_page_t m_run, m_runs[NRUNS];
2553 	u_long count, reclaimed;
2554 	int error, i, options, req_class;
2555 
2556 	KASSERT(npages > 0, ("npages is 0"));
2557 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2558 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2559 	req_class = req & VM_ALLOC_CLASS_MASK;
2560 
2561 	/*
2562 	 * The page daemon is allowed to dig deeper into the free page list.
2563 	 */
2564 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2565 		req_class = VM_ALLOC_SYSTEM;
2566 
2567 	/*
2568 	 * Return if the number of free pages cannot satisfy the requested
2569 	 * allocation.
2570 	 */
2571 	count = vm_cnt.v_free_count;
2572 	if (count < npages + vm_cnt.v_free_reserved || (count < npages +
2573 	    vm_cnt.v_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
2574 	    (count < npages && req_class == VM_ALLOC_INTERRUPT))
2575 		return (false);
2576 
2577 	/*
2578 	 * Scan up to three times, relaxing the restrictions ("options") on
2579 	 * the reclamation of reservations and superpages each time.
2580 	 */
2581 	for (options = VPSC_NORESERV;;) {
2582 		/*
2583 		 * Find the highest runs that satisfy the given constraints
2584 		 * and restrictions, and record them in "m_runs".
2585 		 */
2586 		curr_low = low;
2587 		count = 0;
2588 		for (;;) {
2589 			m_run = vm_phys_scan_contig(domain, npages, curr_low,
2590 			    high, alignment, boundary, options);
2591 			if (m_run == NULL)
2592 				break;
2593 			curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages);
2594 			m_runs[RUN_INDEX(count)] = m_run;
2595 			count++;
2596 		}
2597 
2598 		/*
2599 		 * Reclaim the highest runs in LIFO (descending) order until
2600 		 * the number of reclaimed pages, "reclaimed", is at least
2601 		 * MIN_RECLAIM.  Reset "reclaimed" each time because each
2602 		 * reclamation is idempotent, and runs will (likely) recur
2603 		 * from one scan to the next as restrictions are relaxed.
2604 		 */
2605 		reclaimed = 0;
2606 		for (i = 0; count > 0 && i < NRUNS; i++) {
2607 			count--;
2608 			m_run = m_runs[RUN_INDEX(count)];
2609 			error = vm_page_reclaim_run(req_class, npages, m_run,
2610 			    high);
2611 			if (error == 0) {
2612 				reclaimed += npages;
2613 				if (reclaimed >= MIN_RECLAIM)
2614 					return (true);
2615 			}
2616 		}
2617 
2618 		/*
2619 		 * Either relax the restrictions on the next scan or return if
2620 		 * the last scan had no restrictions.
2621 		 */
2622 		if (options == VPSC_NORESERV)
2623 			options = VPSC_NOSUPER;
2624 		else if (options == VPSC_NOSUPER)
2625 			options = VPSC_ANY;
2626 		else if (options == VPSC_ANY)
2627 			return (reclaimed != 0);
2628 	}
2629 }
2630 
2631 bool
2632 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
2633     u_long alignment, vm_paddr_t boundary)
2634 {
2635 	struct vm_domainset_iter di;
2636 	int domain;
2637 	bool ret;
2638 
2639 	vm_domainset_iter_page_init(&di, kernel_object, &domain, &req);
2640 	do {
2641 		ret = vm_page_reclaim_contig_domain(domain, req, npages, low,
2642 		    high, alignment, boundary);
2643 		if (ret)
2644 			break;
2645 	} while (vm_domainset_iter_page(&di, &domain, &req) == 0);
2646 
2647 	return (ret);
2648 }
2649 
2650 
2651 /*
2652  *	vm_wait:	(also see VM_WAIT macro)
2653  *
2654  *	Sleep until free pages are available for allocation.
2655  *	- Called in various places before memory allocations.
2656  */
2657 static void
2658 _vm_wait(void)
2659 {
2660 
2661 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2662 	if (curproc == pageproc) {
2663 		vm_pageout_pages_needed = 1;
2664 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
2665 		    PDROP | PSWP, "VMWait", 0);
2666 	} else {
2667 		if (pageproc == NULL)
2668 			panic("vm_wait in early boot");
2669 		pagedaemon_wait(PVM, "vmwait");
2670 	}
2671 }
2672 
2673 void
2674 vm_wait(void)
2675 {
2676 
2677 	mtx_lock(&vm_page_queue_free_mtx);
2678 	_vm_wait();
2679 }
2680 
2681 /*
2682  *	vm_page_alloc_fail:
2683  *
2684  *	Called when a page allocation function fails.  Informs the
2685  *	pagedaemon and performs the requested wait.  Requires the
2686  *	page_queue_free and object lock on entry.  Returns with the
2687  *	object lock held and free lock released.  Returns an error when
2688  *	retry is necessary.
2689  *
2690  */
2691 static int
2692 vm_page_alloc_fail(vm_object_t object, int req)
2693 {
2694 
2695 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2696 
2697 	atomic_add_int(&vm_pageout_deficit,
2698 	    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
2699 	if (req & (VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL)) {
2700 		if (object != NULL)
2701 			VM_OBJECT_WUNLOCK(object);
2702 		_vm_wait();
2703 		if (object != NULL)
2704 			VM_OBJECT_WLOCK(object);
2705 		if (req & VM_ALLOC_WAITOK)
2706 			return (EAGAIN);
2707 	} else {
2708 		mtx_unlock(&vm_page_queue_free_mtx);
2709 		pagedaemon_wakeup();
2710 	}
2711 	return (0);
2712 }
2713 
2714 /*
2715  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
2716  *
2717  *	Sleep until free pages are available for allocation.
2718  *	- Called only in vm_fault so that processes page faulting
2719  *	  can be easily tracked.
2720  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2721  *	  processes will be able to grab memory first.  Do not change
2722  *	  this balance without careful testing first.
2723  */
2724 void
2725 vm_waitpfault(void)
2726 {
2727 
2728 	mtx_lock(&vm_page_queue_free_mtx);
2729 	pagedaemon_wait(PUSER, "pfault");
2730 }
2731 
2732 struct vm_pagequeue *
2733 vm_page_pagequeue(vm_page_t m)
2734 {
2735 
2736 	if (vm_page_in_laundry(m))
2737 		return (&vm_dom[0].vmd_pagequeues[m->queue]);
2738 	else
2739 		return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2740 }
2741 
2742 /*
2743  *	vm_page_dequeue:
2744  *
2745  *	Remove the given page from its current page queue.
2746  *
2747  *	The page must be locked.
2748  */
2749 void
2750 vm_page_dequeue(vm_page_t m)
2751 {
2752 	struct vm_pagequeue *pq;
2753 
2754 	vm_page_assert_locked(m);
2755 	KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2756 	    m));
2757 	pq = vm_page_pagequeue(m);
2758 	vm_pagequeue_lock(pq);
2759 	m->queue = PQ_NONE;
2760 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2761 	vm_pagequeue_cnt_dec(pq);
2762 	vm_pagequeue_unlock(pq);
2763 }
2764 
2765 /*
2766  *	vm_page_dequeue_locked:
2767  *
2768  *	Remove the given page from its current page queue.
2769  *
2770  *	The page and page queue must be locked.
2771  */
2772 void
2773 vm_page_dequeue_locked(vm_page_t m)
2774 {
2775 	struct vm_pagequeue *pq;
2776 
2777 	vm_page_lock_assert(m, MA_OWNED);
2778 	pq = vm_page_pagequeue(m);
2779 	vm_pagequeue_assert_locked(pq);
2780 	m->queue = PQ_NONE;
2781 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2782 	vm_pagequeue_cnt_dec(pq);
2783 }
2784 
2785 /*
2786  *	vm_page_enqueue:
2787  *
2788  *	Add the given page to the specified page queue.
2789  *
2790  *	The page must be locked.
2791  */
2792 static void
2793 vm_page_enqueue(uint8_t queue, vm_page_t m)
2794 {
2795 	struct vm_pagequeue *pq;
2796 
2797 	vm_page_lock_assert(m, MA_OWNED);
2798 	KASSERT(queue < PQ_COUNT,
2799 	    ("vm_page_enqueue: invalid queue %u request for page %p",
2800 	    queue, m));
2801 	if (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE)
2802 		pq = &vm_dom[0].vmd_pagequeues[queue];
2803 	else
2804 		pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2805 	vm_pagequeue_lock(pq);
2806 	m->queue = queue;
2807 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2808 	vm_pagequeue_cnt_inc(pq);
2809 	vm_pagequeue_unlock(pq);
2810 }
2811 
2812 /*
2813  *	vm_page_requeue:
2814  *
2815  *	Move the given page to the tail of its current page queue.
2816  *
2817  *	The page must be locked.
2818  */
2819 void
2820 vm_page_requeue(vm_page_t m)
2821 {
2822 	struct vm_pagequeue *pq;
2823 
2824 	vm_page_lock_assert(m, MA_OWNED);
2825 	KASSERT(m->queue != PQ_NONE,
2826 	    ("vm_page_requeue: page %p is not queued", m));
2827 	pq = vm_page_pagequeue(m);
2828 	vm_pagequeue_lock(pq);
2829 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2830 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2831 	vm_pagequeue_unlock(pq);
2832 }
2833 
2834 /*
2835  *	vm_page_requeue_locked:
2836  *
2837  *	Move the given page to the tail of its current page queue.
2838  *
2839  *	The page queue must be locked.
2840  */
2841 void
2842 vm_page_requeue_locked(vm_page_t m)
2843 {
2844 	struct vm_pagequeue *pq;
2845 
2846 	KASSERT(m->queue != PQ_NONE,
2847 	    ("vm_page_requeue_locked: page %p is not queued", m));
2848 	pq = vm_page_pagequeue(m);
2849 	vm_pagequeue_assert_locked(pq);
2850 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2851 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2852 }
2853 
2854 /*
2855  *	vm_page_activate:
2856  *
2857  *	Put the specified page on the active list (if appropriate).
2858  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2859  *	mess with it.
2860  *
2861  *	The page must be locked.
2862  */
2863 void
2864 vm_page_activate(vm_page_t m)
2865 {
2866 	int queue;
2867 
2868 	vm_page_lock_assert(m, MA_OWNED);
2869 	if ((queue = m->queue) != PQ_ACTIVE) {
2870 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2871 			if (m->act_count < ACT_INIT)
2872 				m->act_count = ACT_INIT;
2873 			if (queue != PQ_NONE)
2874 				vm_page_dequeue(m);
2875 			vm_page_enqueue(PQ_ACTIVE, m);
2876 		} else
2877 			KASSERT(queue == PQ_NONE,
2878 			    ("vm_page_activate: wired page %p is queued", m));
2879 	} else {
2880 		if (m->act_count < ACT_INIT)
2881 			m->act_count = ACT_INIT;
2882 	}
2883 }
2884 
2885 /*
2886  *	vm_page_free_wakeup:
2887  *
2888  *	Helper routine for vm_page_free_toq().  This routine is called
2889  *	when a page is added to the free queues.
2890  *
2891  *	The page queues must be locked.
2892  */
2893 static void
2894 vm_page_free_wakeup(void)
2895 {
2896 
2897 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2898 	/*
2899 	 * if pageout daemon needs pages, then tell it that there are
2900 	 * some free.
2901 	 */
2902 	if (vm_pageout_pages_needed &&
2903 	    vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) {
2904 		wakeup(&vm_pageout_pages_needed);
2905 		vm_pageout_pages_needed = 0;
2906 	}
2907 	/*
2908 	 * wakeup processes that are waiting on memory if we hit a
2909 	 * high water mark. And wakeup scheduler process if we have
2910 	 * lots of memory. this process will swapin processes.
2911 	 */
2912 	if (vm_pages_needed && !vm_page_count_min()) {
2913 		vm_pages_needed = false;
2914 		wakeup(&vm_cnt.v_free_count);
2915 	}
2916 }
2917 
2918 /*
2919  *	vm_page_free_prep:
2920  *
2921  *	Prepares the given page to be put on the free list,
2922  *	disassociating it from any VM object. The caller may return
2923  *	the page to the free list only if this function returns true.
2924  *
2925  *	The object must be locked.  The page must be locked if it is
2926  *	managed.  For a queued managed page, the pagequeue_locked
2927  *	argument specifies whether the page queue is already locked.
2928  */
2929 bool
2930 vm_page_free_prep(vm_page_t m, bool pagequeue_locked)
2931 {
2932 
2933 #if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP)
2934 	if (PMAP_HAS_DMAP && (m->flags & PG_ZERO) != 0) {
2935 		uint64_t *p;
2936 		int i;
2937 		p = (uint64_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2938 		for (i = 0; i < PAGE_SIZE / sizeof(uint64_t); i++, p++)
2939 			KASSERT(*p == 0, ("vm_page_free_prep %p PG_ZERO %d %jx",
2940 			    m, i, (uintmax_t)*p));
2941 	}
2942 #endif
2943 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2944 		vm_page_lock_assert(m, MA_OWNED);
2945 		KASSERT(!pmap_page_is_mapped(m),
2946 		    ("vm_page_free_toq: freeing mapped page %p", m));
2947 	} else
2948 		KASSERT(m->queue == PQ_NONE,
2949 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2950 	VM_CNT_INC(v_tfree);
2951 
2952 	if (vm_page_sbusied(m))
2953 		panic("vm_page_free: freeing busy page %p", m);
2954 
2955 	vm_page_remove(m);
2956 
2957 	/*
2958 	 * If fictitious remove object association and
2959 	 * return.
2960 	 */
2961 	if ((m->flags & PG_FICTITIOUS) != 0) {
2962 		KASSERT(m->wire_count == 1,
2963 		    ("fictitious page %p is not wired", m));
2964 		KASSERT(m->queue == PQ_NONE,
2965 		    ("fictitious page %p is queued", m));
2966 		return (false);
2967 	}
2968 
2969 	if (m->queue != PQ_NONE) {
2970 		if (pagequeue_locked)
2971 			vm_page_dequeue_locked(m);
2972 		else
2973 			vm_page_dequeue(m);
2974 	}
2975 	m->valid = 0;
2976 	vm_page_undirty(m);
2977 
2978 	if (m->wire_count != 0)
2979 		panic("vm_page_free: freeing wired page %p", m);
2980 	if (m->hold_count != 0) {
2981 		m->flags &= ~PG_ZERO;
2982 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2983 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
2984 		m->flags |= PG_UNHOLDFREE;
2985 		return (false);
2986 	}
2987 
2988 	/*
2989 	 * Restore the default memory attribute to the page.
2990 	 */
2991 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2992 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2993 
2994 	return (true);
2995 }
2996 
2997 /*
2998  * Insert the page into the physical memory allocator's free page
2999  * queues.  This is the last step to free a page.
3000  */
3001 static void
3002 vm_page_free_phys(vm_page_t m)
3003 {
3004 
3005 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
3006 
3007 	vm_phys_freecnt_adj(m, 1);
3008 #if VM_NRESERVLEVEL > 0
3009 	if (!vm_reserv_free_page(m))
3010 #endif
3011 		vm_phys_free_pages(m, 0);
3012 }
3013 
3014 void
3015 vm_page_free_phys_pglist(struct pglist *tq)
3016 {
3017 	vm_page_t m;
3018 
3019 	if (TAILQ_EMPTY(tq))
3020 		return;
3021 	mtx_lock(&vm_page_queue_free_mtx);
3022 	TAILQ_FOREACH(m, tq, listq)
3023 		vm_page_free_phys(m);
3024 	vm_page_free_wakeup();
3025 	mtx_unlock(&vm_page_queue_free_mtx);
3026 }
3027 
3028 /*
3029  *	vm_page_free_toq:
3030  *
3031  *	Returns the given page to the free list, disassociating it
3032  *	from any VM object.
3033  *
3034  *	The object must be locked.  The page must be locked if it is
3035  *	managed.
3036  */
3037 void
3038 vm_page_free_toq(vm_page_t m)
3039 {
3040 
3041 	if (!vm_page_free_prep(m, false))
3042 		return;
3043 	mtx_lock(&vm_page_queue_free_mtx);
3044 	vm_page_free_phys(m);
3045 	vm_page_free_wakeup();
3046 	mtx_unlock(&vm_page_queue_free_mtx);
3047 }
3048 
3049 /*
3050  *	vm_page_wire:
3051  *
3052  *	Mark this page as wired down by yet
3053  *	another map, removing it from paging queues
3054  *	as necessary.
3055  *
3056  *	If the page is fictitious, then its wire count must remain one.
3057  *
3058  *	The page must be locked.
3059  */
3060 void
3061 vm_page_wire(vm_page_t m)
3062 {
3063 
3064 	/*
3065 	 * Only bump the wire statistics if the page is not already wired,
3066 	 * and only unqueue the page if it is on some queue (if it is unmanaged
3067 	 * it is already off the queues).
3068 	 */
3069 	vm_page_lock_assert(m, MA_OWNED);
3070 	if ((m->flags & PG_FICTITIOUS) != 0) {
3071 		KASSERT(m->wire_count == 1,
3072 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
3073 		    m));
3074 		return;
3075 	}
3076 	if (m->wire_count == 0) {
3077 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
3078 		    m->queue == PQ_NONE,
3079 		    ("vm_page_wire: unmanaged page %p is queued", m));
3080 		vm_page_remque(m);
3081 		atomic_add_int(&vm_cnt.v_wire_count, 1);
3082 	}
3083 	m->wire_count++;
3084 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
3085 }
3086 
3087 /*
3088  * vm_page_unwire:
3089  *
3090  * Release one wiring of the specified page, potentially allowing it to be
3091  * paged out.  Returns TRUE if the number of wirings transitions to zero and
3092  * FALSE otherwise.
3093  *
3094  * Only managed pages belonging to an object can be paged out.  If the number
3095  * of wirings transitions to zero and the page is eligible for page out, then
3096  * the page is added to the specified paging queue (unless PQ_NONE is
3097  * specified).
3098  *
3099  * If a page is fictitious, then its wire count must always be one.
3100  *
3101  * A managed page must be locked.
3102  */
3103 boolean_t
3104 vm_page_unwire(vm_page_t m, uint8_t queue)
3105 {
3106 
3107 	KASSERT(queue < PQ_COUNT || queue == PQ_NONE,
3108 	    ("vm_page_unwire: invalid queue %u request for page %p",
3109 	    queue, m));
3110 	if ((m->oflags & VPO_UNMANAGED) == 0)
3111 		vm_page_assert_locked(m);
3112 	if ((m->flags & PG_FICTITIOUS) != 0) {
3113 		KASSERT(m->wire_count == 1,
3114 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
3115 		return (FALSE);
3116 	}
3117 	if (m->wire_count > 0) {
3118 		m->wire_count--;
3119 		if (m->wire_count == 0) {
3120 			atomic_subtract_int(&vm_cnt.v_wire_count, 1);
3121 			if ((m->oflags & VPO_UNMANAGED) == 0 &&
3122 			    m->object != NULL && queue != PQ_NONE)
3123 				vm_page_enqueue(queue, m);
3124 			return (TRUE);
3125 		} else
3126 			return (FALSE);
3127 	} else
3128 		panic("vm_page_unwire: page %p's wire count is zero", m);
3129 }
3130 
3131 /*
3132  * Move the specified page to the inactive queue.
3133  *
3134  * Normally, "noreuse" is FALSE, resulting in LRU ordering of the inactive
3135  * queue.  However, setting "noreuse" to TRUE will accelerate the specified
3136  * page's reclamation, but it will not unmap the page from any address space.
3137  * This is implemented by inserting the page near the head of the inactive
3138  * queue, using a marker page to guide FIFO insertion ordering.
3139  *
3140  * The page must be locked.
3141  */
3142 static inline void
3143 _vm_page_deactivate(vm_page_t m, boolean_t noreuse)
3144 {
3145 	struct vm_pagequeue *pq;
3146 	int queue;
3147 
3148 	vm_page_assert_locked(m);
3149 
3150 	/*
3151 	 * Ignore if the page is already inactive, unless it is unlikely to be
3152 	 * reactivated.
3153 	 */
3154 	if ((queue = m->queue) == PQ_INACTIVE && !noreuse)
3155 		return;
3156 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
3157 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
3158 		/* Avoid multiple acquisitions of the inactive queue lock. */
3159 		if (queue == PQ_INACTIVE) {
3160 			vm_pagequeue_lock(pq);
3161 			vm_page_dequeue_locked(m);
3162 		} else {
3163 			if (queue != PQ_NONE)
3164 				vm_page_dequeue(m);
3165 			vm_pagequeue_lock(pq);
3166 		}
3167 		m->queue = PQ_INACTIVE;
3168 		if (noreuse)
3169 			TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead,
3170 			    m, plinks.q);
3171 		else
3172 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3173 		vm_pagequeue_cnt_inc(pq);
3174 		vm_pagequeue_unlock(pq);
3175 	}
3176 }
3177 
3178 /*
3179  * Move the specified page to the inactive queue.
3180  *
3181  * The page must be locked.
3182  */
3183 void
3184 vm_page_deactivate(vm_page_t m)
3185 {
3186 
3187 	_vm_page_deactivate(m, FALSE);
3188 }
3189 
3190 /*
3191  * Move the specified page to the inactive queue with the expectation
3192  * that it is unlikely to be reused.
3193  *
3194  * The page must be locked.
3195  */
3196 void
3197 vm_page_deactivate_noreuse(vm_page_t m)
3198 {
3199 
3200 	_vm_page_deactivate(m, TRUE);
3201 }
3202 
3203 /*
3204  * vm_page_launder
3205  *
3206  * 	Put a page in the laundry.
3207  */
3208 void
3209 vm_page_launder(vm_page_t m)
3210 {
3211 	int queue;
3212 
3213 	vm_page_assert_locked(m);
3214 	if ((queue = m->queue) != PQ_LAUNDRY) {
3215 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
3216 			if (queue != PQ_NONE)
3217 				vm_page_dequeue(m);
3218 			vm_page_enqueue(PQ_LAUNDRY, m);
3219 		} else
3220 			KASSERT(queue == PQ_NONE,
3221 			    ("wired page %p is queued", m));
3222 	}
3223 }
3224 
3225 /*
3226  * vm_page_unswappable
3227  *
3228  *	Put a page in the PQ_UNSWAPPABLE holding queue.
3229  */
3230 void
3231 vm_page_unswappable(vm_page_t m)
3232 {
3233 
3234 	vm_page_assert_locked(m);
3235 	KASSERT(m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0,
3236 	    ("page %p already unswappable", m));
3237 	if (m->queue != PQ_NONE)
3238 		vm_page_dequeue(m);
3239 	vm_page_enqueue(PQ_UNSWAPPABLE, m);
3240 }
3241 
3242 /*
3243  * Attempt to free the page.  If it cannot be freed, do nothing.  Returns true
3244  * if the page is freed and false otherwise.
3245  *
3246  * The page must be managed.  The page and its containing object must be
3247  * locked.
3248  */
3249 bool
3250 vm_page_try_to_free(vm_page_t m)
3251 {
3252 
3253 	vm_page_assert_locked(m);
3254 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3255 	KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("page %p is unmanaged", m));
3256 	if (m->dirty != 0 || m->hold_count != 0 || m->wire_count != 0 ||
3257 	    vm_page_busied(m))
3258 		return (false);
3259 	if (m->object->ref_count != 0) {
3260 		pmap_remove_all(m);
3261 		if (m->dirty != 0)
3262 			return (false);
3263 	}
3264 	vm_page_free(m);
3265 	return (true);
3266 }
3267 
3268 /*
3269  * vm_page_advise
3270  *
3271  * 	Apply the specified advice to the given page.
3272  *
3273  *	The object and page must be locked.
3274  */
3275 void
3276 vm_page_advise(vm_page_t m, int advice)
3277 {
3278 
3279 	vm_page_assert_locked(m);
3280 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3281 	if (advice == MADV_FREE)
3282 		/*
3283 		 * Mark the page clean.  This will allow the page to be freed
3284 		 * without first paging it out.  MADV_FREE pages are often
3285 		 * quickly reused by malloc(3), so we do not do anything that
3286 		 * would result in a page fault on a later access.
3287 		 */
3288 		vm_page_undirty(m);
3289 	else if (advice != MADV_DONTNEED) {
3290 		if (advice == MADV_WILLNEED)
3291 			vm_page_activate(m);
3292 		return;
3293 	}
3294 
3295 	/*
3296 	 * Clear any references to the page.  Otherwise, the page daemon will
3297 	 * immediately reactivate the page.
3298 	 */
3299 	vm_page_aflag_clear(m, PGA_REFERENCED);
3300 
3301 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
3302 		vm_page_dirty(m);
3303 
3304 	/*
3305 	 * Place clean pages near the head of the inactive queue rather than
3306 	 * the tail, thus defeating the queue's LRU operation and ensuring that
3307 	 * the page will be reused quickly.  Dirty pages not already in the
3308 	 * laundry are moved there.
3309 	 */
3310 	if (m->dirty == 0)
3311 		vm_page_deactivate_noreuse(m);
3312 	else
3313 		vm_page_launder(m);
3314 }
3315 
3316 /*
3317  * Grab a page, waiting until we are waken up due to the page
3318  * changing state.  We keep on waiting, if the page continues
3319  * to be in the object.  If the page doesn't exist, first allocate it
3320  * and then conditionally zero it.
3321  *
3322  * This routine may sleep.
3323  *
3324  * The object must be locked on entry.  The lock will, however, be released
3325  * and reacquired if the routine sleeps.
3326  */
3327 vm_page_t
3328 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
3329 {
3330 	vm_page_t m;
3331 	int sleep;
3332 	int pflags;
3333 
3334 	VM_OBJECT_ASSERT_WLOCKED(object);
3335 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3336 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3337 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
3338 	pflags = allocflags &
3339 	    ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
3340 	if ((allocflags & VM_ALLOC_NOWAIT) == 0)
3341 		pflags |= VM_ALLOC_WAITFAIL;
3342 retrylookup:
3343 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
3344 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3345 		    vm_page_xbusied(m) : vm_page_busied(m);
3346 		if (sleep) {
3347 			if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3348 				return (NULL);
3349 			/*
3350 			 * Reference the page before unlocking and
3351 			 * sleeping so that the page daemon is less
3352 			 * likely to reclaim it.
3353 			 */
3354 			vm_page_aflag_set(m, PGA_REFERENCED);
3355 			vm_page_lock(m);
3356 			VM_OBJECT_WUNLOCK(object);
3357 			vm_page_busy_sleep(m, "pgrbwt", (allocflags &
3358 			    VM_ALLOC_IGN_SBUSY) != 0);
3359 			VM_OBJECT_WLOCK(object);
3360 			goto retrylookup;
3361 		} else {
3362 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
3363 				vm_page_lock(m);
3364 				vm_page_wire(m);
3365 				vm_page_unlock(m);
3366 			}
3367 			if ((allocflags &
3368 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
3369 				vm_page_xbusy(m);
3370 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
3371 				vm_page_sbusy(m);
3372 			return (m);
3373 		}
3374 	}
3375 	m = vm_page_alloc(object, pindex, pflags);
3376 	if (m == NULL) {
3377 		if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3378 			return (NULL);
3379 		goto retrylookup;
3380 	}
3381 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
3382 		pmap_zero_page(m);
3383 	return (m);
3384 }
3385 
3386 /*
3387  * Return the specified range of pages from the given object.  For each
3388  * page offset within the range, if a page already exists within the object
3389  * at that offset and it is busy, then wait for it to change state.  If,
3390  * instead, the page doesn't exist, then allocate it.
3391  *
3392  * The caller must always specify an allocation class.
3393  *
3394  * allocation classes:
3395  *	VM_ALLOC_NORMAL		normal process request
3396  *	VM_ALLOC_SYSTEM		system *really* needs the pages
3397  *
3398  * The caller must always specify that the pages are to be busied and/or
3399  * wired.
3400  *
3401  * optional allocation flags:
3402  *	VM_ALLOC_IGN_SBUSY	do not sleep on soft busy pages
3403  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
3404  *	VM_ALLOC_NOWAIT		do not sleep
3405  *	VM_ALLOC_SBUSY		set page to sbusy state
3406  *	VM_ALLOC_WIRED		wire the pages
3407  *	VM_ALLOC_ZERO		zero and validate any invalid pages
3408  *
3409  * If VM_ALLOC_NOWAIT is not specified, this routine may sleep.  Otherwise, it
3410  * may return a partial prefix of the requested range.
3411  */
3412 int
3413 vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
3414     vm_page_t *ma, int count)
3415 {
3416 	vm_page_t m, mpred;
3417 	int pflags;
3418 	int i;
3419 	bool sleep;
3420 
3421 	VM_OBJECT_ASSERT_WLOCKED(object);
3422 	KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0,
3423 	    ("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed"));
3424 	KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 ||
3425 	    (allocflags & VM_ALLOC_WIRED) != 0,
3426 	    ("vm_page_grab_pages: the pages must be busied or wired"));
3427 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3428 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3429 	    ("vm_page_grab_pages: VM_ALLOC_SBUSY/IGN_SBUSY mismatch"));
3430 	if (count == 0)
3431 		return (0);
3432 	pflags = allocflags & ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK |
3433 	    VM_ALLOC_WAITFAIL | VM_ALLOC_IGN_SBUSY);
3434 	if ((allocflags & VM_ALLOC_NOWAIT) == 0)
3435 		pflags |= VM_ALLOC_WAITFAIL;
3436 	i = 0;
3437 retrylookup:
3438 	m = vm_radix_lookup_le(&object->rtree, pindex + i);
3439 	if (m == NULL || m->pindex != pindex + i) {
3440 		mpred = m;
3441 		m = NULL;
3442 	} else
3443 		mpred = TAILQ_PREV(m, pglist, listq);
3444 	for (; i < count; i++) {
3445 		if (m != NULL) {
3446 			sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3447 			    vm_page_xbusied(m) : vm_page_busied(m);
3448 			if (sleep) {
3449 				if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3450 					break;
3451 				/*
3452 				 * Reference the page before unlocking and
3453 				 * sleeping so that the page daemon is less
3454 				 * likely to reclaim it.
3455 				 */
3456 				vm_page_aflag_set(m, PGA_REFERENCED);
3457 				vm_page_lock(m);
3458 				VM_OBJECT_WUNLOCK(object);
3459 				vm_page_busy_sleep(m, "grbmaw", (allocflags &
3460 				    VM_ALLOC_IGN_SBUSY) != 0);
3461 				VM_OBJECT_WLOCK(object);
3462 				goto retrylookup;
3463 			}
3464 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
3465 				vm_page_lock(m);
3466 				vm_page_wire(m);
3467 				vm_page_unlock(m);
3468 			}
3469 			if ((allocflags & (VM_ALLOC_NOBUSY |
3470 			    VM_ALLOC_SBUSY)) == 0)
3471 				vm_page_xbusy(m);
3472 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
3473 				vm_page_sbusy(m);
3474 		} else {
3475 			m = vm_page_alloc_after(object, pindex + i,
3476 			    pflags | VM_ALLOC_COUNT(count - i), mpred);
3477 			if (m == NULL) {
3478 				if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3479 					break;
3480 				goto retrylookup;
3481 			}
3482 		}
3483 		if (m->valid == 0 && (allocflags & VM_ALLOC_ZERO) != 0) {
3484 			if ((m->flags & PG_ZERO) == 0)
3485 				pmap_zero_page(m);
3486 			m->valid = VM_PAGE_BITS_ALL;
3487 		}
3488 		ma[i] = mpred = m;
3489 		m = vm_page_next(m);
3490 	}
3491 	return (i);
3492 }
3493 
3494 /*
3495  * Mapping function for valid or dirty bits in a page.
3496  *
3497  * Inputs are required to range within a page.
3498  */
3499 vm_page_bits_t
3500 vm_page_bits(int base, int size)
3501 {
3502 	int first_bit;
3503 	int last_bit;
3504 
3505 	KASSERT(
3506 	    base + size <= PAGE_SIZE,
3507 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
3508 	);
3509 
3510 	if (size == 0)		/* handle degenerate case */
3511 		return (0);
3512 
3513 	first_bit = base >> DEV_BSHIFT;
3514 	last_bit = (base + size - 1) >> DEV_BSHIFT;
3515 
3516 	return (((vm_page_bits_t)2 << last_bit) -
3517 	    ((vm_page_bits_t)1 << first_bit));
3518 }
3519 
3520 /*
3521  *	vm_page_set_valid_range:
3522  *
3523  *	Sets portions of a page valid.  The arguments are expected
3524  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3525  *	of any partial chunks touched by the range.  The invalid portion of
3526  *	such chunks will be zeroed.
3527  *
3528  *	(base + size) must be less then or equal to PAGE_SIZE.
3529  */
3530 void
3531 vm_page_set_valid_range(vm_page_t m, int base, int size)
3532 {
3533 	int endoff, frag;
3534 
3535 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3536 	if (size == 0)	/* handle degenerate case */
3537 		return;
3538 
3539 	/*
3540 	 * If the base is not DEV_BSIZE aligned and the valid
3541 	 * bit is clear, we have to zero out a portion of the
3542 	 * first block.
3543 	 */
3544 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3545 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
3546 		pmap_zero_page_area(m, frag, base - frag);
3547 
3548 	/*
3549 	 * If the ending offset is not DEV_BSIZE aligned and the
3550 	 * valid bit is clear, we have to zero out a portion of
3551 	 * the last block.
3552 	 */
3553 	endoff = base + size;
3554 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3555 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
3556 		pmap_zero_page_area(m, endoff,
3557 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3558 
3559 	/*
3560 	 * Assert that no previously invalid block that is now being validated
3561 	 * is already dirty.
3562 	 */
3563 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
3564 	    ("vm_page_set_valid_range: page %p is dirty", m));
3565 
3566 	/*
3567 	 * Set valid bits inclusive of any overlap.
3568 	 */
3569 	m->valid |= vm_page_bits(base, size);
3570 }
3571 
3572 /*
3573  * Clear the given bits from the specified page's dirty field.
3574  */
3575 static __inline void
3576 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
3577 {
3578 	uintptr_t addr;
3579 #if PAGE_SIZE < 16384
3580 	int shift;
3581 #endif
3582 
3583 	/*
3584 	 * If the object is locked and the page is neither exclusive busy nor
3585 	 * write mapped, then the page's dirty field cannot possibly be
3586 	 * set by a concurrent pmap operation.
3587 	 */
3588 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3589 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
3590 		m->dirty &= ~pagebits;
3591 	else {
3592 		/*
3593 		 * The pmap layer can call vm_page_dirty() without
3594 		 * holding a distinguished lock.  The combination of
3595 		 * the object's lock and an atomic operation suffice
3596 		 * to guarantee consistency of the page dirty field.
3597 		 *
3598 		 * For PAGE_SIZE == 32768 case, compiler already
3599 		 * properly aligns the dirty field, so no forcible
3600 		 * alignment is needed. Only require existence of
3601 		 * atomic_clear_64 when page size is 32768.
3602 		 */
3603 		addr = (uintptr_t)&m->dirty;
3604 #if PAGE_SIZE == 32768
3605 		atomic_clear_64((uint64_t *)addr, pagebits);
3606 #elif PAGE_SIZE == 16384
3607 		atomic_clear_32((uint32_t *)addr, pagebits);
3608 #else		/* PAGE_SIZE <= 8192 */
3609 		/*
3610 		 * Use a trick to perform a 32-bit atomic on the
3611 		 * containing aligned word, to not depend on the existence
3612 		 * of atomic_clear_{8, 16}.
3613 		 */
3614 		shift = addr & (sizeof(uint32_t) - 1);
3615 #if BYTE_ORDER == BIG_ENDIAN
3616 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
3617 #else
3618 		shift *= NBBY;
3619 #endif
3620 		addr &= ~(sizeof(uint32_t) - 1);
3621 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
3622 #endif		/* PAGE_SIZE */
3623 	}
3624 }
3625 
3626 /*
3627  *	vm_page_set_validclean:
3628  *
3629  *	Sets portions of a page valid and clean.  The arguments are expected
3630  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3631  *	of any partial chunks touched by the range.  The invalid portion of
3632  *	such chunks will be zero'd.
3633  *
3634  *	(base + size) must be less then or equal to PAGE_SIZE.
3635  */
3636 void
3637 vm_page_set_validclean(vm_page_t m, int base, int size)
3638 {
3639 	vm_page_bits_t oldvalid, pagebits;
3640 	int endoff, frag;
3641 
3642 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3643 	if (size == 0)	/* handle degenerate case */
3644 		return;
3645 
3646 	/*
3647 	 * If the base is not DEV_BSIZE aligned and the valid
3648 	 * bit is clear, we have to zero out a portion of the
3649 	 * first block.
3650 	 */
3651 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3652 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3653 		pmap_zero_page_area(m, frag, base - frag);
3654 
3655 	/*
3656 	 * If the ending offset is not DEV_BSIZE aligned and the
3657 	 * valid bit is clear, we have to zero out a portion of
3658 	 * the last block.
3659 	 */
3660 	endoff = base + size;
3661 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3662 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3663 		pmap_zero_page_area(m, endoff,
3664 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3665 
3666 	/*
3667 	 * Set valid, clear dirty bits.  If validating the entire
3668 	 * page we can safely clear the pmap modify bit.  We also
3669 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
3670 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
3671 	 * be set again.
3672 	 *
3673 	 * We set valid bits inclusive of any overlap, but we can only
3674 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
3675 	 * the range.
3676 	 */
3677 	oldvalid = m->valid;
3678 	pagebits = vm_page_bits(base, size);
3679 	m->valid |= pagebits;
3680 #if 0	/* NOT YET */
3681 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3682 		frag = DEV_BSIZE - frag;
3683 		base += frag;
3684 		size -= frag;
3685 		if (size < 0)
3686 			size = 0;
3687 	}
3688 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3689 #endif
3690 	if (base == 0 && size == PAGE_SIZE) {
3691 		/*
3692 		 * The page can only be modified within the pmap if it is
3693 		 * mapped, and it can only be mapped if it was previously
3694 		 * fully valid.
3695 		 */
3696 		if (oldvalid == VM_PAGE_BITS_ALL)
3697 			/*
3698 			 * Perform the pmap_clear_modify() first.  Otherwise,
3699 			 * a concurrent pmap operation, such as
3700 			 * pmap_protect(), could clear a modification in the
3701 			 * pmap and set the dirty field on the page before
3702 			 * pmap_clear_modify() had begun and after the dirty
3703 			 * field was cleared here.
3704 			 */
3705 			pmap_clear_modify(m);
3706 		m->dirty = 0;
3707 		m->oflags &= ~VPO_NOSYNC;
3708 	} else if (oldvalid != VM_PAGE_BITS_ALL)
3709 		m->dirty &= ~pagebits;
3710 	else
3711 		vm_page_clear_dirty_mask(m, pagebits);
3712 }
3713 
3714 void
3715 vm_page_clear_dirty(vm_page_t m, int base, int size)
3716 {
3717 
3718 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3719 }
3720 
3721 /*
3722  *	vm_page_set_invalid:
3723  *
3724  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
3725  *	valid and dirty bits for the effected areas are cleared.
3726  */
3727 void
3728 vm_page_set_invalid(vm_page_t m, int base, int size)
3729 {
3730 	vm_page_bits_t bits;
3731 	vm_object_t object;
3732 
3733 	object = m->object;
3734 	VM_OBJECT_ASSERT_WLOCKED(object);
3735 	if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3736 	    size >= object->un_pager.vnp.vnp_size)
3737 		bits = VM_PAGE_BITS_ALL;
3738 	else
3739 		bits = vm_page_bits(base, size);
3740 	if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
3741 	    bits != 0)
3742 		pmap_remove_all(m);
3743 	KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3744 	    !pmap_page_is_mapped(m),
3745 	    ("vm_page_set_invalid: page %p is mapped", m));
3746 	m->valid &= ~bits;
3747 	m->dirty &= ~bits;
3748 }
3749 
3750 /*
3751  * vm_page_zero_invalid()
3752  *
3753  *	The kernel assumes that the invalid portions of a page contain
3754  *	garbage, but such pages can be mapped into memory by user code.
3755  *	When this occurs, we must zero out the non-valid portions of the
3756  *	page so user code sees what it expects.
3757  *
3758  *	Pages are most often semi-valid when the end of a file is mapped
3759  *	into memory and the file's size is not page aligned.
3760  */
3761 void
3762 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3763 {
3764 	int b;
3765 	int i;
3766 
3767 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3768 	/*
3769 	 * Scan the valid bits looking for invalid sections that
3770 	 * must be zeroed.  Invalid sub-DEV_BSIZE'd areas ( where the
3771 	 * valid bit may be set ) have already been zeroed by
3772 	 * vm_page_set_validclean().
3773 	 */
3774 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3775 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3776 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3777 			if (i > b) {
3778 				pmap_zero_page_area(m,
3779 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3780 			}
3781 			b = i + 1;
3782 		}
3783 	}
3784 
3785 	/*
3786 	 * setvalid is TRUE when we can safely set the zero'd areas
3787 	 * as being valid.  We can do this if there are no cache consistancy
3788 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3789 	 */
3790 	if (setvalid)
3791 		m->valid = VM_PAGE_BITS_ALL;
3792 }
3793 
3794 /*
3795  *	vm_page_is_valid:
3796  *
3797  *	Is (partial) page valid?  Note that the case where size == 0
3798  *	will return FALSE in the degenerate case where the page is
3799  *	entirely invalid, and TRUE otherwise.
3800  */
3801 int
3802 vm_page_is_valid(vm_page_t m, int base, int size)
3803 {
3804 	vm_page_bits_t bits;
3805 
3806 	VM_OBJECT_ASSERT_LOCKED(m->object);
3807 	bits = vm_page_bits(base, size);
3808 	return (m->valid != 0 && (m->valid & bits) == bits);
3809 }
3810 
3811 /*
3812  * Returns true if all of the specified predicates are true for the entire
3813  * (super)page and false otherwise.
3814  */
3815 bool
3816 vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m)
3817 {
3818 	vm_object_t object;
3819 	int i, npages;
3820 
3821 	object = m->object;
3822 	VM_OBJECT_ASSERT_LOCKED(object);
3823 	npages = atop(pagesizes[m->psind]);
3824 
3825 	/*
3826 	 * The physically contiguous pages that make up a superpage, i.e., a
3827 	 * page with a page size index ("psind") greater than zero, will
3828 	 * occupy adjacent entries in vm_page_array[].
3829 	 */
3830 	for (i = 0; i < npages; i++) {
3831 		/* Always test object consistency, including "skip_m". */
3832 		if (m[i].object != object)
3833 			return (false);
3834 		if (&m[i] == skip_m)
3835 			continue;
3836 		if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i]))
3837 			return (false);
3838 		if ((flags & PS_ALL_DIRTY) != 0) {
3839 			/*
3840 			 * Calling vm_page_test_dirty() or pmap_is_modified()
3841 			 * might stop this case from spuriously returning
3842 			 * "false".  However, that would require a write lock
3843 			 * on the object containing "m[i]".
3844 			 */
3845 			if (m[i].dirty != VM_PAGE_BITS_ALL)
3846 				return (false);
3847 		}
3848 		if ((flags & PS_ALL_VALID) != 0 &&
3849 		    m[i].valid != VM_PAGE_BITS_ALL)
3850 			return (false);
3851 	}
3852 	return (true);
3853 }
3854 
3855 /*
3856  * Set the page's dirty bits if the page is modified.
3857  */
3858 void
3859 vm_page_test_dirty(vm_page_t m)
3860 {
3861 
3862 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3863 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3864 		vm_page_dirty(m);
3865 }
3866 
3867 void
3868 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3869 {
3870 
3871 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3872 }
3873 
3874 void
3875 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3876 {
3877 
3878 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3879 }
3880 
3881 int
3882 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3883 {
3884 
3885 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3886 }
3887 
3888 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3889 void
3890 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3891 {
3892 
3893 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3894 }
3895 
3896 void
3897 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3898 {
3899 
3900 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3901 }
3902 #endif
3903 
3904 #ifdef INVARIANTS
3905 void
3906 vm_page_object_lock_assert(vm_page_t m)
3907 {
3908 
3909 	/*
3910 	 * Certain of the page's fields may only be modified by the
3911 	 * holder of the containing object's lock or the exclusive busy.
3912 	 * holder.  Unfortunately, the holder of the write busy is
3913 	 * not recorded, and thus cannot be checked here.
3914 	 */
3915 	if (m->object != NULL && !vm_page_xbusied(m))
3916 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3917 }
3918 
3919 void
3920 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
3921 {
3922 
3923 	if ((bits & PGA_WRITEABLE) == 0)
3924 		return;
3925 
3926 	/*
3927 	 * The PGA_WRITEABLE flag can only be set if the page is
3928 	 * managed, is exclusively busied or the object is locked.
3929 	 * Currently, this flag is only set by pmap_enter().
3930 	 */
3931 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3932 	    ("PGA_WRITEABLE on unmanaged page"));
3933 	if (!vm_page_xbusied(m))
3934 		VM_OBJECT_ASSERT_LOCKED(m->object);
3935 }
3936 #endif
3937 
3938 #include "opt_ddb.h"
3939 #ifdef DDB
3940 #include <sys/kernel.h>
3941 
3942 #include <ddb/ddb.h>
3943 
3944 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3945 {
3946 
3947 	db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count);
3948 	db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count);
3949 	db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count);
3950 	db_printf("vm_cnt.v_laundry_count: %d\n", vm_cnt.v_laundry_count);
3951 	db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count);
3952 	db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
3953 	db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
3954 	db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
3955 	db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
3956 }
3957 
3958 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3959 {
3960 	int dom;
3961 
3962 	db_printf("pq_free %d\n", vm_cnt.v_free_count);
3963 	for (dom = 0; dom < vm_ndomains; dom++) {
3964 		db_printf(
3965     "dom %d page_cnt %d free %d pq_act %d pq_inact %d pq_laund %d pq_unsw %d\n",
3966 		    dom,
3967 		    vm_dom[dom].vmd_page_count,
3968 		    vm_dom[dom].vmd_free_count,
3969 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3970 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3971 		    vm_dom[dom].vmd_pagequeues[PQ_LAUNDRY].pq_cnt,
3972 		    vm_dom[dom].vmd_pagequeues[PQ_UNSWAPPABLE].pq_cnt);
3973 	}
3974 }
3975 
3976 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3977 {
3978 	vm_page_t m;
3979 	boolean_t phys;
3980 
3981 	if (!have_addr) {
3982 		db_printf("show pginfo addr\n");
3983 		return;
3984 	}
3985 
3986 	phys = strchr(modif, 'p') != NULL;
3987 	if (phys)
3988 		m = PHYS_TO_VM_PAGE(addr);
3989 	else
3990 		m = (vm_page_t)addr;
3991 	db_printf(
3992     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3993     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3994 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3995 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3996 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3997 }
3998 #endif /* DDB */
3999