xref: /freebsd/sys/vm/vm_page.c (revision 91ebf7d76582a73ac4c0a32964e4d3912010671e)
1 /*-
2  * Copyright (c) 1991 Regents of the University of California.
3  * All rights reserved.
4  * Copyright (c) 1998 Matthew Dillon.  All Rights Reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * The Mach Operating System project at Carnegie-Mellon University.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	from: @(#)vm_page.c	7.4 (Berkeley) 5/7/91
34  */
35 
36 /*-
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  */
62 
63 /*
64  *			GENERAL RULES ON VM_PAGE MANIPULATION
65  *
66  *	- A page queue lock is required when adding or removing a page from a
67  *	  page queue regardless of other locks or the busy state of a page.
68  *
69  *		* In general, no thread besides the page daemon can acquire or
70  *		  hold more than one page queue lock at a time.
71  *
72  *		* The page daemon can acquire and hold any pair of page queue
73  *		  locks in any order.
74  *
75  *	- The object lock is required when inserting or removing
76  *	  pages from an object (vm_page_insert() or vm_page_remove()).
77  *
78  */
79 
80 /*
81  *	Resident memory management module.
82  */
83 
84 #include <sys/cdefs.h>
85 __FBSDID("$FreeBSD$");
86 
87 #include "opt_vm.h"
88 
89 #include <sys/param.h>
90 #include <sys/systm.h>
91 #include <sys/lock.h>
92 #include <sys/kernel.h>
93 #include <sys/limits.h>
94 #include <sys/linker.h>
95 #include <sys/malloc.h>
96 #include <sys/mman.h>
97 #include <sys/msgbuf.h>
98 #include <sys/mutex.h>
99 #include <sys/proc.h>
100 #include <sys/rwlock.h>
101 #include <sys/sbuf.h>
102 #include <sys/sysctl.h>
103 #include <sys/vmmeter.h>
104 #include <sys/vnode.h>
105 
106 #include <vm/vm.h>
107 #include <vm/pmap.h>
108 #include <vm/vm_param.h>
109 #include <vm/vm_kern.h>
110 #include <vm/vm_object.h>
111 #include <vm/vm_page.h>
112 #include <vm/vm_pageout.h>
113 #include <vm/vm_pager.h>
114 #include <vm/vm_phys.h>
115 #include <vm/vm_radix.h>
116 #include <vm/vm_reserv.h>
117 #include <vm/vm_extern.h>
118 #include <vm/uma.h>
119 #include <vm/uma_int.h>
120 
121 #include <machine/md_var.h>
122 
123 /*
124  *	Associated with page of user-allocatable memory is a
125  *	page structure.
126  */
127 
128 struct vm_domain vm_dom[MAXMEMDOM];
129 struct mtx_padalign vm_page_queue_free_mtx;
130 
131 struct mtx_padalign pa_lock[PA_LOCK_COUNT];
132 
133 vm_page_t vm_page_array;
134 long vm_page_array_size;
135 long first_page;
136 int vm_page_zero_count;
137 
138 static int boot_pages = UMA_BOOT_PAGES;
139 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
140     &boot_pages, 0,
141     "number of pages allocated for bootstrapping the VM system");
142 
143 static int pa_tryrelock_restart;
144 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
145     &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
146 
147 static TAILQ_HEAD(, vm_page) blacklist_head;
148 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
149 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
150     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
151 
152 /* Is the page daemon waiting for free pages? */
153 static int vm_pageout_pages_needed;
154 
155 static uma_zone_t fakepg_zone;
156 
157 static struct vnode *vm_page_alloc_init(vm_page_t m);
158 static void vm_page_cache_turn_free(vm_page_t m);
159 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
160 static void vm_page_enqueue(uint8_t queue, vm_page_t m);
161 static void vm_page_free_wakeup(void);
162 static void vm_page_init_fakepg(void *dummy);
163 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
164     vm_pindex_t pindex, vm_page_t mpred);
165 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
166     vm_page_t mpred);
167 static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
168     vm_paddr_t high);
169 
170 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL);
171 
172 static void
173 vm_page_init_fakepg(void *dummy)
174 {
175 
176 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
177 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
178 }
179 
180 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
181 #if PAGE_SIZE == 32768
182 #ifdef CTASSERT
183 CTASSERT(sizeof(u_long) >= 8);
184 #endif
185 #endif
186 
187 /*
188  * Try to acquire a physical address lock while a pmap is locked.  If we
189  * fail to trylock we unlock and lock the pmap directly and cache the
190  * locked pa in *locked.  The caller should then restart their loop in case
191  * the virtual to physical mapping has changed.
192  */
193 int
194 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
195 {
196 	vm_paddr_t lockpa;
197 
198 	lockpa = *locked;
199 	*locked = pa;
200 	if (lockpa) {
201 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
202 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
203 			return (0);
204 		PA_UNLOCK(lockpa);
205 	}
206 	if (PA_TRYLOCK(pa))
207 		return (0);
208 	PMAP_UNLOCK(pmap);
209 	atomic_add_int(&pa_tryrelock_restart, 1);
210 	PA_LOCK(pa);
211 	PMAP_LOCK(pmap);
212 	return (EAGAIN);
213 }
214 
215 /*
216  *	vm_set_page_size:
217  *
218  *	Sets the page size, perhaps based upon the memory
219  *	size.  Must be called before any use of page-size
220  *	dependent functions.
221  */
222 void
223 vm_set_page_size(void)
224 {
225 	if (vm_cnt.v_page_size == 0)
226 		vm_cnt.v_page_size = PAGE_SIZE;
227 	if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
228 		panic("vm_set_page_size: page size not a power of two");
229 }
230 
231 /*
232  *	vm_page_blacklist_next:
233  *
234  *	Find the next entry in the provided string of blacklist
235  *	addresses.  Entries are separated by space, comma, or newline.
236  *	If an invalid integer is encountered then the rest of the
237  *	string is skipped.  Updates the list pointer to the next
238  *	character, or NULL if the string is exhausted or invalid.
239  */
240 static vm_paddr_t
241 vm_page_blacklist_next(char **list, char *end)
242 {
243 	vm_paddr_t bad;
244 	char *cp, *pos;
245 
246 	if (list == NULL || *list == NULL)
247 		return (0);
248 	if (**list =='\0') {
249 		*list = NULL;
250 		return (0);
251 	}
252 
253 	/*
254 	 * If there's no end pointer then the buffer is coming from
255 	 * the kenv and we know it's null-terminated.
256 	 */
257 	if (end == NULL)
258 		end = *list + strlen(*list);
259 
260 	/* Ensure that strtoq() won't walk off the end */
261 	if (*end != '\0') {
262 		if (*end == '\n' || *end == ' ' || *end  == ',')
263 			*end = '\0';
264 		else {
265 			printf("Blacklist not terminated, skipping\n");
266 			*list = NULL;
267 			return (0);
268 		}
269 	}
270 
271 	for (pos = *list; *pos != '\0'; pos = cp) {
272 		bad = strtoq(pos, &cp, 0);
273 		if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
274 			if (bad == 0) {
275 				if (++cp < end)
276 					continue;
277 				else
278 					break;
279 			}
280 		} else
281 			break;
282 		if (*cp == '\0' || ++cp >= end)
283 			*list = NULL;
284 		else
285 			*list = cp;
286 		return (trunc_page(bad));
287 	}
288 	printf("Garbage in RAM blacklist, skipping\n");
289 	*list = NULL;
290 	return (0);
291 }
292 
293 /*
294  *	vm_page_blacklist_check:
295  *
296  *	Iterate through the provided string of blacklist addresses, pulling
297  *	each entry out of the physical allocator free list and putting it
298  *	onto a list for reporting via the vm.page_blacklist sysctl.
299  */
300 static void
301 vm_page_blacklist_check(char *list, char *end)
302 {
303 	vm_paddr_t pa;
304 	vm_page_t m;
305 	char *next;
306 	int ret;
307 
308 	next = list;
309 	while (next != NULL) {
310 		if ((pa = vm_page_blacklist_next(&next, end)) == 0)
311 			continue;
312 		m = vm_phys_paddr_to_vm_page(pa);
313 		if (m == NULL)
314 			continue;
315 		mtx_lock(&vm_page_queue_free_mtx);
316 		ret = vm_phys_unfree_page(m);
317 		mtx_unlock(&vm_page_queue_free_mtx);
318 		if (ret == TRUE) {
319 			TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
320 			if (bootverbose)
321 				printf("Skipping page with pa 0x%jx\n",
322 				    (uintmax_t)pa);
323 		}
324 	}
325 }
326 
327 /*
328  *	vm_page_blacklist_load:
329  *
330  *	Search for a special module named "ram_blacklist".  It'll be a
331  *	plain text file provided by the user via the loader directive
332  *	of the same name.
333  */
334 static void
335 vm_page_blacklist_load(char **list, char **end)
336 {
337 	void *mod;
338 	u_char *ptr;
339 	u_int len;
340 
341 	mod = NULL;
342 	ptr = NULL;
343 
344 	mod = preload_search_by_type("ram_blacklist");
345 	if (mod != NULL) {
346 		ptr = preload_fetch_addr(mod);
347 		len = preload_fetch_size(mod);
348         }
349 	*list = ptr;
350 	if (ptr != NULL)
351 		*end = ptr + len;
352 	else
353 		*end = NULL;
354 	return;
355 }
356 
357 static int
358 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
359 {
360 	vm_page_t m;
361 	struct sbuf sbuf;
362 	int error, first;
363 
364 	first = 1;
365 	error = sysctl_wire_old_buffer(req, 0);
366 	if (error != 0)
367 		return (error);
368 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
369 	TAILQ_FOREACH(m, &blacklist_head, listq) {
370 		sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
371 		    (uintmax_t)m->phys_addr);
372 		first = 0;
373 	}
374 	error = sbuf_finish(&sbuf);
375 	sbuf_delete(&sbuf);
376 	return (error);
377 }
378 
379 static void
380 vm_page_domain_init(struct vm_domain *vmd)
381 {
382 	struct vm_pagequeue *pq;
383 	int i;
384 
385 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
386 	    "vm inactive pagequeue";
387 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
388 	    &vm_cnt.v_inactive_count;
389 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
390 	    "vm active pagequeue";
391 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
392 	    &vm_cnt.v_active_count;
393 	vmd->vmd_page_count = 0;
394 	vmd->vmd_free_count = 0;
395 	vmd->vmd_segs = 0;
396 	vmd->vmd_oom = FALSE;
397 	vmd->vmd_pass = 0;
398 	for (i = 0; i < PQ_COUNT; i++) {
399 		pq = &vmd->vmd_pagequeues[i];
400 		TAILQ_INIT(&pq->pq_pl);
401 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
402 		    MTX_DEF | MTX_DUPOK);
403 	}
404 }
405 
406 /*
407  *	vm_page_startup:
408  *
409  *	Initializes the resident memory module.
410  *
411  *	Allocates memory for the page cells, and
412  *	for the object/offset-to-page hash table headers.
413  *	Each page cell is initialized and placed on the free list.
414  */
415 vm_offset_t
416 vm_page_startup(vm_offset_t vaddr)
417 {
418 	vm_offset_t mapped;
419 	vm_paddr_t page_range;
420 	vm_paddr_t new_end;
421 	int i;
422 	vm_paddr_t pa;
423 	vm_paddr_t last_pa;
424 	char *list, *listend;
425 	vm_paddr_t end;
426 	vm_paddr_t biggestsize;
427 	vm_paddr_t low_water, high_water;
428 	int biggestone;
429 
430 	biggestsize = 0;
431 	biggestone = 0;
432 	vaddr = round_page(vaddr);
433 
434 	for (i = 0; phys_avail[i + 1]; i += 2) {
435 		phys_avail[i] = round_page(phys_avail[i]);
436 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
437 	}
438 
439 	low_water = phys_avail[0];
440 	high_water = phys_avail[1];
441 
442 	for (i = 0; i < vm_phys_nsegs; i++) {
443 		if (vm_phys_segs[i].start < low_water)
444 			low_water = vm_phys_segs[i].start;
445 		if (vm_phys_segs[i].end > high_water)
446 			high_water = vm_phys_segs[i].end;
447 	}
448 	for (i = 0; phys_avail[i + 1]; i += 2) {
449 		vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
450 
451 		if (size > biggestsize) {
452 			biggestone = i;
453 			biggestsize = size;
454 		}
455 		if (phys_avail[i] < low_water)
456 			low_water = phys_avail[i];
457 		if (phys_avail[i + 1] > high_water)
458 			high_water = phys_avail[i + 1];
459 	}
460 
461 	end = phys_avail[biggestone+1];
462 
463 	/*
464 	 * Initialize the page and queue locks.
465 	 */
466 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
467 	for (i = 0; i < PA_LOCK_COUNT; i++)
468 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
469 	for (i = 0; i < vm_ndomains; i++)
470 		vm_page_domain_init(&vm_dom[i]);
471 
472 	/*
473 	 * Allocate memory for use when boot strapping the kernel memory
474 	 * allocator.
475 	 *
476 	 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
477 	 * manually fetch the value.
478 	 */
479 	TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
480 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
481 	new_end = trunc_page(new_end);
482 	mapped = pmap_map(&vaddr, new_end, end,
483 	    VM_PROT_READ | VM_PROT_WRITE);
484 	bzero((void *)mapped, end - new_end);
485 	uma_startup((void *)mapped, boot_pages);
486 
487 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
488     defined(__i386__) || defined(__mips__)
489 	/*
490 	 * Allocate a bitmap to indicate that a random physical page
491 	 * needs to be included in a minidump.
492 	 *
493 	 * The amd64 port needs this to indicate which direct map pages
494 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
495 	 *
496 	 * However, i386 still needs this workspace internally within the
497 	 * minidump code.  In theory, they are not needed on i386, but are
498 	 * included should the sf_buf code decide to use them.
499 	 */
500 	last_pa = 0;
501 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
502 		if (dump_avail[i + 1] > last_pa)
503 			last_pa = dump_avail[i + 1];
504 	page_range = last_pa / PAGE_SIZE;
505 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
506 	new_end -= vm_page_dump_size;
507 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
508 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
509 	bzero((void *)vm_page_dump, vm_page_dump_size);
510 #endif
511 #ifdef __amd64__
512 	/*
513 	 * Request that the physical pages underlying the message buffer be
514 	 * included in a crash dump.  Since the message buffer is accessed
515 	 * through the direct map, they are not automatically included.
516 	 */
517 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
518 	last_pa = pa + round_page(msgbufsize);
519 	while (pa < last_pa) {
520 		dump_add_page(pa);
521 		pa += PAGE_SIZE;
522 	}
523 #endif
524 	/*
525 	 * Compute the number of pages of memory that will be available for
526 	 * use (taking into account the overhead of a page structure per
527 	 * page).
528 	 */
529 	first_page = low_water / PAGE_SIZE;
530 #ifdef VM_PHYSSEG_SPARSE
531 	page_range = 0;
532 	for (i = 0; i < vm_phys_nsegs; i++) {
533 		page_range += atop(vm_phys_segs[i].end -
534 		    vm_phys_segs[i].start);
535 	}
536 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
537 		page_range += atop(phys_avail[i + 1] - phys_avail[i]);
538 #elif defined(VM_PHYSSEG_DENSE)
539 	page_range = high_water / PAGE_SIZE - first_page;
540 #else
541 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
542 #endif
543 	end = new_end;
544 
545 	/*
546 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
547 	 */
548 	vaddr += PAGE_SIZE;
549 
550 	/*
551 	 * Initialize the mem entry structures now, and put them in the free
552 	 * queue.
553 	 */
554 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
555 	mapped = pmap_map(&vaddr, new_end, end,
556 	    VM_PROT_READ | VM_PROT_WRITE);
557 	vm_page_array = (vm_page_t) mapped;
558 #if VM_NRESERVLEVEL > 0
559 	/*
560 	 * Allocate memory for the reservation management system's data
561 	 * structures.
562 	 */
563 	new_end = vm_reserv_startup(&vaddr, new_end, high_water);
564 #endif
565 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
566 	/*
567 	 * pmap_map on arm64, amd64, and mips can come out of the direct-map,
568 	 * not kvm like i386, so the pages must be tracked for a crashdump to
569 	 * include this data.  This includes the vm_page_array and the early
570 	 * UMA bootstrap pages.
571 	 */
572 	for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
573 		dump_add_page(pa);
574 #endif
575 	phys_avail[biggestone + 1] = new_end;
576 
577 	/*
578 	 * Add physical memory segments corresponding to the available
579 	 * physical pages.
580 	 */
581 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
582 		vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
583 
584 	/*
585 	 * Clear all of the page structures
586 	 */
587 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
588 	for (i = 0; i < page_range; i++)
589 		vm_page_array[i].order = VM_NFREEORDER;
590 	vm_page_array_size = page_range;
591 
592 	/*
593 	 * Initialize the physical memory allocator.
594 	 */
595 	vm_phys_init();
596 
597 	/*
598 	 * Add every available physical page that is not blacklisted to
599 	 * the free lists.
600 	 */
601 	vm_cnt.v_page_count = 0;
602 	vm_cnt.v_free_count = 0;
603 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
604 		pa = phys_avail[i];
605 		last_pa = phys_avail[i + 1];
606 		while (pa < last_pa) {
607 			vm_phys_add_page(pa);
608 			pa += PAGE_SIZE;
609 		}
610 	}
611 
612 	TAILQ_INIT(&blacklist_head);
613 	vm_page_blacklist_load(&list, &listend);
614 	vm_page_blacklist_check(list, listend);
615 
616 	list = kern_getenv("vm.blacklist");
617 	vm_page_blacklist_check(list, NULL);
618 
619 	freeenv(list);
620 #if VM_NRESERVLEVEL > 0
621 	/*
622 	 * Initialize the reservation management system.
623 	 */
624 	vm_reserv_init();
625 #endif
626 	return (vaddr);
627 }
628 
629 void
630 vm_page_reference(vm_page_t m)
631 {
632 
633 	vm_page_aflag_set(m, PGA_REFERENCED);
634 }
635 
636 /*
637  *	vm_page_busy_downgrade:
638  *
639  *	Downgrade an exclusive busy page into a single shared busy page.
640  */
641 void
642 vm_page_busy_downgrade(vm_page_t m)
643 {
644 	u_int x;
645 
646 	vm_page_assert_xbusied(m);
647 
648 	for (;;) {
649 		x = m->busy_lock;
650 		x &= VPB_BIT_WAITERS;
651 		if (atomic_cmpset_rel_int(&m->busy_lock,
652 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x))
653 			break;
654 	}
655 }
656 
657 /*
658  *	vm_page_sbusied:
659  *
660  *	Return a positive value if the page is shared busied, 0 otherwise.
661  */
662 int
663 vm_page_sbusied(vm_page_t m)
664 {
665 	u_int x;
666 
667 	x = m->busy_lock;
668 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
669 }
670 
671 /*
672  *	vm_page_sunbusy:
673  *
674  *	Shared unbusy a page.
675  */
676 void
677 vm_page_sunbusy(vm_page_t m)
678 {
679 	u_int x;
680 
681 	vm_page_assert_sbusied(m);
682 
683 	for (;;) {
684 		x = m->busy_lock;
685 		if (VPB_SHARERS(x) > 1) {
686 			if (atomic_cmpset_int(&m->busy_lock, x,
687 			    x - VPB_ONE_SHARER))
688 				break;
689 			continue;
690 		}
691 		if ((x & VPB_BIT_WAITERS) == 0) {
692 			KASSERT(x == VPB_SHARERS_WORD(1),
693 			    ("vm_page_sunbusy: invalid lock state"));
694 			if (atomic_cmpset_int(&m->busy_lock,
695 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
696 				break;
697 			continue;
698 		}
699 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
700 		    ("vm_page_sunbusy: invalid lock state for waiters"));
701 
702 		vm_page_lock(m);
703 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
704 			vm_page_unlock(m);
705 			continue;
706 		}
707 		wakeup(m);
708 		vm_page_unlock(m);
709 		break;
710 	}
711 }
712 
713 /*
714  *	vm_page_busy_sleep:
715  *
716  *	Sleep and release the page lock, using the page pointer as wchan.
717  *	This is used to implement the hard-path of busying mechanism.
718  *
719  *	The given page must be locked.
720  */
721 void
722 vm_page_busy_sleep(vm_page_t m, const char *wmesg)
723 {
724 	u_int x;
725 
726 	vm_page_lock_assert(m, MA_OWNED);
727 
728 	x = m->busy_lock;
729 	if (x == VPB_UNBUSIED) {
730 		vm_page_unlock(m);
731 		return;
732 	}
733 	if ((x & VPB_BIT_WAITERS) == 0 &&
734 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) {
735 		vm_page_unlock(m);
736 		return;
737 	}
738 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
739 }
740 
741 /*
742  *	vm_page_trysbusy:
743  *
744  *	Try to shared busy a page.
745  *	If the operation succeeds 1 is returned otherwise 0.
746  *	The operation never sleeps.
747  */
748 int
749 vm_page_trysbusy(vm_page_t m)
750 {
751 	u_int x;
752 
753 	for (;;) {
754 		x = m->busy_lock;
755 		if ((x & VPB_BIT_SHARED) == 0)
756 			return (0);
757 		if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
758 			return (1);
759 	}
760 }
761 
762 static void
763 vm_page_xunbusy_locked(vm_page_t m)
764 {
765 
766 	vm_page_assert_xbusied(m);
767 	vm_page_assert_locked(m);
768 
769 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
770 	/* There is a waiter, do wakeup() instead of vm_page_flash(). */
771 	wakeup(m);
772 }
773 
774 static void
775 vm_page_xunbusy_maybelocked(vm_page_t m)
776 {
777 	bool lockacq;
778 
779 	vm_page_assert_xbusied(m);
780 
781 	/*
782 	 * Fast path for unbusy.  If it succeeds, we know that there
783 	 * are no waiters, so we do not need a wakeup.
784 	 */
785 	if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER,
786 	    VPB_UNBUSIED))
787 		return;
788 
789 	lockacq = !mtx_owned(vm_page_lockptr(m));
790 	if (lockacq)
791 		vm_page_lock(m);
792 	vm_page_xunbusy_locked(m);
793 	if (lockacq)
794 		vm_page_unlock(m);
795 }
796 
797 /*
798  *	vm_page_xunbusy_hard:
799  *
800  *	Called after the first try the exclusive unbusy of a page failed.
801  *	It is assumed that the waiters bit is on.
802  */
803 void
804 vm_page_xunbusy_hard(vm_page_t m)
805 {
806 
807 	vm_page_assert_xbusied(m);
808 
809 	vm_page_lock(m);
810 	vm_page_xunbusy_locked(m);
811 	vm_page_unlock(m);
812 }
813 
814 /*
815  *	vm_page_flash:
816  *
817  *	Wakeup anyone waiting for the page.
818  *	The ownership bits do not change.
819  *
820  *	The given page must be locked.
821  */
822 void
823 vm_page_flash(vm_page_t m)
824 {
825 	u_int x;
826 
827 	vm_page_lock_assert(m, MA_OWNED);
828 
829 	for (;;) {
830 		x = m->busy_lock;
831 		if ((x & VPB_BIT_WAITERS) == 0)
832 			return;
833 		if (atomic_cmpset_int(&m->busy_lock, x,
834 		    x & (~VPB_BIT_WAITERS)))
835 			break;
836 	}
837 	wakeup(m);
838 }
839 
840 /*
841  * Keep page from being freed by the page daemon
842  * much of the same effect as wiring, except much lower
843  * overhead and should be used only for *very* temporary
844  * holding ("wiring").
845  */
846 void
847 vm_page_hold(vm_page_t mem)
848 {
849 
850 	vm_page_lock_assert(mem, MA_OWNED);
851         mem->hold_count++;
852 }
853 
854 void
855 vm_page_unhold(vm_page_t mem)
856 {
857 
858 	vm_page_lock_assert(mem, MA_OWNED);
859 	KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
860 	--mem->hold_count;
861 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
862 		vm_page_free_toq(mem);
863 }
864 
865 /*
866  *	vm_page_unhold_pages:
867  *
868  *	Unhold each of the pages that is referenced by the given array.
869  */
870 void
871 vm_page_unhold_pages(vm_page_t *ma, int count)
872 {
873 	struct mtx *mtx, *new_mtx;
874 
875 	mtx = NULL;
876 	for (; count != 0; count--) {
877 		/*
878 		 * Avoid releasing and reacquiring the same page lock.
879 		 */
880 		new_mtx = vm_page_lockptr(*ma);
881 		if (mtx != new_mtx) {
882 			if (mtx != NULL)
883 				mtx_unlock(mtx);
884 			mtx = new_mtx;
885 			mtx_lock(mtx);
886 		}
887 		vm_page_unhold(*ma);
888 		ma++;
889 	}
890 	if (mtx != NULL)
891 		mtx_unlock(mtx);
892 }
893 
894 vm_page_t
895 PHYS_TO_VM_PAGE(vm_paddr_t pa)
896 {
897 	vm_page_t m;
898 
899 #ifdef VM_PHYSSEG_SPARSE
900 	m = vm_phys_paddr_to_vm_page(pa);
901 	if (m == NULL)
902 		m = vm_phys_fictitious_to_vm_page(pa);
903 	return (m);
904 #elif defined(VM_PHYSSEG_DENSE)
905 	long pi;
906 
907 	pi = atop(pa);
908 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
909 		m = &vm_page_array[pi - first_page];
910 		return (m);
911 	}
912 	return (vm_phys_fictitious_to_vm_page(pa));
913 #else
914 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
915 #endif
916 }
917 
918 /*
919  *	vm_page_getfake:
920  *
921  *	Create a fictitious page with the specified physical address and
922  *	memory attribute.  The memory attribute is the only the machine-
923  *	dependent aspect of a fictitious page that must be initialized.
924  */
925 vm_page_t
926 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
927 {
928 	vm_page_t m;
929 
930 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
931 	vm_page_initfake(m, paddr, memattr);
932 	return (m);
933 }
934 
935 void
936 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
937 {
938 
939 	if ((m->flags & PG_FICTITIOUS) != 0) {
940 		/*
941 		 * The page's memattr might have changed since the
942 		 * previous initialization.  Update the pmap to the
943 		 * new memattr.
944 		 */
945 		goto memattr;
946 	}
947 	m->phys_addr = paddr;
948 	m->queue = PQ_NONE;
949 	/* Fictitious pages don't use "segind". */
950 	m->flags = PG_FICTITIOUS;
951 	/* Fictitious pages don't use "order" or "pool". */
952 	m->oflags = VPO_UNMANAGED;
953 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
954 	m->wire_count = 1;
955 	pmap_page_init(m);
956 memattr:
957 	pmap_page_set_memattr(m, memattr);
958 }
959 
960 /*
961  *	vm_page_putfake:
962  *
963  *	Release a fictitious page.
964  */
965 void
966 vm_page_putfake(vm_page_t m)
967 {
968 
969 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
970 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
971 	    ("vm_page_putfake: bad page %p", m));
972 	uma_zfree(fakepg_zone, m);
973 }
974 
975 /*
976  *	vm_page_updatefake:
977  *
978  *	Update the given fictitious page to the specified physical address and
979  *	memory attribute.
980  */
981 void
982 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
983 {
984 
985 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
986 	    ("vm_page_updatefake: bad page %p", m));
987 	m->phys_addr = paddr;
988 	pmap_page_set_memattr(m, memattr);
989 }
990 
991 /*
992  *	vm_page_free:
993  *
994  *	Free a page.
995  */
996 void
997 vm_page_free(vm_page_t m)
998 {
999 
1000 	m->flags &= ~PG_ZERO;
1001 	vm_page_free_toq(m);
1002 }
1003 
1004 /*
1005  *	vm_page_free_zero:
1006  *
1007  *	Free a page to the zerod-pages queue
1008  */
1009 void
1010 vm_page_free_zero(vm_page_t m)
1011 {
1012 
1013 	m->flags |= PG_ZERO;
1014 	vm_page_free_toq(m);
1015 }
1016 
1017 /*
1018  * Unbusy and handle the page queueing for a page from the VOP_GETPAGES()
1019  * array which was optionally read ahead or behind.
1020  */
1021 void
1022 vm_page_readahead_finish(vm_page_t m)
1023 {
1024 
1025 	/* We shouldn't put invalid pages on queues. */
1026 	KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m));
1027 
1028 	/*
1029 	 * Since the page is not the actually needed one, whether it should
1030 	 * be activated or deactivated is not obvious.  Empirical results
1031 	 * have shown that deactivating the page is usually the best choice,
1032 	 * unless the page is wanted by another thread.
1033 	 */
1034 	vm_page_lock(m);
1035 	if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1036 		vm_page_activate(m);
1037 	else
1038 		vm_page_deactivate(m);
1039 	vm_page_unlock(m);
1040 	vm_page_xunbusy(m);
1041 }
1042 
1043 /*
1044  *	vm_page_sleep_if_busy:
1045  *
1046  *	Sleep and release the page queues lock if the page is busied.
1047  *	Returns TRUE if the thread slept.
1048  *
1049  *	The given page must be unlocked and object containing it must
1050  *	be locked.
1051  */
1052 int
1053 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1054 {
1055 	vm_object_t obj;
1056 
1057 	vm_page_lock_assert(m, MA_NOTOWNED);
1058 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1059 
1060 	if (vm_page_busied(m)) {
1061 		/*
1062 		 * The page-specific object must be cached because page
1063 		 * identity can change during the sleep, causing the
1064 		 * re-lock of a different object.
1065 		 * It is assumed that a reference to the object is already
1066 		 * held by the callers.
1067 		 */
1068 		obj = m->object;
1069 		vm_page_lock(m);
1070 		VM_OBJECT_WUNLOCK(obj);
1071 		vm_page_busy_sleep(m, msg);
1072 		VM_OBJECT_WLOCK(obj);
1073 		return (TRUE);
1074 	}
1075 	return (FALSE);
1076 }
1077 
1078 /*
1079  *	vm_page_dirty_KBI:		[ internal use only ]
1080  *
1081  *	Set all bits in the page's dirty field.
1082  *
1083  *	The object containing the specified page must be locked if the
1084  *	call is made from the machine-independent layer.
1085  *
1086  *	See vm_page_clear_dirty_mask().
1087  *
1088  *	This function should only be called by vm_page_dirty().
1089  */
1090 void
1091 vm_page_dirty_KBI(vm_page_t m)
1092 {
1093 
1094 	/* These assertions refer to this operation by its public name. */
1095 	KASSERT((m->flags & PG_CACHED) == 0,
1096 	    ("vm_page_dirty: page in cache!"));
1097 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
1098 	    ("vm_page_dirty: page is invalid!"));
1099 	m->dirty = VM_PAGE_BITS_ALL;
1100 }
1101 
1102 /*
1103  *	vm_page_insert:		[ internal use only ]
1104  *
1105  *	Inserts the given mem entry into the object and object list.
1106  *
1107  *	The object must be locked.
1108  */
1109 int
1110 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1111 {
1112 	vm_page_t mpred;
1113 
1114 	VM_OBJECT_ASSERT_WLOCKED(object);
1115 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
1116 	return (vm_page_insert_after(m, object, pindex, mpred));
1117 }
1118 
1119 /*
1120  *	vm_page_insert_after:
1121  *
1122  *	Inserts the page "m" into the specified object at offset "pindex".
1123  *
1124  *	The page "mpred" must immediately precede the offset "pindex" within
1125  *	the specified object.
1126  *
1127  *	The object must be locked.
1128  */
1129 static int
1130 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1131     vm_page_t mpred)
1132 {
1133 	vm_page_t msucc;
1134 
1135 	VM_OBJECT_ASSERT_WLOCKED(object);
1136 	KASSERT(m->object == NULL,
1137 	    ("vm_page_insert_after: page already inserted"));
1138 	if (mpred != NULL) {
1139 		KASSERT(mpred->object == object,
1140 		    ("vm_page_insert_after: object doesn't contain mpred"));
1141 		KASSERT(mpred->pindex < pindex,
1142 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1143 		msucc = TAILQ_NEXT(mpred, listq);
1144 	} else
1145 		msucc = TAILQ_FIRST(&object->memq);
1146 	if (msucc != NULL)
1147 		KASSERT(msucc->pindex > pindex,
1148 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
1149 
1150 	/*
1151 	 * Record the object/offset pair in this page
1152 	 */
1153 	m->object = object;
1154 	m->pindex = pindex;
1155 
1156 	/*
1157 	 * Now link into the object's ordered list of backed pages.
1158 	 */
1159 	if (vm_radix_insert(&object->rtree, m)) {
1160 		m->object = NULL;
1161 		m->pindex = 0;
1162 		return (1);
1163 	}
1164 	vm_page_insert_radixdone(m, object, mpred);
1165 	return (0);
1166 }
1167 
1168 /*
1169  *	vm_page_insert_radixdone:
1170  *
1171  *	Complete page "m" insertion into the specified object after the
1172  *	radix trie hooking.
1173  *
1174  *	The page "mpred" must precede the offset "m->pindex" within the
1175  *	specified object.
1176  *
1177  *	The object must be locked.
1178  */
1179 static void
1180 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1181 {
1182 
1183 	VM_OBJECT_ASSERT_WLOCKED(object);
1184 	KASSERT(object != NULL && m->object == object,
1185 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1186 	if (mpred != NULL) {
1187 		KASSERT(mpred->object == object,
1188 		    ("vm_page_insert_after: object doesn't contain mpred"));
1189 		KASSERT(mpred->pindex < m->pindex,
1190 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1191 	}
1192 
1193 	if (mpred != NULL)
1194 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1195 	else
1196 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1197 
1198 	/*
1199 	 * Show that the object has one more resident page.
1200 	 */
1201 	object->resident_page_count++;
1202 
1203 	/*
1204 	 * Hold the vnode until the last page is released.
1205 	 */
1206 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1207 		vhold(object->handle);
1208 
1209 	/*
1210 	 * Since we are inserting a new and possibly dirty page,
1211 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1212 	 */
1213 	if (pmap_page_is_write_mapped(m))
1214 		vm_object_set_writeable_dirty(object);
1215 }
1216 
1217 /*
1218  *	vm_page_remove:
1219  *
1220  *	Removes the given mem entry from the object/offset-page
1221  *	table and the object page list, but do not invalidate/terminate
1222  *	the backing store.
1223  *
1224  *	The object must be locked.  The page must be locked if it is managed.
1225  */
1226 void
1227 vm_page_remove(vm_page_t m)
1228 {
1229 	vm_object_t object;
1230 
1231 	if ((m->oflags & VPO_UNMANAGED) == 0)
1232 		vm_page_assert_locked(m);
1233 	if ((object = m->object) == NULL)
1234 		return;
1235 	VM_OBJECT_ASSERT_WLOCKED(object);
1236 	if (vm_page_xbusied(m))
1237 		vm_page_xunbusy_maybelocked(m);
1238 
1239 	/*
1240 	 * Now remove from the object's list of backed pages.
1241 	 */
1242 	vm_radix_remove(&object->rtree, m->pindex);
1243 	TAILQ_REMOVE(&object->memq, m, listq);
1244 
1245 	/*
1246 	 * And show that the object has one fewer resident page.
1247 	 */
1248 	object->resident_page_count--;
1249 
1250 	/*
1251 	 * The vnode may now be recycled.
1252 	 */
1253 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1254 		vdrop(object->handle);
1255 
1256 	m->object = NULL;
1257 }
1258 
1259 /*
1260  *	vm_page_lookup:
1261  *
1262  *	Returns the page associated with the object/offset
1263  *	pair specified; if none is found, NULL is returned.
1264  *
1265  *	The object must be locked.
1266  */
1267 vm_page_t
1268 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1269 {
1270 
1271 	VM_OBJECT_ASSERT_LOCKED(object);
1272 	return (vm_radix_lookup(&object->rtree, pindex));
1273 }
1274 
1275 /*
1276  *	vm_page_find_least:
1277  *
1278  *	Returns the page associated with the object with least pindex
1279  *	greater than or equal to the parameter pindex, or NULL.
1280  *
1281  *	The object must be locked.
1282  */
1283 vm_page_t
1284 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1285 {
1286 	vm_page_t m;
1287 
1288 	VM_OBJECT_ASSERT_LOCKED(object);
1289 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1290 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1291 	return (m);
1292 }
1293 
1294 /*
1295  * Returns the given page's successor (by pindex) within the object if it is
1296  * resident; if none is found, NULL is returned.
1297  *
1298  * The object must be locked.
1299  */
1300 vm_page_t
1301 vm_page_next(vm_page_t m)
1302 {
1303 	vm_page_t next;
1304 
1305 	VM_OBJECT_ASSERT_LOCKED(m->object);
1306 	if ((next = TAILQ_NEXT(m, listq)) != NULL &&
1307 	    next->pindex != m->pindex + 1)
1308 		next = NULL;
1309 	return (next);
1310 }
1311 
1312 /*
1313  * Returns the given page's predecessor (by pindex) within the object if it is
1314  * resident; if none is found, NULL is returned.
1315  *
1316  * The object must be locked.
1317  */
1318 vm_page_t
1319 vm_page_prev(vm_page_t m)
1320 {
1321 	vm_page_t prev;
1322 
1323 	VM_OBJECT_ASSERT_LOCKED(m->object);
1324 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL &&
1325 	    prev->pindex != m->pindex - 1)
1326 		prev = NULL;
1327 	return (prev);
1328 }
1329 
1330 /*
1331  * Uses the page mnew as a replacement for an existing page at index
1332  * pindex which must be already present in the object.
1333  *
1334  * The existing page must not be on a paging queue.
1335  */
1336 vm_page_t
1337 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1338 {
1339 	vm_page_t mold;
1340 
1341 	VM_OBJECT_ASSERT_WLOCKED(object);
1342 	KASSERT(mnew->object == NULL,
1343 	    ("vm_page_replace: page already in object"));
1344 
1345 	/*
1346 	 * This function mostly follows vm_page_insert() and
1347 	 * vm_page_remove() without the radix, object count and vnode
1348 	 * dance.  Double check such functions for more comments.
1349 	 */
1350 
1351 	mnew->object = object;
1352 	mnew->pindex = pindex;
1353 	mold = vm_radix_replace(&object->rtree, mnew);
1354 	KASSERT(mold->queue == PQ_NONE,
1355 	    ("vm_page_replace: mold is on a paging queue"));
1356 
1357 	/* Keep the resident page list in sorted order. */
1358 	TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1359 	TAILQ_REMOVE(&object->memq, mold, listq);
1360 
1361 	mold->object = NULL;
1362 	vm_page_xunbusy_maybelocked(mold);
1363 
1364 	/*
1365 	 * The object's resident_page_count does not change because we have
1366 	 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1367 	 */
1368 	if (pmap_page_is_write_mapped(mnew))
1369 		vm_object_set_writeable_dirty(object);
1370 	return (mold);
1371 }
1372 
1373 /*
1374  *	vm_page_rename:
1375  *
1376  *	Move the given memory entry from its
1377  *	current object to the specified target object/offset.
1378  *
1379  *	Note: swap associated with the page must be invalidated by the move.  We
1380  *	      have to do this for several reasons:  (1) we aren't freeing the
1381  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1382  *	      moving the page from object A to B, and will then later move
1383  *	      the backing store from A to B and we can't have a conflict.
1384  *
1385  *	Note: we *always* dirty the page.  It is necessary both for the
1386  *	      fact that we moved it, and because we may be invalidating
1387  *	      swap.  If the page is on the cache, we have to deactivate it
1388  *	      or vm_page_dirty() will panic.  Dirty pages are not allowed
1389  *	      on the cache.
1390  *
1391  *	The objects must be locked.
1392  */
1393 int
1394 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1395 {
1396 	vm_page_t mpred;
1397 	vm_pindex_t opidx;
1398 
1399 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1400 
1401 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1402 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1403 	    ("vm_page_rename: pindex already renamed"));
1404 
1405 	/*
1406 	 * Create a custom version of vm_page_insert() which does not depend
1407 	 * by m_prev and can cheat on the implementation aspects of the
1408 	 * function.
1409 	 */
1410 	opidx = m->pindex;
1411 	m->pindex = new_pindex;
1412 	if (vm_radix_insert(&new_object->rtree, m)) {
1413 		m->pindex = opidx;
1414 		return (1);
1415 	}
1416 
1417 	/*
1418 	 * The operation cannot fail anymore.  The removal must happen before
1419 	 * the listq iterator is tainted.
1420 	 */
1421 	m->pindex = opidx;
1422 	vm_page_lock(m);
1423 	vm_page_remove(m);
1424 
1425 	/* Return back to the new pindex to complete vm_page_insert(). */
1426 	m->pindex = new_pindex;
1427 	m->object = new_object;
1428 	vm_page_unlock(m);
1429 	vm_page_insert_radixdone(m, new_object, mpred);
1430 	vm_page_dirty(m);
1431 	return (0);
1432 }
1433 
1434 /*
1435  *	Convert all of the given object's cached pages that have a
1436  *	pindex within the given range into free pages.  If the value
1437  *	zero is given for "end", then the range's upper bound is
1438  *	infinity.  If the given object is backed by a vnode and it
1439  *	transitions from having one or more cached pages to none, the
1440  *	vnode's hold count is reduced.
1441  */
1442 void
1443 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1444 {
1445 	vm_page_t m;
1446 	boolean_t empty;
1447 
1448 	mtx_lock(&vm_page_queue_free_mtx);
1449 	if (__predict_false(vm_radix_is_empty(&object->cache))) {
1450 		mtx_unlock(&vm_page_queue_free_mtx);
1451 		return;
1452 	}
1453 	while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) {
1454 		if (end != 0 && m->pindex >= end)
1455 			break;
1456 		vm_radix_remove(&object->cache, m->pindex);
1457 		vm_page_cache_turn_free(m);
1458 	}
1459 	empty = vm_radix_is_empty(&object->cache);
1460 	mtx_unlock(&vm_page_queue_free_mtx);
1461 	if (object->type == OBJT_VNODE && empty)
1462 		vdrop(object->handle);
1463 }
1464 
1465 /*
1466  *	Returns the cached page that is associated with the given
1467  *	object and offset.  If, however, none exists, returns NULL.
1468  *
1469  *	The free page queue must be locked.
1470  */
1471 static inline vm_page_t
1472 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
1473 {
1474 
1475 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1476 	return (vm_radix_lookup(&object->cache, pindex));
1477 }
1478 
1479 /*
1480  *	Remove the given cached page from its containing object's
1481  *	collection of cached pages.
1482  *
1483  *	The free page queue must be locked.
1484  */
1485 static void
1486 vm_page_cache_remove(vm_page_t m)
1487 {
1488 
1489 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1490 	KASSERT((m->flags & PG_CACHED) != 0,
1491 	    ("vm_page_cache_remove: page %p is not cached", m));
1492 	vm_radix_remove(&m->object->cache, m->pindex);
1493 	m->object = NULL;
1494 	vm_cnt.v_cache_count--;
1495 }
1496 
1497 /*
1498  *	Transfer all of the cached pages with offset greater than or
1499  *	equal to 'offidxstart' from the original object's cache to the
1500  *	new object's cache.  However, any cached pages with offset
1501  *	greater than or equal to the new object's size are kept in the
1502  *	original object.  Initially, the new object's cache must be
1503  *	empty.  Offset 'offidxstart' in the original object must
1504  *	correspond to offset zero in the new object.
1505  *
1506  *	The new object must be locked.
1507  */
1508 void
1509 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
1510     vm_object_t new_object)
1511 {
1512 	vm_page_t m;
1513 
1514 	/*
1515 	 * Insertion into an object's collection of cached pages
1516 	 * requires the object to be locked.  In contrast, removal does
1517 	 * not.
1518 	 */
1519 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1520 	KASSERT(vm_radix_is_empty(&new_object->cache),
1521 	    ("vm_page_cache_transfer: object %p has cached pages",
1522 	    new_object));
1523 	mtx_lock(&vm_page_queue_free_mtx);
1524 	while ((m = vm_radix_lookup_ge(&orig_object->cache,
1525 	    offidxstart)) != NULL) {
1526 		/*
1527 		 * Transfer all of the pages with offset greater than or
1528 		 * equal to 'offidxstart' from the original object's
1529 		 * cache to the new object's cache.
1530 		 */
1531 		if ((m->pindex - offidxstart) >= new_object->size)
1532 			break;
1533 		vm_radix_remove(&orig_object->cache, m->pindex);
1534 		/* Update the page's object and offset. */
1535 		m->object = new_object;
1536 		m->pindex -= offidxstart;
1537 		if (vm_radix_insert(&new_object->cache, m))
1538 			vm_page_cache_turn_free(m);
1539 	}
1540 	mtx_unlock(&vm_page_queue_free_mtx);
1541 }
1542 
1543 /*
1544  *	Returns TRUE if a cached page is associated with the given object and
1545  *	offset, and FALSE otherwise.
1546  *
1547  *	The object must be locked.
1548  */
1549 boolean_t
1550 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex)
1551 {
1552 	vm_page_t m;
1553 
1554 	/*
1555 	 * Insertion into an object's collection of cached pages requires the
1556 	 * object to be locked.  Therefore, if the object is locked and the
1557 	 * object's collection is empty, there is no need to acquire the free
1558 	 * page queues lock in order to prove that the specified page doesn't
1559 	 * exist.
1560 	 */
1561 	VM_OBJECT_ASSERT_WLOCKED(object);
1562 	if (__predict_true(vm_object_cache_is_empty(object)))
1563 		return (FALSE);
1564 	mtx_lock(&vm_page_queue_free_mtx);
1565 	m = vm_page_cache_lookup(object, pindex);
1566 	mtx_unlock(&vm_page_queue_free_mtx);
1567 	return (m != NULL);
1568 }
1569 
1570 /*
1571  *	vm_page_alloc:
1572  *
1573  *	Allocate and return a page that is associated with the specified
1574  *	object and offset pair.  By default, this page is exclusive busied.
1575  *
1576  *	The caller must always specify an allocation class.
1577  *
1578  *	allocation classes:
1579  *	VM_ALLOC_NORMAL		normal process request
1580  *	VM_ALLOC_SYSTEM		system *really* needs a page
1581  *	VM_ALLOC_INTERRUPT	interrupt time request
1582  *
1583  *	optional allocation flags:
1584  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1585  *				intends to allocate
1586  *	VM_ALLOC_IFCACHED	return page only if it is cached
1587  *	VM_ALLOC_IFNOTCACHED	return NULL, do not reactivate if the page
1588  *				is cached
1589  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1590  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1591  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1592  *				should not be exclusive busy
1593  *	VM_ALLOC_SBUSY		shared busy the allocated page
1594  *	VM_ALLOC_WIRED		wire the allocated page
1595  *	VM_ALLOC_ZERO		prefer a zeroed page
1596  *
1597  *	This routine may not sleep.
1598  */
1599 vm_page_t
1600 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1601 {
1602 	struct vnode *vp = NULL;
1603 	vm_object_t m_object;
1604 	vm_page_t m, mpred;
1605 	int flags, req_class;
1606 
1607 	mpred = 0;	/* XXX: pacify gcc */
1608 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1609 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1610 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1611 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1612 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1613 	    req));
1614 	if (object != NULL)
1615 		VM_OBJECT_ASSERT_WLOCKED(object);
1616 
1617 	req_class = req & VM_ALLOC_CLASS_MASK;
1618 
1619 	/*
1620 	 * The page daemon is allowed to dig deeper into the free page list.
1621 	 */
1622 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1623 		req_class = VM_ALLOC_SYSTEM;
1624 
1625 	if (object != NULL) {
1626 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1627 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1628 		   ("vm_page_alloc: pindex already allocated"));
1629 	}
1630 
1631 	/*
1632 	 * The page allocation request can came from consumers which already
1633 	 * hold the free page queue mutex, like vm_page_insert() in
1634 	 * vm_page_cache().
1635 	 */
1636 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
1637 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
1638 	    (req_class == VM_ALLOC_SYSTEM &&
1639 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
1640 	    (req_class == VM_ALLOC_INTERRUPT &&
1641 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) {
1642 		/*
1643 		 * Allocate from the free queue if the number of free pages
1644 		 * exceeds the minimum for the request class.
1645 		 */
1646 		if (object != NULL &&
1647 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1648 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1649 				mtx_unlock(&vm_page_queue_free_mtx);
1650 				return (NULL);
1651 			}
1652 			if (vm_phys_unfree_page(m))
1653 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1654 #if VM_NRESERVLEVEL > 0
1655 			else if (!vm_reserv_reactivate_page(m))
1656 #else
1657 			else
1658 #endif
1659 				panic("vm_page_alloc: cache page %p is missing"
1660 				    " from the free queue", m);
1661 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1662 			mtx_unlock(&vm_page_queue_free_mtx);
1663 			return (NULL);
1664 #if VM_NRESERVLEVEL > 0
1665 		} else if (object == NULL || (object->flags & (OBJ_COLORED |
1666 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1667 		    vm_reserv_alloc_page(object, pindex, mpred)) == NULL) {
1668 #else
1669 		} else {
1670 #endif
1671 			m = vm_phys_alloc_pages(object != NULL ?
1672 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1673 #if VM_NRESERVLEVEL > 0
1674 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1675 				m = vm_phys_alloc_pages(object != NULL ?
1676 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1677 				    0);
1678 			}
1679 #endif
1680 		}
1681 	} else {
1682 		/*
1683 		 * Not allocatable, give up.
1684 		 */
1685 		mtx_unlock(&vm_page_queue_free_mtx);
1686 		atomic_add_int(&vm_pageout_deficit,
1687 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1688 		pagedaemon_wakeup();
1689 		return (NULL);
1690 	}
1691 
1692 	/*
1693 	 *  At this point we had better have found a good page.
1694 	 */
1695 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1696 	KASSERT(m->queue == PQ_NONE,
1697 	    ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
1698 	KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
1699 	KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
1700 	KASSERT(!vm_page_sbusied(m),
1701 	    ("vm_page_alloc: page %p is busy", m));
1702 	KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
1703 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1704 	    ("vm_page_alloc: page %p has unexpected memattr %d", m,
1705 	    pmap_page_get_memattr(m)));
1706 	if ((m->flags & PG_CACHED) != 0) {
1707 		KASSERT((m->flags & PG_ZERO) == 0,
1708 		    ("vm_page_alloc: cached page %p is PG_ZERO", m));
1709 		KASSERT(m->valid != 0,
1710 		    ("vm_page_alloc: cached page %p is invalid", m));
1711 		if (m->object == object && m->pindex == pindex)
1712 			vm_cnt.v_reactivated++;
1713 		else
1714 			m->valid = 0;
1715 		m_object = m->object;
1716 		vm_page_cache_remove(m);
1717 		if (m_object->type == OBJT_VNODE &&
1718 		    vm_object_cache_is_empty(m_object))
1719 			vp = m_object->handle;
1720 	} else {
1721 		KASSERT(m->valid == 0,
1722 		    ("vm_page_alloc: free page %p is valid", m));
1723 		vm_phys_freecnt_adj(m, -1);
1724 		if ((m->flags & PG_ZERO) != 0)
1725 			vm_page_zero_count--;
1726 	}
1727 	mtx_unlock(&vm_page_queue_free_mtx);
1728 
1729 	/*
1730 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1731 	 */
1732 	flags = 0;
1733 	if ((req & VM_ALLOC_ZERO) != 0)
1734 		flags = PG_ZERO;
1735 	flags &= m->flags;
1736 	if ((req & VM_ALLOC_NODUMP) != 0)
1737 		flags |= PG_NODUMP;
1738 	m->flags = flags;
1739 	m->aflags = 0;
1740 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1741 	    VPO_UNMANAGED : 0;
1742 	m->busy_lock = VPB_UNBUSIED;
1743 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1744 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1745 	if ((req & VM_ALLOC_SBUSY) != 0)
1746 		m->busy_lock = VPB_SHARERS_WORD(1);
1747 	if (req & VM_ALLOC_WIRED) {
1748 		/*
1749 		 * The page lock is not required for wiring a page until that
1750 		 * page is inserted into the object.
1751 		 */
1752 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1753 		m->wire_count = 1;
1754 	}
1755 	m->act_count = 0;
1756 
1757 	if (object != NULL) {
1758 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1759 			/* See the comment below about hold count. */
1760 			if (vp != NULL)
1761 				vdrop(vp);
1762 			pagedaemon_wakeup();
1763 			if (req & VM_ALLOC_WIRED) {
1764 				atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1765 				m->wire_count = 0;
1766 			}
1767 			m->object = NULL;
1768 			m->oflags = VPO_UNMANAGED;
1769 			m->busy_lock = VPB_UNBUSIED;
1770 			vm_page_free(m);
1771 			return (NULL);
1772 		}
1773 
1774 		/* Ignore device objects; the pager sets "memattr" for them. */
1775 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1776 		    (object->flags & OBJ_FICTITIOUS) == 0)
1777 			pmap_page_set_memattr(m, object->memattr);
1778 	} else
1779 		m->pindex = pindex;
1780 
1781 	/*
1782 	 * The following call to vdrop() must come after the above call
1783 	 * to vm_page_insert() in case both affect the same object and
1784 	 * vnode.  Otherwise, the affected vnode's hold count could
1785 	 * temporarily become zero.
1786 	 */
1787 	if (vp != NULL)
1788 		vdrop(vp);
1789 
1790 	/*
1791 	 * Don't wakeup too often - wakeup the pageout daemon when
1792 	 * we would be nearly out of memory.
1793 	 */
1794 	if (vm_paging_needed())
1795 		pagedaemon_wakeup();
1796 
1797 	return (m);
1798 }
1799 
1800 static void
1801 vm_page_alloc_contig_vdrop(struct spglist *lst)
1802 {
1803 
1804 	while (!SLIST_EMPTY(lst)) {
1805 		vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv);
1806 		SLIST_REMOVE_HEAD(lst, plinks.s.ss);
1807 	}
1808 }
1809 
1810 /*
1811  *	vm_page_alloc_contig:
1812  *
1813  *	Allocate a contiguous set of physical pages of the given size "npages"
1814  *	from the free lists.  All of the physical pages must be at or above
1815  *	the given physical address "low" and below the given physical address
1816  *	"high".  The given value "alignment" determines the alignment of the
1817  *	first physical page in the set.  If the given value "boundary" is
1818  *	non-zero, then the set of physical pages cannot cross any physical
1819  *	address boundary that is a multiple of that value.  Both "alignment"
1820  *	and "boundary" must be a power of two.
1821  *
1822  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1823  *	then the memory attribute setting for the physical pages is configured
1824  *	to the object's memory attribute setting.  Otherwise, the memory
1825  *	attribute setting for the physical pages is configured to "memattr",
1826  *	overriding the object's memory attribute setting.  However, if the
1827  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1828  *	memory attribute setting for the physical pages cannot be configured
1829  *	to VM_MEMATTR_DEFAULT.
1830  *
1831  *	The caller must always specify an allocation class.
1832  *
1833  *	allocation classes:
1834  *	VM_ALLOC_NORMAL		normal process request
1835  *	VM_ALLOC_SYSTEM		system *really* needs a page
1836  *	VM_ALLOC_INTERRUPT	interrupt time request
1837  *
1838  *	optional allocation flags:
1839  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1840  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1841  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1842  *				should not be exclusive busy
1843  *	VM_ALLOC_SBUSY		shared busy the allocated page
1844  *	VM_ALLOC_WIRED		wire the allocated page
1845  *	VM_ALLOC_ZERO		prefer a zeroed page
1846  *
1847  *	This routine may not sleep.
1848  */
1849 vm_page_t
1850 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1851     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1852     vm_paddr_t boundary, vm_memattr_t memattr)
1853 {
1854 	struct vnode *drop;
1855 	struct spglist deferred_vdrop_list;
1856 	vm_page_t m, m_tmp, m_ret;
1857 	u_int flags;
1858 	int req_class;
1859 
1860 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1861 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1862 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1863 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1864 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1865 	    req));
1866 	if (object != NULL) {
1867 		VM_OBJECT_ASSERT_WLOCKED(object);
1868 		KASSERT(object->type == OBJT_PHYS,
1869 		    ("vm_page_alloc_contig: object %p isn't OBJT_PHYS",
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 	SLIST_INIT(&deferred_vdrop_list);
1882 	mtx_lock(&vm_page_queue_free_mtx);
1883 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1884 	    vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM &&
1885 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1886 	    vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT &&
1887 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) {
1888 #if VM_NRESERVLEVEL > 0
1889 retry:
1890 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1891 		    (m_ret = vm_reserv_alloc_contig(object, pindex, npages,
1892 		    low, high, alignment, boundary)) == NULL)
1893 #endif
1894 			m_ret = vm_phys_alloc_contig(npages, low, high,
1895 			    alignment, boundary);
1896 	} else {
1897 		mtx_unlock(&vm_page_queue_free_mtx);
1898 		atomic_add_int(&vm_pageout_deficit, npages);
1899 		pagedaemon_wakeup();
1900 		return (NULL);
1901 	}
1902 	if (m_ret != NULL)
1903 		for (m = m_ret; m < &m_ret[npages]; m++) {
1904 			drop = vm_page_alloc_init(m);
1905 			if (drop != NULL) {
1906 				/*
1907 				 * Enqueue the vnode for deferred vdrop().
1908 				 */
1909 				m->plinks.s.pv = drop;
1910 				SLIST_INSERT_HEAD(&deferred_vdrop_list, m,
1911 				    plinks.s.ss);
1912 			}
1913 		}
1914 	else {
1915 #if VM_NRESERVLEVEL > 0
1916 		if (vm_reserv_reclaim_contig(npages, low, high, alignment,
1917 		    boundary))
1918 			goto retry;
1919 #endif
1920 	}
1921 	mtx_unlock(&vm_page_queue_free_mtx);
1922 	if (m_ret == NULL)
1923 		return (NULL);
1924 
1925 	/*
1926 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1927 	 */
1928 	flags = 0;
1929 	if ((req & VM_ALLOC_ZERO) != 0)
1930 		flags = PG_ZERO;
1931 	if ((req & VM_ALLOC_NODUMP) != 0)
1932 		flags |= PG_NODUMP;
1933 	if ((req & VM_ALLOC_WIRED) != 0)
1934 		atomic_add_int(&vm_cnt.v_wire_count, npages);
1935 	if (object != NULL) {
1936 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1937 		    memattr == VM_MEMATTR_DEFAULT)
1938 			memattr = object->memattr;
1939 	}
1940 	for (m = m_ret; m < &m_ret[npages]; m++) {
1941 		m->aflags = 0;
1942 		m->flags = (m->flags | PG_NODUMP) & flags;
1943 		m->busy_lock = VPB_UNBUSIED;
1944 		if (object != NULL) {
1945 			if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
1946 				m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1947 			if ((req & VM_ALLOC_SBUSY) != 0)
1948 				m->busy_lock = VPB_SHARERS_WORD(1);
1949 		}
1950 		if ((req & VM_ALLOC_WIRED) != 0)
1951 			m->wire_count = 1;
1952 		/* Unmanaged pages don't use "act_count". */
1953 		m->oflags = VPO_UNMANAGED;
1954 		if (object != NULL) {
1955 			if (vm_page_insert(m, object, pindex)) {
1956 				vm_page_alloc_contig_vdrop(
1957 				    &deferred_vdrop_list);
1958 				if (vm_paging_needed())
1959 					pagedaemon_wakeup();
1960 				if ((req & VM_ALLOC_WIRED) != 0)
1961 					atomic_subtract_int(&vm_cnt.v_wire_count,
1962 					    npages);
1963 				for (m_tmp = m, m = m_ret;
1964 				    m < &m_ret[npages]; m++) {
1965 					if ((req & VM_ALLOC_WIRED) != 0)
1966 						m->wire_count = 0;
1967 					if (m >= m_tmp) {
1968 						m->object = NULL;
1969 						m->oflags |= VPO_UNMANAGED;
1970 					}
1971 					m->busy_lock = VPB_UNBUSIED;
1972 					vm_page_free(m);
1973 				}
1974 				return (NULL);
1975 			}
1976 		} else
1977 			m->pindex = pindex;
1978 		if (memattr != VM_MEMATTR_DEFAULT)
1979 			pmap_page_set_memattr(m, memattr);
1980 		pindex++;
1981 	}
1982 	vm_page_alloc_contig_vdrop(&deferred_vdrop_list);
1983 	if (vm_paging_needed())
1984 		pagedaemon_wakeup();
1985 	return (m_ret);
1986 }
1987 
1988 /*
1989  * Initialize a page that has been freshly dequeued from a freelist.
1990  * The caller has to drop the vnode returned, if it is not NULL.
1991  *
1992  * This function may only be used to initialize unmanaged pages.
1993  *
1994  * To be called with vm_page_queue_free_mtx held.
1995  */
1996 static struct vnode *
1997 vm_page_alloc_init(vm_page_t m)
1998 {
1999 	struct vnode *drop;
2000 	vm_object_t m_object;
2001 
2002 	KASSERT(m->queue == PQ_NONE,
2003 	    ("vm_page_alloc_init: page %p has unexpected queue %d",
2004 	    m, m->queue));
2005 	KASSERT(m->wire_count == 0,
2006 	    ("vm_page_alloc_init: page %p is wired", m));
2007 	KASSERT(m->hold_count == 0,
2008 	    ("vm_page_alloc_init: page %p is held", m));
2009 	KASSERT(!vm_page_sbusied(m),
2010 	    ("vm_page_alloc_init: page %p is busy", m));
2011 	KASSERT(m->dirty == 0,
2012 	    ("vm_page_alloc_init: page %p is dirty", m));
2013 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
2014 	    ("vm_page_alloc_init: page %p has unexpected memattr %d",
2015 	    m, pmap_page_get_memattr(m)));
2016 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2017 	drop = NULL;
2018 	if ((m->flags & PG_CACHED) != 0) {
2019 		KASSERT((m->flags & PG_ZERO) == 0,
2020 		    ("vm_page_alloc_init: cached page %p is PG_ZERO", m));
2021 		m->valid = 0;
2022 		m_object = m->object;
2023 		vm_page_cache_remove(m);
2024 		if (m_object->type == OBJT_VNODE &&
2025 		    vm_object_cache_is_empty(m_object))
2026 			drop = m_object->handle;
2027 	} else {
2028 		KASSERT(m->valid == 0,
2029 		    ("vm_page_alloc_init: free page %p is valid", m));
2030 		vm_phys_freecnt_adj(m, -1);
2031 		if ((m->flags & PG_ZERO) != 0)
2032 			vm_page_zero_count--;
2033 	}
2034 	return (drop);
2035 }
2036 
2037 /*
2038  * 	vm_page_alloc_freelist:
2039  *
2040  *	Allocate a physical page from the specified free page list.
2041  *
2042  *	The caller must always specify an allocation class.
2043  *
2044  *	allocation classes:
2045  *	VM_ALLOC_NORMAL		normal process request
2046  *	VM_ALLOC_SYSTEM		system *really* needs a page
2047  *	VM_ALLOC_INTERRUPT	interrupt time request
2048  *
2049  *	optional allocation flags:
2050  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
2051  *				intends to allocate
2052  *	VM_ALLOC_WIRED		wire the allocated page
2053  *	VM_ALLOC_ZERO		prefer a zeroed page
2054  *
2055  *	This routine may not sleep.
2056  */
2057 vm_page_t
2058 vm_page_alloc_freelist(int flind, int req)
2059 {
2060 	struct vnode *drop;
2061 	vm_page_t m;
2062 	u_int flags;
2063 	int req_class;
2064 
2065 	req_class = req & VM_ALLOC_CLASS_MASK;
2066 
2067 	/*
2068 	 * The page daemon is allowed to dig deeper into the free page list.
2069 	 */
2070 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2071 		req_class = VM_ALLOC_SYSTEM;
2072 
2073 	/*
2074 	 * Do not allocate reserved pages unless the req has asked for it.
2075 	 */
2076 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
2077 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
2078 	    (req_class == VM_ALLOC_SYSTEM &&
2079 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
2080 	    (req_class == VM_ALLOC_INTERRUPT &&
2081 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0))
2082 		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0);
2083 	else {
2084 		mtx_unlock(&vm_page_queue_free_mtx);
2085 		atomic_add_int(&vm_pageout_deficit,
2086 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
2087 		pagedaemon_wakeup();
2088 		return (NULL);
2089 	}
2090 	if (m == NULL) {
2091 		mtx_unlock(&vm_page_queue_free_mtx);
2092 		return (NULL);
2093 	}
2094 	drop = vm_page_alloc_init(m);
2095 	mtx_unlock(&vm_page_queue_free_mtx);
2096 
2097 	/*
2098 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
2099 	 */
2100 	m->aflags = 0;
2101 	flags = 0;
2102 	if ((req & VM_ALLOC_ZERO) != 0)
2103 		flags = PG_ZERO;
2104 	m->flags &= flags;
2105 	if ((req & VM_ALLOC_WIRED) != 0) {
2106 		/*
2107 		 * The page lock is not required for wiring a page that does
2108 		 * not belong to an object.
2109 		 */
2110 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2111 		m->wire_count = 1;
2112 	}
2113 	/* Unmanaged pages don't use "act_count". */
2114 	m->oflags = VPO_UNMANAGED;
2115 	if (drop != NULL)
2116 		vdrop(drop);
2117 	if (vm_paging_needed())
2118 		pagedaemon_wakeup();
2119 	return (m);
2120 }
2121 
2122 #define	VPSC_ANY	0	/* No restrictions. */
2123 #define	VPSC_NORESERV	1	/* Skip reservations; implies VPSC_NOSUPER. */
2124 #define	VPSC_NOSUPER	2	/* Skip superpages. */
2125 
2126 /*
2127  *	vm_page_scan_contig:
2128  *
2129  *	Scan vm_page_array[] between the specified entries "m_start" and
2130  *	"m_end" for a run of contiguous physical pages that satisfy the
2131  *	specified conditions, and return the lowest page in the run.  The
2132  *	specified "alignment" determines the alignment of the lowest physical
2133  *	page in the run.  If the specified "boundary" is non-zero, then the
2134  *	run of physical pages cannot span a physical address that is a
2135  *	multiple of "boundary".
2136  *
2137  *	"m_end" is never dereferenced, so it need not point to a vm_page
2138  *	structure within vm_page_array[].
2139  *
2140  *	"npages" must be greater than zero.  "m_start" and "m_end" must not
2141  *	span a hole (or discontiguity) in the physical address space.  Both
2142  *	"alignment" and "boundary" must be a power of two.
2143  */
2144 vm_page_t
2145 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
2146     u_long alignment, vm_paddr_t boundary, int options)
2147 {
2148 	struct mtx *m_mtx, *new_mtx;
2149 	vm_object_t object;
2150 	vm_paddr_t pa;
2151 	vm_page_t m, m_run;
2152 #if VM_NRESERVLEVEL > 0
2153 	int level;
2154 #endif
2155 	int m_inc, order, run_ext, run_len;
2156 
2157 	KASSERT(npages > 0, ("npages is 0"));
2158 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2159 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2160 	m_run = NULL;
2161 	run_len = 0;
2162 	m_mtx = NULL;
2163 	for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
2164 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2165 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2166 
2167 		/*
2168 		 * If the current page would be the start of a run, check its
2169 		 * physical address against the end, alignment, and boundary
2170 		 * conditions.  If it doesn't satisfy these conditions, either
2171 		 * terminate the scan or advance to the next page that
2172 		 * satisfies the failed condition.
2173 		 */
2174 		if (run_len == 0) {
2175 			KASSERT(m_run == NULL, ("m_run != NULL"));
2176 			if (m + npages > m_end)
2177 				break;
2178 			pa = VM_PAGE_TO_PHYS(m);
2179 			if ((pa & (alignment - 1)) != 0) {
2180 				m_inc = atop(roundup2(pa, alignment) - pa);
2181 				continue;
2182 			}
2183 			if (rounddown2(pa ^ (pa + ptoa(npages) - 1),
2184 			    boundary) != 0) {
2185 				m_inc = atop(roundup2(pa, boundary) - pa);
2186 				continue;
2187 			}
2188 		} else
2189 			KASSERT(m_run != NULL, ("m_run == NULL"));
2190 
2191 		/*
2192 		 * Avoid releasing and reacquiring the same page lock.
2193 		 */
2194 		new_mtx = vm_page_lockptr(m);
2195 		if (m_mtx != new_mtx) {
2196 			if (m_mtx != NULL)
2197 				mtx_unlock(m_mtx);
2198 			m_mtx = new_mtx;
2199 			mtx_lock(m_mtx);
2200 		}
2201 		m_inc = 1;
2202 retry:
2203 		if (m->wire_count != 0 || m->hold_count != 0)
2204 			run_ext = 0;
2205 #if VM_NRESERVLEVEL > 0
2206 		else if ((level = vm_reserv_level(m)) >= 0 &&
2207 		    (options & VPSC_NORESERV) != 0) {
2208 			run_ext = 0;
2209 			/* Advance to the end of the reservation. */
2210 			pa = VM_PAGE_TO_PHYS(m);
2211 			m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
2212 			    pa);
2213 		}
2214 #endif
2215 		else if ((object = m->object) != NULL) {
2216 			/*
2217 			 * The page is considered eligible for relocation if
2218 			 * and only if it could be laundered or reclaimed by
2219 			 * the page daemon.
2220 			 */
2221 			if (!VM_OBJECT_TRYRLOCK(object)) {
2222 				mtx_unlock(m_mtx);
2223 				VM_OBJECT_RLOCK(object);
2224 				mtx_lock(m_mtx);
2225 				if (m->object != object) {
2226 					/*
2227 					 * The page may have been freed.
2228 					 */
2229 					VM_OBJECT_RUNLOCK(object);
2230 					goto retry;
2231 				} else if (m->wire_count != 0 ||
2232 				    m->hold_count != 0) {
2233 					run_ext = 0;
2234 					goto unlock;
2235 				}
2236 			}
2237 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2238 			    ("page %p is PG_UNHOLDFREE", m));
2239 			/* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */
2240 			if (object->type != OBJT_DEFAULT &&
2241 			    object->type != OBJT_SWAP &&
2242 			    object->type != OBJT_VNODE)
2243 				run_ext = 0;
2244 			else if ((m->flags & PG_CACHED) != 0 ||
2245 			    m != vm_page_lookup(object, m->pindex)) {
2246 				/*
2247 				 * The page is cached or recently converted
2248 				 * from cached to free.
2249 				 */
2250 #if VM_NRESERVLEVEL > 0
2251 				if (level >= 0) {
2252 					/*
2253 					 * The page is reserved.  Extend the
2254 					 * current run by one page.
2255 					 */
2256 					run_ext = 1;
2257 				} else
2258 #endif
2259 				if ((order = m->order) < VM_NFREEORDER) {
2260 					/*
2261 					 * The page is enqueued in the
2262 					 * physical memory allocator's cache/
2263 					 * free page queues.  Moreover, it is
2264 					 * the first page in a power-of-two-
2265 					 * sized run of contiguous cache/free
2266 					 * pages.  Add these pages to the end
2267 					 * of the current run, and jump
2268 					 * ahead.
2269 					 */
2270 					run_ext = 1 << order;
2271 					m_inc = 1 << order;
2272 				} else
2273 					run_ext = 0;
2274 #if VM_NRESERVLEVEL > 0
2275 			} else if ((options & VPSC_NOSUPER) != 0 &&
2276 			    (level = vm_reserv_level_iffullpop(m)) >= 0) {
2277 				run_ext = 0;
2278 				/* Advance to the end of the superpage. */
2279 				pa = VM_PAGE_TO_PHYS(m);
2280 				m_inc = atop(roundup2(pa + 1,
2281 				    vm_reserv_size(level)) - pa);
2282 #endif
2283 			} else if (object->memattr == VM_MEMATTR_DEFAULT &&
2284 			    m->queue != PQ_NONE && !vm_page_busied(m)) {
2285 				/*
2286 				 * The page is allocated but eligible for
2287 				 * relocation.  Extend the current run by one
2288 				 * page.
2289 				 */
2290 				KASSERT(pmap_page_get_memattr(m) ==
2291 				    VM_MEMATTR_DEFAULT,
2292 				    ("page %p has an unexpected memattr", m));
2293 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2294 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2295 				    ("page %p has unexpected oflags", m));
2296 				/* Don't care: VPO_NOSYNC. */
2297 				run_ext = 1;
2298 			} else
2299 				run_ext = 0;
2300 unlock:
2301 			VM_OBJECT_RUNLOCK(object);
2302 #if VM_NRESERVLEVEL > 0
2303 		} else if (level >= 0) {
2304 			/*
2305 			 * The page is reserved but not yet allocated.  In
2306 			 * other words, it is still cached or free.  Extend
2307 			 * the current run by one page.
2308 			 */
2309 			run_ext = 1;
2310 #endif
2311 		} else if ((order = m->order) < VM_NFREEORDER) {
2312 			/*
2313 			 * The page is enqueued in the physical memory
2314 			 * allocator's cache/free page queues.  Moreover, it
2315 			 * is the first page in a power-of-two-sized run of
2316 			 * contiguous cache/free pages.  Add these pages to
2317 			 * the end of the current run, and jump ahead.
2318 			 */
2319 			run_ext = 1 << order;
2320 			m_inc = 1 << order;
2321 		} else {
2322 			/*
2323 			 * Skip the page for one of the following reasons: (1)
2324 			 * It is enqueued in the physical memory allocator's
2325 			 * cache/free page queues.  However, it is not the
2326 			 * first page in a run of contiguous cache/free pages.
2327 			 * (This case rarely occurs because the scan is
2328 			 * performed in ascending order.) (2) It is not
2329 			 * reserved, and it is transitioning from free to
2330 			 * allocated.  (Conversely, the transition from
2331 			 * allocated to free for managed pages is blocked by
2332 			 * the page lock.) (3) It is allocated but not
2333 			 * contained by an object and not wired, e.g.,
2334 			 * allocated by Xen's balloon driver.
2335 			 */
2336 			run_ext = 0;
2337 		}
2338 
2339 		/*
2340 		 * Extend or reset the current run of pages.
2341 		 */
2342 		if (run_ext > 0) {
2343 			if (run_len == 0)
2344 				m_run = m;
2345 			run_len += run_ext;
2346 		} else {
2347 			if (run_len > 0) {
2348 				m_run = NULL;
2349 				run_len = 0;
2350 			}
2351 		}
2352 	}
2353 	if (m_mtx != NULL)
2354 		mtx_unlock(m_mtx);
2355 	if (run_len >= npages)
2356 		return (m_run);
2357 	return (NULL);
2358 }
2359 
2360 /*
2361  *	vm_page_reclaim_run:
2362  *
2363  *	Try to relocate each of the allocated virtual pages within the
2364  *	specified run of physical pages to a new physical address.  Free the
2365  *	physical pages underlying the relocated virtual pages.  A virtual page
2366  *	is relocatable if and only if it could be laundered or reclaimed by
2367  *	the page daemon.  Whenever possible, a virtual page is relocated to a
2368  *	physical address above "high".
2369  *
2370  *	Returns 0 if every physical page within the run was already free or
2371  *	just freed by a successful relocation.  Otherwise, returns a non-zero
2372  *	value indicating why the last attempt to relocate a virtual page was
2373  *	unsuccessful.
2374  *
2375  *	"req_class" must be an allocation class.
2376  */
2377 static int
2378 vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
2379     vm_paddr_t high)
2380 {
2381 	struct mtx *m_mtx, *new_mtx;
2382 	struct spglist free;
2383 	vm_object_t object;
2384 	vm_paddr_t pa;
2385 	vm_page_t m, m_end, m_new;
2386 	int error, order, req;
2387 
2388 	KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
2389 	    ("req_class is not an allocation class"));
2390 	SLIST_INIT(&free);
2391 	error = 0;
2392 	m = m_run;
2393 	m_end = m_run + npages;
2394 	m_mtx = NULL;
2395 	for (; error == 0 && m < m_end; m++) {
2396 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2397 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2398 
2399 		/*
2400 		 * Avoid releasing and reacquiring the same page lock.
2401 		 */
2402 		new_mtx = vm_page_lockptr(m);
2403 		if (m_mtx != new_mtx) {
2404 			if (m_mtx != NULL)
2405 				mtx_unlock(m_mtx);
2406 			m_mtx = new_mtx;
2407 			mtx_lock(m_mtx);
2408 		}
2409 retry:
2410 		if (m->wire_count != 0 || m->hold_count != 0)
2411 			error = EBUSY;
2412 		else if ((object = m->object) != NULL) {
2413 			/*
2414 			 * The page is relocated if and only if it could be
2415 			 * laundered or reclaimed by the page daemon.
2416 			 */
2417 			if (!VM_OBJECT_TRYWLOCK(object)) {
2418 				mtx_unlock(m_mtx);
2419 				VM_OBJECT_WLOCK(object);
2420 				mtx_lock(m_mtx);
2421 				if (m->object != object) {
2422 					/*
2423 					 * The page may have been freed.
2424 					 */
2425 					VM_OBJECT_WUNLOCK(object);
2426 					goto retry;
2427 				} else if (m->wire_count != 0 ||
2428 				    m->hold_count != 0) {
2429 					error = EBUSY;
2430 					goto unlock;
2431 				}
2432 			}
2433 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2434 			    ("page %p is PG_UNHOLDFREE", m));
2435 			/* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */
2436 			if (object->type != OBJT_DEFAULT &&
2437 			    object->type != OBJT_SWAP &&
2438 			    object->type != OBJT_VNODE)
2439 				error = EINVAL;
2440 			else if ((m->flags & PG_CACHED) != 0 ||
2441 			    m != vm_page_lookup(object, m->pindex)) {
2442 				/*
2443 				 * The page is cached or recently converted
2444 				 * from cached to free.
2445 				 */
2446 				VM_OBJECT_WUNLOCK(object);
2447 				goto cached;
2448 			} else if (object->memattr != VM_MEMATTR_DEFAULT)
2449 				error = EINVAL;
2450 			else if (m->queue != PQ_NONE && !vm_page_busied(m)) {
2451 				KASSERT(pmap_page_get_memattr(m) ==
2452 				    VM_MEMATTR_DEFAULT,
2453 				    ("page %p has an unexpected memattr", m));
2454 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2455 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2456 				    ("page %p has unexpected oflags", m));
2457 				/* Don't care: VPO_NOSYNC. */
2458 				if (m->valid != 0) {
2459 					/*
2460 					 * First, try to allocate a new page
2461 					 * that is above "high".  Failing
2462 					 * that, try to allocate a new page
2463 					 * that is below "m_run".  Allocate
2464 					 * the new page between the end of
2465 					 * "m_run" and "high" only as a last
2466 					 * resort.
2467 					 */
2468 					req = req_class | VM_ALLOC_NOOBJ;
2469 					if ((m->flags & PG_NODUMP) != 0)
2470 						req |= VM_ALLOC_NODUMP;
2471 					if (trunc_page(high) !=
2472 					    ~(vm_paddr_t)PAGE_MASK) {
2473 						m_new = vm_page_alloc_contig(
2474 						    NULL, 0, req, 1,
2475 						    round_page(high),
2476 						    ~(vm_paddr_t)0,
2477 						    PAGE_SIZE, 0,
2478 						    VM_MEMATTR_DEFAULT);
2479 					} else
2480 						m_new = NULL;
2481 					if (m_new == NULL) {
2482 						pa = VM_PAGE_TO_PHYS(m_run);
2483 						m_new = vm_page_alloc_contig(
2484 						    NULL, 0, req, 1,
2485 						    0, pa - 1, PAGE_SIZE, 0,
2486 						    VM_MEMATTR_DEFAULT);
2487 					}
2488 					if (m_new == NULL) {
2489 						pa += ptoa(npages);
2490 						m_new = vm_page_alloc_contig(
2491 						    NULL, 0, req, 1,
2492 						    pa, high, PAGE_SIZE, 0,
2493 						    VM_MEMATTR_DEFAULT);
2494 					}
2495 					if (m_new == NULL) {
2496 						error = ENOMEM;
2497 						goto unlock;
2498 					}
2499 					KASSERT(m_new->wire_count == 0,
2500 					    ("page %p is wired", m));
2501 
2502 					/*
2503 					 * Replace "m" with the new page.  For
2504 					 * vm_page_replace(), "m" must be busy
2505 					 * and dequeued.  Finally, change "m"
2506 					 * as if vm_page_free() was called.
2507 					 */
2508 					if (object->ref_count != 0)
2509 						pmap_remove_all(m);
2510 					m_new->aflags = m->aflags;
2511 					KASSERT(m_new->oflags == VPO_UNMANAGED,
2512 					    ("page %p is managed", m));
2513 					m_new->oflags = m->oflags & VPO_NOSYNC;
2514 					pmap_copy_page(m, m_new);
2515 					m_new->valid = m->valid;
2516 					m_new->dirty = m->dirty;
2517 					m->flags &= ~PG_ZERO;
2518 					vm_page_xbusy(m);
2519 					vm_page_remque(m);
2520 					vm_page_replace_checked(m_new, object,
2521 					    m->pindex, m);
2522 					m->valid = 0;
2523 					vm_page_undirty(m);
2524 
2525 					/*
2526 					 * The new page must be deactivated
2527 					 * before the object is unlocked.
2528 					 */
2529 					new_mtx = vm_page_lockptr(m_new);
2530 					if (m_mtx != new_mtx) {
2531 						mtx_unlock(m_mtx);
2532 						m_mtx = new_mtx;
2533 						mtx_lock(m_mtx);
2534 					}
2535 					vm_page_deactivate(m_new);
2536 				} else {
2537 					m->flags &= ~PG_ZERO;
2538 					vm_page_remque(m);
2539 					vm_page_remove(m);
2540 					KASSERT(m->dirty == 0,
2541 					    ("page %p is dirty", m));
2542 				}
2543 				SLIST_INSERT_HEAD(&free, m, plinks.s.ss);
2544 			} else
2545 				error = EBUSY;
2546 unlock:
2547 			VM_OBJECT_WUNLOCK(object);
2548 		} else {
2549 cached:
2550 			mtx_lock(&vm_page_queue_free_mtx);
2551 			order = m->order;
2552 			if (order < VM_NFREEORDER) {
2553 				/*
2554 				 * The page is enqueued in the physical memory
2555 				 * allocator's cache/free page queues.
2556 				 * Moreover, it is the first page in a power-
2557 				 * of-two-sized run of contiguous cache/free
2558 				 * pages.  Jump ahead to the last page within
2559 				 * that run, and continue from there.
2560 				 */
2561 				m += (1 << order) - 1;
2562 			}
2563 #if VM_NRESERVLEVEL > 0
2564 			else if (vm_reserv_is_page_free(m))
2565 				order = 0;
2566 #endif
2567 			mtx_unlock(&vm_page_queue_free_mtx);
2568 			if (order == VM_NFREEORDER)
2569 				error = EINVAL;
2570 		}
2571 	}
2572 	if (m_mtx != NULL)
2573 		mtx_unlock(m_mtx);
2574 	if ((m = SLIST_FIRST(&free)) != NULL) {
2575 		mtx_lock(&vm_page_queue_free_mtx);
2576 		do {
2577 			SLIST_REMOVE_HEAD(&free, plinks.s.ss);
2578 			vm_phys_freecnt_adj(m, 1);
2579 #if VM_NRESERVLEVEL > 0
2580 			if (!vm_reserv_free_page(m))
2581 #else
2582 			if (true)
2583 #endif
2584 				vm_phys_free_pages(m, 0);
2585 		} while ((m = SLIST_FIRST(&free)) != NULL);
2586 		vm_page_zero_idle_wakeup();
2587 		vm_page_free_wakeup();
2588 		mtx_unlock(&vm_page_queue_free_mtx);
2589 	}
2590 	return (error);
2591 }
2592 
2593 #define	NRUNS	16
2594 
2595 CTASSERT(powerof2(NRUNS));
2596 
2597 #define	RUN_INDEX(count)	((count) & (NRUNS - 1))
2598 
2599 #define	MIN_RECLAIM	8
2600 
2601 /*
2602  *	vm_page_reclaim_contig:
2603  *
2604  *	Reclaim allocated, contiguous physical memory satisfying the specified
2605  *	conditions by relocating the virtual pages using that physical memory.
2606  *	Returns true if reclamation is successful and false otherwise.  Since
2607  *	relocation requires the allocation of physical pages, reclamation may
2608  *	fail due to a shortage of cache/free pages.  When reclamation fails,
2609  *	callers are expected to perform VM_WAIT before retrying a failed
2610  *	allocation operation, e.g., vm_page_alloc_contig().
2611  *
2612  *	The caller must always specify an allocation class through "req".
2613  *
2614  *	allocation classes:
2615  *	VM_ALLOC_NORMAL		normal process request
2616  *	VM_ALLOC_SYSTEM		system *really* needs a page
2617  *	VM_ALLOC_INTERRUPT	interrupt time request
2618  *
2619  *	The optional allocation flags are ignored.
2620  *
2621  *	"npages" must be greater than zero.  Both "alignment" and "boundary"
2622  *	must be a power of two.
2623  */
2624 bool
2625 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
2626     u_long alignment, vm_paddr_t boundary)
2627 {
2628 	vm_paddr_t curr_low;
2629 	vm_page_t m_run, m_runs[NRUNS];
2630 	u_long count, reclaimed;
2631 	int error, i, options, req_class;
2632 
2633 	KASSERT(npages > 0, ("npages is 0"));
2634 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2635 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2636 	req_class = req & VM_ALLOC_CLASS_MASK;
2637 
2638 	/*
2639 	 * The page daemon is allowed to dig deeper into the free page list.
2640 	 */
2641 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2642 		req_class = VM_ALLOC_SYSTEM;
2643 
2644 	/*
2645 	 * Return if the number of cached and free pages cannot satisfy the
2646 	 * requested allocation.
2647 	 */
2648 	count = vm_cnt.v_free_count + vm_cnt.v_cache_count;
2649 	if (count < npages + vm_cnt.v_free_reserved || (count < npages +
2650 	    vm_cnt.v_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
2651 	    (count < npages && req_class == VM_ALLOC_INTERRUPT))
2652 		return (false);
2653 
2654 	/*
2655 	 * Scan up to three times, relaxing the restrictions ("options") on
2656 	 * the reclamation of reservations and superpages each time.
2657 	 */
2658 	for (options = VPSC_NORESERV;;) {
2659 		/*
2660 		 * Find the highest runs that satisfy the given constraints
2661 		 * and restrictions, and record them in "m_runs".
2662 		 */
2663 		curr_low = low;
2664 		count = 0;
2665 		for (;;) {
2666 			m_run = vm_phys_scan_contig(npages, curr_low, high,
2667 			    alignment, boundary, options);
2668 			if (m_run == NULL)
2669 				break;
2670 			curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages);
2671 			m_runs[RUN_INDEX(count)] = m_run;
2672 			count++;
2673 		}
2674 
2675 		/*
2676 		 * Reclaim the highest runs in LIFO (descending) order until
2677 		 * the number of reclaimed pages, "reclaimed", is at least
2678 		 * MIN_RECLAIM.  Reset "reclaimed" each time because each
2679 		 * reclamation is idempotent, and runs will (likely) recur
2680 		 * from one scan to the next as restrictions are relaxed.
2681 		 */
2682 		reclaimed = 0;
2683 		for (i = 0; count > 0 && i < NRUNS; i++) {
2684 			count--;
2685 			m_run = m_runs[RUN_INDEX(count)];
2686 			error = vm_page_reclaim_run(req_class, npages, m_run,
2687 			    high);
2688 			if (error == 0) {
2689 				reclaimed += npages;
2690 				if (reclaimed >= MIN_RECLAIM)
2691 					return (true);
2692 			}
2693 		}
2694 
2695 		/*
2696 		 * Either relax the restrictions on the next scan or return if
2697 		 * the last scan had no restrictions.
2698 		 */
2699 		if (options == VPSC_NORESERV)
2700 			options = VPSC_NOSUPER;
2701 		else if (options == VPSC_NOSUPER)
2702 			options = VPSC_ANY;
2703 		else if (options == VPSC_ANY)
2704 			return (reclaimed != 0);
2705 	}
2706 }
2707 
2708 /*
2709  *	vm_wait:	(also see VM_WAIT macro)
2710  *
2711  *	Sleep until free pages are available for allocation.
2712  *	- Called in various places before memory allocations.
2713  */
2714 void
2715 vm_wait(void)
2716 {
2717 
2718 	mtx_lock(&vm_page_queue_free_mtx);
2719 	if (curproc == pageproc) {
2720 		vm_pageout_pages_needed = 1;
2721 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
2722 		    PDROP | PSWP, "VMWait", 0);
2723 	} else {
2724 		if (!vm_pageout_wanted) {
2725 			vm_pageout_wanted = true;
2726 			wakeup(&vm_pageout_wanted);
2727 		}
2728 		vm_pages_needed = true;
2729 		msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
2730 		    "vmwait", 0);
2731 	}
2732 }
2733 
2734 /*
2735  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
2736  *
2737  *	Sleep until free pages are available for allocation.
2738  *	- Called only in vm_fault so that processes page faulting
2739  *	  can be easily tracked.
2740  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2741  *	  processes will be able to grab memory first.  Do not change
2742  *	  this balance without careful testing first.
2743  */
2744 void
2745 vm_waitpfault(void)
2746 {
2747 
2748 	mtx_lock(&vm_page_queue_free_mtx);
2749 	if (!vm_pageout_wanted) {
2750 		vm_pageout_wanted = true;
2751 		wakeup(&vm_pageout_wanted);
2752 	}
2753 	vm_pages_needed = true;
2754 	msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
2755 	    "pfault", 0);
2756 }
2757 
2758 struct vm_pagequeue *
2759 vm_page_pagequeue(vm_page_t m)
2760 {
2761 
2762 	return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2763 }
2764 
2765 /*
2766  *	vm_page_dequeue:
2767  *
2768  *	Remove the given page from its current page queue.
2769  *
2770  *	The page must be locked.
2771  */
2772 void
2773 vm_page_dequeue(vm_page_t m)
2774 {
2775 	struct vm_pagequeue *pq;
2776 
2777 	vm_page_assert_locked(m);
2778 	KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2779 	    m));
2780 	pq = vm_page_pagequeue(m);
2781 	vm_pagequeue_lock(pq);
2782 	m->queue = PQ_NONE;
2783 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2784 	vm_pagequeue_cnt_dec(pq);
2785 	vm_pagequeue_unlock(pq);
2786 }
2787 
2788 /*
2789  *	vm_page_dequeue_locked:
2790  *
2791  *	Remove the given page from its current page queue.
2792  *
2793  *	The page and page queue must be locked.
2794  */
2795 void
2796 vm_page_dequeue_locked(vm_page_t m)
2797 {
2798 	struct vm_pagequeue *pq;
2799 
2800 	vm_page_lock_assert(m, MA_OWNED);
2801 	pq = vm_page_pagequeue(m);
2802 	vm_pagequeue_assert_locked(pq);
2803 	m->queue = PQ_NONE;
2804 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2805 	vm_pagequeue_cnt_dec(pq);
2806 }
2807 
2808 /*
2809  *	vm_page_enqueue:
2810  *
2811  *	Add the given page to the specified page queue.
2812  *
2813  *	The page must be locked.
2814  */
2815 static void
2816 vm_page_enqueue(uint8_t queue, vm_page_t m)
2817 {
2818 	struct vm_pagequeue *pq;
2819 
2820 	vm_page_lock_assert(m, MA_OWNED);
2821 	KASSERT(queue < PQ_COUNT,
2822 	    ("vm_page_enqueue: invalid queue %u request for page %p",
2823 	    queue, m));
2824 	pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2825 	vm_pagequeue_lock(pq);
2826 	m->queue = queue;
2827 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2828 	vm_pagequeue_cnt_inc(pq);
2829 	vm_pagequeue_unlock(pq);
2830 }
2831 
2832 /*
2833  *	vm_page_requeue:
2834  *
2835  *	Move the given page to the tail of its current page queue.
2836  *
2837  *	The page must be locked.
2838  */
2839 void
2840 vm_page_requeue(vm_page_t m)
2841 {
2842 	struct vm_pagequeue *pq;
2843 
2844 	vm_page_lock_assert(m, MA_OWNED);
2845 	KASSERT(m->queue != PQ_NONE,
2846 	    ("vm_page_requeue: page %p is not queued", m));
2847 	pq = vm_page_pagequeue(m);
2848 	vm_pagequeue_lock(pq);
2849 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2850 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2851 	vm_pagequeue_unlock(pq);
2852 }
2853 
2854 /*
2855  *	vm_page_requeue_locked:
2856  *
2857  *	Move the given page to the tail of its current page queue.
2858  *
2859  *	The page queue must be locked.
2860  */
2861 void
2862 vm_page_requeue_locked(vm_page_t m)
2863 {
2864 	struct vm_pagequeue *pq;
2865 
2866 	KASSERT(m->queue != PQ_NONE,
2867 	    ("vm_page_requeue_locked: page %p is not queued", m));
2868 	pq = vm_page_pagequeue(m);
2869 	vm_pagequeue_assert_locked(pq);
2870 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2871 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2872 }
2873 
2874 /*
2875  *	vm_page_activate:
2876  *
2877  *	Put the specified page on the active list (if appropriate).
2878  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2879  *	mess with it.
2880  *
2881  *	The page must be locked.
2882  */
2883 void
2884 vm_page_activate(vm_page_t m)
2885 {
2886 	int queue;
2887 
2888 	vm_page_lock_assert(m, MA_OWNED);
2889 	if ((queue = m->queue) != PQ_ACTIVE) {
2890 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2891 			if (m->act_count < ACT_INIT)
2892 				m->act_count = ACT_INIT;
2893 			if (queue != PQ_NONE)
2894 				vm_page_dequeue(m);
2895 			vm_page_enqueue(PQ_ACTIVE, m);
2896 		} else
2897 			KASSERT(queue == PQ_NONE,
2898 			    ("vm_page_activate: wired page %p is queued", m));
2899 	} else {
2900 		if (m->act_count < ACT_INIT)
2901 			m->act_count = ACT_INIT;
2902 	}
2903 }
2904 
2905 /*
2906  *	vm_page_free_wakeup:
2907  *
2908  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
2909  *	routine is called when a page has been added to the cache or free
2910  *	queues.
2911  *
2912  *	The page queues must be locked.
2913  */
2914 static inline void
2915 vm_page_free_wakeup(void)
2916 {
2917 
2918 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2919 	/*
2920 	 * if pageout daemon needs pages, then tell it that there are
2921 	 * some free.
2922 	 */
2923 	if (vm_pageout_pages_needed &&
2924 	    vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) {
2925 		wakeup(&vm_pageout_pages_needed);
2926 		vm_pageout_pages_needed = 0;
2927 	}
2928 	/*
2929 	 * wakeup processes that are waiting on memory if we hit a
2930 	 * high water mark. And wakeup scheduler process if we have
2931 	 * lots of memory. this process will swapin processes.
2932 	 */
2933 	if (vm_pages_needed && !vm_page_count_min()) {
2934 		vm_pages_needed = false;
2935 		wakeup(&vm_cnt.v_free_count);
2936 	}
2937 }
2938 
2939 /*
2940  *	Turn a cached page into a free page, by changing its attributes.
2941  *	Keep the statistics up-to-date.
2942  *
2943  *	The free page queue must be locked.
2944  */
2945 static void
2946 vm_page_cache_turn_free(vm_page_t m)
2947 {
2948 
2949 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2950 
2951 	m->object = NULL;
2952 	m->valid = 0;
2953 	KASSERT((m->flags & PG_CACHED) != 0,
2954 	    ("vm_page_cache_turn_free: page %p is not cached", m));
2955 	m->flags &= ~PG_CACHED;
2956 	vm_cnt.v_cache_count--;
2957 	vm_phys_freecnt_adj(m, 1);
2958 }
2959 
2960 /*
2961  *	vm_page_free_toq:
2962  *
2963  *	Returns the given page to the free list,
2964  *	disassociating it with any VM object.
2965  *
2966  *	The object must be locked.  The page must be locked if it is managed.
2967  */
2968 void
2969 vm_page_free_toq(vm_page_t m)
2970 {
2971 
2972 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2973 		vm_page_lock_assert(m, MA_OWNED);
2974 		KASSERT(!pmap_page_is_mapped(m),
2975 		    ("vm_page_free_toq: freeing mapped page %p", m));
2976 	} else
2977 		KASSERT(m->queue == PQ_NONE,
2978 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2979 	PCPU_INC(cnt.v_tfree);
2980 
2981 	if (vm_page_sbusied(m))
2982 		panic("vm_page_free: freeing busy page %p", m);
2983 
2984 	/*
2985 	 * Unqueue, then remove page.  Note that we cannot destroy
2986 	 * the page here because we do not want to call the pager's
2987 	 * callback routine until after we've put the page on the
2988 	 * appropriate free queue.
2989 	 */
2990 	vm_page_remque(m);
2991 	vm_page_remove(m);
2992 
2993 	/*
2994 	 * If fictitious remove object association and
2995 	 * return, otherwise delay object association removal.
2996 	 */
2997 	if ((m->flags & PG_FICTITIOUS) != 0) {
2998 		return;
2999 	}
3000 
3001 	m->valid = 0;
3002 	vm_page_undirty(m);
3003 
3004 	if (m->wire_count != 0)
3005 		panic("vm_page_free: freeing wired page %p", m);
3006 	if (m->hold_count != 0) {
3007 		m->flags &= ~PG_ZERO;
3008 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
3009 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
3010 		m->flags |= PG_UNHOLDFREE;
3011 	} else {
3012 		/*
3013 		 * Restore the default memory attribute to the page.
3014 		 */
3015 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
3016 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
3017 
3018 		/*
3019 		 * Insert the page into the physical memory allocator's
3020 		 * cache/free page queues.
3021 		 */
3022 		mtx_lock(&vm_page_queue_free_mtx);
3023 		vm_phys_freecnt_adj(m, 1);
3024 #if VM_NRESERVLEVEL > 0
3025 		if (!vm_reserv_free_page(m))
3026 #else
3027 		if (TRUE)
3028 #endif
3029 			vm_phys_free_pages(m, 0);
3030 		if ((m->flags & PG_ZERO) != 0)
3031 			++vm_page_zero_count;
3032 		else
3033 			vm_page_zero_idle_wakeup();
3034 		vm_page_free_wakeup();
3035 		mtx_unlock(&vm_page_queue_free_mtx);
3036 	}
3037 }
3038 
3039 /*
3040  *	vm_page_wire:
3041  *
3042  *	Mark this page as wired down by yet
3043  *	another map, removing it from paging queues
3044  *	as necessary.
3045  *
3046  *	If the page is fictitious, then its wire count must remain one.
3047  *
3048  *	The page must be locked.
3049  */
3050 void
3051 vm_page_wire(vm_page_t m)
3052 {
3053 
3054 	/*
3055 	 * Only bump the wire statistics if the page is not already wired,
3056 	 * and only unqueue the page if it is on some queue (if it is unmanaged
3057 	 * it is already off the queues).
3058 	 */
3059 	vm_page_lock_assert(m, MA_OWNED);
3060 	if ((m->flags & PG_FICTITIOUS) != 0) {
3061 		KASSERT(m->wire_count == 1,
3062 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
3063 		    m));
3064 		return;
3065 	}
3066 	if (m->wire_count == 0) {
3067 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
3068 		    m->queue == PQ_NONE,
3069 		    ("vm_page_wire: unmanaged page %p is queued", m));
3070 		vm_page_remque(m);
3071 		atomic_add_int(&vm_cnt.v_wire_count, 1);
3072 	}
3073 	m->wire_count++;
3074 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
3075 }
3076 
3077 /*
3078  * vm_page_unwire:
3079  *
3080  * Release one wiring of the specified page, potentially allowing it to be
3081  * paged out.  Returns TRUE if the number of wirings transitions to zero and
3082  * FALSE otherwise.
3083  *
3084  * Only managed pages belonging to an object can be paged out.  If the number
3085  * of wirings transitions to zero and the page is eligible for page out, then
3086  * the page is added to the specified paging queue (unless PQ_NONE is
3087  * specified).
3088  *
3089  * If a page is fictitious, then its wire count must always be one.
3090  *
3091  * A managed page must be locked.
3092  */
3093 boolean_t
3094 vm_page_unwire(vm_page_t m, uint8_t queue)
3095 {
3096 
3097 	KASSERT(queue < PQ_COUNT || queue == PQ_NONE,
3098 	    ("vm_page_unwire: invalid queue %u request for page %p",
3099 	    queue, m));
3100 	if ((m->oflags & VPO_UNMANAGED) == 0)
3101 		vm_page_assert_locked(m);
3102 	if ((m->flags & PG_FICTITIOUS) != 0) {
3103 		KASSERT(m->wire_count == 1,
3104 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
3105 		return (FALSE);
3106 	}
3107 	if (m->wire_count > 0) {
3108 		m->wire_count--;
3109 		if (m->wire_count == 0) {
3110 			atomic_subtract_int(&vm_cnt.v_wire_count, 1);
3111 			if ((m->oflags & VPO_UNMANAGED) == 0 &&
3112 			    m->object != NULL && queue != PQ_NONE) {
3113 				if (queue == PQ_INACTIVE)
3114 					m->flags &= ~PG_WINATCFLS;
3115 				vm_page_enqueue(queue, m);
3116 			}
3117 			return (TRUE);
3118 		} else
3119 			return (FALSE);
3120 	} else
3121 		panic("vm_page_unwire: page %p's wire count is zero", m);
3122 }
3123 
3124 /*
3125  * Move the specified page to the inactive queue.
3126  *
3127  * Many pages placed on the inactive queue should actually go
3128  * into the cache, but it is difficult to figure out which.  What
3129  * we do instead, if the inactive target is well met, is to put
3130  * clean pages at the head of the inactive queue instead of the tail.
3131  * This will cause them to be moved to the cache more quickly and
3132  * if not actively re-referenced, reclaimed more quickly.  If we just
3133  * stick these pages at the end of the inactive queue, heavy filesystem
3134  * meta-data accesses can cause an unnecessary paging load on memory bound
3135  * processes.  This optimization causes one-time-use metadata to be
3136  * reused more quickly.
3137  *
3138  * Normally noreuse is FALSE, resulting in LRU operation.  noreuse is set
3139  * to TRUE if we want this page to be 'as if it were placed in the cache',
3140  * except without unmapping it from the process address space.  In
3141  * practice this is implemented by inserting the page at the head of the
3142  * queue, using a marker page to guide FIFO insertion ordering.
3143  *
3144  * The page must be locked.
3145  */
3146 static inline void
3147 _vm_page_deactivate(vm_page_t m, boolean_t noreuse)
3148 {
3149 	struct vm_pagequeue *pq;
3150 	int queue;
3151 
3152 	vm_page_assert_locked(m);
3153 
3154 	/*
3155 	 * Ignore if the page is already inactive, unless it is unlikely to be
3156 	 * reactivated.
3157 	 */
3158 	if ((queue = m->queue) == PQ_INACTIVE && !noreuse)
3159 		return;
3160 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
3161 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
3162 		/* Avoid multiple acquisitions of the inactive queue lock. */
3163 		if (queue == PQ_INACTIVE) {
3164 			vm_pagequeue_lock(pq);
3165 			vm_page_dequeue_locked(m);
3166 		} else {
3167 			if (queue != PQ_NONE)
3168 				vm_page_dequeue(m);
3169 			m->flags &= ~PG_WINATCFLS;
3170 			vm_pagequeue_lock(pq);
3171 		}
3172 		m->queue = PQ_INACTIVE;
3173 		if (noreuse)
3174 			TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead,
3175 			    m, plinks.q);
3176 		else
3177 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3178 		vm_pagequeue_cnt_inc(pq);
3179 		vm_pagequeue_unlock(pq);
3180 	}
3181 }
3182 
3183 /*
3184  * Move the specified page to the inactive queue.
3185  *
3186  * The page must be locked.
3187  */
3188 void
3189 vm_page_deactivate(vm_page_t m)
3190 {
3191 
3192 	_vm_page_deactivate(m, FALSE);
3193 }
3194 
3195 /*
3196  * Move the specified page to the inactive queue with the expectation
3197  * that it is unlikely to be reused.
3198  *
3199  * The page must be locked.
3200  */
3201 void
3202 vm_page_deactivate_noreuse(vm_page_t m)
3203 {
3204 
3205 	_vm_page_deactivate(m, TRUE);
3206 }
3207 
3208 /*
3209  * vm_page_try_to_cache:
3210  *
3211  * Returns 0 on failure, 1 on success
3212  */
3213 int
3214 vm_page_try_to_cache(vm_page_t m)
3215 {
3216 
3217 	vm_page_lock_assert(m, MA_OWNED);
3218 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3219 	if (m->dirty || m->hold_count || m->wire_count ||
3220 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
3221 		return (0);
3222 	pmap_remove_all(m);
3223 	if (m->dirty)
3224 		return (0);
3225 	vm_page_cache(m);
3226 	return (1);
3227 }
3228 
3229 /*
3230  * vm_page_try_to_free()
3231  *
3232  *	Attempt to free the page.  If we cannot free it, we do nothing.
3233  *	1 is returned on success, 0 on failure.
3234  */
3235 int
3236 vm_page_try_to_free(vm_page_t m)
3237 {
3238 
3239 	vm_page_lock_assert(m, MA_OWNED);
3240 	if (m->object != NULL)
3241 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3242 	if (m->dirty || m->hold_count || m->wire_count ||
3243 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
3244 		return (0);
3245 	pmap_remove_all(m);
3246 	if (m->dirty)
3247 		return (0);
3248 	vm_page_free(m);
3249 	return (1);
3250 }
3251 
3252 /*
3253  * vm_page_cache
3254  *
3255  * Put the specified page onto the page cache queue (if appropriate).
3256  *
3257  * The object and page must be locked.
3258  */
3259 void
3260 vm_page_cache(vm_page_t m)
3261 {
3262 	vm_object_t object;
3263 	boolean_t cache_was_empty;
3264 
3265 	vm_page_lock_assert(m, MA_OWNED);
3266 	object = m->object;
3267 	VM_OBJECT_ASSERT_WLOCKED(object);
3268 	if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) ||
3269 	    m->hold_count || m->wire_count)
3270 		panic("vm_page_cache: attempting to cache busy page");
3271 	KASSERT(!pmap_page_is_mapped(m),
3272 	    ("vm_page_cache: page %p is mapped", m));
3273 	KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m));
3274 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
3275 	    (object->type == OBJT_SWAP &&
3276 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
3277 		/*
3278 		 * Hypothesis: A cache-eligible page belonging to a
3279 		 * default object or swap object but without a backing
3280 		 * store must be zero filled.
3281 		 */
3282 		vm_page_free(m);
3283 		return;
3284 	}
3285 	KASSERT((m->flags & PG_CACHED) == 0,
3286 	    ("vm_page_cache: page %p is already cached", m));
3287 
3288 	/*
3289 	 * Remove the page from the paging queues.
3290 	 */
3291 	vm_page_remque(m);
3292 
3293 	/*
3294 	 * Remove the page from the object's collection of resident
3295 	 * pages.
3296 	 */
3297 	vm_radix_remove(&object->rtree, m->pindex);
3298 	TAILQ_REMOVE(&object->memq, m, listq);
3299 	object->resident_page_count--;
3300 
3301 	/*
3302 	 * Restore the default memory attribute to the page.
3303 	 */
3304 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
3305 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
3306 
3307 	/*
3308 	 * Insert the page into the object's collection of cached pages
3309 	 * and the physical memory allocator's cache/free page queues.
3310 	 */
3311 	m->flags &= ~PG_ZERO;
3312 	mtx_lock(&vm_page_queue_free_mtx);
3313 	cache_was_empty = vm_radix_is_empty(&object->cache);
3314 	if (vm_radix_insert(&object->cache, m)) {
3315 		mtx_unlock(&vm_page_queue_free_mtx);
3316 		if (object->type == OBJT_VNODE &&
3317 		    object->resident_page_count == 0)
3318 			vdrop(object->handle);
3319 		m->object = NULL;
3320 		vm_page_free(m);
3321 		return;
3322 	}
3323 
3324 	/*
3325 	 * The above call to vm_radix_insert() could reclaim the one pre-
3326 	 * existing cached page from this object, resulting in a call to
3327 	 * vdrop().
3328 	 */
3329 	if (!cache_was_empty)
3330 		cache_was_empty = vm_radix_is_singleton(&object->cache);
3331 
3332 	m->flags |= PG_CACHED;
3333 	vm_cnt.v_cache_count++;
3334 	PCPU_INC(cnt.v_tcached);
3335 #if VM_NRESERVLEVEL > 0
3336 	if (!vm_reserv_free_page(m)) {
3337 #else
3338 	if (TRUE) {
3339 #endif
3340 		vm_phys_free_pages(m, 0);
3341 	}
3342 	vm_page_free_wakeup();
3343 	mtx_unlock(&vm_page_queue_free_mtx);
3344 
3345 	/*
3346 	 * Increment the vnode's hold count if this is the object's only
3347 	 * cached page.  Decrement the vnode's hold count if this was
3348 	 * the object's only resident page.
3349 	 */
3350 	if (object->type == OBJT_VNODE) {
3351 		if (cache_was_empty && object->resident_page_count != 0)
3352 			vhold(object->handle);
3353 		else if (!cache_was_empty && object->resident_page_count == 0)
3354 			vdrop(object->handle);
3355 	}
3356 }
3357 
3358 /*
3359  * vm_page_advise
3360  *
3361  * 	Deactivate or do nothing, as appropriate.
3362  *
3363  *	The object and page must be locked.
3364  */
3365 void
3366 vm_page_advise(vm_page_t m, int advice)
3367 {
3368 
3369 	vm_page_assert_locked(m);
3370 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3371 	if (advice == MADV_FREE)
3372 		/*
3373 		 * Mark the page clean.  This will allow the page to be freed
3374 		 * up by the system.  However, such pages are often reused
3375 		 * quickly by malloc() so we do not do anything that would
3376 		 * cause a page fault if we can help it.
3377 		 *
3378 		 * Specifically, we do not try to actually free the page now
3379 		 * nor do we try to put it in the cache (which would cause a
3380 		 * page fault on reuse).
3381 		 *
3382 		 * But we do make the page as freeable as we can without
3383 		 * actually taking the step of unmapping it.
3384 		 */
3385 		m->dirty = 0;
3386 	else if (advice != MADV_DONTNEED)
3387 		return;
3388 
3389 	/*
3390 	 * Clear any references to the page.  Otherwise, the page daemon will
3391 	 * immediately reactivate the page.
3392 	 */
3393 	vm_page_aflag_clear(m, PGA_REFERENCED);
3394 
3395 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
3396 		vm_page_dirty(m);
3397 
3398 	/*
3399 	 * Place clean pages at the head of the inactive queue rather than the
3400 	 * tail, thus defeating the queue's LRU operation and ensuring that the
3401 	 * page will be reused quickly.
3402 	 */
3403 	_vm_page_deactivate(m, m->dirty == 0);
3404 }
3405 
3406 /*
3407  * Grab a page, waiting until we are waken up due to the page
3408  * changing state.  We keep on waiting, if the page continues
3409  * to be in the object.  If the page doesn't exist, first allocate it
3410  * and then conditionally zero it.
3411  *
3412  * This routine may sleep.
3413  *
3414  * The object must be locked on entry.  The lock will, however, be released
3415  * and reacquired if the routine sleeps.
3416  */
3417 vm_page_t
3418 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
3419 {
3420 	vm_page_t m;
3421 	int sleep;
3422 
3423 	VM_OBJECT_ASSERT_WLOCKED(object);
3424 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3425 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3426 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
3427 retrylookup:
3428 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
3429 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3430 		    vm_page_xbusied(m) : vm_page_busied(m);
3431 		if (sleep) {
3432 			if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3433 				return (NULL);
3434 			/*
3435 			 * Reference the page before unlocking and
3436 			 * sleeping so that the page daemon is less
3437 			 * likely to reclaim it.
3438 			 */
3439 			vm_page_aflag_set(m, PGA_REFERENCED);
3440 			vm_page_lock(m);
3441 			VM_OBJECT_WUNLOCK(object);
3442 			vm_page_busy_sleep(m, "pgrbwt");
3443 			VM_OBJECT_WLOCK(object);
3444 			goto retrylookup;
3445 		} else {
3446 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
3447 				vm_page_lock(m);
3448 				vm_page_wire(m);
3449 				vm_page_unlock(m);
3450 			}
3451 			if ((allocflags &
3452 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
3453 				vm_page_xbusy(m);
3454 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
3455 				vm_page_sbusy(m);
3456 			return (m);
3457 		}
3458 	}
3459 	m = vm_page_alloc(object, pindex, allocflags);
3460 	if (m == NULL) {
3461 		if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3462 			return (NULL);
3463 		VM_OBJECT_WUNLOCK(object);
3464 		VM_WAIT;
3465 		VM_OBJECT_WLOCK(object);
3466 		goto retrylookup;
3467 	} else if (m->valid != 0)
3468 		return (m);
3469 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
3470 		pmap_zero_page(m);
3471 	return (m);
3472 }
3473 
3474 /*
3475  * Mapping function for valid or dirty bits in a page.
3476  *
3477  * Inputs are required to range within a page.
3478  */
3479 vm_page_bits_t
3480 vm_page_bits(int base, int size)
3481 {
3482 	int first_bit;
3483 	int last_bit;
3484 
3485 	KASSERT(
3486 	    base + size <= PAGE_SIZE,
3487 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
3488 	);
3489 
3490 	if (size == 0)		/* handle degenerate case */
3491 		return (0);
3492 
3493 	first_bit = base >> DEV_BSHIFT;
3494 	last_bit = (base + size - 1) >> DEV_BSHIFT;
3495 
3496 	return (((vm_page_bits_t)2 << last_bit) -
3497 	    ((vm_page_bits_t)1 << first_bit));
3498 }
3499 
3500 /*
3501  *	vm_page_set_valid_range:
3502  *
3503  *	Sets portions of a page valid.  The arguments are expected
3504  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3505  *	of any partial chunks touched by the range.  The invalid portion of
3506  *	such chunks will be zeroed.
3507  *
3508  *	(base + size) must be less then or equal to PAGE_SIZE.
3509  */
3510 void
3511 vm_page_set_valid_range(vm_page_t m, int base, int size)
3512 {
3513 	int endoff, frag;
3514 
3515 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3516 	if (size == 0)	/* handle degenerate case */
3517 		return;
3518 
3519 	/*
3520 	 * If the base is not DEV_BSIZE aligned and the valid
3521 	 * bit is clear, we have to zero out a portion of the
3522 	 * first block.
3523 	 */
3524 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3525 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
3526 		pmap_zero_page_area(m, frag, base - frag);
3527 
3528 	/*
3529 	 * If the ending offset is not DEV_BSIZE aligned and the
3530 	 * valid bit is clear, we have to zero out a portion of
3531 	 * the last block.
3532 	 */
3533 	endoff = base + size;
3534 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3535 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
3536 		pmap_zero_page_area(m, endoff,
3537 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3538 
3539 	/*
3540 	 * Assert that no previously invalid block that is now being validated
3541 	 * is already dirty.
3542 	 */
3543 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
3544 	    ("vm_page_set_valid_range: page %p is dirty", m));
3545 
3546 	/*
3547 	 * Set valid bits inclusive of any overlap.
3548 	 */
3549 	m->valid |= vm_page_bits(base, size);
3550 }
3551 
3552 /*
3553  * Clear the given bits from the specified page's dirty field.
3554  */
3555 static __inline void
3556 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
3557 {
3558 	uintptr_t addr;
3559 #if PAGE_SIZE < 16384
3560 	int shift;
3561 #endif
3562 
3563 	/*
3564 	 * If the object is locked and the page is neither exclusive busy nor
3565 	 * write mapped, then the page's dirty field cannot possibly be
3566 	 * set by a concurrent pmap operation.
3567 	 */
3568 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3569 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
3570 		m->dirty &= ~pagebits;
3571 	else {
3572 		/*
3573 		 * The pmap layer can call vm_page_dirty() without
3574 		 * holding a distinguished lock.  The combination of
3575 		 * the object's lock and an atomic operation suffice
3576 		 * to guarantee consistency of the page dirty field.
3577 		 *
3578 		 * For PAGE_SIZE == 32768 case, compiler already
3579 		 * properly aligns the dirty field, so no forcible
3580 		 * alignment is needed. Only require existence of
3581 		 * atomic_clear_64 when page size is 32768.
3582 		 */
3583 		addr = (uintptr_t)&m->dirty;
3584 #if PAGE_SIZE == 32768
3585 		atomic_clear_64((uint64_t *)addr, pagebits);
3586 #elif PAGE_SIZE == 16384
3587 		atomic_clear_32((uint32_t *)addr, pagebits);
3588 #else		/* PAGE_SIZE <= 8192 */
3589 		/*
3590 		 * Use a trick to perform a 32-bit atomic on the
3591 		 * containing aligned word, to not depend on the existence
3592 		 * of atomic_clear_{8, 16}.
3593 		 */
3594 		shift = addr & (sizeof(uint32_t) - 1);
3595 #if BYTE_ORDER == BIG_ENDIAN
3596 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
3597 #else
3598 		shift *= NBBY;
3599 #endif
3600 		addr &= ~(sizeof(uint32_t) - 1);
3601 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
3602 #endif		/* PAGE_SIZE */
3603 	}
3604 }
3605 
3606 /*
3607  *	vm_page_set_validclean:
3608  *
3609  *	Sets portions of a page valid and clean.  The arguments are expected
3610  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3611  *	of any partial chunks touched by the range.  The invalid portion of
3612  *	such chunks will be zero'd.
3613  *
3614  *	(base + size) must be less then or equal to PAGE_SIZE.
3615  */
3616 void
3617 vm_page_set_validclean(vm_page_t m, int base, int size)
3618 {
3619 	vm_page_bits_t oldvalid, pagebits;
3620 	int endoff, frag;
3621 
3622 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3623 	if (size == 0)	/* handle degenerate case */
3624 		return;
3625 
3626 	/*
3627 	 * If the base is not DEV_BSIZE aligned and the valid
3628 	 * bit is clear, we have to zero out a portion of the
3629 	 * first block.
3630 	 */
3631 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3632 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3633 		pmap_zero_page_area(m, frag, base - frag);
3634 
3635 	/*
3636 	 * If the ending offset is not DEV_BSIZE aligned and the
3637 	 * valid bit is clear, we have to zero out a portion of
3638 	 * the last block.
3639 	 */
3640 	endoff = base + size;
3641 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3642 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3643 		pmap_zero_page_area(m, endoff,
3644 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3645 
3646 	/*
3647 	 * Set valid, clear dirty bits.  If validating the entire
3648 	 * page we can safely clear the pmap modify bit.  We also
3649 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
3650 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
3651 	 * be set again.
3652 	 *
3653 	 * We set valid bits inclusive of any overlap, but we can only
3654 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
3655 	 * the range.
3656 	 */
3657 	oldvalid = m->valid;
3658 	pagebits = vm_page_bits(base, size);
3659 	m->valid |= pagebits;
3660 #if 0	/* NOT YET */
3661 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3662 		frag = DEV_BSIZE - frag;
3663 		base += frag;
3664 		size -= frag;
3665 		if (size < 0)
3666 			size = 0;
3667 	}
3668 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3669 #endif
3670 	if (base == 0 && size == PAGE_SIZE) {
3671 		/*
3672 		 * The page can only be modified within the pmap if it is
3673 		 * mapped, and it can only be mapped if it was previously
3674 		 * fully valid.
3675 		 */
3676 		if (oldvalid == VM_PAGE_BITS_ALL)
3677 			/*
3678 			 * Perform the pmap_clear_modify() first.  Otherwise,
3679 			 * a concurrent pmap operation, such as
3680 			 * pmap_protect(), could clear a modification in the
3681 			 * pmap and set the dirty field on the page before
3682 			 * pmap_clear_modify() had begun and after the dirty
3683 			 * field was cleared here.
3684 			 */
3685 			pmap_clear_modify(m);
3686 		m->dirty = 0;
3687 		m->oflags &= ~VPO_NOSYNC;
3688 	} else if (oldvalid != VM_PAGE_BITS_ALL)
3689 		m->dirty &= ~pagebits;
3690 	else
3691 		vm_page_clear_dirty_mask(m, pagebits);
3692 }
3693 
3694 void
3695 vm_page_clear_dirty(vm_page_t m, int base, int size)
3696 {
3697 
3698 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3699 }
3700 
3701 /*
3702  *	vm_page_set_invalid:
3703  *
3704  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
3705  *	valid and dirty bits for the effected areas are cleared.
3706  */
3707 void
3708 vm_page_set_invalid(vm_page_t m, int base, int size)
3709 {
3710 	vm_page_bits_t bits;
3711 	vm_object_t object;
3712 
3713 	object = m->object;
3714 	VM_OBJECT_ASSERT_WLOCKED(object);
3715 	if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3716 	    size >= object->un_pager.vnp.vnp_size)
3717 		bits = VM_PAGE_BITS_ALL;
3718 	else
3719 		bits = vm_page_bits(base, size);
3720 	if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
3721 	    bits != 0)
3722 		pmap_remove_all(m);
3723 	KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3724 	    !pmap_page_is_mapped(m),
3725 	    ("vm_page_set_invalid: page %p is mapped", m));
3726 	m->valid &= ~bits;
3727 	m->dirty &= ~bits;
3728 }
3729 
3730 /*
3731  * vm_page_zero_invalid()
3732  *
3733  *	The kernel assumes that the invalid portions of a page contain
3734  *	garbage, but such pages can be mapped into memory by user code.
3735  *	When this occurs, we must zero out the non-valid portions of the
3736  *	page so user code sees what it expects.
3737  *
3738  *	Pages are most often semi-valid when the end of a file is mapped
3739  *	into memory and the file's size is not page aligned.
3740  */
3741 void
3742 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3743 {
3744 	int b;
3745 	int i;
3746 
3747 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3748 	/*
3749 	 * Scan the valid bits looking for invalid sections that
3750 	 * must be zeroed.  Invalid sub-DEV_BSIZE'd areas ( where the
3751 	 * valid bit may be set ) have already been zeroed by
3752 	 * vm_page_set_validclean().
3753 	 */
3754 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3755 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3756 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3757 			if (i > b) {
3758 				pmap_zero_page_area(m,
3759 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3760 			}
3761 			b = i + 1;
3762 		}
3763 	}
3764 
3765 	/*
3766 	 * setvalid is TRUE when we can safely set the zero'd areas
3767 	 * as being valid.  We can do this if there are no cache consistancy
3768 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3769 	 */
3770 	if (setvalid)
3771 		m->valid = VM_PAGE_BITS_ALL;
3772 }
3773 
3774 /*
3775  *	vm_page_is_valid:
3776  *
3777  *	Is (partial) page valid?  Note that the case where size == 0
3778  *	will return FALSE in the degenerate case where the page is
3779  *	entirely invalid, and TRUE otherwise.
3780  */
3781 int
3782 vm_page_is_valid(vm_page_t m, int base, int size)
3783 {
3784 	vm_page_bits_t bits;
3785 
3786 	VM_OBJECT_ASSERT_LOCKED(m->object);
3787 	bits = vm_page_bits(base, size);
3788 	return (m->valid != 0 && (m->valid & bits) == bits);
3789 }
3790 
3791 /*
3792  *	vm_page_ps_is_valid:
3793  *
3794  *	Returns TRUE if the entire (super)page is valid and FALSE otherwise.
3795  */
3796 boolean_t
3797 vm_page_ps_is_valid(vm_page_t m)
3798 {
3799 	int i, npages;
3800 
3801 	VM_OBJECT_ASSERT_LOCKED(m->object);
3802 	npages = atop(pagesizes[m->psind]);
3803 
3804 	/*
3805 	 * The physically contiguous pages that make up a superpage, i.e., a
3806 	 * page with a page size index ("psind") greater than zero, will
3807 	 * occupy adjacent entries in vm_page_array[].
3808 	 */
3809 	for (i = 0; i < npages; i++) {
3810 		if (m[i].valid != VM_PAGE_BITS_ALL)
3811 			return (FALSE);
3812 	}
3813 	return (TRUE);
3814 }
3815 
3816 /*
3817  * Set the page's dirty bits if the page is modified.
3818  */
3819 void
3820 vm_page_test_dirty(vm_page_t m)
3821 {
3822 
3823 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3824 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3825 		vm_page_dirty(m);
3826 }
3827 
3828 void
3829 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3830 {
3831 
3832 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3833 }
3834 
3835 void
3836 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3837 {
3838 
3839 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3840 }
3841 
3842 int
3843 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3844 {
3845 
3846 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3847 }
3848 
3849 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3850 void
3851 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3852 {
3853 
3854 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3855 }
3856 
3857 void
3858 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3859 {
3860 
3861 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3862 }
3863 #endif
3864 
3865 #ifdef INVARIANTS
3866 void
3867 vm_page_object_lock_assert(vm_page_t m)
3868 {
3869 
3870 	/*
3871 	 * Certain of the page's fields may only be modified by the
3872 	 * holder of the containing object's lock or the exclusive busy.
3873 	 * holder.  Unfortunately, the holder of the write busy is
3874 	 * not recorded, and thus cannot be checked here.
3875 	 */
3876 	if (m->object != NULL && !vm_page_xbusied(m))
3877 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3878 }
3879 
3880 void
3881 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
3882 {
3883 
3884 	if ((bits & PGA_WRITEABLE) == 0)
3885 		return;
3886 
3887 	/*
3888 	 * The PGA_WRITEABLE flag can only be set if the page is
3889 	 * managed, is exclusively busied or the object is locked.
3890 	 * Currently, this flag is only set by pmap_enter().
3891 	 */
3892 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3893 	    ("PGA_WRITEABLE on unmanaged page"));
3894 	if (!vm_page_xbusied(m))
3895 		VM_OBJECT_ASSERT_LOCKED(m->object);
3896 }
3897 #endif
3898 
3899 #include "opt_ddb.h"
3900 #ifdef DDB
3901 #include <sys/kernel.h>
3902 
3903 #include <ddb/ddb.h>
3904 
3905 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3906 {
3907 	db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count);
3908 	db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count);
3909 	db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count);
3910 	db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count);
3911 	db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count);
3912 	db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
3913 	db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
3914 	db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
3915 	db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
3916 }
3917 
3918 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3919 {
3920 	int dom;
3921 
3922 	db_printf("pq_free %d pq_cache %d\n",
3923 	    vm_cnt.v_free_count, vm_cnt.v_cache_count);
3924 	for (dom = 0; dom < vm_ndomains; dom++) {
3925 		db_printf(
3926 	"dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n",
3927 		    dom,
3928 		    vm_dom[dom].vmd_page_count,
3929 		    vm_dom[dom].vmd_free_count,
3930 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3931 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3932 		    vm_dom[dom].vmd_pass);
3933 	}
3934 }
3935 
3936 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3937 {
3938 	vm_page_t m;
3939 	boolean_t phys;
3940 
3941 	if (!have_addr) {
3942 		db_printf("show pginfo addr\n");
3943 		return;
3944 	}
3945 
3946 	phys = strchr(modif, 'p') != NULL;
3947 	if (phys)
3948 		m = PHYS_TO_VM_PAGE(addr);
3949 	else
3950 		m = (vm_page_t)addr;
3951 	db_printf(
3952     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3953     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3954 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3955 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3956 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3957 }
3958 #endif /* DDB */
3959