xref: /freebsd/sys/vm/vm_page.c (revision 1ce1c6895245648ba022f7187df1626904dc1f89)
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_init_fakepg(void *dummy);
162 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
163     vm_pindex_t pindex, vm_page_t mpred);
164 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
165     vm_page_t mpred);
166 
167 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL);
168 
169 static void
170 vm_page_init_fakepg(void *dummy)
171 {
172 
173 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
174 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
175 }
176 
177 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
178 #if PAGE_SIZE == 32768
179 #ifdef CTASSERT
180 CTASSERT(sizeof(u_long) >= 8);
181 #endif
182 #endif
183 
184 /*
185  * Try to acquire a physical address lock while a pmap is locked.  If we
186  * fail to trylock we unlock and lock the pmap directly and cache the
187  * locked pa in *locked.  The caller should then restart their loop in case
188  * the virtual to physical mapping has changed.
189  */
190 int
191 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
192 {
193 	vm_paddr_t lockpa;
194 
195 	lockpa = *locked;
196 	*locked = pa;
197 	if (lockpa) {
198 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
199 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
200 			return (0);
201 		PA_UNLOCK(lockpa);
202 	}
203 	if (PA_TRYLOCK(pa))
204 		return (0);
205 	PMAP_UNLOCK(pmap);
206 	atomic_add_int(&pa_tryrelock_restart, 1);
207 	PA_LOCK(pa);
208 	PMAP_LOCK(pmap);
209 	return (EAGAIN);
210 }
211 
212 /*
213  *	vm_set_page_size:
214  *
215  *	Sets the page size, perhaps based upon the memory
216  *	size.  Must be called before any use of page-size
217  *	dependent functions.
218  */
219 void
220 vm_set_page_size(void)
221 {
222 	if (vm_cnt.v_page_size == 0)
223 		vm_cnt.v_page_size = PAGE_SIZE;
224 	if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
225 		panic("vm_set_page_size: page size not a power of two");
226 }
227 
228 /*
229  *	vm_page_blacklist_next:
230  *
231  *	Find the next entry in the provided string of blacklist
232  *	addresses.  Entries are separated by space, comma, or newline.
233  *	If an invalid integer is encountered then the rest of the
234  *	string is skipped.  Updates the list pointer to the next
235  *	character, or NULL if the string is exhausted or invalid.
236  */
237 static vm_paddr_t
238 vm_page_blacklist_next(char **list, char *end)
239 {
240 	vm_paddr_t bad;
241 	char *cp, *pos;
242 
243 	if (list == NULL || *list == NULL)
244 		return (0);
245 	if (**list =='\0') {
246 		*list = NULL;
247 		return (0);
248 	}
249 
250 	/*
251 	 * If there's no end pointer then the buffer is coming from
252 	 * the kenv and we know it's null-terminated.
253 	 */
254 	if (end == NULL)
255 		end = *list + strlen(*list);
256 
257 	/* Ensure that strtoq() won't walk off the end */
258 	if (*end != '\0') {
259 		if (*end == '\n' || *end == ' ' || *end  == ',')
260 			*end = '\0';
261 		else {
262 			printf("Blacklist not terminated, skipping\n");
263 			*list = NULL;
264 			return (0);
265 		}
266 	}
267 
268 	for (pos = *list; *pos != '\0'; pos = cp) {
269 		bad = strtoq(pos, &cp, 0);
270 		if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
271 			if (bad == 0) {
272 				if (++cp < end)
273 					continue;
274 				else
275 					break;
276 			}
277 		} else
278 			break;
279 		if (*cp == '\0' || ++cp >= end)
280 			*list = NULL;
281 		else
282 			*list = cp;
283 		return (trunc_page(bad));
284 	}
285 	printf("Garbage in RAM blacklist, skipping\n");
286 	*list = NULL;
287 	return (0);
288 }
289 
290 /*
291  *	vm_page_blacklist_check:
292  *
293  *	Iterate through the provided string of blacklist addresses, pulling
294  *	each entry out of the physical allocator free list and putting it
295  *	onto a list for reporting via the vm.page_blacklist sysctl.
296  */
297 static void
298 vm_page_blacklist_check(char *list, char *end)
299 {
300 	vm_paddr_t pa;
301 	vm_page_t m;
302 	char *next;
303 	int ret;
304 
305 	next = list;
306 	while (next != NULL) {
307 		if ((pa = vm_page_blacklist_next(&next, end)) == 0)
308 			continue;
309 		m = vm_phys_paddr_to_vm_page(pa);
310 		if (m == NULL)
311 			continue;
312 		mtx_lock(&vm_page_queue_free_mtx);
313 		ret = vm_phys_unfree_page(m);
314 		mtx_unlock(&vm_page_queue_free_mtx);
315 		if (ret == TRUE) {
316 			TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
317 			if (bootverbose)
318 				printf("Skipping page with pa 0x%jx\n",
319 				    (uintmax_t)pa);
320 		}
321 	}
322 }
323 
324 /*
325  *	vm_page_blacklist_load:
326  *
327  *	Search for a special module named "ram_blacklist".  It'll be a
328  *	plain text file provided by the user via the loader directive
329  *	of the same name.
330  */
331 static void
332 vm_page_blacklist_load(char **list, char **end)
333 {
334 	void *mod;
335 	u_char *ptr;
336 	u_int len;
337 
338 	mod = NULL;
339 	ptr = NULL;
340 
341 	mod = preload_search_by_type("ram_blacklist");
342 	if (mod != NULL) {
343 		ptr = preload_fetch_addr(mod);
344 		len = preload_fetch_size(mod);
345         }
346 	*list = ptr;
347 	if (ptr != NULL)
348 		*end = ptr + len;
349 	else
350 		*end = NULL;
351 	return;
352 }
353 
354 static int
355 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
356 {
357 	vm_page_t m;
358 	struct sbuf sbuf;
359 	int error, first;
360 
361 	first = 1;
362 	error = sysctl_wire_old_buffer(req, 0);
363 	if (error != 0)
364 		return (error);
365 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
366 	TAILQ_FOREACH(m, &blacklist_head, listq) {
367 		sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
368 		    (uintmax_t)m->phys_addr);
369 		first = 0;
370 	}
371 	error = sbuf_finish(&sbuf);
372 	sbuf_delete(&sbuf);
373 	return (error);
374 }
375 
376 static void
377 vm_page_domain_init(struct vm_domain *vmd)
378 {
379 	struct vm_pagequeue *pq;
380 	int i;
381 
382 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
383 	    "vm inactive pagequeue";
384 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
385 	    &vm_cnt.v_inactive_count;
386 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
387 	    "vm active pagequeue";
388 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
389 	    &vm_cnt.v_active_count;
390 	vmd->vmd_page_count = 0;
391 	vmd->vmd_free_count = 0;
392 	vmd->vmd_segs = 0;
393 	vmd->vmd_oom = FALSE;
394 	vmd->vmd_pass = 0;
395 	for (i = 0; i < PQ_COUNT; i++) {
396 		pq = &vmd->vmd_pagequeues[i];
397 		TAILQ_INIT(&pq->pq_pl);
398 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
399 		    MTX_DEF | MTX_DUPOK);
400 	}
401 }
402 
403 /*
404  *	vm_page_startup:
405  *
406  *	Initializes the resident memory module.
407  *
408  *	Allocates memory for the page cells, and
409  *	for the object/offset-to-page hash table headers.
410  *	Each page cell is initialized and placed on the free list.
411  */
412 vm_offset_t
413 vm_page_startup(vm_offset_t vaddr)
414 {
415 	vm_offset_t mapped;
416 	vm_paddr_t page_range;
417 	vm_paddr_t new_end;
418 	int i;
419 	vm_paddr_t pa;
420 	vm_paddr_t last_pa;
421 	char *list, *listend;
422 	vm_paddr_t end;
423 	vm_paddr_t biggestsize;
424 	vm_paddr_t low_water, high_water;
425 	int biggestone;
426 
427 	biggestsize = 0;
428 	biggestone = 0;
429 	vaddr = round_page(vaddr);
430 
431 	for (i = 0; phys_avail[i + 1]; i += 2) {
432 		phys_avail[i] = round_page(phys_avail[i]);
433 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
434 	}
435 
436 	low_water = phys_avail[0];
437 	high_water = phys_avail[1];
438 
439 	for (i = 0; i < vm_phys_nsegs; i++) {
440 		if (vm_phys_segs[i].start < low_water)
441 			low_water = vm_phys_segs[i].start;
442 		if (vm_phys_segs[i].end > high_water)
443 			high_water = vm_phys_segs[i].end;
444 	}
445 	for (i = 0; phys_avail[i + 1]; i += 2) {
446 		vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
447 
448 		if (size > biggestsize) {
449 			biggestone = i;
450 			biggestsize = size;
451 		}
452 		if (phys_avail[i] < low_water)
453 			low_water = phys_avail[i];
454 		if (phys_avail[i + 1] > high_water)
455 			high_water = phys_avail[i + 1];
456 	}
457 
458 	end = phys_avail[biggestone+1];
459 
460 	/*
461 	 * Initialize the page and queue locks.
462 	 */
463 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
464 	for (i = 0; i < PA_LOCK_COUNT; i++)
465 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
466 	for (i = 0; i < vm_ndomains; i++)
467 		vm_page_domain_init(&vm_dom[i]);
468 
469 	/*
470 	 * Allocate memory for use when boot strapping the kernel memory
471 	 * allocator.
472 	 *
473 	 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
474 	 * manually fetch the value.
475 	 */
476 	TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
477 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
478 	new_end = trunc_page(new_end);
479 	mapped = pmap_map(&vaddr, new_end, end,
480 	    VM_PROT_READ | VM_PROT_WRITE);
481 	bzero((void *)mapped, end - new_end);
482 	uma_startup((void *)mapped, boot_pages);
483 
484 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
485     defined(__i386__) || defined(__mips__)
486 	/*
487 	 * Allocate a bitmap to indicate that a random physical page
488 	 * needs to be included in a minidump.
489 	 *
490 	 * The amd64 port needs this to indicate which direct map pages
491 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
492 	 *
493 	 * However, i386 still needs this workspace internally within the
494 	 * minidump code.  In theory, they are not needed on i386, but are
495 	 * included should the sf_buf code decide to use them.
496 	 */
497 	last_pa = 0;
498 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
499 		if (dump_avail[i + 1] > last_pa)
500 			last_pa = dump_avail[i + 1];
501 	page_range = last_pa / PAGE_SIZE;
502 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
503 	new_end -= vm_page_dump_size;
504 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
505 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
506 	bzero((void *)vm_page_dump, vm_page_dump_size);
507 #endif
508 #ifdef __amd64__
509 	/*
510 	 * Request that the physical pages underlying the message buffer be
511 	 * included in a crash dump.  Since the message buffer is accessed
512 	 * through the direct map, they are not automatically included.
513 	 */
514 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
515 	last_pa = pa + round_page(msgbufsize);
516 	while (pa < last_pa) {
517 		dump_add_page(pa);
518 		pa += PAGE_SIZE;
519 	}
520 #endif
521 	/*
522 	 * Compute the number of pages of memory that will be available for
523 	 * use (taking into account the overhead of a page structure per
524 	 * page).
525 	 */
526 	first_page = low_water / PAGE_SIZE;
527 #ifdef VM_PHYSSEG_SPARSE
528 	page_range = 0;
529 	for (i = 0; i < vm_phys_nsegs; i++) {
530 		page_range += atop(vm_phys_segs[i].end -
531 		    vm_phys_segs[i].start);
532 	}
533 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
534 		page_range += atop(phys_avail[i + 1] - phys_avail[i]);
535 #elif defined(VM_PHYSSEG_DENSE)
536 	page_range = high_water / PAGE_SIZE - first_page;
537 #else
538 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
539 #endif
540 	end = new_end;
541 
542 	/*
543 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
544 	 */
545 	vaddr += PAGE_SIZE;
546 
547 	/*
548 	 * Initialize the mem entry structures now, and put them in the free
549 	 * queue.
550 	 */
551 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
552 	mapped = pmap_map(&vaddr, new_end, end,
553 	    VM_PROT_READ | VM_PROT_WRITE);
554 	vm_page_array = (vm_page_t) mapped;
555 #if VM_NRESERVLEVEL > 0
556 	/*
557 	 * Allocate memory for the reservation management system's data
558 	 * structures.
559 	 */
560 	new_end = vm_reserv_startup(&vaddr, new_end, high_water);
561 #endif
562 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
563 	/*
564 	 * pmap_map on arm64, amd64, and mips can come out of the direct-map,
565 	 * not kvm like i386, so the pages must be tracked for a crashdump to
566 	 * include this data.  This includes the vm_page_array and the early
567 	 * UMA bootstrap pages.
568 	 */
569 	for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
570 		dump_add_page(pa);
571 #endif
572 	phys_avail[biggestone + 1] = new_end;
573 
574 	/*
575 	 * Add physical memory segments corresponding to the available
576 	 * physical pages.
577 	 */
578 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
579 		vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
580 
581 	/*
582 	 * Clear all of the page structures
583 	 */
584 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
585 	for (i = 0; i < page_range; i++)
586 		vm_page_array[i].order = VM_NFREEORDER;
587 	vm_page_array_size = page_range;
588 
589 	/*
590 	 * Initialize the physical memory allocator.
591 	 */
592 	vm_phys_init();
593 
594 	/*
595 	 * Add every available physical page that is not blacklisted to
596 	 * the free lists.
597 	 */
598 	vm_cnt.v_page_count = 0;
599 	vm_cnt.v_free_count = 0;
600 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
601 		pa = phys_avail[i];
602 		last_pa = phys_avail[i + 1];
603 		while (pa < last_pa) {
604 			vm_phys_add_page(pa);
605 			pa += PAGE_SIZE;
606 		}
607 	}
608 
609 	TAILQ_INIT(&blacklist_head);
610 	vm_page_blacklist_load(&list, &listend);
611 	vm_page_blacklist_check(list, listend);
612 
613 	list = kern_getenv("vm.blacklist");
614 	vm_page_blacklist_check(list, NULL);
615 
616 	freeenv(list);
617 #if VM_NRESERVLEVEL > 0
618 	/*
619 	 * Initialize the reservation management system.
620 	 */
621 	vm_reserv_init();
622 #endif
623 	return (vaddr);
624 }
625 
626 void
627 vm_page_reference(vm_page_t m)
628 {
629 
630 	vm_page_aflag_set(m, PGA_REFERENCED);
631 }
632 
633 /*
634  *	vm_page_busy_downgrade:
635  *
636  *	Downgrade an exclusive busy page into a single shared busy page.
637  */
638 void
639 vm_page_busy_downgrade(vm_page_t m)
640 {
641 	u_int x;
642 
643 	vm_page_assert_xbusied(m);
644 
645 	for (;;) {
646 		x = m->busy_lock;
647 		x &= VPB_BIT_WAITERS;
648 		if (atomic_cmpset_rel_int(&m->busy_lock,
649 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x))
650 			break;
651 	}
652 }
653 
654 /*
655  *	vm_page_sbusied:
656  *
657  *	Return a positive value if the page is shared busied, 0 otherwise.
658  */
659 int
660 vm_page_sbusied(vm_page_t m)
661 {
662 	u_int x;
663 
664 	x = m->busy_lock;
665 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
666 }
667 
668 /*
669  *	vm_page_sunbusy:
670  *
671  *	Shared unbusy a page.
672  */
673 void
674 vm_page_sunbusy(vm_page_t m)
675 {
676 	u_int x;
677 
678 	vm_page_assert_sbusied(m);
679 
680 	for (;;) {
681 		x = m->busy_lock;
682 		if (VPB_SHARERS(x) > 1) {
683 			if (atomic_cmpset_int(&m->busy_lock, x,
684 			    x - VPB_ONE_SHARER))
685 				break;
686 			continue;
687 		}
688 		if ((x & VPB_BIT_WAITERS) == 0) {
689 			KASSERT(x == VPB_SHARERS_WORD(1),
690 			    ("vm_page_sunbusy: invalid lock state"));
691 			if (atomic_cmpset_int(&m->busy_lock,
692 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
693 				break;
694 			continue;
695 		}
696 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
697 		    ("vm_page_sunbusy: invalid lock state for waiters"));
698 
699 		vm_page_lock(m);
700 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
701 			vm_page_unlock(m);
702 			continue;
703 		}
704 		wakeup(m);
705 		vm_page_unlock(m);
706 		break;
707 	}
708 }
709 
710 /*
711  *	vm_page_busy_sleep:
712  *
713  *	Sleep and release the page lock, using the page pointer as wchan.
714  *	This is used to implement the hard-path of busying mechanism.
715  *
716  *	The given page must be locked.
717  */
718 void
719 vm_page_busy_sleep(vm_page_t m, const char *wmesg)
720 {
721 	u_int x;
722 
723 	vm_page_lock_assert(m, MA_OWNED);
724 
725 	x = m->busy_lock;
726 	if (x == VPB_UNBUSIED) {
727 		vm_page_unlock(m);
728 		return;
729 	}
730 	if ((x & VPB_BIT_WAITERS) == 0 &&
731 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) {
732 		vm_page_unlock(m);
733 		return;
734 	}
735 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
736 }
737 
738 /*
739  *	vm_page_trysbusy:
740  *
741  *	Try to shared busy a page.
742  *	If the operation succeeds 1 is returned otherwise 0.
743  *	The operation never sleeps.
744  */
745 int
746 vm_page_trysbusy(vm_page_t m)
747 {
748 	u_int x;
749 
750 	for (;;) {
751 		x = m->busy_lock;
752 		if ((x & VPB_BIT_SHARED) == 0)
753 			return (0);
754 		if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
755 			return (1);
756 	}
757 }
758 
759 /*
760  *	vm_page_xunbusy_hard:
761  *
762  *	Called after the first try the exclusive unbusy of a page failed.
763  *	It is assumed that the waiters bit is on.
764  */
765 void
766 vm_page_xunbusy_hard(vm_page_t m)
767 {
768 
769 	vm_page_assert_xbusied(m);
770 
771 	vm_page_lock(m);
772 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
773 	wakeup(m);
774 	vm_page_unlock(m);
775 }
776 
777 /*
778  *	vm_page_flash:
779  *
780  *	Wakeup anyone waiting for the page.
781  *	The ownership bits do not change.
782  *
783  *	The given page must be locked.
784  */
785 void
786 vm_page_flash(vm_page_t m)
787 {
788 	u_int x;
789 
790 	vm_page_lock_assert(m, MA_OWNED);
791 
792 	for (;;) {
793 		x = m->busy_lock;
794 		if ((x & VPB_BIT_WAITERS) == 0)
795 			return;
796 		if (atomic_cmpset_int(&m->busy_lock, x,
797 		    x & (~VPB_BIT_WAITERS)))
798 			break;
799 	}
800 	wakeup(m);
801 }
802 
803 /*
804  * Keep page from being freed by the page daemon
805  * much of the same effect as wiring, except much lower
806  * overhead and should be used only for *very* temporary
807  * holding ("wiring").
808  */
809 void
810 vm_page_hold(vm_page_t mem)
811 {
812 
813 	vm_page_lock_assert(mem, MA_OWNED);
814         mem->hold_count++;
815 }
816 
817 void
818 vm_page_unhold(vm_page_t mem)
819 {
820 
821 	vm_page_lock_assert(mem, MA_OWNED);
822 	KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
823 	--mem->hold_count;
824 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
825 		vm_page_free_toq(mem);
826 }
827 
828 /*
829  *	vm_page_unhold_pages:
830  *
831  *	Unhold each of the pages that is referenced by the given array.
832  */
833 void
834 vm_page_unhold_pages(vm_page_t *ma, int count)
835 {
836 	struct mtx *mtx, *new_mtx;
837 
838 	mtx = NULL;
839 	for (; count != 0; count--) {
840 		/*
841 		 * Avoid releasing and reacquiring the same page lock.
842 		 */
843 		new_mtx = vm_page_lockptr(*ma);
844 		if (mtx != new_mtx) {
845 			if (mtx != NULL)
846 				mtx_unlock(mtx);
847 			mtx = new_mtx;
848 			mtx_lock(mtx);
849 		}
850 		vm_page_unhold(*ma);
851 		ma++;
852 	}
853 	if (mtx != NULL)
854 		mtx_unlock(mtx);
855 }
856 
857 vm_page_t
858 PHYS_TO_VM_PAGE(vm_paddr_t pa)
859 {
860 	vm_page_t m;
861 
862 #ifdef VM_PHYSSEG_SPARSE
863 	m = vm_phys_paddr_to_vm_page(pa);
864 	if (m == NULL)
865 		m = vm_phys_fictitious_to_vm_page(pa);
866 	return (m);
867 #elif defined(VM_PHYSSEG_DENSE)
868 	long pi;
869 
870 	pi = atop(pa);
871 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
872 		m = &vm_page_array[pi - first_page];
873 		return (m);
874 	}
875 	return (vm_phys_fictitious_to_vm_page(pa));
876 #else
877 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
878 #endif
879 }
880 
881 /*
882  *	vm_page_getfake:
883  *
884  *	Create a fictitious page with the specified physical address and
885  *	memory attribute.  The memory attribute is the only the machine-
886  *	dependent aspect of a fictitious page that must be initialized.
887  */
888 vm_page_t
889 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
890 {
891 	vm_page_t m;
892 
893 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
894 	vm_page_initfake(m, paddr, memattr);
895 	return (m);
896 }
897 
898 void
899 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
900 {
901 
902 	if ((m->flags & PG_FICTITIOUS) != 0) {
903 		/*
904 		 * The page's memattr might have changed since the
905 		 * previous initialization.  Update the pmap to the
906 		 * new memattr.
907 		 */
908 		goto memattr;
909 	}
910 	m->phys_addr = paddr;
911 	m->queue = PQ_NONE;
912 	/* Fictitious pages don't use "segind". */
913 	m->flags = PG_FICTITIOUS;
914 	/* Fictitious pages don't use "order" or "pool". */
915 	m->oflags = VPO_UNMANAGED;
916 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
917 	m->wire_count = 1;
918 	pmap_page_init(m);
919 memattr:
920 	pmap_page_set_memattr(m, memattr);
921 }
922 
923 /*
924  *	vm_page_putfake:
925  *
926  *	Release a fictitious page.
927  */
928 void
929 vm_page_putfake(vm_page_t m)
930 {
931 
932 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
933 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
934 	    ("vm_page_putfake: bad page %p", m));
935 	uma_zfree(fakepg_zone, m);
936 }
937 
938 /*
939  *	vm_page_updatefake:
940  *
941  *	Update the given fictitious page to the specified physical address and
942  *	memory attribute.
943  */
944 void
945 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
946 {
947 
948 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
949 	    ("vm_page_updatefake: bad page %p", m));
950 	m->phys_addr = paddr;
951 	pmap_page_set_memattr(m, memattr);
952 }
953 
954 /*
955  *	vm_page_free:
956  *
957  *	Free a page.
958  */
959 void
960 vm_page_free(vm_page_t m)
961 {
962 
963 	m->flags &= ~PG_ZERO;
964 	vm_page_free_toq(m);
965 }
966 
967 /*
968  *	vm_page_free_zero:
969  *
970  *	Free a page to the zerod-pages queue
971  */
972 void
973 vm_page_free_zero(vm_page_t m)
974 {
975 
976 	m->flags |= PG_ZERO;
977 	vm_page_free_toq(m);
978 }
979 
980 /*
981  * Unbusy and handle the page queueing for a page from the VOP_GETPAGES()
982  * array which is not the request page.
983  */
984 void
985 vm_page_readahead_finish(vm_page_t m)
986 {
987 
988 	if (m->valid != 0) {
989 		/*
990 		 * Since the page is not the requested page, whether
991 		 * it should be activated or deactivated is not
992 		 * obvious.  Empirical results have shown that
993 		 * deactivating the page is usually the best choice,
994 		 * unless the page is wanted by another thread.
995 		 */
996 		vm_page_lock(m);
997 		if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
998 			vm_page_activate(m);
999 		else
1000 			vm_page_deactivate(m);
1001 		vm_page_unlock(m);
1002 		vm_page_xunbusy(m);
1003 	} else {
1004 		/*
1005 		 * Free the completely invalid page.  Such page state
1006 		 * occurs due to the short read operation which did
1007 		 * not covered our page at all, or in case when a read
1008 		 * error happens.
1009 		 */
1010 		vm_page_lock(m);
1011 		vm_page_free(m);
1012 		vm_page_unlock(m);
1013 	}
1014 }
1015 
1016 /*
1017  *	vm_page_sleep_if_busy:
1018  *
1019  *	Sleep and release the page queues lock if the page is busied.
1020  *	Returns TRUE if the thread slept.
1021  *
1022  *	The given page must be unlocked and object containing it must
1023  *	be locked.
1024  */
1025 int
1026 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1027 {
1028 	vm_object_t obj;
1029 
1030 	vm_page_lock_assert(m, MA_NOTOWNED);
1031 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1032 
1033 	if (vm_page_busied(m)) {
1034 		/*
1035 		 * The page-specific object must be cached because page
1036 		 * identity can change during the sleep, causing the
1037 		 * re-lock of a different object.
1038 		 * It is assumed that a reference to the object is already
1039 		 * held by the callers.
1040 		 */
1041 		obj = m->object;
1042 		vm_page_lock(m);
1043 		VM_OBJECT_WUNLOCK(obj);
1044 		vm_page_busy_sleep(m, msg);
1045 		VM_OBJECT_WLOCK(obj);
1046 		return (TRUE);
1047 	}
1048 	return (FALSE);
1049 }
1050 
1051 /*
1052  *	vm_page_dirty_KBI:		[ internal use only ]
1053  *
1054  *	Set all bits in the page's dirty field.
1055  *
1056  *	The object containing the specified page must be locked if the
1057  *	call is made from the machine-independent layer.
1058  *
1059  *	See vm_page_clear_dirty_mask().
1060  *
1061  *	This function should only be called by vm_page_dirty().
1062  */
1063 void
1064 vm_page_dirty_KBI(vm_page_t m)
1065 {
1066 
1067 	/* These assertions refer to this operation by its public name. */
1068 	KASSERT((m->flags & PG_CACHED) == 0,
1069 	    ("vm_page_dirty: page in cache!"));
1070 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
1071 	    ("vm_page_dirty: page is invalid!"));
1072 	m->dirty = VM_PAGE_BITS_ALL;
1073 }
1074 
1075 /*
1076  *	vm_page_insert:		[ internal use only ]
1077  *
1078  *	Inserts the given mem entry into the object and object list.
1079  *
1080  *	The object must be locked.
1081  */
1082 int
1083 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1084 {
1085 	vm_page_t mpred;
1086 
1087 	VM_OBJECT_ASSERT_WLOCKED(object);
1088 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
1089 	return (vm_page_insert_after(m, object, pindex, mpred));
1090 }
1091 
1092 /*
1093  *	vm_page_insert_after:
1094  *
1095  *	Inserts the page "m" into the specified object at offset "pindex".
1096  *
1097  *	The page "mpred" must immediately precede the offset "pindex" within
1098  *	the specified object.
1099  *
1100  *	The object must be locked.
1101  */
1102 static int
1103 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1104     vm_page_t mpred)
1105 {
1106 	vm_pindex_t sidx;
1107 	vm_object_t sobj;
1108 	vm_page_t msucc;
1109 
1110 	VM_OBJECT_ASSERT_WLOCKED(object);
1111 	KASSERT(m->object == NULL,
1112 	    ("vm_page_insert_after: page already inserted"));
1113 	if (mpred != NULL) {
1114 		KASSERT(mpred->object == object,
1115 		    ("vm_page_insert_after: object doesn't contain mpred"));
1116 		KASSERT(mpred->pindex < pindex,
1117 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1118 		msucc = TAILQ_NEXT(mpred, listq);
1119 	} else
1120 		msucc = TAILQ_FIRST(&object->memq);
1121 	if (msucc != NULL)
1122 		KASSERT(msucc->pindex > pindex,
1123 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
1124 
1125 	/*
1126 	 * Record the object/offset pair in this page
1127 	 */
1128 	sobj = m->object;
1129 	sidx = m->pindex;
1130 	m->object = object;
1131 	m->pindex = pindex;
1132 
1133 	/*
1134 	 * Now link into the object's ordered list of backed pages.
1135 	 */
1136 	if (vm_radix_insert(&object->rtree, m)) {
1137 		m->object = sobj;
1138 		m->pindex = sidx;
1139 		return (1);
1140 	}
1141 	vm_page_insert_radixdone(m, object, mpred);
1142 	return (0);
1143 }
1144 
1145 /*
1146  *	vm_page_insert_radixdone:
1147  *
1148  *	Complete page "m" insertion into the specified object after the
1149  *	radix trie hooking.
1150  *
1151  *	The page "mpred" must precede the offset "m->pindex" within the
1152  *	specified object.
1153  *
1154  *	The object must be locked.
1155  */
1156 static void
1157 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1158 {
1159 
1160 	VM_OBJECT_ASSERT_WLOCKED(object);
1161 	KASSERT(object != NULL && m->object == object,
1162 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1163 	if (mpred != NULL) {
1164 		KASSERT(mpred->object == object,
1165 		    ("vm_page_insert_after: object doesn't contain mpred"));
1166 		KASSERT(mpred->pindex < m->pindex,
1167 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1168 	}
1169 
1170 	if (mpred != NULL)
1171 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1172 	else
1173 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1174 
1175 	/*
1176 	 * Show that the object has one more resident page.
1177 	 */
1178 	object->resident_page_count++;
1179 
1180 	/*
1181 	 * Hold the vnode until the last page is released.
1182 	 */
1183 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1184 		vhold(object->handle);
1185 
1186 	/*
1187 	 * Since we are inserting a new and possibly dirty page,
1188 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1189 	 */
1190 	if (pmap_page_is_write_mapped(m))
1191 		vm_object_set_writeable_dirty(object);
1192 }
1193 
1194 /*
1195  *	vm_page_remove:
1196  *
1197  *	Removes the given mem entry from the object/offset-page
1198  *	table and the object page list, but do not invalidate/terminate
1199  *	the backing store.
1200  *
1201  *	The object must be locked.  The page must be locked if it is managed.
1202  */
1203 void
1204 vm_page_remove(vm_page_t m)
1205 {
1206 	vm_object_t object;
1207 	boolean_t lockacq;
1208 
1209 	if ((m->oflags & VPO_UNMANAGED) == 0)
1210 		vm_page_lock_assert(m, MA_OWNED);
1211 	if ((object = m->object) == NULL)
1212 		return;
1213 	VM_OBJECT_ASSERT_WLOCKED(object);
1214 	if (vm_page_xbusied(m)) {
1215 		lockacq = FALSE;
1216 		if ((m->oflags & VPO_UNMANAGED) != 0 &&
1217 		    !mtx_owned(vm_page_lockptr(m))) {
1218 			lockacq = TRUE;
1219 			vm_page_lock(m);
1220 		}
1221 		vm_page_flash(m);
1222 		atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
1223 		if (lockacq)
1224 			vm_page_unlock(m);
1225 	}
1226 
1227 	/*
1228 	 * Now remove from the object's list of backed pages.
1229 	 */
1230 	vm_radix_remove(&object->rtree, m->pindex);
1231 	TAILQ_REMOVE(&object->memq, m, listq);
1232 
1233 	/*
1234 	 * And show that the object has one fewer resident page.
1235 	 */
1236 	object->resident_page_count--;
1237 
1238 	/*
1239 	 * The vnode may now be recycled.
1240 	 */
1241 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1242 		vdrop(object->handle);
1243 
1244 	m->object = NULL;
1245 }
1246 
1247 /*
1248  *	vm_page_lookup:
1249  *
1250  *	Returns the page associated with the object/offset
1251  *	pair specified; if none is found, NULL is returned.
1252  *
1253  *	The object must be locked.
1254  */
1255 vm_page_t
1256 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1257 {
1258 
1259 	VM_OBJECT_ASSERT_LOCKED(object);
1260 	return (vm_radix_lookup(&object->rtree, pindex));
1261 }
1262 
1263 /*
1264  *	vm_page_find_least:
1265  *
1266  *	Returns the page associated with the object with least pindex
1267  *	greater than or equal to the parameter pindex, or NULL.
1268  *
1269  *	The object must be locked.
1270  */
1271 vm_page_t
1272 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1273 {
1274 	vm_page_t m;
1275 
1276 	VM_OBJECT_ASSERT_LOCKED(object);
1277 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1278 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1279 	return (m);
1280 }
1281 
1282 /*
1283  * Returns the given page's successor (by pindex) within the object if it is
1284  * resident; if none is found, NULL is returned.
1285  *
1286  * The object must be locked.
1287  */
1288 vm_page_t
1289 vm_page_next(vm_page_t m)
1290 {
1291 	vm_page_t next;
1292 
1293 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1294 	if ((next = TAILQ_NEXT(m, listq)) != NULL &&
1295 	    next->pindex != m->pindex + 1)
1296 		next = NULL;
1297 	return (next);
1298 }
1299 
1300 /*
1301  * Returns the given page's predecessor (by pindex) within the object if it is
1302  * resident; if none is found, NULL is returned.
1303  *
1304  * The object must be locked.
1305  */
1306 vm_page_t
1307 vm_page_prev(vm_page_t m)
1308 {
1309 	vm_page_t prev;
1310 
1311 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1312 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL &&
1313 	    prev->pindex != m->pindex - 1)
1314 		prev = NULL;
1315 	return (prev);
1316 }
1317 
1318 /*
1319  * Uses the page mnew as a replacement for an existing page at index
1320  * pindex which must be already present in the object.
1321  *
1322  * The existing page must not be on a paging queue.
1323  */
1324 vm_page_t
1325 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1326 {
1327 	vm_page_t mold;
1328 
1329 	VM_OBJECT_ASSERT_WLOCKED(object);
1330 	KASSERT(mnew->object == NULL,
1331 	    ("vm_page_replace: page already in object"));
1332 
1333 	/*
1334 	 * This function mostly follows vm_page_insert() and
1335 	 * vm_page_remove() without the radix, object count and vnode
1336 	 * dance.  Double check such functions for more comments.
1337 	 */
1338 
1339 	mnew->object = object;
1340 	mnew->pindex = pindex;
1341 	mold = vm_radix_replace(&object->rtree, mnew);
1342 	KASSERT(mold->queue == PQ_NONE,
1343 	    ("vm_page_replace: mold is on a paging queue"));
1344 
1345 	/* Keep the resident page list in sorted order. */
1346 	TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1347 	TAILQ_REMOVE(&object->memq, mold, listq);
1348 
1349 	mold->object = NULL;
1350 	vm_page_xunbusy(mold);
1351 
1352 	/*
1353 	 * The object's resident_page_count does not change because we have
1354 	 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1355 	 */
1356 	if (pmap_page_is_write_mapped(mnew))
1357 		vm_object_set_writeable_dirty(object);
1358 	return (mold);
1359 }
1360 
1361 /*
1362  *	vm_page_rename:
1363  *
1364  *	Move the given memory entry from its
1365  *	current object to the specified target object/offset.
1366  *
1367  *	Note: swap associated with the page must be invalidated by the move.  We
1368  *	      have to do this for several reasons:  (1) we aren't freeing the
1369  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1370  *	      moving the page from object A to B, and will then later move
1371  *	      the backing store from A to B and we can't have a conflict.
1372  *
1373  *	Note: we *always* dirty the page.  It is necessary both for the
1374  *	      fact that we moved it, and because we may be invalidating
1375  *	      swap.  If the page is on the cache, we have to deactivate it
1376  *	      or vm_page_dirty() will panic.  Dirty pages are not allowed
1377  *	      on the cache.
1378  *
1379  *	The objects must be locked.
1380  */
1381 int
1382 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1383 {
1384 	vm_page_t mpred;
1385 	vm_pindex_t opidx;
1386 
1387 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1388 
1389 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1390 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1391 	    ("vm_page_rename: pindex already renamed"));
1392 
1393 	/*
1394 	 * Create a custom version of vm_page_insert() which does not depend
1395 	 * by m_prev and can cheat on the implementation aspects of the
1396 	 * function.
1397 	 */
1398 	opidx = m->pindex;
1399 	m->pindex = new_pindex;
1400 	if (vm_radix_insert(&new_object->rtree, m)) {
1401 		m->pindex = opidx;
1402 		return (1);
1403 	}
1404 
1405 	/*
1406 	 * The operation cannot fail anymore.  The removal must happen before
1407 	 * the listq iterator is tainted.
1408 	 */
1409 	m->pindex = opidx;
1410 	vm_page_lock(m);
1411 	vm_page_remove(m);
1412 
1413 	/* Return back to the new pindex to complete vm_page_insert(). */
1414 	m->pindex = new_pindex;
1415 	m->object = new_object;
1416 	vm_page_unlock(m);
1417 	vm_page_insert_radixdone(m, new_object, mpred);
1418 	vm_page_dirty(m);
1419 	return (0);
1420 }
1421 
1422 /*
1423  *	Convert all of the given object's cached pages that have a
1424  *	pindex within the given range into free pages.  If the value
1425  *	zero is given for "end", then the range's upper bound is
1426  *	infinity.  If the given object is backed by a vnode and it
1427  *	transitions from having one or more cached pages to none, the
1428  *	vnode's hold count is reduced.
1429  */
1430 void
1431 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1432 {
1433 	vm_page_t m;
1434 	boolean_t empty;
1435 
1436 	mtx_lock(&vm_page_queue_free_mtx);
1437 	if (__predict_false(vm_radix_is_empty(&object->cache))) {
1438 		mtx_unlock(&vm_page_queue_free_mtx);
1439 		return;
1440 	}
1441 	while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) {
1442 		if (end != 0 && m->pindex >= end)
1443 			break;
1444 		vm_radix_remove(&object->cache, m->pindex);
1445 		vm_page_cache_turn_free(m);
1446 	}
1447 	empty = vm_radix_is_empty(&object->cache);
1448 	mtx_unlock(&vm_page_queue_free_mtx);
1449 	if (object->type == OBJT_VNODE && empty)
1450 		vdrop(object->handle);
1451 }
1452 
1453 /*
1454  *	Returns the cached page that is associated with the given
1455  *	object and offset.  If, however, none exists, returns NULL.
1456  *
1457  *	The free page queue must be locked.
1458  */
1459 static inline vm_page_t
1460 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
1461 {
1462 
1463 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1464 	return (vm_radix_lookup(&object->cache, pindex));
1465 }
1466 
1467 /*
1468  *	Remove the given cached page from its containing object's
1469  *	collection of cached pages.
1470  *
1471  *	The free page queue must be locked.
1472  */
1473 static void
1474 vm_page_cache_remove(vm_page_t m)
1475 {
1476 
1477 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1478 	KASSERT((m->flags & PG_CACHED) != 0,
1479 	    ("vm_page_cache_remove: page %p is not cached", m));
1480 	vm_radix_remove(&m->object->cache, m->pindex);
1481 	m->object = NULL;
1482 	vm_cnt.v_cache_count--;
1483 }
1484 
1485 /*
1486  *	Transfer all of the cached pages with offset greater than or
1487  *	equal to 'offidxstart' from the original object's cache to the
1488  *	new object's cache.  However, any cached pages with offset
1489  *	greater than or equal to the new object's size are kept in the
1490  *	original object.  Initially, the new object's cache must be
1491  *	empty.  Offset 'offidxstart' in the original object must
1492  *	correspond to offset zero in the new object.
1493  *
1494  *	The new object must be locked.
1495  */
1496 void
1497 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
1498     vm_object_t new_object)
1499 {
1500 	vm_page_t m;
1501 
1502 	/*
1503 	 * Insertion into an object's collection of cached pages
1504 	 * requires the object to be locked.  In contrast, removal does
1505 	 * not.
1506 	 */
1507 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1508 	KASSERT(vm_radix_is_empty(&new_object->cache),
1509 	    ("vm_page_cache_transfer: object %p has cached pages",
1510 	    new_object));
1511 	mtx_lock(&vm_page_queue_free_mtx);
1512 	while ((m = vm_radix_lookup_ge(&orig_object->cache,
1513 	    offidxstart)) != NULL) {
1514 		/*
1515 		 * Transfer all of the pages with offset greater than or
1516 		 * equal to 'offidxstart' from the original object's
1517 		 * cache to the new object's cache.
1518 		 */
1519 		if ((m->pindex - offidxstart) >= new_object->size)
1520 			break;
1521 		vm_radix_remove(&orig_object->cache, m->pindex);
1522 		/* Update the page's object and offset. */
1523 		m->object = new_object;
1524 		m->pindex -= offidxstart;
1525 		if (vm_radix_insert(&new_object->cache, m))
1526 			vm_page_cache_turn_free(m);
1527 	}
1528 	mtx_unlock(&vm_page_queue_free_mtx);
1529 }
1530 
1531 /*
1532  *	Returns TRUE if a cached page is associated with the given object and
1533  *	offset, and FALSE otherwise.
1534  *
1535  *	The object must be locked.
1536  */
1537 boolean_t
1538 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex)
1539 {
1540 	vm_page_t m;
1541 
1542 	/*
1543 	 * Insertion into an object's collection of cached pages requires the
1544 	 * object to be locked.  Therefore, if the object is locked and the
1545 	 * object's collection is empty, there is no need to acquire the free
1546 	 * page queues lock in order to prove that the specified page doesn't
1547 	 * exist.
1548 	 */
1549 	VM_OBJECT_ASSERT_WLOCKED(object);
1550 	if (__predict_true(vm_object_cache_is_empty(object)))
1551 		return (FALSE);
1552 	mtx_lock(&vm_page_queue_free_mtx);
1553 	m = vm_page_cache_lookup(object, pindex);
1554 	mtx_unlock(&vm_page_queue_free_mtx);
1555 	return (m != NULL);
1556 }
1557 
1558 /*
1559  *	vm_page_alloc:
1560  *
1561  *	Allocate and return a page that is associated with the specified
1562  *	object and offset pair.  By default, this page is exclusive busied.
1563  *
1564  *	The caller must always specify an allocation class.
1565  *
1566  *	allocation classes:
1567  *	VM_ALLOC_NORMAL		normal process request
1568  *	VM_ALLOC_SYSTEM		system *really* needs a page
1569  *	VM_ALLOC_INTERRUPT	interrupt time request
1570  *
1571  *	optional allocation flags:
1572  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1573  *				intends to allocate
1574  *	VM_ALLOC_IFCACHED	return page only if it is cached
1575  *	VM_ALLOC_IFNOTCACHED	return NULL, do not reactivate if the page
1576  *				is cached
1577  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1578  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1579  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1580  *				should not be exclusive busy
1581  *	VM_ALLOC_SBUSY		shared busy the allocated page
1582  *	VM_ALLOC_WIRED		wire the allocated page
1583  *	VM_ALLOC_ZERO		prefer a zeroed page
1584  *
1585  *	This routine may not sleep.
1586  */
1587 vm_page_t
1588 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1589 {
1590 	struct vnode *vp = NULL;
1591 	vm_object_t m_object;
1592 	vm_page_t m, mpred;
1593 	int flags, req_class;
1594 
1595 	mpred = 0;	/* XXX: pacify gcc */
1596 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1597 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1598 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1599 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1600 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1601 	    req));
1602 	if (object != NULL)
1603 		VM_OBJECT_ASSERT_WLOCKED(object);
1604 
1605 	req_class = req & VM_ALLOC_CLASS_MASK;
1606 
1607 	/*
1608 	 * The page daemon is allowed to dig deeper into the free page list.
1609 	 */
1610 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1611 		req_class = VM_ALLOC_SYSTEM;
1612 
1613 	if (object != NULL) {
1614 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1615 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1616 		   ("vm_page_alloc: pindex already allocated"));
1617 	}
1618 
1619 	/*
1620 	 * The page allocation request can came from consumers which already
1621 	 * hold the free page queue mutex, like vm_page_insert() in
1622 	 * vm_page_cache().
1623 	 */
1624 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
1625 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
1626 	    (req_class == VM_ALLOC_SYSTEM &&
1627 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
1628 	    (req_class == VM_ALLOC_INTERRUPT &&
1629 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) {
1630 		/*
1631 		 * Allocate from the free queue if the number of free pages
1632 		 * exceeds the minimum for the request class.
1633 		 */
1634 		if (object != NULL &&
1635 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1636 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1637 				mtx_unlock(&vm_page_queue_free_mtx);
1638 				return (NULL);
1639 			}
1640 			if (vm_phys_unfree_page(m))
1641 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1642 #if VM_NRESERVLEVEL > 0
1643 			else if (!vm_reserv_reactivate_page(m))
1644 #else
1645 			else
1646 #endif
1647 				panic("vm_page_alloc: cache page %p is missing"
1648 				    " from the free queue", m);
1649 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1650 			mtx_unlock(&vm_page_queue_free_mtx);
1651 			return (NULL);
1652 #if VM_NRESERVLEVEL > 0
1653 		} else if (object == NULL || (object->flags & (OBJ_COLORED |
1654 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1655 		    vm_reserv_alloc_page(object, pindex, mpred)) == NULL) {
1656 #else
1657 		} else {
1658 #endif
1659 			m = vm_phys_alloc_pages(object != NULL ?
1660 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1661 #if VM_NRESERVLEVEL > 0
1662 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1663 				m = vm_phys_alloc_pages(object != NULL ?
1664 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1665 				    0);
1666 			}
1667 #endif
1668 		}
1669 	} else {
1670 		/*
1671 		 * Not allocatable, give up.
1672 		 */
1673 		mtx_unlock(&vm_page_queue_free_mtx);
1674 		atomic_add_int(&vm_pageout_deficit,
1675 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1676 		pagedaemon_wakeup();
1677 		return (NULL);
1678 	}
1679 
1680 	/*
1681 	 *  At this point we had better have found a good page.
1682 	 */
1683 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1684 	KASSERT(m->queue == PQ_NONE,
1685 	    ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
1686 	KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
1687 	KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
1688 	KASSERT(!vm_page_sbusied(m),
1689 	    ("vm_page_alloc: page %p is busy", m));
1690 	KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
1691 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1692 	    ("vm_page_alloc: page %p has unexpected memattr %d", m,
1693 	    pmap_page_get_memattr(m)));
1694 	if ((m->flags & PG_CACHED) != 0) {
1695 		KASSERT((m->flags & PG_ZERO) == 0,
1696 		    ("vm_page_alloc: cached page %p is PG_ZERO", m));
1697 		KASSERT(m->valid != 0,
1698 		    ("vm_page_alloc: cached page %p is invalid", m));
1699 		if (m->object == object && m->pindex == pindex)
1700 			vm_cnt.v_reactivated++;
1701 		else
1702 			m->valid = 0;
1703 		m_object = m->object;
1704 		vm_page_cache_remove(m);
1705 		if (m_object->type == OBJT_VNODE &&
1706 		    vm_object_cache_is_empty(m_object))
1707 			vp = m_object->handle;
1708 	} else {
1709 		KASSERT(m->valid == 0,
1710 		    ("vm_page_alloc: free page %p is valid", m));
1711 		vm_phys_freecnt_adj(m, -1);
1712 		if ((m->flags & PG_ZERO) != 0)
1713 			vm_page_zero_count--;
1714 	}
1715 	mtx_unlock(&vm_page_queue_free_mtx);
1716 
1717 	/*
1718 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1719 	 */
1720 	flags = 0;
1721 	if ((req & VM_ALLOC_ZERO) != 0)
1722 		flags = PG_ZERO;
1723 	flags &= m->flags;
1724 	if ((req & VM_ALLOC_NODUMP) != 0)
1725 		flags |= PG_NODUMP;
1726 	m->flags = flags;
1727 	m->aflags = 0;
1728 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1729 	    VPO_UNMANAGED : 0;
1730 	m->busy_lock = VPB_UNBUSIED;
1731 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1732 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1733 	if ((req & VM_ALLOC_SBUSY) != 0)
1734 		m->busy_lock = VPB_SHARERS_WORD(1);
1735 	if (req & VM_ALLOC_WIRED) {
1736 		/*
1737 		 * The page lock is not required for wiring a page until that
1738 		 * page is inserted into the object.
1739 		 */
1740 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1741 		m->wire_count = 1;
1742 	}
1743 	m->act_count = 0;
1744 
1745 	if (object != NULL) {
1746 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1747 			/* See the comment below about hold count. */
1748 			if (vp != NULL)
1749 				vdrop(vp);
1750 			pagedaemon_wakeup();
1751 			if (req & VM_ALLOC_WIRED) {
1752 				atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1753 				m->wire_count = 0;
1754 			}
1755 			m->object = NULL;
1756 			m->oflags = VPO_UNMANAGED;
1757 			vm_page_free(m);
1758 			return (NULL);
1759 		}
1760 
1761 		/* Ignore device objects; the pager sets "memattr" for them. */
1762 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1763 		    (object->flags & OBJ_FICTITIOUS) == 0)
1764 			pmap_page_set_memattr(m, object->memattr);
1765 	} else
1766 		m->pindex = pindex;
1767 
1768 	/*
1769 	 * The following call to vdrop() must come after the above call
1770 	 * to vm_page_insert() in case both affect the same object and
1771 	 * vnode.  Otherwise, the affected vnode's hold count could
1772 	 * temporarily become zero.
1773 	 */
1774 	if (vp != NULL)
1775 		vdrop(vp);
1776 
1777 	/*
1778 	 * Don't wakeup too often - wakeup the pageout daemon when
1779 	 * we would be nearly out of memory.
1780 	 */
1781 	if (vm_paging_needed())
1782 		pagedaemon_wakeup();
1783 
1784 	return (m);
1785 }
1786 
1787 static void
1788 vm_page_alloc_contig_vdrop(struct spglist *lst)
1789 {
1790 
1791 	while (!SLIST_EMPTY(lst)) {
1792 		vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv);
1793 		SLIST_REMOVE_HEAD(lst, plinks.s.ss);
1794 	}
1795 }
1796 
1797 /*
1798  *	vm_page_alloc_contig:
1799  *
1800  *	Allocate a contiguous set of physical pages of the given size "npages"
1801  *	from the free lists.  All of the physical pages must be at or above
1802  *	the given physical address "low" and below the given physical address
1803  *	"high".  The given value "alignment" determines the alignment of the
1804  *	first physical page in the set.  If the given value "boundary" is
1805  *	non-zero, then the set of physical pages cannot cross any physical
1806  *	address boundary that is a multiple of that value.  Both "alignment"
1807  *	and "boundary" must be a power of two.
1808  *
1809  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1810  *	then the memory attribute setting for the physical pages is configured
1811  *	to the object's memory attribute setting.  Otherwise, the memory
1812  *	attribute setting for the physical pages is configured to "memattr",
1813  *	overriding the object's memory attribute setting.  However, if the
1814  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1815  *	memory attribute setting for the physical pages cannot be configured
1816  *	to VM_MEMATTR_DEFAULT.
1817  *
1818  *	The caller must always specify an allocation class.
1819  *
1820  *	allocation classes:
1821  *	VM_ALLOC_NORMAL		normal process request
1822  *	VM_ALLOC_SYSTEM		system *really* needs a page
1823  *	VM_ALLOC_INTERRUPT	interrupt time request
1824  *
1825  *	optional allocation flags:
1826  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1827  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1828  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1829  *				should not be exclusive busy
1830  *	VM_ALLOC_SBUSY		shared busy the allocated page
1831  *	VM_ALLOC_WIRED		wire the allocated page
1832  *	VM_ALLOC_ZERO		prefer a zeroed page
1833  *
1834  *	This routine may not sleep.
1835  */
1836 vm_page_t
1837 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1838     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1839     vm_paddr_t boundary, vm_memattr_t memattr)
1840 {
1841 	struct vnode *drop;
1842 	struct spglist deferred_vdrop_list;
1843 	vm_page_t m, m_tmp, m_ret;
1844 	u_int flags;
1845 	int req_class;
1846 
1847 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1848 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1849 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1850 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1851 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1852 	    req));
1853 	if (object != NULL) {
1854 		VM_OBJECT_ASSERT_WLOCKED(object);
1855 		KASSERT(object->type == OBJT_PHYS,
1856 		    ("vm_page_alloc_contig: object %p isn't OBJT_PHYS",
1857 		    object));
1858 	}
1859 	KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1860 	req_class = req & VM_ALLOC_CLASS_MASK;
1861 
1862 	/*
1863 	 * The page daemon is allowed to dig deeper into the free page list.
1864 	 */
1865 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1866 		req_class = VM_ALLOC_SYSTEM;
1867 
1868 	SLIST_INIT(&deferred_vdrop_list);
1869 	mtx_lock(&vm_page_queue_free_mtx);
1870 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1871 	    vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM &&
1872 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1873 	    vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT &&
1874 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) {
1875 #if VM_NRESERVLEVEL > 0
1876 retry:
1877 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1878 		    (m_ret = vm_reserv_alloc_contig(object, pindex, npages,
1879 		    low, high, alignment, boundary)) == NULL)
1880 #endif
1881 			m_ret = vm_phys_alloc_contig(npages, low, high,
1882 			    alignment, boundary);
1883 	} else {
1884 		mtx_unlock(&vm_page_queue_free_mtx);
1885 		atomic_add_int(&vm_pageout_deficit, npages);
1886 		pagedaemon_wakeup();
1887 		return (NULL);
1888 	}
1889 	if (m_ret != NULL)
1890 		for (m = m_ret; m < &m_ret[npages]; m++) {
1891 			drop = vm_page_alloc_init(m);
1892 			if (drop != NULL) {
1893 				/*
1894 				 * Enqueue the vnode for deferred vdrop().
1895 				 */
1896 				m->plinks.s.pv = drop;
1897 				SLIST_INSERT_HEAD(&deferred_vdrop_list, m,
1898 				    plinks.s.ss);
1899 			}
1900 		}
1901 	else {
1902 #if VM_NRESERVLEVEL > 0
1903 		if (vm_reserv_reclaim_contig(npages, low, high, alignment,
1904 		    boundary))
1905 			goto retry;
1906 #endif
1907 	}
1908 	mtx_unlock(&vm_page_queue_free_mtx);
1909 	if (m_ret == NULL)
1910 		return (NULL);
1911 
1912 	/*
1913 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1914 	 */
1915 	flags = 0;
1916 	if ((req & VM_ALLOC_ZERO) != 0)
1917 		flags = PG_ZERO;
1918 	if ((req & VM_ALLOC_NODUMP) != 0)
1919 		flags |= PG_NODUMP;
1920 	if ((req & VM_ALLOC_WIRED) != 0)
1921 		atomic_add_int(&vm_cnt.v_wire_count, npages);
1922 	if (object != NULL) {
1923 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1924 		    memattr == VM_MEMATTR_DEFAULT)
1925 			memattr = object->memattr;
1926 	}
1927 	for (m = m_ret; m < &m_ret[npages]; m++) {
1928 		m->aflags = 0;
1929 		m->flags = (m->flags | PG_NODUMP) & flags;
1930 		m->busy_lock = VPB_UNBUSIED;
1931 		if (object != NULL) {
1932 			if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
1933 				m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1934 			if ((req & VM_ALLOC_SBUSY) != 0)
1935 				m->busy_lock = VPB_SHARERS_WORD(1);
1936 		}
1937 		if ((req & VM_ALLOC_WIRED) != 0)
1938 			m->wire_count = 1;
1939 		/* Unmanaged pages don't use "act_count". */
1940 		m->oflags = VPO_UNMANAGED;
1941 		if (object != NULL) {
1942 			if (vm_page_insert(m, object, pindex)) {
1943 				vm_page_alloc_contig_vdrop(
1944 				    &deferred_vdrop_list);
1945 				if (vm_paging_needed())
1946 					pagedaemon_wakeup();
1947 				if ((req & VM_ALLOC_WIRED) != 0)
1948 					atomic_subtract_int(&vm_cnt.v_wire_count,
1949 					    npages);
1950 				for (m_tmp = m, m = m_ret;
1951 				    m < &m_ret[npages]; m++) {
1952 					if ((req & VM_ALLOC_WIRED) != 0)
1953 						m->wire_count = 0;
1954 					if (m >= m_tmp)
1955 						m->object = NULL;
1956 					vm_page_free(m);
1957 				}
1958 				return (NULL);
1959 			}
1960 		} else
1961 			m->pindex = pindex;
1962 		if (memattr != VM_MEMATTR_DEFAULT)
1963 			pmap_page_set_memattr(m, memattr);
1964 		pindex++;
1965 	}
1966 	vm_page_alloc_contig_vdrop(&deferred_vdrop_list);
1967 	if (vm_paging_needed())
1968 		pagedaemon_wakeup();
1969 	return (m_ret);
1970 }
1971 
1972 /*
1973  * Initialize a page that has been freshly dequeued from a freelist.
1974  * The caller has to drop the vnode returned, if it is not NULL.
1975  *
1976  * This function may only be used to initialize unmanaged pages.
1977  *
1978  * To be called with vm_page_queue_free_mtx held.
1979  */
1980 static struct vnode *
1981 vm_page_alloc_init(vm_page_t m)
1982 {
1983 	struct vnode *drop;
1984 	vm_object_t m_object;
1985 
1986 	KASSERT(m->queue == PQ_NONE,
1987 	    ("vm_page_alloc_init: page %p has unexpected queue %d",
1988 	    m, m->queue));
1989 	KASSERT(m->wire_count == 0,
1990 	    ("vm_page_alloc_init: page %p is wired", m));
1991 	KASSERT(m->hold_count == 0,
1992 	    ("vm_page_alloc_init: page %p is held", m));
1993 	KASSERT(!vm_page_sbusied(m),
1994 	    ("vm_page_alloc_init: page %p is busy", m));
1995 	KASSERT(m->dirty == 0,
1996 	    ("vm_page_alloc_init: page %p is dirty", m));
1997 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1998 	    ("vm_page_alloc_init: page %p has unexpected memattr %d",
1999 	    m, pmap_page_get_memattr(m)));
2000 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2001 	drop = NULL;
2002 	if ((m->flags & PG_CACHED) != 0) {
2003 		KASSERT((m->flags & PG_ZERO) == 0,
2004 		    ("vm_page_alloc_init: cached page %p is PG_ZERO", m));
2005 		m->valid = 0;
2006 		m_object = m->object;
2007 		vm_page_cache_remove(m);
2008 		if (m_object->type == OBJT_VNODE &&
2009 		    vm_object_cache_is_empty(m_object))
2010 			drop = m_object->handle;
2011 	} else {
2012 		KASSERT(m->valid == 0,
2013 		    ("vm_page_alloc_init: free page %p is valid", m));
2014 		vm_phys_freecnt_adj(m, -1);
2015 		if ((m->flags & PG_ZERO) != 0)
2016 			vm_page_zero_count--;
2017 	}
2018 	return (drop);
2019 }
2020 
2021 /*
2022  * 	vm_page_alloc_freelist:
2023  *
2024  *	Allocate a physical page from the specified free page list.
2025  *
2026  *	The caller must always specify an allocation class.
2027  *
2028  *	allocation classes:
2029  *	VM_ALLOC_NORMAL		normal process request
2030  *	VM_ALLOC_SYSTEM		system *really* needs a page
2031  *	VM_ALLOC_INTERRUPT	interrupt time request
2032  *
2033  *	optional allocation flags:
2034  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
2035  *				intends to allocate
2036  *	VM_ALLOC_WIRED		wire the allocated page
2037  *	VM_ALLOC_ZERO		prefer a zeroed page
2038  *
2039  *	This routine may not sleep.
2040  */
2041 vm_page_t
2042 vm_page_alloc_freelist(int flind, int req)
2043 {
2044 	struct vnode *drop;
2045 	vm_page_t m;
2046 	u_int flags;
2047 	int req_class;
2048 
2049 	req_class = req & VM_ALLOC_CLASS_MASK;
2050 
2051 	/*
2052 	 * The page daemon is allowed to dig deeper into the free page list.
2053 	 */
2054 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2055 		req_class = VM_ALLOC_SYSTEM;
2056 
2057 	/*
2058 	 * Do not allocate reserved pages unless the req has asked for it.
2059 	 */
2060 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
2061 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
2062 	    (req_class == VM_ALLOC_SYSTEM &&
2063 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
2064 	    (req_class == VM_ALLOC_INTERRUPT &&
2065 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0))
2066 		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0);
2067 	else {
2068 		mtx_unlock(&vm_page_queue_free_mtx);
2069 		atomic_add_int(&vm_pageout_deficit,
2070 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
2071 		pagedaemon_wakeup();
2072 		return (NULL);
2073 	}
2074 	if (m == NULL) {
2075 		mtx_unlock(&vm_page_queue_free_mtx);
2076 		return (NULL);
2077 	}
2078 	drop = vm_page_alloc_init(m);
2079 	mtx_unlock(&vm_page_queue_free_mtx);
2080 
2081 	/*
2082 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
2083 	 */
2084 	m->aflags = 0;
2085 	flags = 0;
2086 	if ((req & VM_ALLOC_ZERO) != 0)
2087 		flags = PG_ZERO;
2088 	m->flags &= flags;
2089 	if ((req & VM_ALLOC_WIRED) != 0) {
2090 		/*
2091 		 * The page lock is not required for wiring a page that does
2092 		 * not belong to an object.
2093 		 */
2094 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2095 		m->wire_count = 1;
2096 	}
2097 	/* Unmanaged pages don't use "act_count". */
2098 	m->oflags = VPO_UNMANAGED;
2099 	if (drop != NULL)
2100 		vdrop(drop);
2101 	if (vm_paging_needed())
2102 		pagedaemon_wakeup();
2103 	return (m);
2104 }
2105 
2106 /*
2107  *	vm_wait:	(also see VM_WAIT macro)
2108  *
2109  *	Sleep until free pages are available for allocation.
2110  *	- Called in various places before memory allocations.
2111  */
2112 void
2113 vm_wait(void)
2114 {
2115 
2116 	mtx_lock(&vm_page_queue_free_mtx);
2117 	if (curproc == pageproc) {
2118 		vm_pageout_pages_needed = 1;
2119 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
2120 		    PDROP | PSWP, "VMWait", 0);
2121 	} else {
2122 		if (!vm_pages_needed) {
2123 			vm_pages_needed = 1;
2124 			wakeup(&vm_pages_needed);
2125 		}
2126 		msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
2127 		    "vmwait", 0);
2128 	}
2129 }
2130 
2131 /*
2132  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
2133  *
2134  *	Sleep until free pages are available for allocation.
2135  *	- Called only in vm_fault so that processes page faulting
2136  *	  can be easily tracked.
2137  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2138  *	  processes will be able to grab memory first.  Do not change
2139  *	  this balance without careful testing first.
2140  */
2141 void
2142 vm_waitpfault(void)
2143 {
2144 
2145 	mtx_lock(&vm_page_queue_free_mtx);
2146 	if (!vm_pages_needed) {
2147 		vm_pages_needed = 1;
2148 		wakeup(&vm_pages_needed);
2149 	}
2150 	msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
2151 	    "pfault", 0);
2152 }
2153 
2154 struct vm_pagequeue *
2155 vm_page_pagequeue(vm_page_t m)
2156 {
2157 
2158 	return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2159 }
2160 
2161 /*
2162  *	vm_page_dequeue:
2163  *
2164  *	Remove the given page from its current page queue.
2165  *
2166  *	The page must be locked.
2167  */
2168 void
2169 vm_page_dequeue(vm_page_t m)
2170 {
2171 	struct vm_pagequeue *pq;
2172 
2173 	vm_page_assert_locked(m);
2174 	KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2175 	    m));
2176 	pq = vm_page_pagequeue(m);
2177 	vm_pagequeue_lock(pq);
2178 	m->queue = PQ_NONE;
2179 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2180 	vm_pagequeue_cnt_dec(pq);
2181 	vm_pagequeue_unlock(pq);
2182 }
2183 
2184 /*
2185  *	vm_page_dequeue_locked:
2186  *
2187  *	Remove the given page from its current page queue.
2188  *
2189  *	The page and page queue must be locked.
2190  */
2191 void
2192 vm_page_dequeue_locked(vm_page_t m)
2193 {
2194 	struct vm_pagequeue *pq;
2195 
2196 	vm_page_lock_assert(m, MA_OWNED);
2197 	pq = vm_page_pagequeue(m);
2198 	vm_pagequeue_assert_locked(pq);
2199 	m->queue = PQ_NONE;
2200 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2201 	vm_pagequeue_cnt_dec(pq);
2202 }
2203 
2204 /*
2205  *	vm_page_enqueue:
2206  *
2207  *	Add the given page to the specified page queue.
2208  *
2209  *	The page must be locked.
2210  */
2211 static void
2212 vm_page_enqueue(uint8_t queue, vm_page_t m)
2213 {
2214 	struct vm_pagequeue *pq;
2215 
2216 	vm_page_lock_assert(m, MA_OWNED);
2217 	KASSERT(queue < PQ_COUNT,
2218 	    ("vm_page_enqueue: invalid queue %u request for page %p",
2219 	    queue, m));
2220 	pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2221 	vm_pagequeue_lock(pq);
2222 	m->queue = queue;
2223 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2224 	vm_pagequeue_cnt_inc(pq);
2225 	vm_pagequeue_unlock(pq);
2226 }
2227 
2228 /*
2229  *	vm_page_requeue:
2230  *
2231  *	Move the given page to the tail of its current page queue.
2232  *
2233  *	The page must be locked.
2234  */
2235 void
2236 vm_page_requeue(vm_page_t m)
2237 {
2238 	struct vm_pagequeue *pq;
2239 
2240 	vm_page_lock_assert(m, MA_OWNED);
2241 	KASSERT(m->queue != PQ_NONE,
2242 	    ("vm_page_requeue: page %p is not queued", m));
2243 	pq = vm_page_pagequeue(m);
2244 	vm_pagequeue_lock(pq);
2245 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2246 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2247 	vm_pagequeue_unlock(pq);
2248 }
2249 
2250 /*
2251  *	vm_page_requeue_locked:
2252  *
2253  *	Move the given page to the tail of its current page queue.
2254  *
2255  *	The page queue must be locked.
2256  */
2257 void
2258 vm_page_requeue_locked(vm_page_t m)
2259 {
2260 	struct vm_pagequeue *pq;
2261 
2262 	KASSERT(m->queue != PQ_NONE,
2263 	    ("vm_page_requeue_locked: page %p is not queued", m));
2264 	pq = vm_page_pagequeue(m);
2265 	vm_pagequeue_assert_locked(pq);
2266 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2267 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2268 }
2269 
2270 /*
2271  *	vm_page_activate:
2272  *
2273  *	Put the specified page on the active list (if appropriate).
2274  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2275  *	mess with it.
2276  *
2277  *	The page must be locked.
2278  */
2279 void
2280 vm_page_activate(vm_page_t m)
2281 {
2282 	int queue;
2283 
2284 	vm_page_lock_assert(m, MA_OWNED);
2285 	if ((queue = m->queue) != PQ_ACTIVE) {
2286 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2287 			if (m->act_count < ACT_INIT)
2288 				m->act_count = ACT_INIT;
2289 			if (queue != PQ_NONE)
2290 				vm_page_dequeue(m);
2291 			vm_page_enqueue(PQ_ACTIVE, m);
2292 		} else
2293 			KASSERT(queue == PQ_NONE,
2294 			    ("vm_page_activate: wired page %p is queued", m));
2295 	} else {
2296 		if (m->act_count < ACT_INIT)
2297 			m->act_count = ACT_INIT;
2298 	}
2299 }
2300 
2301 /*
2302  *	vm_page_free_wakeup:
2303  *
2304  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
2305  *	routine is called when a page has been added to the cache or free
2306  *	queues.
2307  *
2308  *	The page queues must be locked.
2309  */
2310 static inline void
2311 vm_page_free_wakeup(void)
2312 {
2313 
2314 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2315 	/*
2316 	 * if pageout daemon needs pages, then tell it that there are
2317 	 * some free.
2318 	 */
2319 	if (vm_pageout_pages_needed &&
2320 	    vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) {
2321 		wakeup(&vm_pageout_pages_needed);
2322 		vm_pageout_pages_needed = 0;
2323 	}
2324 	/*
2325 	 * wakeup processes that are waiting on memory if we hit a
2326 	 * high water mark. And wakeup scheduler process if we have
2327 	 * lots of memory. this process will swapin processes.
2328 	 */
2329 	if (vm_pages_needed && !vm_page_count_min()) {
2330 		vm_pages_needed = 0;
2331 		wakeup(&vm_cnt.v_free_count);
2332 	}
2333 }
2334 
2335 /*
2336  *	Turn a cached page into a free page, by changing its attributes.
2337  *	Keep the statistics up-to-date.
2338  *
2339  *	The free page queue must be locked.
2340  */
2341 static void
2342 vm_page_cache_turn_free(vm_page_t m)
2343 {
2344 
2345 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2346 
2347 	m->object = NULL;
2348 	m->valid = 0;
2349 	KASSERT((m->flags & PG_CACHED) != 0,
2350 	    ("vm_page_cache_turn_free: page %p is not cached", m));
2351 	m->flags &= ~PG_CACHED;
2352 	vm_cnt.v_cache_count--;
2353 	vm_phys_freecnt_adj(m, 1);
2354 }
2355 
2356 /*
2357  *	vm_page_free_toq:
2358  *
2359  *	Returns the given page to the free list,
2360  *	disassociating it with any VM object.
2361  *
2362  *	The object must be locked.  The page must be locked if it is managed.
2363  */
2364 void
2365 vm_page_free_toq(vm_page_t m)
2366 {
2367 
2368 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2369 		vm_page_lock_assert(m, MA_OWNED);
2370 		KASSERT(!pmap_page_is_mapped(m),
2371 		    ("vm_page_free_toq: freeing mapped page %p", m));
2372 	} else
2373 		KASSERT(m->queue == PQ_NONE,
2374 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2375 	PCPU_INC(cnt.v_tfree);
2376 
2377 	if (vm_page_sbusied(m))
2378 		panic("vm_page_free: freeing busy page %p", m);
2379 
2380 	/*
2381 	 * Unqueue, then remove page.  Note that we cannot destroy
2382 	 * the page here because we do not want to call the pager's
2383 	 * callback routine until after we've put the page on the
2384 	 * appropriate free queue.
2385 	 */
2386 	vm_page_remque(m);
2387 	vm_page_remove(m);
2388 
2389 	/*
2390 	 * If fictitious remove object association and
2391 	 * return, otherwise delay object association removal.
2392 	 */
2393 	if ((m->flags & PG_FICTITIOUS) != 0) {
2394 		return;
2395 	}
2396 
2397 	m->valid = 0;
2398 	vm_page_undirty(m);
2399 
2400 	if (m->wire_count != 0)
2401 		panic("vm_page_free: freeing wired page %p", m);
2402 	if (m->hold_count != 0) {
2403 		m->flags &= ~PG_ZERO;
2404 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2405 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
2406 		m->flags |= PG_UNHOLDFREE;
2407 	} else {
2408 		/*
2409 		 * Restore the default memory attribute to the page.
2410 		 */
2411 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2412 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2413 
2414 		/*
2415 		 * Insert the page into the physical memory allocator's
2416 		 * cache/free page queues.
2417 		 */
2418 		mtx_lock(&vm_page_queue_free_mtx);
2419 		vm_phys_freecnt_adj(m, 1);
2420 #if VM_NRESERVLEVEL > 0
2421 		if (!vm_reserv_free_page(m))
2422 #else
2423 		if (TRUE)
2424 #endif
2425 			vm_phys_free_pages(m, 0);
2426 		if ((m->flags & PG_ZERO) != 0)
2427 			++vm_page_zero_count;
2428 		else
2429 			vm_page_zero_idle_wakeup();
2430 		vm_page_free_wakeup();
2431 		mtx_unlock(&vm_page_queue_free_mtx);
2432 	}
2433 }
2434 
2435 /*
2436  *	vm_page_wire:
2437  *
2438  *	Mark this page as wired down by yet
2439  *	another map, removing it from paging queues
2440  *	as necessary.
2441  *
2442  *	If the page is fictitious, then its wire count must remain one.
2443  *
2444  *	The page must be locked.
2445  */
2446 void
2447 vm_page_wire(vm_page_t m)
2448 {
2449 
2450 	/*
2451 	 * Only bump the wire statistics if the page is not already wired,
2452 	 * and only unqueue the page if it is on some queue (if it is unmanaged
2453 	 * it is already off the queues).
2454 	 */
2455 	vm_page_lock_assert(m, MA_OWNED);
2456 	if ((m->flags & PG_FICTITIOUS) != 0) {
2457 		KASSERT(m->wire_count == 1,
2458 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
2459 		    m));
2460 		return;
2461 	}
2462 	if (m->wire_count == 0) {
2463 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
2464 		    m->queue == PQ_NONE,
2465 		    ("vm_page_wire: unmanaged page %p is queued", m));
2466 		vm_page_remque(m);
2467 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2468 	}
2469 	m->wire_count++;
2470 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
2471 }
2472 
2473 /*
2474  * vm_page_unwire:
2475  *
2476  * Release one wiring of the specified page, potentially allowing it to be
2477  * paged out.  Returns TRUE if the number of wirings transitions to zero and
2478  * FALSE otherwise.
2479  *
2480  * Only managed pages belonging to an object can be paged out.  If the number
2481  * of wirings transitions to zero and the page is eligible for page out, then
2482  * the page is added to the specified paging queue (unless PQ_NONE is
2483  * specified).
2484  *
2485  * If a page is fictitious, then its wire count must always be one.
2486  *
2487  * A managed page must be locked.
2488  */
2489 boolean_t
2490 vm_page_unwire(vm_page_t m, uint8_t queue)
2491 {
2492 
2493 	KASSERT(queue < PQ_COUNT || queue == PQ_NONE,
2494 	    ("vm_page_unwire: invalid queue %u request for page %p",
2495 	    queue, m));
2496 	if ((m->oflags & VPO_UNMANAGED) == 0)
2497 		vm_page_assert_locked(m);
2498 	if ((m->flags & PG_FICTITIOUS) != 0) {
2499 		KASSERT(m->wire_count == 1,
2500 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
2501 		return (FALSE);
2502 	}
2503 	if (m->wire_count > 0) {
2504 		m->wire_count--;
2505 		if (m->wire_count == 0) {
2506 			atomic_subtract_int(&vm_cnt.v_wire_count, 1);
2507 			if ((m->oflags & VPO_UNMANAGED) == 0 &&
2508 			    m->object != NULL && queue != PQ_NONE) {
2509 				if (queue == PQ_INACTIVE)
2510 					m->flags &= ~PG_WINATCFLS;
2511 				vm_page_enqueue(queue, m);
2512 			}
2513 			return (TRUE);
2514 		} else
2515 			return (FALSE);
2516 	} else
2517 		panic("vm_page_unwire: page %p's wire count is zero", m);
2518 }
2519 
2520 /*
2521  * Move the specified page to the inactive queue.
2522  *
2523  * Many pages placed on the inactive queue should actually go
2524  * into the cache, but it is difficult to figure out which.  What
2525  * we do instead, if the inactive target is well met, is to put
2526  * clean pages at the head of the inactive queue instead of the tail.
2527  * This will cause them to be moved to the cache more quickly and
2528  * if not actively re-referenced, reclaimed more quickly.  If we just
2529  * stick these pages at the end of the inactive queue, heavy filesystem
2530  * meta-data accesses can cause an unnecessary paging load on memory bound
2531  * processes.  This optimization causes one-time-use metadata to be
2532  * reused more quickly.
2533  *
2534  * Normally noreuse is FALSE, resulting in LRU operation.  noreuse is set
2535  * to TRUE if we want this page to be 'as if it were placed in the cache',
2536  * except without unmapping it from the process address space.  In
2537  * practice this is implemented by inserting the page at the head of the
2538  * queue, using a marker page to guide FIFO insertion ordering.
2539  *
2540  * The page must be locked.
2541  */
2542 static inline void
2543 _vm_page_deactivate(vm_page_t m, boolean_t noreuse)
2544 {
2545 	struct vm_pagequeue *pq;
2546 	int queue;
2547 
2548 	vm_page_assert_locked(m);
2549 
2550 	/*
2551 	 * Ignore if the page is already inactive, unless it is unlikely to be
2552 	 * reactivated.
2553 	 */
2554 	if ((queue = m->queue) == PQ_INACTIVE && !noreuse)
2555 		return;
2556 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2557 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
2558 		/* Avoid multiple acquisitions of the inactive queue lock. */
2559 		if (queue == PQ_INACTIVE) {
2560 			vm_pagequeue_lock(pq);
2561 			vm_page_dequeue_locked(m);
2562 		} else {
2563 			if (queue != PQ_NONE)
2564 				vm_page_dequeue(m);
2565 			m->flags &= ~PG_WINATCFLS;
2566 			vm_pagequeue_lock(pq);
2567 		}
2568 		m->queue = PQ_INACTIVE;
2569 		if (noreuse)
2570 			TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead,
2571 			    m, plinks.q);
2572 		else
2573 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2574 		vm_pagequeue_cnt_inc(pq);
2575 		vm_pagequeue_unlock(pq);
2576 	}
2577 }
2578 
2579 /*
2580  * Move the specified page to the inactive queue.
2581  *
2582  * The page must be locked.
2583  */
2584 void
2585 vm_page_deactivate(vm_page_t m)
2586 {
2587 
2588 	_vm_page_deactivate(m, FALSE);
2589 }
2590 
2591 /*
2592  * Move the specified page to the inactive queue with the expectation
2593  * that it is unlikely to be reused.
2594  *
2595  * The page must be locked.
2596  */
2597 void
2598 vm_page_deactivate_noreuse(vm_page_t m)
2599 {
2600 
2601 	_vm_page_deactivate(m, TRUE);
2602 }
2603 
2604 /*
2605  * vm_page_try_to_cache:
2606  *
2607  * Returns 0 on failure, 1 on success
2608  */
2609 int
2610 vm_page_try_to_cache(vm_page_t m)
2611 {
2612 
2613 	vm_page_lock_assert(m, MA_OWNED);
2614 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2615 	if (m->dirty || m->hold_count || m->wire_count ||
2616 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
2617 		return (0);
2618 	pmap_remove_all(m);
2619 	if (m->dirty)
2620 		return (0);
2621 	vm_page_cache(m);
2622 	return (1);
2623 }
2624 
2625 /*
2626  * vm_page_try_to_free()
2627  *
2628  *	Attempt to free the page.  If we cannot free it, we do nothing.
2629  *	1 is returned on success, 0 on failure.
2630  */
2631 int
2632 vm_page_try_to_free(vm_page_t m)
2633 {
2634 
2635 	vm_page_lock_assert(m, MA_OWNED);
2636 	if (m->object != NULL)
2637 		VM_OBJECT_ASSERT_WLOCKED(m->object);
2638 	if (m->dirty || m->hold_count || m->wire_count ||
2639 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
2640 		return (0);
2641 	pmap_remove_all(m);
2642 	if (m->dirty)
2643 		return (0);
2644 	vm_page_free(m);
2645 	return (1);
2646 }
2647 
2648 /*
2649  * vm_page_cache
2650  *
2651  * Put the specified page onto the page cache queue (if appropriate).
2652  *
2653  * The object and page must be locked.
2654  */
2655 void
2656 vm_page_cache(vm_page_t m)
2657 {
2658 	vm_object_t object;
2659 	boolean_t cache_was_empty;
2660 
2661 	vm_page_lock_assert(m, MA_OWNED);
2662 	object = m->object;
2663 	VM_OBJECT_ASSERT_WLOCKED(object);
2664 	if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) ||
2665 	    m->hold_count || m->wire_count)
2666 		panic("vm_page_cache: attempting to cache busy page");
2667 	KASSERT(!pmap_page_is_mapped(m),
2668 	    ("vm_page_cache: page %p is mapped", m));
2669 	KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m));
2670 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
2671 	    (object->type == OBJT_SWAP &&
2672 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
2673 		/*
2674 		 * Hypothesis: A cache-eligible page belonging to a
2675 		 * default object or swap object but without a backing
2676 		 * store must be zero filled.
2677 		 */
2678 		vm_page_free(m);
2679 		return;
2680 	}
2681 	KASSERT((m->flags & PG_CACHED) == 0,
2682 	    ("vm_page_cache: page %p is already cached", m));
2683 
2684 	/*
2685 	 * Remove the page from the paging queues.
2686 	 */
2687 	vm_page_remque(m);
2688 
2689 	/*
2690 	 * Remove the page from the object's collection of resident
2691 	 * pages.
2692 	 */
2693 	vm_radix_remove(&object->rtree, m->pindex);
2694 	TAILQ_REMOVE(&object->memq, m, listq);
2695 	object->resident_page_count--;
2696 
2697 	/*
2698 	 * Restore the default memory attribute to the page.
2699 	 */
2700 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2701 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2702 
2703 	/*
2704 	 * Insert the page into the object's collection of cached pages
2705 	 * and the physical memory allocator's cache/free page queues.
2706 	 */
2707 	m->flags &= ~PG_ZERO;
2708 	mtx_lock(&vm_page_queue_free_mtx);
2709 	cache_was_empty = vm_radix_is_empty(&object->cache);
2710 	if (vm_radix_insert(&object->cache, m)) {
2711 		mtx_unlock(&vm_page_queue_free_mtx);
2712 		if (object->resident_page_count == 0)
2713 			vdrop(object->handle);
2714 		m->object = NULL;
2715 		vm_page_free(m);
2716 		return;
2717 	}
2718 
2719 	/*
2720 	 * The above call to vm_radix_insert() could reclaim the one pre-
2721 	 * existing cached page from this object, resulting in a call to
2722 	 * vdrop().
2723 	 */
2724 	if (!cache_was_empty)
2725 		cache_was_empty = vm_radix_is_singleton(&object->cache);
2726 
2727 	m->flags |= PG_CACHED;
2728 	vm_cnt.v_cache_count++;
2729 	PCPU_INC(cnt.v_tcached);
2730 #if VM_NRESERVLEVEL > 0
2731 	if (!vm_reserv_free_page(m)) {
2732 #else
2733 	if (TRUE) {
2734 #endif
2735 		vm_phys_free_pages(m, 0);
2736 	}
2737 	vm_page_free_wakeup();
2738 	mtx_unlock(&vm_page_queue_free_mtx);
2739 
2740 	/*
2741 	 * Increment the vnode's hold count if this is the object's only
2742 	 * cached page.  Decrement the vnode's hold count if this was
2743 	 * the object's only resident page.
2744 	 */
2745 	if (object->type == OBJT_VNODE) {
2746 		if (cache_was_empty && object->resident_page_count != 0)
2747 			vhold(object->handle);
2748 		else if (!cache_was_empty && object->resident_page_count == 0)
2749 			vdrop(object->handle);
2750 	}
2751 }
2752 
2753 /*
2754  * vm_page_advise
2755  *
2756  * 	Deactivate or do nothing, as appropriate.
2757  *
2758  *	The object and page must be locked.
2759  */
2760 void
2761 vm_page_advise(vm_page_t m, int advice)
2762 {
2763 
2764 	vm_page_assert_locked(m);
2765 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2766 	if (advice == MADV_FREE)
2767 		/*
2768 		 * Mark the page clean.  This will allow the page to be freed
2769 		 * up by the system.  However, such pages are often reused
2770 		 * quickly by malloc() so we do not do anything that would
2771 		 * cause a page fault if we can help it.
2772 		 *
2773 		 * Specifically, we do not try to actually free the page now
2774 		 * nor do we try to put it in the cache (which would cause a
2775 		 * page fault on reuse).
2776 		 *
2777 		 * But we do make the page as freeable as we can without
2778 		 * actually taking the step of unmapping it.
2779 		 */
2780 		m->dirty = 0;
2781 	else if (advice != MADV_DONTNEED)
2782 		return;
2783 
2784 	/*
2785 	 * Clear any references to the page.  Otherwise, the page daemon will
2786 	 * immediately reactivate the page.
2787 	 */
2788 	vm_page_aflag_clear(m, PGA_REFERENCED);
2789 
2790 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
2791 		vm_page_dirty(m);
2792 
2793 	/*
2794 	 * Place clean pages at the head of the inactive queue rather than the
2795 	 * tail, thus defeating the queue's LRU operation and ensuring that the
2796 	 * page will be reused quickly.
2797 	 */
2798 	_vm_page_deactivate(m, m->dirty == 0);
2799 }
2800 
2801 /*
2802  * Grab a page, waiting until we are waken up due to the page
2803  * changing state.  We keep on waiting, if the page continues
2804  * to be in the object.  If the page doesn't exist, first allocate it
2805  * and then conditionally zero it.
2806  *
2807  * This routine may sleep.
2808  *
2809  * The object must be locked on entry.  The lock will, however, be released
2810  * and reacquired if the routine sleeps.
2811  */
2812 vm_page_t
2813 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
2814 {
2815 	vm_page_t m;
2816 	int sleep;
2817 
2818 	VM_OBJECT_ASSERT_WLOCKED(object);
2819 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
2820 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
2821 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
2822 retrylookup:
2823 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
2824 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
2825 		    vm_page_xbusied(m) : vm_page_busied(m);
2826 		if (sleep) {
2827 			if ((allocflags & VM_ALLOC_NOWAIT) != 0)
2828 				return (NULL);
2829 			/*
2830 			 * Reference the page before unlocking and
2831 			 * sleeping so that the page daemon is less
2832 			 * likely to reclaim it.
2833 			 */
2834 			vm_page_aflag_set(m, PGA_REFERENCED);
2835 			vm_page_lock(m);
2836 			VM_OBJECT_WUNLOCK(object);
2837 			vm_page_busy_sleep(m, "pgrbwt");
2838 			VM_OBJECT_WLOCK(object);
2839 			goto retrylookup;
2840 		} else {
2841 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
2842 				vm_page_lock(m);
2843 				vm_page_wire(m);
2844 				vm_page_unlock(m);
2845 			}
2846 			if ((allocflags &
2847 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
2848 				vm_page_xbusy(m);
2849 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
2850 				vm_page_sbusy(m);
2851 			return (m);
2852 		}
2853 	}
2854 	m = vm_page_alloc(object, pindex, allocflags);
2855 	if (m == NULL) {
2856 		if ((allocflags & VM_ALLOC_NOWAIT) != 0)
2857 			return (NULL);
2858 		VM_OBJECT_WUNLOCK(object);
2859 		VM_WAIT;
2860 		VM_OBJECT_WLOCK(object);
2861 		goto retrylookup;
2862 	} else if (m->valid != 0)
2863 		return (m);
2864 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
2865 		pmap_zero_page(m);
2866 	return (m);
2867 }
2868 
2869 /*
2870  * Mapping function for valid or dirty bits in a page.
2871  *
2872  * Inputs are required to range within a page.
2873  */
2874 vm_page_bits_t
2875 vm_page_bits(int base, int size)
2876 {
2877 	int first_bit;
2878 	int last_bit;
2879 
2880 	KASSERT(
2881 	    base + size <= PAGE_SIZE,
2882 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
2883 	);
2884 
2885 	if (size == 0)		/* handle degenerate case */
2886 		return (0);
2887 
2888 	first_bit = base >> DEV_BSHIFT;
2889 	last_bit = (base + size - 1) >> DEV_BSHIFT;
2890 
2891 	return (((vm_page_bits_t)2 << last_bit) -
2892 	    ((vm_page_bits_t)1 << first_bit));
2893 }
2894 
2895 /*
2896  *	vm_page_set_valid_range:
2897  *
2898  *	Sets portions of a page valid.  The arguments are expected
2899  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
2900  *	of any partial chunks touched by the range.  The invalid portion of
2901  *	such chunks will be zeroed.
2902  *
2903  *	(base + size) must be less then or equal to PAGE_SIZE.
2904  */
2905 void
2906 vm_page_set_valid_range(vm_page_t m, int base, int size)
2907 {
2908 	int endoff, frag;
2909 
2910 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2911 	if (size == 0)	/* handle degenerate case */
2912 		return;
2913 
2914 	/*
2915 	 * If the base is not DEV_BSIZE aligned and the valid
2916 	 * bit is clear, we have to zero out a portion of the
2917 	 * first block.
2918 	 */
2919 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
2920 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
2921 		pmap_zero_page_area(m, frag, base - frag);
2922 
2923 	/*
2924 	 * If the ending offset is not DEV_BSIZE aligned and the
2925 	 * valid bit is clear, we have to zero out a portion of
2926 	 * the last block.
2927 	 */
2928 	endoff = base + size;
2929 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
2930 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
2931 		pmap_zero_page_area(m, endoff,
2932 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
2933 
2934 	/*
2935 	 * Assert that no previously invalid block that is now being validated
2936 	 * is already dirty.
2937 	 */
2938 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
2939 	    ("vm_page_set_valid_range: page %p is dirty", m));
2940 
2941 	/*
2942 	 * Set valid bits inclusive of any overlap.
2943 	 */
2944 	m->valid |= vm_page_bits(base, size);
2945 }
2946 
2947 /*
2948  * Clear the given bits from the specified page's dirty field.
2949  */
2950 static __inline void
2951 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
2952 {
2953 	uintptr_t addr;
2954 #if PAGE_SIZE < 16384
2955 	int shift;
2956 #endif
2957 
2958 	/*
2959 	 * If the object is locked and the page is neither exclusive busy nor
2960 	 * write mapped, then the page's dirty field cannot possibly be
2961 	 * set by a concurrent pmap operation.
2962 	 */
2963 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2964 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
2965 		m->dirty &= ~pagebits;
2966 	else {
2967 		/*
2968 		 * The pmap layer can call vm_page_dirty() without
2969 		 * holding a distinguished lock.  The combination of
2970 		 * the object's lock and an atomic operation suffice
2971 		 * to guarantee consistency of the page dirty field.
2972 		 *
2973 		 * For PAGE_SIZE == 32768 case, compiler already
2974 		 * properly aligns the dirty field, so no forcible
2975 		 * alignment is needed. Only require existence of
2976 		 * atomic_clear_64 when page size is 32768.
2977 		 */
2978 		addr = (uintptr_t)&m->dirty;
2979 #if PAGE_SIZE == 32768
2980 		atomic_clear_64((uint64_t *)addr, pagebits);
2981 #elif PAGE_SIZE == 16384
2982 		atomic_clear_32((uint32_t *)addr, pagebits);
2983 #else		/* PAGE_SIZE <= 8192 */
2984 		/*
2985 		 * Use a trick to perform a 32-bit atomic on the
2986 		 * containing aligned word, to not depend on the existence
2987 		 * of atomic_clear_{8, 16}.
2988 		 */
2989 		shift = addr & (sizeof(uint32_t) - 1);
2990 #if BYTE_ORDER == BIG_ENDIAN
2991 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
2992 #else
2993 		shift *= NBBY;
2994 #endif
2995 		addr &= ~(sizeof(uint32_t) - 1);
2996 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
2997 #endif		/* PAGE_SIZE */
2998 	}
2999 }
3000 
3001 /*
3002  *	vm_page_set_validclean:
3003  *
3004  *	Sets portions of a page valid and clean.  The arguments are expected
3005  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3006  *	of any partial chunks touched by the range.  The invalid portion of
3007  *	such chunks will be zero'd.
3008  *
3009  *	(base + size) must be less then or equal to PAGE_SIZE.
3010  */
3011 void
3012 vm_page_set_validclean(vm_page_t m, int base, int size)
3013 {
3014 	vm_page_bits_t oldvalid, pagebits;
3015 	int endoff, frag;
3016 
3017 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3018 	if (size == 0)	/* handle degenerate case */
3019 		return;
3020 
3021 	/*
3022 	 * If the base is not DEV_BSIZE aligned and the valid
3023 	 * bit is clear, we have to zero out a portion of the
3024 	 * first block.
3025 	 */
3026 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
3027 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3028 		pmap_zero_page_area(m, frag, base - frag);
3029 
3030 	/*
3031 	 * If the ending offset is not DEV_BSIZE aligned and the
3032 	 * valid bit is clear, we have to zero out a portion of
3033 	 * the last block.
3034 	 */
3035 	endoff = base + size;
3036 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
3037 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3038 		pmap_zero_page_area(m, endoff,
3039 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3040 
3041 	/*
3042 	 * Set valid, clear dirty bits.  If validating the entire
3043 	 * page we can safely clear the pmap modify bit.  We also
3044 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
3045 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
3046 	 * be set again.
3047 	 *
3048 	 * We set valid bits inclusive of any overlap, but we can only
3049 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
3050 	 * the range.
3051 	 */
3052 	oldvalid = m->valid;
3053 	pagebits = vm_page_bits(base, size);
3054 	m->valid |= pagebits;
3055 #if 0	/* NOT YET */
3056 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3057 		frag = DEV_BSIZE - frag;
3058 		base += frag;
3059 		size -= frag;
3060 		if (size < 0)
3061 			size = 0;
3062 	}
3063 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3064 #endif
3065 	if (base == 0 && size == PAGE_SIZE) {
3066 		/*
3067 		 * The page can only be modified within the pmap if it is
3068 		 * mapped, and it can only be mapped if it was previously
3069 		 * fully valid.
3070 		 */
3071 		if (oldvalid == VM_PAGE_BITS_ALL)
3072 			/*
3073 			 * Perform the pmap_clear_modify() first.  Otherwise,
3074 			 * a concurrent pmap operation, such as
3075 			 * pmap_protect(), could clear a modification in the
3076 			 * pmap and set the dirty field on the page before
3077 			 * pmap_clear_modify() had begun and after the dirty
3078 			 * field was cleared here.
3079 			 */
3080 			pmap_clear_modify(m);
3081 		m->dirty = 0;
3082 		m->oflags &= ~VPO_NOSYNC;
3083 	} else if (oldvalid != VM_PAGE_BITS_ALL)
3084 		m->dirty &= ~pagebits;
3085 	else
3086 		vm_page_clear_dirty_mask(m, pagebits);
3087 }
3088 
3089 void
3090 vm_page_clear_dirty(vm_page_t m, int base, int size)
3091 {
3092 
3093 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3094 }
3095 
3096 /*
3097  *	vm_page_set_invalid:
3098  *
3099  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
3100  *	valid and dirty bits for the effected areas are cleared.
3101  */
3102 void
3103 vm_page_set_invalid(vm_page_t m, int base, int size)
3104 {
3105 	vm_page_bits_t bits;
3106 	vm_object_t object;
3107 
3108 	object = m->object;
3109 	VM_OBJECT_ASSERT_WLOCKED(object);
3110 	if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3111 	    size >= object->un_pager.vnp.vnp_size)
3112 		bits = VM_PAGE_BITS_ALL;
3113 	else
3114 		bits = vm_page_bits(base, size);
3115 	if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
3116 	    bits != 0)
3117 		pmap_remove_all(m);
3118 	KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3119 	    !pmap_page_is_mapped(m),
3120 	    ("vm_page_set_invalid: page %p is mapped", m));
3121 	m->valid &= ~bits;
3122 	m->dirty &= ~bits;
3123 }
3124 
3125 /*
3126  * vm_page_zero_invalid()
3127  *
3128  *	The kernel assumes that the invalid portions of a page contain
3129  *	garbage, but such pages can be mapped into memory by user code.
3130  *	When this occurs, we must zero out the non-valid portions of the
3131  *	page so user code sees what it expects.
3132  *
3133  *	Pages are most often semi-valid when the end of a file is mapped
3134  *	into memory and the file's size is not page aligned.
3135  */
3136 void
3137 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3138 {
3139 	int b;
3140 	int i;
3141 
3142 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3143 	/*
3144 	 * Scan the valid bits looking for invalid sections that
3145 	 * must be zeroed.  Invalid sub-DEV_BSIZE'd areas ( where the
3146 	 * valid bit may be set ) have already been zeroed by
3147 	 * vm_page_set_validclean().
3148 	 */
3149 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3150 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3151 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3152 			if (i > b) {
3153 				pmap_zero_page_area(m,
3154 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3155 			}
3156 			b = i + 1;
3157 		}
3158 	}
3159 
3160 	/*
3161 	 * setvalid is TRUE when we can safely set the zero'd areas
3162 	 * as being valid.  We can do this if there are no cache consistancy
3163 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3164 	 */
3165 	if (setvalid)
3166 		m->valid = VM_PAGE_BITS_ALL;
3167 }
3168 
3169 /*
3170  *	vm_page_is_valid:
3171  *
3172  *	Is (partial) page valid?  Note that the case where size == 0
3173  *	will return FALSE in the degenerate case where the page is
3174  *	entirely invalid, and TRUE otherwise.
3175  */
3176 int
3177 vm_page_is_valid(vm_page_t m, int base, int size)
3178 {
3179 	vm_page_bits_t bits;
3180 
3181 	VM_OBJECT_ASSERT_LOCKED(m->object);
3182 	bits = vm_page_bits(base, size);
3183 	return (m->valid != 0 && (m->valid & bits) == bits);
3184 }
3185 
3186 /*
3187  *	vm_page_ps_is_valid:
3188  *
3189  *	Returns TRUE if the entire (super)page is valid and FALSE otherwise.
3190  */
3191 boolean_t
3192 vm_page_ps_is_valid(vm_page_t m)
3193 {
3194 	int i, npages;
3195 
3196 	VM_OBJECT_ASSERT_LOCKED(m->object);
3197 	npages = atop(pagesizes[m->psind]);
3198 
3199 	/*
3200 	 * The physically contiguous pages that make up a superpage, i.e., a
3201 	 * page with a page size index ("psind") greater than zero, will
3202 	 * occupy adjacent entries in vm_page_array[].
3203 	 */
3204 	for (i = 0; i < npages; i++) {
3205 		if (m[i].valid != VM_PAGE_BITS_ALL)
3206 			return (FALSE);
3207 	}
3208 	return (TRUE);
3209 }
3210 
3211 /*
3212  * Set the page's dirty bits if the page is modified.
3213  */
3214 void
3215 vm_page_test_dirty(vm_page_t m)
3216 {
3217 
3218 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3219 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3220 		vm_page_dirty(m);
3221 }
3222 
3223 void
3224 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3225 {
3226 
3227 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3228 }
3229 
3230 void
3231 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3232 {
3233 
3234 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3235 }
3236 
3237 int
3238 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3239 {
3240 
3241 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3242 }
3243 
3244 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3245 void
3246 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3247 {
3248 
3249 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3250 }
3251 
3252 void
3253 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3254 {
3255 
3256 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3257 }
3258 #endif
3259 
3260 #ifdef INVARIANTS
3261 void
3262 vm_page_object_lock_assert(vm_page_t m)
3263 {
3264 
3265 	/*
3266 	 * Certain of the page's fields may only be modified by the
3267 	 * holder of the containing object's lock or the exclusive busy.
3268 	 * holder.  Unfortunately, the holder of the write busy is
3269 	 * not recorded, and thus cannot be checked here.
3270 	 */
3271 	if (m->object != NULL && !vm_page_xbusied(m))
3272 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3273 }
3274 
3275 void
3276 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
3277 {
3278 
3279 	if ((bits & PGA_WRITEABLE) == 0)
3280 		return;
3281 
3282 	/*
3283 	 * The PGA_WRITEABLE flag can only be set if the page is
3284 	 * managed, is exclusively busied or the object is locked.
3285 	 * Currently, this flag is only set by pmap_enter().
3286 	 */
3287 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3288 	    ("PGA_WRITEABLE on unmanaged page"));
3289 	if (!vm_page_xbusied(m))
3290 		VM_OBJECT_ASSERT_LOCKED(m->object);
3291 }
3292 #endif
3293 
3294 #include "opt_ddb.h"
3295 #ifdef DDB
3296 #include <sys/kernel.h>
3297 
3298 #include <ddb/ddb.h>
3299 
3300 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3301 {
3302 	db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count);
3303 	db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count);
3304 	db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count);
3305 	db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count);
3306 	db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count);
3307 	db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
3308 	db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
3309 	db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
3310 	db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
3311 }
3312 
3313 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3314 {
3315 	int dom;
3316 
3317 	db_printf("pq_free %d pq_cache %d\n",
3318 	    vm_cnt.v_free_count, vm_cnt.v_cache_count);
3319 	for (dom = 0; dom < vm_ndomains; dom++) {
3320 		db_printf(
3321 	"dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n",
3322 		    dom,
3323 		    vm_dom[dom].vmd_page_count,
3324 		    vm_dom[dom].vmd_free_count,
3325 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3326 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3327 		    vm_dom[dom].vmd_pass);
3328 	}
3329 }
3330 
3331 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3332 {
3333 	vm_page_t m;
3334 	boolean_t phys;
3335 
3336 	if (!have_addr) {
3337 		db_printf("show pginfo addr\n");
3338 		return;
3339 	}
3340 
3341 	phys = strchr(modif, 'p') != NULL;
3342 	if (phys)
3343 		m = PHYS_TO_VM_PAGE(addr);
3344 	else
3345 		m = (vm_page_t)addr;
3346 	db_printf(
3347     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3348     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3349 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3350 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3351 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3352 }
3353 #endif /* DDB */
3354