xref: /freebsd/sys/vm/vm_page.c (revision d7149f4e5176e24d02ef4cc3a0636623153209f0)
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 
153 static uma_zone_t fakepg_zone;
154 
155 static struct vnode *vm_page_alloc_init(vm_page_t m);
156 static void vm_page_cache_turn_free(vm_page_t m);
157 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
158 static void vm_page_enqueue(uint8_t queue, vm_page_t m);
159 static void vm_page_init_fakepg(void *dummy);
160 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
161     vm_pindex_t pindex, vm_page_t mpred);
162 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
163     vm_page_t mpred);
164 
165 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL);
166 
167 static void
168 vm_page_init_fakepg(void *dummy)
169 {
170 
171 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
172 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
173 }
174 
175 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
176 #if PAGE_SIZE == 32768
177 #ifdef CTASSERT
178 CTASSERT(sizeof(u_long) >= 8);
179 #endif
180 #endif
181 
182 /*
183  * Try to acquire a physical address lock while a pmap is locked.  If we
184  * fail to trylock we unlock and lock the pmap directly and cache the
185  * locked pa in *locked.  The caller should then restart their loop in case
186  * the virtual to physical mapping has changed.
187  */
188 int
189 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
190 {
191 	vm_paddr_t lockpa;
192 
193 	lockpa = *locked;
194 	*locked = pa;
195 	if (lockpa) {
196 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
197 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
198 			return (0);
199 		PA_UNLOCK(lockpa);
200 	}
201 	if (PA_TRYLOCK(pa))
202 		return (0);
203 	PMAP_UNLOCK(pmap);
204 	atomic_add_int(&pa_tryrelock_restart, 1);
205 	PA_LOCK(pa);
206 	PMAP_LOCK(pmap);
207 	return (EAGAIN);
208 }
209 
210 /*
211  *	vm_set_page_size:
212  *
213  *	Sets the page size, perhaps based upon the memory
214  *	size.  Must be called before any use of page-size
215  *	dependent functions.
216  */
217 void
218 vm_set_page_size(void)
219 {
220 	if (vm_cnt.v_page_size == 0)
221 		vm_cnt.v_page_size = PAGE_SIZE;
222 	if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
223 		panic("vm_set_page_size: page size not a power of two");
224 }
225 
226 /*
227  *	vm_page_blacklist_next:
228  *
229  *	Find the next entry in the provided string of blacklist
230  *	addresses.  Entries are separated by space, comma, or newline.
231  *	If an invalid integer is encountered then the rest of the
232  *	string is skipped.  Updates the list pointer to the next
233  *	character, or NULL if the string is exhausted or invalid.
234  */
235 static vm_paddr_t
236 vm_page_blacklist_next(char **list, char *end)
237 {
238 	vm_paddr_t bad;
239 	char *cp, *pos;
240 
241 	if (list == NULL || *list == NULL)
242 		return (0);
243 	if (**list =='\0') {
244 		*list = NULL;
245 		return (0);
246 	}
247 
248 	/*
249 	 * If there's no end pointer then the buffer is coming from
250 	 * the kenv and we know it's null-terminated.
251 	 */
252 	if (end == NULL)
253 		end = *list + strlen(*list);
254 
255 	/* Ensure that strtoq() won't walk off the end */
256 	if (*end != '\0') {
257 		if (*end == '\n' || *end == ' ' || *end  == ',')
258 			*end = '\0';
259 		else {
260 			printf("Blacklist not terminated, skipping\n");
261 			*list = NULL;
262 			return (0);
263 		}
264 	}
265 
266 	for (pos = *list; *pos != '\0'; pos = cp) {
267 		bad = strtoq(pos, &cp, 0);
268 		if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
269 			if (bad == 0) {
270 				if (++cp < end)
271 					continue;
272 				else
273 					break;
274 			}
275 		} else
276 			break;
277 		if (*cp == '\0' || ++cp >= end)
278 			*list = NULL;
279 		else
280 			*list = cp;
281 		return (trunc_page(bad));
282 	}
283 	printf("Garbage in RAM blacklist, skipping\n");
284 	*list = NULL;
285 	return (0);
286 }
287 
288 /*
289  *	vm_page_blacklist_check:
290  *
291  *	Iterate through the provided string of blacklist addresses, pulling
292  *	each entry out of the physical allocator free list and putting it
293  *	onto a list for reporting via the vm.page_blacklist sysctl.
294  */
295 static void
296 vm_page_blacklist_check(char *list, char *end)
297 {
298 	vm_paddr_t pa;
299 	vm_page_t m;
300 	char *next;
301 	int ret;
302 
303 	next = list;
304 	while (next != NULL) {
305 		if ((pa = vm_page_blacklist_next(&next, end)) == 0)
306 			continue;
307 		m = vm_phys_paddr_to_vm_page(pa);
308 		if (m == NULL)
309 			continue;
310 		mtx_lock(&vm_page_queue_free_mtx);
311 		ret = vm_phys_unfree_page(m);
312 		mtx_unlock(&vm_page_queue_free_mtx);
313 		if (ret == TRUE) {
314 			TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
315 			if (bootverbose)
316 				printf("Skipping page with pa 0x%jx\n",
317 				    (uintmax_t)pa);
318 		}
319 	}
320 }
321 
322 /*
323  *	vm_page_blacklist_load:
324  *
325  *	Search for a special module named "ram_blacklist".  It'll be a
326  *	plain text file provided by the user via the loader directive
327  *	of the same name.
328  */
329 static void
330 vm_page_blacklist_load(char **list, char **end)
331 {
332 	void *mod;
333 	u_char *ptr;
334 	u_int len;
335 
336 	mod = NULL;
337 	ptr = NULL;
338 
339 	mod = preload_search_by_type("ram_blacklist");
340 	if (mod != NULL) {
341 		ptr = preload_fetch_addr(mod);
342 		len = preload_fetch_size(mod);
343         }
344 	*list = ptr;
345 	if (ptr != NULL)
346 		*end = ptr + len;
347 	else
348 		*end = NULL;
349 	return;
350 }
351 
352 static int
353 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
354 {
355 	vm_page_t m;
356 	struct sbuf sbuf;
357 	int error, first;
358 
359 	first = 1;
360 	error = sysctl_wire_old_buffer(req, 0);
361 	if (error != 0)
362 		return (error);
363 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
364 	TAILQ_FOREACH(m, &blacklist_head, listq) {
365 		sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
366 		    (uintmax_t)m->phys_addr);
367 		first = 0;
368 	}
369 	error = sbuf_finish(&sbuf);
370 	sbuf_delete(&sbuf);
371 	return (error);
372 }
373 
374 static void
375 vm_page_domain_init(struct vm_domain *vmd)
376 {
377 	struct vm_pagequeue *pq;
378 	int i;
379 
380 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
381 	    "vm inactive pagequeue";
382 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
383 	    &vm_cnt.v_inactive_count;
384 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
385 	    "vm active pagequeue";
386 	*__DECONST(int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
387 	    &vm_cnt.v_active_count;
388 	vmd->vmd_page_count = 0;
389 	vmd->vmd_free_count = 0;
390 	vmd->vmd_segs = 0;
391 	vmd->vmd_oom = FALSE;
392 	vmd->vmd_pass = 0;
393 	for (i = 0; i < PQ_COUNT; i++) {
394 		pq = &vmd->vmd_pagequeues[i];
395 		TAILQ_INIT(&pq->pq_pl);
396 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
397 		    MTX_DEF | MTX_DUPOK);
398 	}
399 }
400 
401 /*
402  *	vm_page_startup:
403  *
404  *	Initializes the resident memory module.
405  *
406  *	Allocates memory for the page cells, and
407  *	for the object/offset-to-page hash table headers.
408  *	Each page cell is initialized and placed on the free list.
409  */
410 vm_offset_t
411 vm_page_startup(vm_offset_t vaddr)
412 {
413 	vm_offset_t mapped;
414 	vm_paddr_t page_range;
415 	vm_paddr_t new_end;
416 	int i;
417 	vm_paddr_t pa;
418 	vm_paddr_t last_pa;
419 	char *list, *listend;
420 	vm_paddr_t end;
421 	vm_paddr_t biggestsize;
422 	vm_paddr_t low_water, high_water;
423 	int biggestone;
424 
425 	biggestsize = 0;
426 	biggestone = 0;
427 	vaddr = round_page(vaddr);
428 
429 	for (i = 0; phys_avail[i + 1]; i += 2) {
430 		phys_avail[i] = round_page(phys_avail[i]);
431 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
432 	}
433 
434 	low_water = phys_avail[0];
435 	high_water = phys_avail[1];
436 
437 	for (i = 0; i < vm_phys_nsegs; i++) {
438 		if (vm_phys_segs[i].start < low_water)
439 			low_water = vm_phys_segs[i].start;
440 		if (vm_phys_segs[i].end > high_water)
441 			high_water = vm_phys_segs[i].end;
442 	}
443 	for (i = 0; phys_avail[i + 1]; i += 2) {
444 		vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
445 
446 		if (size > biggestsize) {
447 			biggestone = i;
448 			biggestsize = size;
449 		}
450 		if (phys_avail[i] < low_water)
451 			low_water = phys_avail[i];
452 		if (phys_avail[i + 1] > high_water)
453 			high_water = phys_avail[i + 1];
454 	}
455 
456 	end = phys_avail[biggestone+1];
457 
458 	/*
459 	 * Initialize the page and queue locks.
460 	 */
461 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
462 	for (i = 0; i < PA_LOCK_COUNT; i++)
463 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
464 	for (i = 0; i < vm_ndomains; i++)
465 		vm_page_domain_init(&vm_dom[i]);
466 
467 	/*
468 	 * Allocate memory for use when boot strapping the kernel memory
469 	 * allocator.
470 	 *
471 	 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
472 	 * manually fetch the value.
473 	 */
474 	TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
475 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
476 	new_end = trunc_page(new_end);
477 	mapped = pmap_map(&vaddr, new_end, end,
478 	    VM_PROT_READ | VM_PROT_WRITE);
479 	bzero((void *)mapped, end - new_end);
480 	uma_startup((void *)mapped, boot_pages);
481 
482 #if defined(__amd64__) || defined(__i386__) || defined(__arm__) || \
483     defined(__mips__)
484 	/*
485 	 * Allocate a bitmap to indicate that a random physical page
486 	 * needs to be included in a minidump.
487 	 *
488 	 * The amd64 port needs this to indicate which direct map pages
489 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
490 	 *
491 	 * However, i386 still needs this workspace internally within the
492 	 * minidump code.  In theory, they are not needed on i386, but are
493 	 * included should the sf_buf code decide to use them.
494 	 */
495 	last_pa = 0;
496 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
497 		if (dump_avail[i + 1] > last_pa)
498 			last_pa = dump_avail[i + 1];
499 	page_range = last_pa / PAGE_SIZE;
500 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
501 	new_end -= vm_page_dump_size;
502 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
503 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
504 	bzero((void *)vm_page_dump, vm_page_dump_size);
505 #endif
506 #ifdef __amd64__
507 	/*
508 	 * Request that the physical pages underlying the message buffer be
509 	 * included in a crash dump.  Since the message buffer is accessed
510 	 * through the direct map, they are not automatically included.
511 	 */
512 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
513 	last_pa = pa + round_page(msgbufsize);
514 	while (pa < last_pa) {
515 		dump_add_page(pa);
516 		pa += PAGE_SIZE;
517 	}
518 #endif
519 	/*
520 	 * Compute the number of pages of memory that will be available for
521 	 * use (taking into account the overhead of a page structure per
522 	 * page).
523 	 */
524 	first_page = low_water / PAGE_SIZE;
525 #ifdef VM_PHYSSEG_SPARSE
526 	page_range = 0;
527 	for (i = 0; i < vm_phys_nsegs; i++) {
528 		page_range += atop(vm_phys_segs[i].end -
529 		    vm_phys_segs[i].start);
530 	}
531 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
532 		page_range += atop(phys_avail[i + 1] - phys_avail[i]);
533 #elif defined(VM_PHYSSEG_DENSE)
534 	page_range = high_water / PAGE_SIZE - first_page;
535 #else
536 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
537 #endif
538 	end = new_end;
539 
540 	/*
541 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
542 	 */
543 	vaddr += PAGE_SIZE;
544 
545 	/*
546 	 * Initialize the mem entry structures now, and put them in the free
547 	 * queue.
548 	 */
549 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
550 	mapped = pmap_map(&vaddr, new_end, end,
551 	    VM_PROT_READ | VM_PROT_WRITE);
552 	vm_page_array = (vm_page_t) mapped;
553 #if VM_NRESERVLEVEL > 0
554 	/*
555 	 * Allocate memory for the reservation management system's data
556 	 * structures.
557 	 */
558 	new_end = vm_reserv_startup(&vaddr, new_end, high_water);
559 #endif
560 #if defined(__amd64__) || defined(__mips__)
561 	/*
562 	 * pmap_map on amd64 and mips can come out of the direct-map, not kvm
563 	 * like i386, so the pages must be tracked for a crashdump to include
564 	 * this data.  This includes the vm_page_array and the early UMA
565 	 * bootstrap pages.
566 	 */
567 	for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
568 		dump_add_page(pa);
569 #endif
570 	phys_avail[biggestone + 1] = new_end;
571 
572 	/*
573 	 * Add physical memory segments corresponding to the available
574 	 * physical pages.
575 	 */
576 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
577 		vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
578 
579 	/*
580 	 * Clear all of the page structures
581 	 */
582 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
583 	for (i = 0; i < page_range; i++)
584 		vm_page_array[i].order = VM_NFREEORDER;
585 	vm_page_array_size = page_range;
586 
587 	/*
588 	 * Initialize the physical memory allocator.
589 	 */
590 	vm_phys_init();
591 
592 	/*
593 	 * Add every available physical page that is not blacklisted to
594 	 * the free lists.
595 	 */
596 	vm_cnt.v_page_count = 0;
597 	vm_cnt.v_free_count = 0;
598 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
599 		pa = phys_avail[i];
600 		last_pa = phys_avail[i + 1];
601 		while (pa < last_pa) {
602 			vm_phys_add_page(pa);
603 			pa += PAGE_SIZE;
604 		}
605 	}
606 
607 	TAILQ_INIT(&blacklist_head);
608 	vm_page_blacklist_load(&list, &listend);
609 	vm_page_blacklist_check(list, listend);
610 
611 	list = kern_getenv("vm.blacklist");
612 	vm_page_blacklist_check(list, NULL);
613 
614 	freeenv(list);
615 #if VM_NRESERVLEVEL > 0
616 	/*
617 	 * Initialize the reservation management system.
618 	 */
619 	vm_reserv_init();
620 #endif
621 	return (vaddr);
622 }
623 
624 void
625 vm_page_reference(vm_page_t m)
626 {
627 
628 	vm_page_aflag_set(m, PGA_REFERENCED);
629 }
630 
631 /*
632  *	vm_page_busy_downgrade:
633  *
634  *	Downgrade an exclusive busy page into a single shared busy page.
635  */
636 void
637 vm_page_busy_downgrade(vm_page_t m)
638 {
639 	u_int x;
640 
641 	vm_page_assert_xbusied(m);
642 
643 	for (;;) {
644 		x = m->busy_lock;
645 		x &= VPB_BIT_WAITERS;
646 		if (atomic_cmpset_rel_int(&m->busy_lock,
647 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x))
648 			break;
649 	}
650 }
651 
652 /*
653  *	vm_page_sbusied:
654  *
655  *	Return a positive value if the page is shared busied, 0 otherwise.
656  */
657 int
658 vm_page_sbusied(vm_page_t m)
659 {
660 	u_int x;
661 
662 	x = m->busy_lock;
663 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
664 }
665 
666 /*
667  *	vm_page_sunbusy:
668  *
669  *	Shared unbusy a page.
670  */
671 void
672 vm_page_sunbusy(vm_page_t m)
673 {
674 	u_int x;
675 
676 	vm_page_assert_sbusied(m);
677 
678 	for (;;) {
679 		x = m->busy_lock;
680 		if (VPB_SHARERS(x) > 1) {
681 			if (atomic_cmpset_int(&m->busy_lock, x,
682 			    x - VPB_ONE_SHARER))
683 				break;
684 			continue;
685 		}
686 		if ((x & VPB_BIT_WAITERS) == 0) {
687 			KASSERT(x == VPB_SHARERS_WORD(1),
688 			    ("vm_page_sunbusy: invalid lock state"));
689 			if (atomic_cmpset_int(&m->busy_lock,
690 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
691 				break;
692 			continue;
693 		}
694 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
695 		    ("vm_page_sunbusy: invalid lock state for waiters"));
696 
697 		vm_page_lock(m);
698 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
699 			vm_page_unlock(m);
700 			continue;
701 		}
702 		wakeup(m);
703 		vm_page_unlock(m);
704 		break;
705 	}
706 }
707 
708 /*
709  *	vm_page_busy_sleep:
710  *
711  *	Sleep and release the page lock, using the page pointer as wchan.
712  *	This is used to implement the hard-path of busying mechanism.
713  *
714  *	The given page must be locked.
715  */
716 void
717 vm_page_busy_sleep(vm_page_t m, const char *wmesg)
718 {
719 	u_int x;
720 
721 	vm_page_lock_assert(m, MA_OWNED);
722 
723 	x = m->busy_lock;
724 	if (x == VPB_UNBUSIED) {
725 		vm_page_unlock(m);
726 		return;
727 	}
728 	if ((x & VPB_BIT_WAITERS) == 0 &&
729 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) {
730 		vm_page_unlock(m);
731 		return;
732 	}
733 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
734 }
735 
736 /*
737  *	vm_page_trysbusy:
738  *
739  *	Try to shared busy a page.
740  *	If the operation succeeds 1 is returned otherwise 0.
741  *	The operation never sleeps.
742  */
743 int
744 vm_page_trysbusy(vm_page_t m)
745 {
746 	u_int x;
747 
748 	for (;;) {
749 		x = m->busy_lock;
750 		if ((x & VPB_BIT_SHARED) == 0)
751 			return (0);
752 		if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
753 			return (1);
754 	}
755 }
756 
757 /*
758  *	vm_page_xunbusy_hard:
759  *
760  *	Called after the first try the exclusive unbusy of a page failed.
761  *	It is assumed that the waiters bit is on.
762  */
763 void
764 vm_page_xunbusy_hard(vm_page_t m)
765 {
766 
767 	vm_page_assert_xbusied(m);
768 
769 	vm_page_lock(m);
770 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
771 	wakeup(m);
772 	vm_page_unlock(m);
773 }
774 
775 /*
776  *	vm_page_flash:
777  *
778  *	Wakeup anyone waiting for the page.
779  *	The ownership bits do not change.
780  *
781  *	The given page must be locked.
782  */
783 void
784 vm_page_flash(vm_page_t m)
785 {
786 	u_int x;
787 
788 	vm_page_lock_assert(m, MA_OWNED);
789 
790 	for (;;) {
791 		x = m->busy_lock;
792 		if ((x & VPB_BIT_WAITERS) == 0)
793 			return;
794 		if (atomic_cmpset_int(&m->busy_lock, x,
795 		    x & (~VPB_BIT_WAITERS)))
796 			break;
797 	}
798 	wakeup(m);
799 }
800 
801 /*
802  * Keep page from being freed by the page daemon
803  * much of the same effect as wiring, except much lower
804  * overhead and should be used only for *very* temporary
805  * holding ("wiring").
806  */
807 void
808 vm_page_hold(vm_page_t mem)
809 {
810 
811 	vm_page_lock_assert(mem, MA_OWNED);
812         mem->hold_count++;
813 }
814 
815 void
816 vm_page_unhold(vm_page_t mem)
817 {
818 
819 	vm_page_lock_assert(mem, MA_OWNED);
820 	KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
821 	--mem->hold_count;
822 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
823 		vm_page_free_toq(mem);
824 }
825 
826 /*
827  *	vm_page_unhold_pages:
828  *
829  *	Unhold each of the pages that is referenced by the given array.
830  */
831 void
832 vm_page_unhold_pages(vm_page_t *ma, int count)
833 {
834 	struct mtx *mtx, *new_mtx;
835 
836 	mtx = NULL;
837 	for (; count != 0; count--) {
838 		/*
839 		 * Avoid releasing and reacquiring the same page lock.
840 		 */
841 		new_mtx = vm_page_lockptr(*ma);
842 		if (mtx != new_mtx) {
843 			if (mtx != NULL)
844 				mtx_unlock(mtx);
845 			mtx = new_mtx;
846 			mtx_lock(mtx);
847 		}
848 		vm_page_unhold(*ma);
849 		ma++;
850 	}
851 	if (mtx != NULL)
852 		mtx_unlock(mtx);
853 }
854 
855 vm_page_t
856 PHYS_TO_VM_PAGE(vm_paddr_t pa)
857 {
858 	vm_page_t m;
859 
860 #ifdef VM_PHYSSEG_SPARSE
861 	m = vm_phys_paddr_to_vm_page(pa);
862 	if (m == NULL)
863 		m = vm_phys_fictitious_to_vm_page(pa);
864 	return (m);
865 #elif defined(VM_PHYSSEG_DENSE)
866 	long pi;
867 
868 	pi = atop(pa);
869 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
870 		m = &vm_page_array[pi - first_page];
871 		return (m);
872 	}
873 	return (vm_phys_fictitious_to_vm_page(pa));
874 #else
875 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
876 #endif
877 }
878 
879 /*
880  *	vm_page_getfake:
881  *
882  *	Create a fictitious page with the specified physical address and
883  *	memory attribute.  The memory attribute is the only the machine-
884  *	dependent aspect of a fictitious page that must be initialized.
885  */
886 vm_page_t
887 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
888 {
889 	vm_page_t m;
890 
891 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
892 	vm_page_initfake(m, paddr, memattr);
893 	return (m);
894 }
895 
896 void
897 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
898 {
899 
900 	if ((m->flags & PG_FICTITIOUS) != 0) {
901 		/*
902 		 * The page's memattr might have changed since the
903 		 * previous initialization.  Update the pmap to the
904 		 * new memattr.
905 		 */
906 		goto memattr;
907 	}
908 	m->phys_addr = paddr;
909 	m->queue = PQ_NONE;
910 	/* Fictitious pages don't use "segind". */
911 	m->flags = PG_FICTITIOUS;
912 	/* Fictitious pages don't use "order" or "pool". */
913 	m->oflags = VPO_UNMANAGED;
914 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
915 	m->wire_count = 1;
916 	pmap_page_init(m);
917 memattr:
918 	pmap_page_set_memattr(m, memattr);
919 }
920 
921 /*
922  *	vm_page_putfake:
923  *
924  *	Release a fictitious page.
925  */
926 void
927 vm_page_putfake(vm_page_t m)
928 {
929 
930 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
931 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
932 	    ("vm_page_putfake: bad page %p", m));
933 	uma_zfree(fakepg_zone, m);
934 }
935 
936 /*
937  *	vm_page_updatefake:
938  *
939  *	Update the given fictitious page to the specified physical address and
940  *	memory attribute.
941  */
942 void
943 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
944 {
945 
946 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
947 	    ("vm_page_updatefake: bad page %p", m));
948 	m->phys_addr = paddr;
949 	pmap_page_set_memattr(m, memattr);
950 }
951 
952 /*
953  *	vm_page_free:
954  *
955  *	Free a page.
956  */
957 void
958 vm_page_free(vm_page_t m)
959 {
960 
961 	m->flags &= ~PG_ZERO;
962 	vm_page_free_toq(m);
963 }
964 
965 /*
966  *	vm_page_free_zero:
967  *
968  *	Free a page to the zerod-pages queue
969  */
970 void
971 vm_page_free_zero(vm_page_t m)
972 {
973 
974 	m->flags |= PG_ZERO;
975 	vm_page_free_toq(m);
976 }
977 
978 /*
979  * Unbusy and handle the page queueing for a page from the VOP_GETPAGES()
980  * array which is not the request page.
981  */
982 void
983 vm_page_readahead_finish(vm_page_t m)
984 {
985 
986 	if (m->valid != 0) {
987 		/*
988 		 * Since the page is not the requested page, whether
989 		 * it should be activated or deactivated is not
990 		 * obvious.  Empirical results have shown that
991 		 * deactivating the page is usually the best choice,
992 		 * unless the page is wanted by another thread.
993 		 */
994 		vm_page_lock(m);
995 		if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
996 			vm_page_activate(m);
997 		else
998 			vm_page_deactivate(m);
999 		vm_page_unlock(m);
1000 		vm_page_xunbusy(m);
1001 	} else {
1002 		/*
1003 		 * Free the completely invalid page.  Such page state
1004 		 * occurs due to the short read operation which did
1005 		 * not covered our page at all, or in case when a read
1006 		 * error happens.
1007 		 */
1008 		vm_page_lock(m);
1009 		vm_page_free(m);
1010 		vm_page_unlock(m);
1011 	}
1012 }
1013 
1014 /*
1015  *	vm_page_sleep_if_busy:
1016  *
1017  *	Sleep and release the page queues lock if the page is busied.
1018  *	Returns TRUE if the thread slept.
1019  *
1020  *	The given page must be unlocked and object containing it must
1021  *	be locked.
1022  */
1023 int
1024 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1025 {
1026 	vm_object_t obj;
1027 
1028 	vm_page_lock_assert(m, MA_NOTOWNED);
1029 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1030 
1031 	if (vm_page_busied(m)) {
1032 		/*
1033 		 * The page-specific object must be cached because page
1034 		 * identity can change during the sleep, causing the
1035 		 * re-lock of a different object.
1036 		 * It is assumed that a reference to the object is already
1037 		 * held by the callers.
1038 		 */
1039 		obj = m->object;
1040 		vm_page_lock(m);
1041 		VM_OBJECT_WUNLOCK(obj);
1042 		vm_page_busy_sleep(m, msg);
1043 		VM_OBJECT_WLOCK(obj);
1044 		return (TRUE);
1045 	}
1046 	return (FALSE);
1047 }
1048 
1049 /*
1050  *	vm_page_dirty_KBI:		[ internal use only ]
1051  *
1052  *	Set all bits in the page's dirty field.
1053  *
1054  *	The object containing the specified page must be locked if the
1055  *	call is made from the machine-independent layer.
1056  *
1057  *	See vm_page_clear_dirty_mask().
1058  *
1059  *	This function should only be called by vm_page_dirty().
1060  */
1061 void
1062 vm_page_dirty_KBI(vm_page_t m)
1063 {
1064 
1065 	/* These assertions refer to this operation by its public name. */
1066 	KASSERT((m->flags & PG_CACHED) == 0,
1067 	    ("vm_page_dirty: page in cache!"));
1068 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
1069 	    ("vm_page_dirty: page is invalid!"));
1070 	m->dirty = VM_PAGE_BITS_ALL;
1071 }
1072 
1073 /*
1074  *	vm_page_insert:		[ internal use only ]
1075  *
1076  *	Inserts the given mem entry into the object and object list.
1077  *
1078  *	The object must be locked.
1079  */
1080 int
1081 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1082 {
1083 	vm_page_t mpred;
1084 
1085 	VM_OBJECT_ASSERT_WLOCKED(object);
1086 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
1087 	return (vm_page_insert_after(m, object, pindex, mpred));
1088 }
1089 
1090 /*
1091  *	vm_page_insert_after:
1092  *
1093  *	Inserts the page "m" into the specified object at offset "pindex".
1094  *
1095  *	The page "mpred" must immediately precede the offset "pindex" within
1096  *	the specified object.
1097  *
1098  *	The object must be locked.
1099  */
1100 static int
1101 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1102     vm_page_t mpred)
1103 {
1104 	vm_pindex_t sidx;
1105 	vm_object_t sobj;
1106 	vm_page_t msucc;
1107 
1108 	VM_OBJECT_ASSERT_WLOCKED(object);
1109 	KASSERT(m->object == NULL,
1110 	    ("vm_page_insert_after: page already inserted"));
1111 	if (mpred != NULL) {
1112 		KASSERT(mpred->object == object,
1113 		    ("vm_page_insert_after: object doesn't contain mpred"));
1114 		KASSERT(mpred->pindex < pindex,
1115 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1116 		msucc = TAILQ_NEXT(mpred, listq);
1117 	} else
1118 		msucc = TAILQ_FIRST(&object->memq);
1119 	if (msucc != NULL)
1120 		KASSERT(msucc->pindex > pindex,
1121 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
1122 
1123 	/*
1124 	 * Record the object/offset pair in this page
1125 	 */
1126 	sobj = m->object;
1127 	sidx = m->pindex;
1128 	m->object = object;
1129 	m->pindex = pindex;
1130 
1131 	/*
1132 	 * Now link into the object's ordered list of backed pages.
1133 	 */
1134 	if (vm_radix_insert(&object->rtree, m)) {
1135 		m->object = sobj;
1136 		m->pindex = sidx;
1137 		return (1);
1138 	}
1139 	vm_page_insert_radixdone(m, object, mpred);
1140 	return (0);
1141 }
1142 
1143 /*
1144  *	vm_page_insert_radixdone:
1145  *
1146  *	Complete page "m" insertion into the specified object after the
1147  *	radix trie hooking.
1148  *
1149  *	The page "mpred" must precede the offset "m->pindex" within the
1150  *	specified object.
1151  *
1152  *	The object must be locked.
1153  */
1154 static void
1155 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1156 {
1157 
1158 	VM_OBJECT_ASSERT_WLOCKED(object);
1159 	KASSERT(object != NULL && m->object == object,
1160 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1161 	if (mpred != NULL) {
1162 		KASSERT(mpred->object == object,
1163 		    ("vm_page_insert_after: object doesn't contain mpred"));
1164 		KASSERT(mpred->pindex < m->pindex,
1165 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1166 	}
1167 
1168 	if (mpred != NULL)
1169 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1170 	else
1171 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1172 
1173 	/*
1174 	 * Show that the object has one more resident page.
1175 	 */
1176 	object->resident_page_count++;
1177 
1178 	/*
1179 	 * Hold the vnode until the last page is released.
1180 	 */
1181 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1182 		vhold(object->handle);
1183 
1184 	/*
1185 	 * Since we are inserting a new and possibly dirty page,
1186 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1187 	 */
1188 	if (pmap_page_is_write_mapped(m))
1189 		vm_object_set_writeable_dirty(object);
1190 }
1191 
1192 /*
1193  *	vm_page_remove:
1194  *
1195  *	Removes the given mem entry from the object/offset-page
1196  *	table and the object page list, but do not invalidate/terminate
1197  *	the backing store.
1198  *
1199  *	The object must be locked.  The page must be locked if it is managed.
1200  */
1201 void
1202 vm_page_remove(vm_page_t m)
1203 {
1204 	vm_object_t object;
1205 	boolean_t lockacq;
1206 
1207 	if ((m->oflags & VPO_UNMANAGED) == 0)
1208 		vm_page_lock_assert(m, MA_OWNED);
1209 	if ((object = m->object) == NULL)
1210 		return;
1211 	VM_OBJECT_ASSERT_WLOCKED(object);
1212 	if (vm_page_xbusied(m)) {
1213 		lockacq = FALSE;
1214 		if ((m->oflags & VPO_UNMANAGED) != 0 &&
1215 		    !mtx_owned(vm_page_lockptr(m))) {
1216 			lockacq = TRUE;
1217 			vm_page_lock(m);
1218 		}
1219 		vm_page_flash(m);
1220 		atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
1221 		if (lockacq)
1222 			vm_page_unlock(m);
1223 	}
1224 
1225 	/*
1226 	 * Now remove from the object's list of backed pages.
1227 	 */
1228 	vm_radix_remove(&object->rtree, m->pindex);
1229 	TAILQ_REMOVE(&object->memq, m, listq);
1230 
1231 	/*
1232 	 * And show that the object has one fewer resident page.
1233 	 */
1234 	object->resident_page_count--;
1235 
1236 	/*
1237 	 * The vnode may now be recycled.
1238 	 */
1239 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1240 		vdrop(object->handle);
1241 
1242 	m->object = NULL;
1243 }
1244 
1245 /*
1246  *	vm_page_lookup:
1247  *
1248  *	Returns the page associated with the object/offset
1249  *	pair specified; if none is found, NULL is returned.
1250  *
1251  *	The object must be locked.
1252  */
1253 vm_page_t
1254 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1255 {
1256 
1257 	VM_OBJECT_ASSERT_LOCKED(object);
1258 	return (vm_radix_lookup(&object->rtree, pindex));
1259 }
1260 
1261 /*
1262  *	vm_page_find_least:
1263  *
1264  *	Returns the page associated with the object with least pindex
1265  *	greater than or equal to the parameter pindex, or NULL.
1266  *
1267  *	The object must be locked.
1268  */
1269 vm_page_t
1270 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1271 {
1272 	vm_page_t m;
1273 
1274 	VM_OBJECT_ASSERT_LOCKED(object);
1275 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1276 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1277 	return (m);
1278 }
1279 
1280 /*
1281  * Returns the given page's successor (by pindex) within the object if it is
1282  * resident; if none is found, NULL is returned.
1283  *
1284  * The object must be locked.
1285  */
1286 vm_page_t
1287 vm_page_next(vm_page_t m)
1288 {
1289 	vm_page_t next;
1290 
1291 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1292 	if ((next = TAILQ_NEXT(m, listq)) != NULL &&
1293 	    next->pindex != m->pindex + 1)
1294 		next = NULL;
1295 	return (next);
1296 }
1297 
1298 /*
1299  * Returns the given page's predecessor (by pindex) within the object if it is
1300  * resident; if none is found, NULL is returned.
1301  *
1302  * The object must be locked.
1303  */
1304 vm_page_t
1305 vm_page_prev(vm_page_t m)
1306 {
1307 	vm_page_t prev;
1308 
1309 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1310 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL &&
1311 	    prev->pindex != m->pindex - 1)
1312 		prev = NULL;
1313 	return (prev);
1314 }
1315 
1316 /*
1317  * Uses the page mnew as a replacement for an existing page at index
1318  * pindex which must be already present in the object.
1319  *
1320  * The existing page must not be on a paging queue.
1321  */
1322 vm_page_t
1323 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1324 {
1325 	vm_page_t mold, mpred;
1326 
1327 	VM_OBJECT_ASSERT_WLOCKED(object);
1328 
1329 	/*
1330 	 * This function mostly follows vm_page_insert() and
1331 	 * vm_page_remove() without the radix, object count and vnode
1332 	 * dance.  Double check such functions for more comments.
1333 	 */
1334 	mpred = vm_radix_lookup(&object->rtree, pindex);
1335 	KASSERT(mpred != NULL,
1336 	    ("vm_page_replace: replacing page not present with pindex"));
1337 	mpred = TAILQ_PREV(mpred, respgs, listq);
1338 	if (mpred != NULL)
1339 		KASSERT(mpred->pindex < pindex,
1340 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1341 
1342 	mnew->object = object;
1343 	mnew->pindex = pindex;
1344 	mold = vm_radix_replace(&object->rtree, mnew);
1345 	KASSERT(mold->queue == PQ_NONE,
1346 	    ("vm_page_replace: mold is on a paging queue"));
1347 
1348 	/* Detach the old page from the resident tailq. */
1349 	TAILQ_REMOVE(&object->memq, mold, listq);
1350 
1351 	mold->object = NULL;
1352 	vm_page_xunbusy(mold);
1353 
1354 	/* Insert the new page in the resident tailq. */
1355 	if (mpred != NULL)
1356 		TAILQ_INSERT_AFTER(&object->memq, mpred, mnew, listq);
1357 	else
1358 		TAILQ_INSERT_HEAD(&object->memq, mnew, listq);
1359 	if (pmap_page_is_write_mapped(mnew))
1360 		vm_object_set_writeable_dirty(object);
1361 	return (mold);
1362 }
1363 
1364 /*
1365  *	vm_page_rename:
1366  *
1367  *	Move the given memory entry from its
1368  *	current object to the specified target object/offset.
1369  *
1370  *	Note: swap associated with the page must be invalidated by the move.  We
1371  *	      have to do this for several reasons:  (1) we aren't freeing the
1372  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1373  *	      moving the page from object A to B, and will then later move
1374  *	      the backing store from A to B and we can't have a conflict.
1375  *
1376  *	Note: we *always* dirty the page.  It is necessary both for the
1377  *	      fact that we moved it, and because we may be invalidating
1378  *	      swap.  If the page is on the cache, we have to deactivate it
1379  *	      or vm_page_dirty() will panic.  Dirty pages are not allowed
1380  *	      on the cache.
1381  *
1382  *	The objects must be locked.
1383  */
1384 int
1385 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1386 {
1387 	vm_page_t mpred;
1388 	vm_pindex_t opidx;
1389 
1390 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1391 
1392 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1393 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1394 	    ("vm_page_rename: pindex already renamed"));
1395 
1396 	/*
1397 	 * Create a custom version of vm_page_insert() which does not depend
1398 	 * by m_prev and can cheat on the implementation aspects of the
1399 	 * function.
1400 	 */
1401 	opidx = m->pindex;
1402 	m->pindex = new_pindex;
1403 	if (vm_radix_insert(&new_object->rtree, m)) {
1404 		m->pindex = opidx;
1405 		return (1);
1406 	}
1407 
1408 	/*
1409 	 * The operation cannot fail anymore.  The removal must happen before
1410 	 * the listq iterator is tainted.
1411 	 */
1412 	m->pindex = opidx;
1413 	vm_page_lock(m);
1414 	vm_page_remove(m);
1415 
1416 	/* Return back to the new pindex to complete vm_page_insert(). */
1417 	m->pindex = new_pindex;
1418 	m->object = new_object;
1419 	vm_page_unlock(m);
1420 	vm_page_insert_radixdone(m, new_object, mpred);
1421 	vm_page_dirty(m);
1422 	return (0);
1423 }
1424 
1425 /*
1426  *	Convert all of the given object's cached pages that have a
1427  *	pindex within the given range into free pages.  If the value
1428  *	zero is given for "end", then the range's upper bound is
1429  *	infinity.  If the given object is backed by a vnode and it
1430  *	transitions from having one or more cached pages to none, the
1431  *	vnode's hold count is reduced.
1432  */
1433 void
1434 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1435 {
1436 	vm_page_t m;
1437 	boolean_t empty;
1438 
1439 	mtx_lock(&vm_page_queue_free_mtx);
1440 	if (__predict_false(vm_radix_is_empty(&object->cache))) {
1441 		mtx_unlock(&vm_page_queue_free_mtx);
1442 		return;
1443 	}
1444 	while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) {
1445 		if (end != 0 && m->pindex >= end)
1446 			break;
1447 		vm_radix_remove(&object->cache, m->pindex);
1448 		vm_page_cache_turn_free(m);
1449 	}
1450 	empty = vm_radix_is_empty(&object->cache);
1451 	mtx_unlock(&vm_page_queue_free_mtx);
1452 	if (object->type == OBJT_VNODE && empty)
1453 		vdrop(object->handle);
1454 }
1455 
1456 /*
1457  *	Returns the cached page that is associated with the given
1458  *	object and offset.  If, however, none exists, returns NULL.
1459  *
1460  *	The free page queue must be locked.
1461  */
1462 static inline vm_page_t
1463 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
1464 {
1465 
1466 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1467 	return (vm_radix_lookup(&object->cache, pindex));
1468 }
1469 
1470 /*
1471  *	Remove the given cached page from its containing object's
1472  *	collection of cached pages.
1473  *
1474  *	The free page queue must be locked.
1475  */
1476 static void
1477 vm_page_cache_remove(vm_page_t m)
1478 {
1479 
1480 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1481 	KASSERT((m->flags & PG_CACHED) != 0,
1482 	    ("vm_page_cache_remove: page %p is not cached", m));
1483 	vm_radix_remove(&m->object->cache, m->pindex);
1484 	m->object = NULL;
1485 	vm_cnt.v_cache_count--;
1486 }
1487 
1488 /*
1489  *	Transfer all of the cached pages with offset greater than or
1490  *	equal to 'offidxstart' from the original object's cache to the
1491  *	new object's cache.  However, any cached pages with offset
1492  *	greater than or equal to the new object's size are kept in the
1493  *	original object.  Initially, the new object's cache must be
1494  *	empty.  Offset 'offidxstart' in the original object must
1495  *	correspond to offset zero in the new object.
1496  *
1497  *	The new object must be locked.
1498  */
1499 void
1500 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
1501     vm_object_t new_object)
1502 {
1503 	vm_page_t m;
1504 
1505 	/*
1506 	 * Insertion into an object's collection of cached pages
1507 	 * requires the object to be locked.  In contrast, removal does
1508 	 * not.
1509 	 */
1510 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1511 	KASSERT(vm_radix_is_empty(&new_object->cache),
1512 	    ("vm_page_cache_transfer: object %p has cached pages",
1513 	    new_object));
1514 	mtx_lock(&vm_page_queue_free_mtx);
1515 	while ((m = vm_radix_lookup_ge(&orig_object->cache,
1516 	    offidxstart)) != NULL) {
1517 		/*
1518 		 * Transfer all of the pages with offset greater than or
1519 		 * equal to 'offidxstart' from the original object's
1520 		 * cache to the new object's cache.
1521 		 */
1522 		if ((m->pindex - offidxstart) >= new_object->size)
1523 			break;
1524 		vm_radix_remove(&orig_object->cache, m->pindex);
1525 		/* Update the page's object and offset. */
1526 		m->object = new_object;
1527 		m->pindex -= offidxstart;
1528 		if (vm_radix_insert(&new_object->cache, m))
1529 			vm_page_cache_turn_free(m);
1530 	}
1531 	mtx_unlock(&vm_page_queue_free_mtx);
1532 }
1533 
1534 /*
1535  *	Returns TRUE if a cached page is associated with the given object and
1536  *	offset, and FALSE otherwise.
1537  *
1538  *	The object must be locked.
1539  */
1540 boolean_t
1541 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex)
1542 {
1543 	vm_page_t m;
1544 
1545 	/*
1546 	 * Insertion into an object's collection of cached pages requires the
1547 	 * object to be locked.  Therefore, if the object is locked and the
1548 	 * object's collection is empty, there is no need to acquire the free
1549 	 * page queues lock in order to prove that the specified page doesn't
1550 	 * exist.
1551 	 */
1552 	VM_OBJECT_ASSERT_WLOCKED(object);
1553 	if (__predict_true(vm_object_cache_is_empty(object)))
1554 		return (FALSE);
1555 	mtx_lock(&vm_page_queue_free_mtx);
1556 	m = vm_page_cache_lookup(object, pindex);
1557 	mtx_unlock(&vm_page_queue_free_mtx);
1558 	return (m != NULL);
1559 }
1560 
1561 /*
1562  *	vm_page_alloc:
1563  *
1564  *	Allocate and return a page that is associated with the specified
1565  *	object and offset pair.  By default, this page is exclusive busied.
1566  *
1567  *	The caller must always specify an allocation class.
1568  *
1569  *	allocation classes:
1570  *	VM_ALLOC_NORMAL		normal process request
1571  *	VM_ALLOC_SYSTEM		system *really* needs a page
1572  *	VM_ALLOC_INTERRUPT	interrupt time request
1573  *
1574  *	optional allocation flags:
1575  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1576  *				intends to allocate
1577  *	VM_ALLOC_IFCACHED	return page only if it is cached
1578  *	VM_ALLOC_IFNOTCACHED	return NULL, do not reactivate if the page
1579  *				is cached
1580  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1581  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1582  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1583  *				should not be exclusive busy
1584  *	VM_ALLOC_SBUSY		shared busy the allocated page
1585  *	VM_ALLOC_WIRED		wire the allocated page
1586  *	VM_ALLOC_ZERO		prefer a zeroed page
1587  *
1588  *	This routine may not sleep.
1589  */
1590 vm_page_t
1591 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1592 {
1593 	struct vnode *vp = NULL;
1594 	vm_object_t m_object;
1595 	vm_page_t m, mpred;
1596 	int flags, req_class;
1597 
1598 	mpred = 0;	/* XXX: pacify gcc */
1599 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1600 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1601 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1602 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1603 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1604 	    req));
1605 	if (object != NULL)
1606 		VM_OBJECT_ASSERT_WLOCKED(object);
1607 
1608 	req_class = req & VM_ALLOC_CLASS_MASK;
1609 
1610 	/*
1611 	 * The page daemon is allowed to dig deeper into the free page list.
1612 	 */
1613 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1614 		req_class = VM_ALLOC_SYSTEM;
1615 
1616 	if (object != NULL) {
1617 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1618 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1619 		   ("vm_page_alloc: pindex already allocated"));
1620 	}
1621 
1622 	/*
1623 	 * The page allocation request can came from consumers which already
1624 	 * hold the free page queue mutex, like vm_page_insert() in
1625 	 * vm_page_cache().
1626 	 */
1627 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
1628 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
1629 	    (req_class == VM_ALLOC_SYSTEM &&
1630 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
1631 	    (req_class == VM_ALLOC_INTERRUPT &&
1632 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) {
1633 		/*
1634 		 * Allocate from the free queue if the number of free pages
1635 		 * exceeds the minimum for the request class.
1636 		 */
1637 		if (object != NULL &&
1638 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1639 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1640 				mtx_unlock(&vm_page_queue_free_mtx);
1641 				return (NULL);
1642 			}
1643 			if (vm_phys_unfree_page(m))
1644 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1645 #if VM_NRESERVLEVEL > 0
1646 			else if (!vm_reserv_reactivate_page(m))
1647 #else
1648 			else
1649 #endif
1650 				panic("vm_page_alloc: cache page %p is missing"
1651 				    " from the free queue", m);
1652 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1653 			mtx_unlock(&vm_page_queue_free_mtx);
1654 			return (NULL);
1655 #if VM_NRESERVLEVEL > 0
1656 		} else if (object == NULL || (object->flags & (OBJ_COLORED |
1657 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1658 		    vm_reserv_alloc_page(object, pindex, mpred)) == NULL) {
1659 #else
1660 		} else {
1661 #endif
1662 			m = vm_phys_alloc_pages(object != NULL ?
1663 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1664 #if VM_NRESERVLEVEL > 0
1665 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1666 				m = vm_phys_alloc_pages(object != NULL ?
1667 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1668 				    0);
1669 			}
1670 #endif
1671 		}
1672 	} else {
1673 		/*
1674 		 * Not allocatable, give up.
1675 		 */
1676 		mtx_unlock(&vm_page_queue_free_mtx);
1677 		atomic_add_int(&vm_pageout_deficit,
1678 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1679 		pagedaemon_wakeup();
1680 		return (NULL);
1681 	}
1682 
1683 	/*
1684 	 *  At this point we had better have found a good page.
1685 	 */
1686 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1687 	KASSERT(m->queue == PQ_NONE,
1688 	    ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
1689 	KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
1690 	KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
1691 	KASSERT(!vm_page_sbusied(m),
1692 	    ("vm_page_alloc: page %p is busy", m));
1693 	KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
1694 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1695 	    ("vm_page_alloc: page %p has unexpected memattr %d", m,
1696 	    pmap_page_get_memattr(m)));
1697 	if ((m->flags & PG_CACHED) != 0) {
1698 		KASSERT((m->flags & PG_ZERO) == 0,
1699 		    ("vm_page_alloc: cached page %p is PG_ZERO", m));
1700 		KASSERT(m->valid != 0,
1701 		    ("vm_page_alloc: cached page %p is invalid", m));
1702 		if (m->object == object && m->pindex == pindex)
1703 			vm_cnt.v_reactivated++;
1704 		else
1705 			m->valid = 0;
1706 		m_object = m->object;
1707 		vm_page_cache_remove(m);
1708 		if (m_object->type == OBJT_VNODE &&
1709 		    vm_object_cache_is_empty(m_object))
1710 			vp = m_object->handle;
1711 	} else {
1712 		KASSERT(m->valid == 0,
1713 		    ("vm_page_alloc: free page %p is valid", m));
1714 		vm_phys_freecnt_adj(m, -1);
1715 		if ((m->flags & PG_ZERO) != 0)
1716 			vm_page_zero_count--;
1717 	}
1718 	mtx_unlock(&vm_page_queue_free_mtx);
1719 
1720 	/*
1721 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1722 	 */
1723 	flags = 0;
1724 	if ((req & VM_ALLOC_ZERO) != 0)
1725 		flags = PG_ZERO;
1726 	flags &= m->flags;
1727 	if ((req & VM_ALLOC_NODUMP) != 0)
1728 		flags |= PG_NODUMP;
1729 	m->flags = flags;
1730 	m->aflags = 0;
1731 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1732 	    VPO_UNMANAGED : 0;
1733 	m->busy_lock = VPB_UNBUSIED;
1734 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1735 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1736 	if ((req & VM_ALLOC_SBUSY) != 0)
1737 		m->busy_lock = VPB_SHARERS_WORD(1);
1738 	if (req & VM_ALLOC_WIRED) {
1739 		/*
1740 		 * The page lock is not required for wiring a page until that
1741 		 * page is inserted into the object.
1742 		 */
1743 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1744 		m->wire_count = 1;
1745 	}
1746 	m->act_count = 0;
1747 
1748 	if (object != NULL) {
1749 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1750 			/* See the comment below about hold count. */
1751 			if (vp != NULL)
1752 				vdrop(vp);
1753 			pagedaemon_wakeup();
1754 			if (req & VM_ALLOC_WIRED) {
1755 				atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1756 				m->wire_count = 0;
1757 			}
1758 			m->object = NULL;
1759 			vm_page_free(m);
1760 			return (NULL);
1761 		}
1762 
1763 		/* Ignore device objects; the pager sets "memattr" for them. */
1764 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1765 		    (object->flags & OBJ_FICTITIOUS) == 0)
1766 			pmap_page_set_memattr(m, object->memattr);
1767 	} else
1768 		m->pindex = pindex;
1769 
1770 	/*
1771 	 * The following call to vdrop() must come after the above call
1772 	 * to vm_page_insert() in case both affect the same object and
1773 	 * vnode.  Otherwise, the affected vnode's hold count could
1774 	 * temporarily become zero.
1775 	 */
1776 	if (vp != NULL)
1777 		vdrop(vp);
1778 
1779 	/*
1780 	 * Don't wakeup too often - wakeup the pageout daemon when
1781 	 * we would be nearly out of memory.
1782 	 */
1783 	if (vm_paging_needed())
1784 		pagedaemon_wakeup();
1785 
1786 	return (m);
1787 }
1788 
1789 static void
1790 vm_page_alloc_contig_vdrop(struct spglist *lst)
1791 {
1792 
1793 	while (!SLIST_EMPTY(lst)) {
1794 		vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv);
1795 		SLIST_REMOVE_HEAD(lst, plinks.s.ss);
1796 	}
1797 }
1798 
1799 /*
1800  *	vm_page_alloc_contig:
1801  *
1802  *	Allocate a contiguous set of physical pages of the given size "npages"
1803  *	from the free lists.  All of the physical pages must be at or above
1804  *	the given physical address "low" and below the given physical address
1805  *	"high".  The given value "alignment" determines the alignment of the
1806  *	first physical page in the set.  If the given value "boundary" is
1807  *	non-zero, then the set of physical pages cannot cross any physical
1808  *	address boundary that is a multiple of that value.  Both "alignment"
1809  *	and "boundary" must be a power of two.
1810  *
1811  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1812  *	then the memory attribute setting for the physical pages is configured
1813  *	to the object's memory attribute setting.  Otherwise, the memory
1814  *	attribute setting for the physical pages is configured to "memattr",
1815  *	overriding the object's memory attribute setting.  However, if the
1816  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1817  *	memory attribute setting for the physical pages cannot be configured
1818  *	to VM_MEMATTR_DEFAULT.
1819  *
1820  *	The caller must always specify an allocation class.
1821  *
1822  *	allocation classes:
1823  *	VM_ALLOC_NORMAL		normal process request
1824  *	VM_ALLOC_SYSTEM		system *really* needs a page
1825  *	VM_ALLOC_INTERRUPT	interrupt time request
1826  *
1827  *	optional allocation flags:
1828  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1829  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1830  *				should not be exclusive busy
1831  *	VM_ALLOC_SBUSY		shared busy the allocated page
1832  *	VM_ALLOC_WIRED		wire the allocated page
1833  *	VM_ALLOC_ZERO		prefer a zeroed page
1834  *
1835  *	This routine may not sleep.
1836  */
1837 vm_page_t
1838 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1839     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1840     vm_paddr_t boundary, vm_memattr_t memattr)
1841 {
1842 	struct vnode *drop;
1843 	struct spglist deferred_vdrop_list;
1844 	vm_page_t m, m_tmp, m_ret;
1845 	u_int flags;
1846 	int req_class;
1847 
1848 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1849 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1850 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1851 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1852 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1853 	    req));
1854 	if (object != NULL) {
1855 		VM_OBJECT_ASSERT_WLOCKED(object);
1856 		KASSERT(object->type == OBJT_PHYS,
1857 		    ("vm_page_alloc_contig: object %p isn't OBJT_PHYS",
1858 		    object));
1859 	}
1860 	KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1861 	req_class = req & VM_ALLOC_CLASS_MASK;
1862 
1863 	/*
1864 	 * The page daemon is allowed to dig deeper into the free page list.
1865 	 */
1866 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1867 		req_class = VM_ALLOC_SYSTEM;
1868 
1869 	SLIST_INIT(&deferred_vdrop_list);
1870 	mtx_lock(&vm_page_queue_free_mtx);
1871 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1872 	    vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM &&
1873 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1874 	    vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT &&
1875 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) {
1876 #if VM_NRESERVLEVEL > 0
1877 retry:
1878 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1879 		    (m_ret = vm_reserv_alloc_contig(object, pindex, npages,
1880 		    low, high, alignment, boundary)) == NULL)
1881 #endif
1882 			m_ret = vm_phys_alloc_contig(npages, low, high,
1883 			    alignment, boundary);
1884 	} else {
1885 		mtx_unlock(&vm_page_queue_free_mtx);
1886 		atomic_add_int(&vm_pageout_deficit, npages);
1887 		pagedaemon_wakeup();
1888 		return (NULL);
1889 	}
1890 	if (m_ret != NULL)
1891 		for (m = m_ret; m < &m_ret[npages]; m++) {
1892 			drop = vm_page_alloc_init(m);
1893 			if (drop != NULL) {
1894 				/*
1895 				 * Enqueue the vnode for deferred vdrop().
1896 				 */
1897 				m->plinks.s.pv = drop;
1898 				SLIST_INSERT_HEAD(&deferred_vdrop_list, m,
1899 				    plinks.s.ss);
1900 			}
1901 		}
1902 	else {
1903 #if VM_NRESERVLEVEL > 0
1904 		if (vm_reserv_reclaim_contig(npages, low, high, alignment,
1905 		    boundary))
1906 			goto retry;
1907 #endif
1908 	}
1909 	mtx_unlock(&vm_page_queue_free_mtx);
1910 	if (m_ret == NULL)
1911 		return (NULL);
1912 
1913 	/*
1914 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1915 	 */
1916 	flags = 0;
1917 	if ((req & VM_ALLOC_ZERO) != 0)
1918 		flags = PG_ZERO;
1919 	if ((req & VM_ALLOC_NODUMP) != 0)
1920 		flags |= PG_NODUMP;
1921 	if ((req & VM_ALLOC_WIRED) != 0)
1922 		atomic_add_int(&vm_cnt.v_wire_count, npages);
1923 	if (object != NULL) {
1924 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1925 		    memattr == VM_MEMATTR_DEFAULT)
1926 			memattr = object->memattr;
1927 	}
1928 	for (m = m_ret; m < &m_ret[npages]; m++) {
1929 		m->aflags = 0;
1930 		m->flags = (m->flags | PG_NODUMP) & flags;
1931 		m->busy_lock = VPB_UNBUSIED;
1932 		if (object != NULL) {
1933 			if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
1934 				m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1935 			if ((req & VM_ALLOC_SBUSY) != 0)
1936 				m->busy_lock = VPB_SHARERS_WORD(1);
1937 		}
1938 		if ((req & VM_ALLOC_WIRED) != 0)
1939 			m->wire_count = 1;
1940 		/* Unmanaged pages don't use "act_count". */
1941 		m->oflags = VPO_UNMANAGED;
1942 		if (object != NULL) {
1943 			if (vm_page_insert(m, object, pindex)) {
1944 				vm_page_alloc_contig_vdrop(
1945 				    &deferred_vdrop_list);
1946 				if (vm_paging_needed())
1947 					pagedaemon_wakeup();
1948 				if ((req & VM_ALLOC_WIRED) != 0)
1949 					atomic_subtract_int(&vm_cnt.v_wire_count,
1950 					    npages);
1951 				for (m_tmp = m, m = m_ret;
1952 				    m < &m_ret[npages]; m++) {
1953 					if ((req & VM_ALLOC_WIRED) != 0)
1954 						m->wire_count = 0;
1955 					if (m >= m_tmp)
1956 						m->object = NULL;
1957 					vm_page_free(m);
1958 				}
1959 				return (NULL);
1960 			}
1961 		} else
1962 			m->pindex = pindex;
1963 		if (memattr != VM_MEMATTR_DEFAULT)
1964 			pmap_page_set_memattr(m, memattr);
1965 		pindex++;
1966 	}
1967 	vm_page_alloc_contig_vdrop(&deferred_vdrop_list);
1968 	if (vm_paging_needed())
1969 		pagedaemon_wakeup();
1970 	return (m_ret);
1971 }
1972 
1973 /*
1974  * Initialize a page that has been freshly dequeued from a freelist.
1975  * The caller has to drop the vnode returned, if it is not NULL.
1976  *
1977  * This function may only be used to initialize unmanaged pages.
1978  *
1979  * To be called with vm_page_queue_free_mtx held.
1980  */
1981 static struct vnode *
1982 vm_page_alloc_init(vm_page_t m)
1983 {
1984 	struct vnode *drop;
1985 	vm_object_t m_object;
1986 
1987 	KASSERT(m->queue == PQ_NONE,
1988 	    ("vm_page_alloc_init: page %p has unexpected queue %d",
1989 	    m, m->queue));
1990 	KASSERT(m->wire_count == 0,
1991 	    ("vm_page_alloc_init: page %p is wired", m));
1992 	KASSERT(m->hold_count == 0,
1993 	    ("vm_page_alloc_init: page %p is held", m));
1994 	KASSERT(!vm_page_sbusied(m),
1995 	    ("vm_page_alloc_init: page %p is busy", m));
1996 	KASSERT(m->dirty == 0,
1997 	    ("vm_page_alloc_init: page %p is dirty", m));
1998 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1999 	    ("vm_page_alloc_init: page %p has unexpected memattr %d",
2000 	    m, pmap_page_get_memattr(m)));
2001 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2002 	drop = NULL;
2003 	if ((m->flags & PG_CACHED) != 0) {
2004 		KASSERT((m->flags & PG_ZERO) == 0,
2005 		    ("vm_page_alloc_init: cached page %p is PG_ZERO", m));
2006 		m->valid = 0;
2007 		m_object = m->object;
2008 		vm_page_cache_remove(m);
2009 		if (m_object->type == OBJT_VNODE &&
2010 		    vm_object_cache_is_empty(m_object))
2011 			drop = m_object->handle;
2012 	} else {
2013 		KASSERT(m->valid == 0,
2014 		    ("vm_page_alloc_init: free page %p is valid", m));
2015 		vm_phys_freecnt_adj(m, -1);
2016 		if ((m->flags & PG_ZERO) != 0)
2017 			vm_page_zero_count--;
2018 	}
2019 	return (drop);
2020 }
2021 
2022 /*
2023  * 	vm_page_alloc_freelist:
2024  *
2025  *	Allocate a physical page from the specified free page list.
2026  *
2027  *	The caller must always specify an allocation class.
2028  *
2029  *	allocation classes:
2030  *	VM_ALLOC_NORMAL		normal process request
2031  *	VM_ALLOC_SYSTEM		system *really* needs a page
2032  *	VM_ALLOC_INTERRUPT	interrupt time request
2033  *
2034  *	optional allocation flags:
2035  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
2036  *				intends to allocate
2037  *	VM_ALLOC_WIRED		wire the allocated page
2038  *	VM_ALLOC_ZERO		prefer a zeroed page
2039  *
2040  *	This routine may not sleep.
2041  */
2042 vm_page_t
2043 vm_page_alloc_freelist(int flind, int req)
2044 {
2045 	struct vnode *drop;
2046 	vm_page_t m;
2047 	u_int flags;
2048 	int req_class;
2049 
2050 	req_class = req & VM_ALLOC_CLASS_MASK;
2051 
2052 	/*
2053 	 * The page daemon is allowed to dig deeper into the free page list.
2054 	 */
2055 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2056 		req_class = VM_ALLOC_SYSTEM;
2057 
2058 	/*
2059 	 * Do not allocate reserved pages unless the req has asked for it.
2060 	 */
2061 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
2062 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
2063 	    (req_class == VM_ALLOC_SYSTEM &&
2064 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
2065 	    (req_class == VM_ALLOC_INTERRUPT &&
2066 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0))
2067 		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0);
2068 	else {
2069 		mtx_unlock(&vm_page_queue_free_mtx);
2070 		atomic_add_int(&vm_pageout_deficit,
2071 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
2072 		pagedaemon_wakeup();
2073 		return (NULL);
2074 	}
2075 	if (m == NULL) {
2076 		mtx_unlock(&vm_page_queue_free_mtx);
2077 		return (NULL);
2078 	}
2079 	drop = vm_page_alloc_init(m);
2080 	mtx_unlock(&vm_page_queue_free_mtx);
2081 
2082 	/*
2083 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
2084 	 */
2085 	m->aflags = 0;
2086 	flags = 0;
2087 	if ((req & VM_ALLOC_ZERO) != 0)
2088 		flags = PG_ZERO;
2089 	m->flags &= flags;
2090 	if ((req & VM_ALLOC_WIRED) != 0) {
2091 		/*
2092 		 * The page lock is not required for wiring a page that does
2093 		 * not belong to an object.
2094 		 */
2095 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2096 		m->wire_count = 1;
2097 	}
2098 	/* Unmanaged pages don't use "act_count". */
2099 	m->oflags = VPO_UNMANAGED;
2100 	if (drop != NULL)
2101 		vdrop(drop);
2102 	if (vm_paging_needed())
2103 		pagedaemon_wakeup();
2104 	return (m);
2105 }
2106 
2107 /*
2108  *	vm_wait:	(also see VM_WAIT macro)
2109  *
2110  *	Sleep until free pages are available for allocation.
2111  *	- Called in various places before memory allocations.
2112  */
2113 void
2114 vm_wait(void)
2115 {
2116 
2117 	mtx_lock(&vm_page_queue_free_mtx);
2118 	if (curproc == pageproc) {
2119 		vm_pageout_pages_needed = 1;
2120 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
2121 		    PDROP | PSWP, "VMWait", 0);
2122 	} else {
2123 		if (!vm_pages_needed) {
2124 			vm_pages_needed = 1;
2125 			wakeup(&vm_pages_needed);
2126 		}
2127 		msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
2128 		    "vmwait", 0);
2129 	}
2130 }
2131 
2132 /*
2133  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
2134  *
2135  *	Sleep until free pages are available for allocation.
2136  *	- Called only in vm_fault so that processes page faulting
2137  *	  can be easily tracked.
2138  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2139  *	  processes will be able to grab memory first.  Do not change
2140  *	  this balance without careful testing first.
2141  */
2142 void
2143 vm_waitpfault(void)
2144 {
2145 
2146 	mtx_lock(&vm_page_queue_free_mtx);
2147 	if (!vm_pages_needed) {
2148 		vm_pages_needed = 1;
2149 		wakeup(&vm_pages_needed);
2150 	}
2151 	msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
2152 	    "pfault", 0);
2153 }
2154 
2155 struct vm_pagequeue *
2156 vm_page_pagequeue(vm_page_t m)
2157 {
2158 
2159 	return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2160 }
2161 
2162 /*
2163  *	vm_page_dequeue:
2164  *
2165  *	Remove the given page from its current page queue.
2166  *
2167  *	The page must be locked.
2168  */
2169 void
2170 vm_page_dequeue(vm_page_t m)
2171 {
2172 	struct vm_pagequeue *pq;
2173 
2174 	vm_page_assert_locked(m);
2175 	KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2176 	    m));
2177 	pq = vm_page_pagequeue(m);
2178 	vm_pagequeue_lock(pq);
2179 	m->queue = PQ_NONE;
2180 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2181 	vm_pagequeue_cnt_dec(pq);
2182 	vm_pagequeue_unlock(pq);
2183 }
2184 
2185 /*
2186  *	vm_page_dequeue_locked:
2187  *
2188  *	Remove the given page from its current page queue.
2189  *
2190  *	The page and page queue must be locked.
2191  */
2192 void
2193 vm_page_dequeue_locked(vm_page_t m)
2194 {
2195 	struct vm_pagequeue *pq;
2196 
2197 	vm_page_lock_assert(m, MA_OWNED);
2198 	pq = vm_page_pagequeue(m);
2199 	vm_pagequeue_assert_locked(pq);
2200 	m->queue = PQ_NONE;
2201 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2202 	vm_pagequeue_cnt_dec(pq);
2203 }
2204 
2205 /*
2206  *	vm_page_enqueue:
2207  *
2208  *	Add the given page to the specified page queue.
2209  *
2210  *	The page must be locked.
2211  */
2212 static void
2213 vm_page_enqueue(uint8_t queue, vm_page_t m)
2214 {
2215 	struct vm_pagequeue *pq;
2216 
2217 	vm_page_lock_assert(m, MA_OWNED);
2218 	KASSERT(queue < PQ_COUNT,
2219 	    ("vm_page_enqueue: invalid queue %u request for page %p",
2220 	    queue, m));
2221 	pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2222 	vm_pagequeue_lock(pq);
2223 	m->queue = queue;
2224 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2225 	vm_pagequeue_cnt_inc(pq);
2226 	vm_pagequeue_unlock(pq);
2227 }
2228 
2229 /*
2230  *	vm_page_requeue:
2231  *
2232  *	Move the given page to the tail of its current page queue.
2233  *
2234  *	The page must be locked.
2235  */
2236 void
2237 vm_page_requeue(vm_page_t m)
2238 {
2239 	struct vm_pagequeue *pq;
2240 
2241 	vm_page_lock_assert(m, MA_OWNED);
2242 	KASSERT(m->queue != PQ_NONE,
2243 	    ("vm_page_requeue: page %p is not queued", m));
2244 	pq = vm_page_pagequeue(m);
2245 	vm_pagequeue_lock(pq);
2246 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2247 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2248 	vm_pagequeue_unlock(pq);
2249 }
2250 
2251 /*
2252  *	vm_page_requeue_locked:
2253  *
2254  *	Move the given page to the tail of its current page queue.
2255  *
2256  *	The page queue must be locked.
2257  */
2258 void
2259 vm_page_requeue_locked(vm_page_t m)
2260 {
2261 	struct vm_pagequeue *pq;
2262 
2263 	KASSERT(m->queue != PQ_NONE,
2264 	    ("vm_page_requeue_locked: page %p is not queued", m));
2265 	pq = vm_page_pagequeue(m);
2266 	vm_pagequeue_assert_locked(pq);
2267 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2268 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2269 }
2270 
2271 /*
2272  *	vm_page_activate:
2273  *
2274  *	Put the specified page on the active list (if appropriate).
2275  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2276  *	mess with it.
2277  *
2278  *	The page must be locked.
2279  */
2280 void
2281 vm_page_activate(vm_page_t m)
2282 {
2283 	int queue;
2284 
2285 	vm_page_lock_assert(m, MA_OWNED);
2286 	if ((queue = m->queue) != PQ_ACTIVE) {
2287 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2288 			if (m->act_count < ACT_INIT)
2289 				m->act_count = ACT_INIT;
2290 			if (queue != PQ_NONE)
2291 				vm_page_dequeue(m);
2292 			vm_page_enqueue(PQ_ACTIVE, m);
2293 		} else
2294 			KASSERT(queue == PQ_NONE,
2295 			    ("vm_page_activate: wired page %p is queued", m));
2296 	} else {
2297 		if (m->act_count < ACT_INIT)
2298 			m->act_count = ACT_INIT;
2299 	}
2300 }
2301 
2302 /*
2303  *	vm_page_free_wakeup:
2304  *
2305  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
2306  *	routine is called when a page has been added to the cache or free
2307  *	queues.
2308  *
2309  *	The page queues must be locked.
2310  */
2311 static inline void
2312 vm_page_free_wakeup(void)
2313 {
2314 
2315 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2316 	/*
2317 	 * if pageout daemon needs pages, then tell it that there are
2318 	 * some free.
2319 	 */
2320 	if (vm_pageout_pages_needed &&
2321 	    vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) {
2322 		wakeup(&vm_pageout_pages_needed);
2323 		vm_pageout_pages_needed = 0;
2324 	}
2325 	/*
2326 	 * wakeup processes that are waiting on memory if we hit a
2327 	 * high water mark. And wakeup scheduler process if we have
2328 	 * lots of memory. this process will swapin processes.
2329 	 */
2330 	if (vm_pages_needed && !vm_page_count_min()) {
2331 		vm_pages_needed = 0;
2332 		wakeup(&vm_cnt.v_free_count);
2333 	}
2334 }
2335 
2336 /*
2337  *	Turn a cached page into a free page, by changing its attributes.
2338  *	Keep the statistics up-to-date.
2339  *
2340  *	The free page queue must be locked.
2341  */
2342 static void
2343 vm_page_cache_turn_free(vm_page_t m)
2344 {
2345 
2346 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2347 
2348 	m->object = NULL;
2349 	m->valid = 0;
2350 	KASSERT((m->flags & PG_CACHED) != 0,
2351 	    ("vm_page_cache_turn_free: page %p is not cached", m));
2352 	m->flags &= ~PG_CACHED;
2353 	vm_cnt.v_cache_count--;
2354 	vm_phys_freecnt_adj(m, 1);
2355 }
2356 
2357 /*
2358  *	vm_page_free_toq:
2359  *
2360  *	Returns the given page to the free list,
2361  *	disassociating it with any VM object.
2362  *
2363  *	The object must be locked.  The page must be locked if it is managed.
2364  */
2365 void
2366 vm_page_free_toq(vm_page_t m)
2367 {
2368 
2369 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2370 		vm_page_lock_assert(m, MA_OWNED);
2371 		KASSERT(!pmap_page_is_mapped(m),
2372 		    ("vm_page_free_toq: freeing mapped page %p", m));
2373 	} else
2374 		KASSERT(m->queue == PQ_NONE,
2375 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2376 	PCPU_INC(cnt.v_tfree);
2377 
2378 	if (vm_page_sbusied(m))
2379 		panic("vm_page_free: freeing busy page %p", m);
2380 
2381 	/*
2382 	 * Unqueue, then remove page.  Note that we cannot destroy
2383 	 * the page here because we do not want to call the pager's
2384 	 * callback routine until after we've put the page on the
2385 	 * appropriate free queue.
2386 	 */
2387 	vm_page_remque(m);
2388 	vm_page_remove(m);
2389 
2390 	/*
2391 	 * If fictitious remove object association and
2392 	 * return, otherwise delay object association removal.
2393 	 */
2394 	if ((m->flags & PG_FICTITIOUS) != 0) {
2395 		return;
2396 	}
2397 
2398 	m->valid = 0;
2399 	vm_page_undirty(m);
2400 
2401 	if (m->wire_count != 0)
2402 		panic("vm_page_free: freeing wired page %p", m);
2403 	if (m->hold_count != 0) {
2404 		m->flags &= ~PG_ZERO;
2405 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2406 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
2407 		m->flags |= PG_UNHOLDFREE;
2408 	} else {
2409 		/*
2410 		 * Restore the default memory attribute to the page.
2411 		 */
2412 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2413 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2414 
2415 		/*
2416 		 * Insert the page into the physical memory allocator's
2417 		 * cache/free page queues.
2418 		 */
2419 		mtx_lock(&vm_page_queue_free_mtx);
2420 		vm_phys_freecnt_adj(m, 1);
2421 #if VM_NRESERVLEVEL > 0
2422 		if (!vm_reserv_free_page(m))
2423 #else
2424 		if (TRUE)
2425 #endif
2426 			vm_phys_free_pages(m, 0);
2427 		if ((m->flags & PG_ZERO) != 0)
2428 			++vm_page_zero_count;
2429 		else
2430 			vm_page_zero_idle_wakeup();
2431 		vm_page_free_wakeup();
2432 		mtx_unlock(&vm_page_queue_free_mtx);
2433 	}
2434 }
2435 
2436 /*
2437  *	vm_page_wire:
2438  *
2439  *	Mark this page as wired down by yet
2440  *	another map, removing it from paging queues
2441  *	as necessary.
2442  *
2443  *	If the page is fictitious, then its wire count must remain one.
2444  *
2445  *	The page must be locked.
2446  */
2447 void
2448 vm_page_wire(vm_page_t m)
2449 {
2450 
2451 	/*
2452 	 * Only bump the wire statistics if the page is not already wired,
2453 	 * and only unqueue the page if it is on some queue (if it is unmanaged
2454 	 * it is already off the queues).
2455 	 */
2456 	vm_page_lock_assert(m, MA_OWNED);
2457 	if ((m->flags & PG_FICTITIOUS) != 0) {
2458 		KASSERT(m->wire_count == 1,
2459 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
2460 		    m));
2461 		return;
2462 	}
2463 	if (m->wire_count == 0) {
2464 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
2465 		    m->queue == PQ_NONE,
2466 		    ("vm_page_wire: unmanaged page %p is queued", m));
2467 		vm_page_remque(m);
2468 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2469 	}
2470 	m->wire_count++;
2471 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
2472 }
2473 
2474 /*
2475  * vm_page_unwire:
2476  *
2477  * Release one wiring of the specified page, potentially enabling it to be
2478  * paged again.  If paging is enabled, then the value of the parameter
2479  * "queue" determines the queue to which the page is added.
2480  *
2481  * However, unless the page belongs to an object, it is not enqueued because
2482  * it cannot be paged out.
2483  *
2484  * If a page is fictitious, then its wire count must always be one.
2485  *
2486  * A managed page must be locked.
2487  */
2488 void
2489 vm_page_unwire(vm_page_t m, uint8_t queue)
2490 {
2491 
2492 	KASSERT(queue < PQ_COUNT,
2493 	    ("vm_page_unwire: invalid queue %u request for page %p",
2494 	    queue, m));
2495 	if ((m->oflags & VPO_UNMANAGED) == 0)
2496 		vm_page_lock_assert(m, MA_OWNED);
2497 	if ((m->flags & PG_FICTITIOUS) != 0) {
2498 		KASSERT(m->wire_count == 1,
2499 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
2500 		return;
2501 	}
2502 	if (m->wire_count > 0) {
2503 		m->wire_count--;
2504 		if (m->wire_count == 0) {
2505 			atomic_subtract_int(&vm_cnt.v_wire_count, 1);
2506 			if ((m->oflags & VPO_UNMANAGED) != 0 ||
2507 			    m->object == NULL)
2508 				return;
2509 			if (queue == PQ_INACTIVE)
2510 				m->flags &= ~PG_WINATCFLS;
2511 			vm_page_enqueue(queue, m);
2512 		}
2513 	} else
2514 		panic("vm_page_unwire: page %p's wire count is zero", m);
2515 }
2516 
2517 /*
2518  * Move the specified page to the inactive queue.
2519  *
2520  * Many pages placed on the inactive queue should actually go
2521  * into the cache, but it is difficult to figure out which.  What
2522  * we do instead, if the inactive target is well met, is to put
2523  * clean pages at the head of the inactive queue instead of the tail.
2524  * This will cause them to be moved to the cache more quickly and
2525  * if not actively re-referenced, reclaimed more quickly.  If we just
2526  * stick these pages at the end of the inactive queue, heavy filesystem
2527  * meta-data accesses can cause an unnecessary paging load on memory bound
2528  * processes.  This optimization causes one-time-use metadata to be
2529  * reused more quickly.
2530  *
2531  * Normally athead is 0 resulting in LRU operation.  athead is set
2532  * to 1 if we want this page to be 'as if it were placed in the cache',
2533  * except without unmapping it from the process address space.
2534  *
2535  * The page must be locked.
2536  */
2537 static inline void
2538 _vm_page_deactivate(vm_page_t m, int athead)
2539 {
2540 	struct vm_pagequeue *pq;
2541 	int queue;
2542 
2543 	vm_page_lock_assert(m, MA_OWNED);
2544 
2545 	/*
2546 	 * Ignore if already inactive.
2547 	 */
2548 	if ((queue = m->queue) == PQ_INACTIVE)
2549 		return;
2550 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2551 		if (queue != PQ_NONE)
2552 			vm_page_dequeue(m);
2553 		m->flags &= ~PG_WINATCFLS;
2554 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
2555 		vm_pagequeue_lock(pq);
2556 		m->queue = PQ_INACTIVE;
2557 		if (athead)
2558 			TAILQ_INSERT_HEAD(&pq->pq_pl, m, plinks.q);
2559 		else
2560 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2561 		vm_pagequeue_cnt_inc(pq);
2562 		vm_pagequeue_unlock(pq);
2563 	}
2564 }
2565 
2566 /*
2567  * Move the specified page to the inactive queue.
2568  *
2569  * The page must be locked.
2570  */
2571 void
2572 vm_page_deactivate(vm_page_t m)
2573 {
2574 
2575 	_vm_page_deactivate(m, 0);
2576 }
2577 
2578 /*
2579  * vm_page_try_to_cache:
2580  *
2581  * Returns 0 on failure, 1 on success
2582  */
2583 int
2584 vm_page_try_to_cache(vm_page_t m)
2585 {
2586 
2587 	vm_page_lock_assert(m, MA_OWNED);
2588 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2589 	if (m->dirty || m->hold_count || m->wire_count ||
2590 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
2591 		return (0);
2592 	pmap_remove_all(m);
2593 	if (m->dirty)
2594 		return (0);
2595 	vm_page_cache(m);
2596 	return (1);
2597 }
2598 
2599 /*
2600  * vm_page_try_to_free()
2601  *
2602  *	Attempt to free the page.  If we cannot free it, we do nothing.
2603  *	1 is returned on success, 0 on failure.
2604  */
2605 int
2606 vm_page_try_to_free(vm_page_t m)
2607 {
2608 
2609 	vm_page_lock_assert(m, MA_OWNED);
2610 	if (m->object != NULL)
2611 		VM_OBJECT_ASSERT_WLOCKED(m->object);
2612 	if (m->dirty || m->hold_count || m->wire_count ||
2613 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
2614 		return (0);
2615 	pmap_remove_all(m);
2616 	if (m->dirty)
2617 		return (0);
2618 	vm_page_free(m);
2619 	return (1);
2620 }
2621 
2622 /*
2623  * vm_page_cache
2624  *
2625  * Put the specified page onto the page cache queue (if appropriate).
2626  *
2627  * The object and page must be locked.
2628  */
2629 void
2630 vm_page_cache(vm_page_t m)
2631 {
2632 	vm_object_t object;
2633 	boolean_t cache_was_empty;
2634 
2635 	vm_page_lock_assert(m, MA_OWNED);
2636 	object = m->object;
2637 	VM_OBJECT_ASSERT_WLOCKED(object);
2638 	if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) ||
2639 	    m->hold_count || m->wire_count)
2640 		panic("vm_page_cache: attempting to cache busy page");
2641 	KASSERT(!pmap_page_is_mapped(m),
2642 	    ("vm_page_cache: page %p is mapped", m));
2643 	KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m));
2644 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
2645 	    (object->type == OBJT_SWAP &&
2646 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
2647 		/*
2648 		 * Hypothesis: A cache-eligible page belonging to a
2649 		 * default object or swap object but without a backing
2650 		 * store must be zero filled.
2651 		 */
2652 		vm_page_free(m);
2653 		return;
2654 	}
2655 	KASSERT((m->flags & PG_CACHED) == 0,
2656 	    ("vm_page_cache: page %p is already cached", m));
2657 
2658 	/*
2659 	 * Remove the page from the paging queues.
2660 	 */
2661 	vm_page_remque(m);
2662 
2663 	/*
2664 	 * Remove the page from the object's collection of resident
2665 	 * pages.
2666 	 */
2667 	vm_radix_remove(&object->rtree, m->pindex);
2668 	TAILQ_REMOVE(&object->memq, m, listq);
2669 	object->resident_page_count--;
2670 
2671 	/*
2672 	 * Restore the default memory attribute to the page.
2673 	 */
2674 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2675 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2676 
2677 	/*
2678 	 * Insert the page into the object's collection of cached pages
2679 	 * and the physical memory allocator's cache/free page queues.
2680 	 */
2681 	m->flags &= ~PG_ZERO;
2682 	mtx_lock(&vm_page_queue_free_mtx);
2683 	cache_was_empty = vm_radix_is_empty(&object->cache);
2684 	if (vm_radix_insert(&object->cache, m)) {
2685 		mtx_unlock(&vm_page_queue_free_mtx);
2686 		if (object->resident_page_count == 0)
2687 			vdrop(object->handle);
2688 		m->object = NULL;
2689 		vm_page_free(m);
2690 		return;
2691 	}
2692 
2693 	/*
2694 	 * The above call to vm_radix_insert() could reclaim the one pre-
2695 	 * existing cached page from this object, resulting in a call to
2696 	 * vdrop().
2697 	 */
2698 	if (!cache_was_empty)
2699 		cache_was_empty = vm_radix_is_singleton(&object->cache);
2700 
2701 	m->flags |= PG_CACHED;
2702 	vm_cnt.v_cache_count++;
2703 	PCPU_INC(cnt.v_tcached);
2704 #if VM_NRESERVLEVEL > 0
2705 	if (!vm_reserv_free_page(m)) {
2706 #else
2707 	if (TRUE) {
2708 #endif
2709 		vm_phys_set_pool(VM_FREEPOOL_CACHE, m, 0);
2710 		vm_phys_free_pages(m, 0);
2711 	}
2712 	vm_page_free_wakeup();
2713 	mtx_unlock(&vm_page_queue_free_mtx);
2714 
2715 	/*
2716 	 * Increment the vnode's hold count if this is the object's only
2717 	 * cached page.  Decrement the vnode's hold count if this was
2718 	 * the object's only resident page.
2719 	 */
2720 	if (object->type == OBJT_VNODE) {
2721 		if (cache_was_empty && object->resident_page_count != 0)
2722 			vhold(object->handle);
2723 		else if (!cache_was_empty && object->resident_page_count == 0)
2724 			vdrop(object->handle);
2725 	}
2726 }
2727 
2728 /*
2729  * vm_page_advise
2730  *
2731  *	Cache, deactivate, or do nothing as appropriate.  This routine
2732  *	is used by madvise().
2733  *
2734  *	Generally speaking we want to move the page into the cache so
2735  *	it gets reused quickly.  However, this can result in a silly syndrome
2736  *	due to the page recycling too quickly.  Small objects will not be
2737  *	fully cached.  On the other hand, if we move the page to the inactive
2738  *	queue we wind up with a problem whereby very large objects
2739  *	unnecessarily blow away our inactive and cache queues.
2740  *
2741  *	The solution is to move the pages based on a fixed weighting.  We
2742  *	either leave them alone, deactivate them, or move them to the cache,
2743  *	where moving them to the cache has the highest weighting.
2744  *	By forcing some pages into other queues we eventually force the
2745  *	system to balance the queues, potentially recovering other unrelated
2746  *	space from active.  The idea is to not force this to happen too
2747  *	often.
2748  *
2749  *	The object and page must be locked.
2750  */
2751 void
2752 vm_page_advise(vm_page_t m, int advice)
2753 {
2754 	int dnw, head;
2755 
2756 	vm_page_assert_locked(m);
2757 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2758 	if (advice == MADV_FREE) {
2759 		/*
2760 		 * Mark the page clean.  This will allow the page to be freed
2761 		 * up by the system.  However, such pages are often reused
2762 		 * quickly by malloc() so we do not do anything that would
2763 		 * cause a page fault if we can help it.
2764 		 *
2765 		 * Specifically, we do not try to actually free the page now
2766 		 * nor do we try to put it in the cache (which would cause a
2767 		 * page fault on reuse).
2768 		 *
2769 		 * But we do make the page is freeable as we can without
2770 		 * actually taking the step of unmapping it.
2771 		 */
2772 		m->dirty = 0;
2773 		m->act_count = 0;
2774 	} else if (advice != MADV_DONTNEED)
2775 		return;
2776 	dnw = PCPU_GET(dnweight);
2777 	PCPU_INC(dnweight);
2778 
2779 	/*
2780 	 * Occasionally leave the page alone.
2781 	 */
2782 	if ((dnw & 0x01F0) == 0 || m->queue == PQ_INACTIVE) {
2783 		if (m->act_count >= ACT_INIT)
2784 			--m->act_count;
2785 		return;
2786 	}
2787 
2788 	/*
2789 	 * Clear any references to the page.  Otherwise, the page daemon will
2790 	 * immediately reactivate the page.
2791 	 */
2792 	vm_page_aflag_clear(m, PGA_REFERENCED);
2793 
2794 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
2795 		vm_page_dirty(m);
2796 
2797 	if (m->dirty || (dnw & 0x0070) == 0) {
2798 		/*
2799 		 * Deactivate the page 3 times out of 32.
2800 		 */
2801 		head = 0;
2802 	} else {
2803 		/*
2804 		 * Cache the page 28 times out of every 32.  Note that
2805 		 * the page is deactivated instead of cached, but placed
2806 		 * at the head of the queue instead of the tail.
2807 		 */
2808 		head = 1;
2809 	}
2810 	_vm_page_deactivate(m, head);
2811 }
2812 
2813 /*
2814  * Grab a page, waiting until we are waken up due to the page
2815  * changing state.  We keep on waiting, if the page continues
2816  * to be in the object.  If the page doesn't exist, first allocate it
2817  * and then conditionally zero it.
2818  *
2819  * This routine may sleep.
2820  *
2821  * The object must be locked on entry.  The lock will, however, be released
2822  * and reacquired if the routine sleeps.
2823  */
2824 vm_page_t
2825 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
2826 {
2827 	vm_page_t m;
2828 	int sleep;
2829 
2830 	VM_OBJECT_ASSERT_WLOCKED(object);
2831 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
2832 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
2833 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
2834 retrylookup:
2835 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
2836 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
2837 		    vm_page_xbusied(m) : vm_page_busied(m);
2838 		if (sleep) {
2839 			if ((allocflags & VM_ALLOC_NOWAIT) != 0)
2840 				return (NULL);
2841 			/*
2842 			 * Reference the page before unlocking and
2843 			 * sleeping so that the page daemon is less
2844 			 * likely to reclaim it.
2845 			 */
2846 			vm_page_aflag_set(m, PGA_REFERENCED);
2847 			vm_page_lock(m);
2848 			VM_OBJECT_WUNLOCK(object);
2849 			vm_page_busy_sleep(m, "pgrbwt");
2850 			VM_OBJECT_WLOCK(object);
2851 			goto retrylookup;
2852 		} else {
2853 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
2854 				vm_page_lock(m);
2855 				vm_page_wire(m);
2856 				vm_page_unlock(m);
2857 			}
2858 			if ((allocflags &
2859 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
2860 				vm_page_xbusy(m);
2861 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
2862 				vm_page_sbusy(m);
2863 			return (m);
2864 		}
2865 	}
2866 	m = vm_page_alloc(object, pindex, allocflags);
2867 	if (m == NULL) {
2868 		if ((allocflags & VM_ALLOC_NOWAIT) != 0)
2869 			return (NULL);
2870 		VM_OBJECT_WUNLOCK(object);
2871 		VM_WAIT;
2872 		VM_OBJECT_WLOCK(object);
2873 		goto retrylookup;
2874 	} else if (m->valid != 0)
2875 		return (m);
2876 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
2877 		pmap_zero_page(m);
2878 	return (m);
2879 }
2880 
2881 /*
2882  * Mapping function for valid or dirty bits in a page.
2883  *
2884  * Inputs are required to range within a page.
2885  */
2886 vm_page_bits_t
2887 vm_page_bits(int base, int size)
2888 {
2889 	int first_bit;
2890 	int last_bit;
2891 
2892 	KASSERT(
2893 	    base + size <= PAGE_SIZE,
2894 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
2895 	);
2896 
2897 	if (size == 0)		/* handle degenerate case */
2898 		return (0);
2899 
2900 	first_bit = base >> DEV_BSHIFT;
2901 	last_bit = (base + size - 1) >> DEV_BSHIFT;
2902 
2903 	return (((vm_page_bits_t)2 << last_bit) -
2904 	    ((vm_page_bits_t)1 << first_bit));
2905 }
2906 
2907 /*
2908  *	vm_page_set_valid_range:
2909  *
2910  *	Sets portions of a page valid.  The arguments are expected
2911  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
2912  *	of any partial chunks touched by the range.  The invalid portion of
2913  *	such chunks will be zeroed.
2914  *
2915  *	(base + size) must be less then or equal to PAGE_SIZE.
2916  */
2917 void
2918 vm_page_set_valid_range(vm_page_t m, int base, int size)
2919 {
2920 	int endoff, frag;
2921 
2922 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2923 	if (size == 0)	/* handle degenerate case */
2924 		return;
2925 
2926 	/*
2927 	 * If the base is not DEV_BSIZE aligned and the valid
2928 	 * bit is clear, we have to zero out a portion of the
2929 	 * first block.
2930 	 */
2931 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
2932 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
2933 		pmap_zero_page_area(m, frag, base - frag);
2934 
2935 	/*
2936 	 * If the ending offset is not DEV_BSIZE aligned and the
2937 	 * valid bit is clear, we have to zero out a portion of
2938 	 * the last block.
2939 	 */
2940 	endoff = base + size;
2941 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
2942 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
2943 		pmap_zero_page_area(m, endoff,
2944 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
2945 
2946 	/*
2947 	 * Assert that no previously invalid block that is now being validated
2948 	 * is already dirty.
2949 	 */
2950 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
2951 	    ("vm_page_set_valid_range: page %p is dirty", m));
2952 
2953 	/*
2954 	 * Set valid bits inclusive of any overlap.
2955 	 */
2956 	m->valid |= vm_page_bits(base, size);
2957 }
2958 
2959 /*
2960  * Clear the given bits from the specified page's dirty field.
2961  */
2962 static __inline void
2963 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
2964 {
2965 	uintptr_t addr;
2966 #if PAGE_SIZE < 16384
2967 	int shift;
2968 #endif
2969 
2970 	/*
2971 	 * If the object is locked and the page is neither exclusive busy nor
2972 	 * write mapped, then the page's dirty field cannot possibly be
2973 	 * set by a concurrent pmap operation.
2974 	 */
2975 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2976 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
2977 		m->dirty &= ~pagebits;
2978 	else {
2979 		/*
2980 		 * The pmap layer can call vm_page_dirty() without
2981 		 * holding a distinguished lock.  The combination of
2982 		 * the object's lock and an atomic operation suffice
2983 		 * to guarantee consistency of the page dirty field.
2984 		 *
2985 		 * For PAGE_SIZE == 32768 case, compiler already
2986 		 * properly aligns the dirty field, so no forcible
2987 		 * alignment is needed. Only require existence of
2988 		 * atomic_clear_64 when page size is 32768.
2989 		 */
2990 		addr = (uintptr_t)&m->dirty;
2991 #if PAGE_SIZE == 32768
2992 		atomic_clear_64((uint64_t *)addr, pagebits);
2993 #elif PAGE_SIZE == 16384
2994 		atomic_clear_32((uint32_t *)addr, pagebits);
2995 #else		/* PAGE_SIZE <= 8192 */
2996 		/*
2997 		 * Use a trick to perform a 32-bit atomic on the
2998 		 * containing aligned word, to not depend on the existence
2999 		 * of atomic_clear_{8, 16}.
3000 		 */
3001 		shift = addr & (sizeof(uint32_t) - 1);
3002 #if BYTE_ORDER == BIG_ENDIAN
3003 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
3004 #else
3005 		shift *= NBBY;
3006 #endif
3007 		addr &= ~(sizeof(uint32_t) - 1);
3008 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
3009 #endif		/* PAGE_SIZE */
3010 	}
3011 }
3012 
3013 /*
3014  *	vm_page_set_validclean:
3015  *
3016  *	Sets portions of a page valid and clean.  The arguments are expected
3017  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3018  *	of any partial chunks touched by the range.  The invalid portion of
3019  *	such chunks will be zero'd.
3020  *
3021  *	(base + size) must be less then or equal to PAGE_SIZE.
3022  */
3023 void
3024 vm_page_set_validclean(vm_page_t m, int base, int size)
3025 {
3026 	vm_page_bits_t oldvalid, pagebits;
3027 	int endoff, frag;
3028 
3029 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3030 	if (size == 0)	/* handle degenerate case */
3031 		return;
3032 
3033 	/*
3034 	 * If the base is not DEV_BSIZE aligned and the valid
3035 	 * bit is clear, we have to zero out a portion of the
3036 	 * first block.
3037 	 */
3038 	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
3039 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3040 		pmap_zero_page_area(m, frag, base - frag);
3041 
3042 	/*
3043 	 * If the ending offset is not DEV_BSIZE aligned and the
3044 	 * valid bit is clear, we have to zero out a portion of
3045 	 * the last block.
3046 	 */
3047 	endoff = base + size;
3048 	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
3049 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3050 		pmap_zero_page_area(m, endoff,
3051 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3052 
3053 	/*
3054 	 * Set valid, clear dirty bits.  If validating the entire
3055 	 * page we can safely clear the pmap modify bit.  We also
3056 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
3057 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
3058 	 * be set again.
3059 	 *
3060 	 * We set valid bits inclusive of any overlap, but we can only
3061 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
3062 	 * the range.
3063 	 */
3064 	oldvalid = m->valid;
3065 	pagebits = vm_page_bits(base, size);
3066 	m->valid |= pagebits;
3067 #if 0	/* NOT YET */
3068 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3069 		frag = DEV_BSIZE - frag;
3070 		base += frag;
3071 		size -= frag;
3072 		if (size < 0)
3073 			size = 0;
3074 	}
3075 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3076 #endif
3077 	if (base == 0 && size == PAGE_SIZE) {
3078 		/*
3079 		 * The page can only be modified within the pmap if it is
3080 		 * mapped, and it can only be mapped if it was previously
3081 		 * fully valid.
3082 		 */
3083 		if (oldvalid == VM_PAGE_BITS_ALL)
3084 			/*
3085 			 * Perform the pmap_clear_modify() first.  Otherwise,
3086 			 * a concurrent pmap operation, such as
3087 			 * pmap_protect(), could clear a modification in the
3088 			 * pmap and set the dirty field on the page before
3089 			 * pmap_clear_modify() had begun and after the dirty
3090 			 * field was cleared here.
3091 			 */
3092 			pmap_clear_modify(m);
3093 		m->dirty = 0;
3094 		m->oflags &= ~VPO_NOSYNC;
3095 	} else if (oldvalid != VM_PAGE_BITS_ALL)
3096 		m->dirty &= ~pagebits;
3097 	else
3098 		vm_page_clear_dirty_mask(m, pagebits);
3099 }
3100 
3101 void
3102 vm_page_clear_dirty(vm_page_t m, int base, int size)
3103 {
3104 
3105 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3106 }
3107 
3108 /*
3109  *	vm_page_set_invalid:
3110  *
3111  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
3112  *	valid and dirty bits for the effected areas are cleared.
3113  */
3114 void
3115 vm_page_set_invalid(vm_page_t m, int base, int size)
3116 {
3117 	vm_page_bits_t bits;
3118 	vm_object_t object;
3119 
3120 	object = m->object;
3121 	VM_OBJECT_ASSERT_WLOCKED(object);
3122 	if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3123 	    size >= object->un_pager.vnp.vnp_size)
3124 		bits = VM_PAGE_BITS_ALL;
3125 	else
3126 		bits = vm_page_bits(base, size);
3127 	if (m->valid == VM_PAGE_BITS_ALL && bits != 0)
3128 		pmap_remove_all(m);
3129 	KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3130 	    !pmap_page_is_mapped(m),
3131 	    ("vm_page_set_invalid: page %p is mapped", m));
3132 	m->valid &= ~bits;
3133 	m->dirty &= ~bits;
3134 }
3135 
3136 /*
3137  * vm_page_zero_invalid()
3138  *
3139  *	The kernel assumes that the invalid portions of a page contain
3140  *	garbage, but such pages can be mapped into memory by user code.
3141  *	When this occurs, we must zero out the non-valid portions of the
3142  *	page so user code sees what it expects.
3143  *
3144  *	Pages are most often semi-valid when the end of a file is mapped
3145  *	into memory and the file's size is not page aligned.
3146  */
3147 void
3148 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3149 {
3150 	int b;
3151 	int i;
3152 
3153 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3154 	/*
3155 	 * Scan the valid bits looking for invalid sections that
3156 	 * must be zerod.  Invalid sub-DEV_BSIZE'd areas ( where the
3157 	 * valid bit may be set ) have already been zerod by
3158 	 * vm_page_set_validclean().
3159 	 */
3160 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3161 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3162 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3163 			if (i > b) {
3164 				pmap_zero_page_area(m,
3165 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3166 			}
3167 			b = i + 1;
3168 		}
3169 	}
3170 
3171 	/*
3172 	 * setvalid is TRUE when we can safely set the zero'd areas
3173 	 * as being valid.  We can do this if there are no cache consistancy
3174 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3175 	 */
3176 	if (setvalid)
3177 		m->valid = VM_PAGE_BITS_ALL;
3178 }
3179 
3180 /*
3181  *	vm_page_is_valid:
3182  *
3183  *	Is (partial) page valid?  Note that the case where size == 0
3184  *	will return FALSE in the degenerate case where the page is
3185  *	entirely invalid, and TRUE otherwise.
3186  */
3187 int
3188 vm_page_is_valid(vm_page_t m, int base, int size)
3189 {
3190 	vm_page_bits_t bits;
3191 
3192 	VM_OBJECT_ASSERT_LOCKED(m->object);
3193 	bits = vm_page_bits(base, size);
3194 	return (m->valid != 0 && (m->valid & bits) == bits);
3195 }
3196 
3197 /*
3198  *	vm_page_ps_is_valid:
3199  *
3200  *	Returns TRUE if the entire (super)page is valid and FALSE otherwise.
3201  */
3202 boolean_t
3203 vm_page_ps_is_valid(vm_page_t m)
3204 {
3205 	int i, npages;
3206 
3207 	VM_OBJECT_ASSERT_LOCKED(m->object);
3208 	npages = atop(pagesizes[m->psind]);
3209 
3210 	/*
3211 	 * The physically contiguous pages that make up a superpage, i.e., a
3212 	 * page with a page size index ("psind") greater than zero, will
3213 	 * occupy adjacent entries in vm_page_array[].
3214 	 */
3215 	for (i = 0; i < npages; i++) {
3216 		if (m[i].valid != VM_PAGE_BITS_ALL)
3217 			return (FALSE);
3218 	}
3219 	return (TRUE);
3220 }
3221 
3222 /*
3223  * Set the page's dirty bits if the page is modified.
3224  */
3225 void
3226 vm_page_test_dirty(vm_page_t m)
3227 {
3228 
3229 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3230 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3231 		vm_page_dirty(m);
3232 }
3233 
3234 void
3235 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3236 {
3237 
3238 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3239 }
3240 
3241 void
3242 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3243 {
3244 
3245 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3246 }
3247 
3248 int
3249 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3250 {
3251 
3252 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3253 }
3254 
3255 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3256 void
3257 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3258 {
3259 
3260 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3261 }
3262 
3263 void
3264 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3265 {
3266 
3267 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3268 }
3269 #endif
3270 
3271 #ifdef INVARIANTS
3272 void
3273 vm_page_object_lock_assert(vm_page_t m)
3274 {
3275 
3276 	/*
3277 	 * Certain of the page's fields may only be modified by the
3278 	 * holder of the containing object's lock or the exclusive busy.
3279 	 * holder.  Unfortunately, the holder of the write busy is
3280 	 * not recorded, and thus cannot be checked here.
3281 	 */
3282 	if (m->object != NULL && !vm_page_xbusied(m))
3283 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3284 }
3285 
3286 void
3287 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
3288 {
3289 
3290 	if ((bits & PGA_WRITEABLE) == 0)
3291 		return;
3292 
3293 	/*
3294 	 * The PGA_WRITEABLE flag can only be set if the page is
3295 	 * managed, is exclusively busied or the object is locked.
3296 	 * Currently, this flag is only set by pmap_enter().
3297 	 */
3298 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3299 	    ("PGA_WRITEABLE on unmanaged page"));
3300 	if (!vm_page_xbusied(m))
3301 		VM_OBJECT_ASSERT_LOCKED(m->object);
3302 }
3303 #endif
3304 
3305 #include "opt_ddb.h"
3306 #ifdef DDB
3307 #include <sys/kernel.h>
3308 
3309 #include <ddb/ddb.h>
3310 
3311 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3312 {
3313 	db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count);
3314 	db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count);
3315 	db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count);
3316 	db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count);
3317 	db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count);
3318 	db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
3319 	db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
3320 	db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
3321 	db_printf("vm_cnt.v_cache_min: %d\n", vm_cnt.v_cache_min);
3322 	db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
3323 }
3324 
3325 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3326 {
3327 	int dom;
3328 
3329 	db_printf("pq_free %d pq_cache %d\n",
3330 	    vm_cnt.v_free_count, vm_cnt.v_cache_count);
3331 	for (dom = 0; dom < vm_ndomains; dom++) {
3332 		db_printf(
3333 	"dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n",
3334 		    dom,
3335 		    vm_dom[dom].vmd_page_count,
3336 		    vm_dom[dom].vmd_free_count,
3337 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3338 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3339 		    vm_dom[dom].vmd_pass);
3340 	}
3341 }
3342 
3343 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3344 {
3345 	vm_page_t m;
3346 	boolean_t phys;
3347 
3348 	if (!have_addr) {
3349 		db_printf("show pginfo addr\n");
3350 		return;
3351 	}
3352 
3353 	phys = strchr(modif, 'p') != NULL;
3354 	if (phys)
3355 		m = PHYS_TO_VM_PAGE(addr);
3356 	else
3357 		m = (vm_page_t)addr;
3358 	db_printf(
3359     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3360     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3361 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3362 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3363 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3364 }
3365 #endif /* DDB */
3366