xref: /freebsd/sys/vm/vm_page.c (revision 55620f43deef5c0eb5b4b0f675de18b30c8d1c2d)
1 /*-
2  * Copyright (c) 1991 Regents of the University of California.
3  * All rights reserved.
4  * Copyright (c) 1998 Matthew Dillon.  All Rights Reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * The Mach Operating System project at Carnegie-Mellon University.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	from: @(#)vm_page.c	7.4 (Berkeley) 5/7/91
34  */
35 
36 /*-
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  */
62 
63 /*
64  *			GENERAL RULES ON VM_PAGE MANIPULATION
65  *
66  *	- A page queue lock is required when adding or removing a page from a
67  *	  page queue regardless of other locks or the busy state of a page.
68  *
69  *		* In general, no thread besides the page daemon can acquire or
70  *		  hold more than one page queue lock at a time.
71  *
72  *		* The page daemon can acquire and hold any pair of page queue
73  *		  locks in any order.
74  *
75  *	- The object lock is required when inserting or removing
76  *	  pages from an object (vm_page_insert() or vm_page_remove()).
77  *
78  */
79 
80 /*
81  *	Resident memory management module.
82  */
83 
84 #include <sys/cdefs.h>
85 __FBSDID("$FreeBSD$");
86 
87 #include "opt_vm.h"
88 
89 #include <sys/param.h>
90 #include <sys/systm.h>
91 #include <sys/lock.h>
92 #include <sys/kernel.h>
93 #include <sys/limits.h>
94 #include <sys/linker.h>
95 #include <sys/malloc.h>
96 #include <sys/mman.h>
97 #include <sys/msgbuf.h>
98 #include <sys/mutex.h>
99 #include <sys/proc.h>
100 #include <sys/rwlock.h>
101 #include <sys/sbuf.h>
102 #include <sys/sysctl.h>
103 #include <sys/vmmeter.h>
104 #include <sys/vnode.h>
105 
106 #include <vm/vm.h>
107 #include <vm/pmap.h>
108 #include <vm/vm_param.h>
109 #include <vm/vm_kern.h>
110 #include <vm/vm_object.h>
111 #include <vm/vm_page.h>
112 #include <vm/vm_pageout.h>
113 #include <vm/vm_pager.h>
114 #include <vm/vm_phys.h>
115 #include <vm/vm_radix.h>
116 #include <vm/vm_reserv.h>
117 #include <vm/vm_extern.h>
118 #include <vm/uma.h>
119 #include <vm/uma_int.h>
120 
121 #include <machine/md_var.h>
122 
123 /*
124  *	Associated with page of user-allocatable memory is a
125  *	page structure.
126  */
127 
128 struct vm_domain vm_dom[MAXMEMDOM];
129 struct mtx_padalign vm_page_queue_free_mtx;
130 
131 struct mtx_padalign pa_lock[PA_LOCK_COUNT];
132 
133 vm_page_t vm_page_array;
134 long vm_page_array_size;
135 long first_page;
136 int vm_page_zero_count;
137 
138 static int boot_pages = UMA_BOOT_PAGES;
139 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
140     &boot_pages, 0,
141     "number of pages allocated for bootstrapping the VM system");
142 
143 static int pa_tryrelock_restart;
144 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
145     &pa_tryrelock_restart, 0, "Number of tryrelock restarts");
146 
147 static TAILQ_HEAD(, vm_page) blacklist_head;
148 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
149 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
150     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
151 
152 /* Is the page daemon waiting for free pages? */
153 static int vm_pageout_pages_needed;
154 
155 static uma_zone_t fakepg_zone;
156 
157 static struct vnode *vm_page_alloc_init(vm_page_t m);
158 static void vm_page_cache_turn_free(vm_page_t m);
159 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
160 static void vm_page_enqueue(uint8_t queue, vm_page_t m);
161 static void vm_page_free_wakeup(void);
162 static void vm_page_init_fakepg(void *dummy);
163 static int vm_page_insert_after(vm_page_t m, vm_object_t object,
164     vm_pindex_t pindex, vm_page_t mpred);
165 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
166     vm_page_t mpred);
167 static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
168     vm_paddr_t high);
169 
170 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL);
171 
172 static void
173 vm_page_init_fakepg(void *dummy)
174 {
175 
176 	fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
177 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
178 }
179 
180 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
181 #if PAGE_SIZE == 32768
182 #ifdef CTASSERT
183 CTASSERT(sizeof(u_long) >= 8);
184 #endif
185 #endif
186 
187 /*
188  * Try to acquire a physical address lock while a pmap is locked.  If we
189  * fail to trylock we unlock and lock the pmap directly and cache the
190  * locked pa in *locked.  The caller should then restart their loop in case
191  * the virtual to physical mapping has changed.
192  */
193 int
194 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
195 {
196 	vm_paddr_t lockpa;
197 
198 	lockpa = *locked;
199 	*locked = pa;
200 	if (lockpa) {
201 		PA_LOCK_ASSERT(lockpa, MA_OWNED);
202 		if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
203 			return (0);
204 		PA_UNLOCK(lockpa);
205 	}
206 	if (PA_TRYLOCK(pa))
207 		return (0);
208 	PMAP_UNLOCK(pmap);
209 	atomic_add_int(&pa_tryrelock_restart, 1);
210 	PA_LOCK(pa);
211 	PMAP_LOCK(pmap);
212 	return (EAGAIN);
213 }
214 
215 /*
216  *	vm_set_page_size:
217  *
218  *	Sets the page size, perhaps based upon the memory
219  *	size.  Must be called before any use of page-size
220  *	dependent functions.
221  */
222 void
223 vm_set_page_size(void)
224 {
225 	if (vm_cnt.v_page_size == 0)
226 		vm_cnt.v_page_size = PAGE_SIZE;
227 	if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
228 		panic("vm_set_page_size: page size not a power of two");
229 }
230 
231 /*
232  *	vm_page_blacklist_next:
233  *
234  *	Find the next entry in the provided string of blacklist
235  *	addresses.  Entries are separated by space, comma, or newline.
236  *	If an invalid integer is encountered then the rest of the
237  *	string is skipped.  Updates the list pointer to the next
238  *	character, or NULL if the string is exhausted or invalid.
239  */
240 static vm_paddr_t
241 vm_page_blacklist_next(char **list, char *end)
242 {
243 	vm_paddr_t bad;
244 	char *cp, *pos;
245 
246 	if (list == NULL || *list == NULL)
247 		return (0);
248 	if (**list =='\0') {
249 		*list = NULL;
250 		return (0);
251 	}
252 
253 	/*
254 	 * If there's no end pointer then the buffer is coming from
255 	 * the kenv and we know it's null-terminated.
256 	 */
257 	if (end == NULL)
258 		end = *list + strlen(*list);
259 
260 	/* Ensure that strtoq() won't walk off the end */
261 	if (*end != '\0') {
262 		if (*end == '\n' || *end == ' ' || *end  == ',')
263 			*end = '\0';
264 		else {
265 			printf("Blacklist not terminated, skipping\n");
266 			*list = NULL;
267 			return (0);
268 		}
269 	}
270 
271 	for (pos = *list; *pos != '\0'; pos = cp) {
272 		bad = strtoq(pos, &cp, 0);
273 		if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
274 			if (bad == 0) {
275 				if (++cp < end)
276 					continue;
277 				else
278 					break;
279 			}
280 		} else
281 			break;
282 		if (*cp == '\0' || ++cp >= end)
283 			*list = NULL;
284 		else
285 			*list = cp;
286 		return (trunc_page(bad));
287 	}
288 	printf("Garbage in RAM blacklist, skipping\n");
289 	*list = NULL;
290 	return (0);
291 }
292 
293 /*
294  *	vm_page_blacklist_check:
295  *
296  *	Iterate through the provided string of blacklist addresses, pulling
297  *	each entry out of the physical allocator free list and putting it
298  *	onto a list for reporting via the vm.page_blacklist sysctl.
299  */
300 static void
301 vm_page_blacklist_check(char *list, char *end)
302 {
303 	vm_paddr_t pa;
304 	vm_page_t m;
305 	char *next;
306 	int ret;
307 
308 	next = list;
309 	while (next != NULL) {
310 		if ((pa = vm_page_blacklist_next(&next, end)) == 0)
311 			continue;
312 		m = vm_phys_paddr_to_vm_page(pa);
313 		if (m == NULL)
314 			continue;
315 		mtx_lock(&vm_page_queue_free_mtx);
316 		ret = vm_phys_unfree_page(m);
317 		mtx_unlock(&vm_page_queue_free_mtx);
318 		if (ret == TRUE) {
319 			TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
320 			if (bootverbose)
321 				printf("Skipping page with pa 0x%jx\n",
322 				    (uintmax_t)pa);
323 		}
324 	}
325 }
326 
327 /*
328  *	vm_page_blacklist_load:
329  *
330  *	Search for a special module named "ram_blacklist".  It'll be a
331  *	plain text file provided by the user via the loader directive
332  *	of the same name.
333  */
334 static void
335 vm_page_blacklist_load(char **list, char **end)
336 {
337 	void *mod;
338 	u_char *ptr;
339 	u_int len;
340 
341 	mod = NULL;
342 	ptr = NULL;
343 
344 	mod = preload_search_by_type("ram_blacklist");
345 	if (mod != NULL) {
346 		ptr = preload_fetch_addr(mod);
347 		len = preload_fetch_size(mod);
348         }
349 	*list = ptr;
350 	if (ptr != NULL)
351 		*end = ptr + len;
352 	else
353 		*end = NULL;
354 	return;
355 }
356 
357 static int
358 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
359 {
360 	vm_page_t m;
361 	struct sbuf sbuf;
362 	int error, first;
363 
364 	first = 1;
365 	error = sysctl_wire_old_buffer(req, 0);
366 	if (error != 0)
367 		return (error);
368 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
369 	TAILQ_FOREACH(m, &blacklist_head, listq) {
370 		sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
371 		    (uintmax_t)m->phys_addr);
372 		first = 0;
373 	}
374 	error = sbuf_finish(&sbuf);
375 	sbuf_delete(&sbuf);
376 	return (error);
377 }
378 
379 static void
380 vm_page_domain_init(struct vm_domain *vmd)
381 {
382 	struct vm_pagequeue *pq;
383 	int i;
384 
385 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
386 	    "vm inactive pagequeue";
387 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) =
388 	    &vm_cnt.v_inactive_count;
389 	*__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
390 	    "vm active pagequeue";
391 	*__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) =
392 	    &vm_cnt.v_active_count;
393 	vmd->vmd_page_count = 0;
394 	vmd->vmd_free_count = 0;
395 	vmd->vmd_segs = 0;
396 	vmd->vmd_oom = FALSE;
397 	vmd->vmd_pass = 0;
398 	for (i = 0; i < PQ_COUNT; i++) {
399 		pq = &vmd->vmd_pagequeues[i];
400 		TAILQ_INIT(&pq->pq_pl);
401 		mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
402 		    MTX_DEF | MTX_DUPOK);
403 	}
404 }
405 
406 /*
407  *	vm_page_startup:
408  *
409  *	Initializes the resident memory module.
410  *
411  *	Allocates memory for the page cells, and
412  *	for the object/offset-to-page hash table headers.
413  *	Each page cell is initialized and placed on the free list.
414  */
415 vm_offset_t
416 vm_page_startup(vm_offset_t vaddr)
417 {
418 	vm_offset_t mapped;
419 	vm_paddr_t page_range;
420 	vm_paddr_t new_end;
421 	int i;
422 	vm_paddr_t pa;
423 	vm_paddr_t last_pa;
424 	char *list, *listend;
425 	vm_paddr_t end;
426 	vm_paddr_t biggestsize;
427 	vm_paddr_t low_water, high_water;
428 	int biggestone;
429 
430 	biggestsize = 0;
431 	biggestone = 0;
432 	vaddr = round_page(vaddr);
433 
434 	for (i = 0; phys_avail[i + 1]; i += 2) {
435 		phys_avail[i] = round_page(phys_avail[i]);
436 		phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
437 	}
438 
439 	low_water = phys_avail[0];
440 	high_water = phys_avail[1];
441 
442 	for (i = 0; i < vm_phys_nsegs; i++) {
443 		if (vm_phys_segs[i].start < low_water)
444 			low_water = vm_phys_segs[i].start;
445 		if (vm_phys_segs[i].end > high_water)
446 			high_water = vm_phys_segs[i].end;
447 	}
448 	for (i = 0; phys_avail[i + 1]; i += 2) {
449 		vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
450 
451 		if (size > biggestsize) {
452 			biggestone = i;
453 			biggestsize = size;
454 		}
455 		if (phys_avail[i] < low_water)
456 			low_water = phys_avail[i];
457 		if (phys_avail[i + 1] > high_water)
458 			high_water = phys_avail[i + 1];
459 	}
460 
461 	end = phys_avail[biggestone+1];
462 
463 	/*
464 	 * Initialize the page and queue locks.
465 	 */
466 	mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF);
467 	for (i = 0; i < PA_LOCK_COUNT; i++)
468 		mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
469 	for (i = 0; i < vm_ndomains; i++)
470 		vm_page_domain_init(&vm_dom[i]);
471 
472 	/*
473 	 * Allocate memory for use when boot strapping the kernel memory
474 	 * allocator.
475 	 *
476 	 * CTFLAG_RDTUN doesn't work during the early boot process, so we must
477 	 * manually fetch the value.
478 	 */
479 	TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages);
480 	new_end = end - (boot_pages * UMA_SLAB_SIZE);
481 	new_end = trunc_page(new_end);
482 	mapped = pmap_map(&vaddr, new_end, end,
483 	    VM_PROT_READ | VM_PROT_WRITE);
484 	bzero((void *)mapped, end - new_end);
485 	uma_startup((void *)mapped, boot_pages);
486 
487 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
488     defined(__i386__) || defined(__mips__)
489 	/*
490 	 * Allocate a bitmap to indicate that a random physical page
491 	 * needs to be included in a minidump.
492 	 *
493 	 * The amd64 port needs this to indicate which direct map pages
494 	 * need to be dumped, via calls to dump_add_page()/dump_drop_page().
495 	 *
496 	 * However, i386 still needs this workspace internally within the
497 	 * minidump code.  In theory, they are not needed on i386, but are
498 	 * included should the sf_buf code decide to use them.
499 	 */
500 	last_pa = 0;
501 	for (i = 0; dump_avail[i + 1] != 0; i += 2)
502 		if (dump_avail[i + 1] > last_pa)
503 			last_pa = dump_avail[i + 1];
504 	page_range = last_pa / PAGE_SIZE;
505 	vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
506 	new_end -= vm_page_dump_size;
507 	vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
508 	    new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
509 	bzero((void *)vm_page_dump, vm_page_dump_size);
510 #endif
511 #ifdef __amd64__
512 	/*
513 	 * Request that the physical pages underlying the message buffer be
514 	 * included in a crash dump.  Since the message buffer is accessed
515 	 * through the direct map, they are not automatically included.
516 	 */
517 	pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
518 	last_pa = pa + round_page(msgbufsize);
519 	while (pa < last_pa) {
520 		dump_add_page(pa);
521 		pa += PAGE_SIZE;
522 	}
523 #endif
524 	/*
525 	 * Compute the number of pages of memory that will be available for
526 	 * use (taking into account the overhead of a page structure per
527 	 * page).
528 	 */
529 	first_page = low_water / PAGE_SIZE;
530 #ifdef VM_PHYSSEG_SPARSE
531 	page_range = 0;
532 	for (i = 0; i < vm_phys_nsegs; i++) {
533 		page_range += atop(vm_phys_segs[i].end -
534 		    vm_phys_segs[i].start);
535 	}
536 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
537 		page_range += atop(phys_avail[i + 1] - phys_avail[i]);
538 #elif defined(VM_PHYSSEG_DENSE)
539 	page_range = high_water / PAGE_SIZE - first_page;
540 #else
541 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
542 #endif
543 	end = new_end;
544 
545 	/*
546 	 * Reserve an unmapped guard page to trap access to vm_page_array[-1].
547 	 */
548 	vaddr += PAGE_SIZE;
549 
550 	/*
551 	 * Initialize the mem entry structures now, and put them in the free
552 	 * queue.
553 	 */
554 	new_end = trunc_page(end - page_range * sizeof(struct vm_page));
555 	mapped = pmap_map(&vaddr, new_end, end,
556 	    VM_PROT_READ | VM_PROT_WRITE);
557 	vm_page_array = (vm_page_t) mapped;
558 #if VM_NRESERVLEVEL > 0
559 	/*
560 	 * Allocate memory for the reservation management system's data
561 	 * structures.
562 	 */
563 	new_end = vm_reserv_startup(&vaddr, new_end, high_water);
564 #endif
565 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
566 	/*
567 	 * pmap_map on arm64, amd64, and mips can come out of the direct-map,
568 	 * not kvm like i386, so the pages must be tracked for a crashdump to
569 	 * include this data.  This includes the vm_page_array and the early
570 	 * UMA bootstrap pages.
571 	 */
572 	for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
573 		dump_add_page(pa);
574 #endif
575 	phys_avail[biggestone + 1] = new_end;
576 
577 	/*
578 	 * Add physical memory segments corresponding to the available
579 	 * physical pages.
580 	 */
581 	for (i = 0; phys_avail[i + 1] != 0; i += 2)
582 		vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
583 
584 	/*
585 	 * Clear all of the page structures
586 	 */
587 	bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
588 	for (i = 0; i < page_range; i++)
589 		vm_page_array[i].order = VM_NFREEORDER;
590 	vm_page_array_size = page_range;
591 
592 	/*
593 	 * Initialize the physical memory allocator.
594 	 */
595 	vm_phys_init();
596 
597 	/*
598 	 * Add every available physical page that is not blacklisted to
599 	 * the free lists.
600 	 */
601 	vm_cnt.v_page_count = 0;
602 	vm_cnt.v_free_count = 0;
603 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
604 		pa = phys_avail[i];
605 		last_pa = phys_avail[i + 1];
606 		while (pa < last_pa) {
607 			vm_phys_add_page(pa);
608 			pa += PAGE_SIZE;
609 		}
610 	}
611 
612 	TAILQ_INIT(&blacklist_head);
613 	vm_page_blacklist_load(&list, &listend);
614 	vm_page_blacklist_check(list, listend);
615 
616 	list = kern_getenv("vm.blacklist");
617 	vm_page_blacklist_check(list, NULL);
618 
619 	freeenv(list);
620 #if VM_NRESERVLEVEL > 0
621 	/*
622 	 * Initialize the reservation management system.
623 	 */
624 	vm_reserv_init();
625 #endif
626 	return (vaddr);
627 }
628 
629 void
630 vm_page_reference(vm_page_t m)
631 {
632 
633 	vm_page_aflag_set(m, PGA_REFERENCED);
634 }
635 
636 /*
637  *	vm_page_busy_downgrade:
638  *
639  *	Downgrade an exclusive busy page into a single shared busy page.
640  */
641 void
642 vm_page_busy_downgrade(vm_page_t m)
643 {
644 	u_int x;
645 
646 	vm_page_assert_xbusied(m);
647 
648 	for (;;) {
649 		x = m->busy_lock;
650 		x &= VPB_BIT_WAITERS;
651 		if (atomic_cmpset_rel_int(&m->busy_lock,
652 		    VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x))
653 			break;
654 	}
655 }
656 
657 /*
658  *	vm_page_sbusied:
659  *
660  *	Return a positive value if the page is shared busied, 0 otherwise.
661  */
662 int
663 vm_page_sbusied(vm_page_t m)
664 {
665 	u_int x;
666 
667 	x = m->busy_lock;
668 	return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
669 }
670 
671 /*
672  *	vm_page_sunbusy:
673  *
674  *	Shared unbusy a page.
675  */
676 void
677 vm_page_sunbusy(vm_page_t m)
678 {
679 	u_int x;
680 
681 	vm_page_assert_sbusied(m);
682 
683 	for (;;) {
684 		x = m->busy_lock;
685 		if (VPB_SHARERS(x) > 1) {
686 			if (atomic_cmpset_int(&m->busy_lock, x,
687 			    x - VPB_ONE_SHARER))
688 				break;
689 			continue;
690 		}
691 		if ((x & VPB_BIT_WAITERS) == 0) {
692 			KASSERT(x == VPB_SHARERS_WORD(1),
693 			    ("vm_page_sunbusy: invalid lock state"));
694 			if (atomic_cmpset_int(&m->busy_lock,
695 			    VPB_SHARERS_WORD(1), VPB_UNBUSIED))
696 				break;
697 			continue;
698 		}
699 		KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS),
700 		    ("vm_page_sunbusy: invalid lock state for waiters"));
701 
702 		vm_page_lock(m);
703 		if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) {
704 			vm_page_unlock(m);
705 			continue;
706 		}
707 		wakeup(m);
708 		vm_page_unlock(m);
709 		break;
710 	}
711 }
712 
713 /*
714  *	vm_page_busy_sleep:
715  *
716  *	Sleep and release the page lock, using the page pointer as wchan.
717  *	This is used to implement the hard-path of busying mechanism.
718  *
719  *	The given page must be locked.
720  */
721 void
722 vm_page_busy_sleep(vm_page_t m, const char *wmesg)
723 {
724 	u_int x;
725 
726 	vm_page_lock_assert(m, MA_OWNED);
727 
728 	x = m->busy_lock;
729 	if (x == VPB_UNBUSIED) {
730 		vm_page_unlock(m);
731 		return;
732 	}
733 	if ((x & VPB_BIT_WAITERS) == 0 &&
734 	    !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) {
735 		vm_page_unlock(m);
736 		return;
737 	}
738 	msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0);
739 }
740 
741 /*
742  *	vm_page_trysbusy:
743  *
744  *	Try to shared busy a page.
745  *	If the operation succeeds 1 is returned otherwise 0.
746  *	The operation never sleeps.
747  */
748 int
749 vm_page_trysbusy(vm_page_t m)
750 {
751 	u_int x;
752 
753 	for (;;) {
754 		x = m->busy_lock;
755 		if ((x & VPB_BIT_SHARED) == 0)
756 			return (0);
757 		if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER))
758 			return (1);
759 	}
760 }
761 
762 static void
763 vm_page_xunbusy_maybelocked(vm_page_t m)
764 {
765 	bool lockacq;
766 
767 	vm_page_assert_xbusied(m);
768 
769 	lockacq = !mtx_owned(vm_page_lockptr(m));
770 	if (lockacq)
771 		vm_page_lock(m);
772 	vm_page_flash(m);
773 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
774 	if (lockacq)
775 		vm_page_unlock(m);
776 }
777 
778 /*
779  *	vm_page_xunbusy_hard:
780  *
781  *	Called after the first try the exclusive unbusy of a page failed.
782  *	It is assumed that the waiters bit is on.
783  */
784 void
785 vm_page_xunbusy_hard(vm_page_t m)
786 {
787 
788 	vm_page_assert_xbusied(m);
789 
790 	vm_page_lock(m);
791 	atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
792 	wakeup(m);
793 	vm_page_unlock(m);
794 }
795 
796 /*
797  *	vm_page_flash:
798  *
799  *	Wakeup anyone waiting for the page.
800  *	The ownership bits do not change.
801  *
802  *	The given page must be locked.
803  */
804 void
805 vm_page_flash(vm_page_t m)
806 {
807 	u_int x;
808 
809 	vm_page_lock_assert(m, MA_OWNED);
810 
811 	for (;;) {
812 		x = m->busy_lock;
813 		if ((x & VPB_BIT_WAITERS) == 0)
814 			return;
815 		if (atomic_cmpset_int(&m->busy_lock, x,
816 		    x & (~VPB_BIT_WAITERS)))
817 			break;
818 	}
819 	wakeup(m);
820 }
821 
822 /*
823  * Keep page from being freed by the page daemon
824  * much of the same effect as wiring, except much lower
825  * overhead and should be used only for *very* temporary
826  * holding ("wiring").
827  */
828 void
829 vm_page_hold(vm_page_t mem)
830 {
831 
832 	vm_page_lock_assert(mem, MA_OWNED);
833         mem->hold_count++;
834 }
835 
836 void
837 vm_page_unhold(vm_page_t mem)
838 {
839 
840 	vm_page_lock_assert(mem, MA_OWNED);
841 	KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!"));
842 	--mem->hold_count;
843 	if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0)
844 		vm_page_free_toq(mem);
845 }
846 
847 /*
848  *	vm_page_unhold_pages:
849  *
850  *	Unhold each of the pages that is referenced by the given array.
851  */
852 void
853 vm_page_unhold_pages(vm_page_t *ma, int count)
854 {
855 	struct mtx *mtx, *new_mtx;
856 
857 	mtx = NULL;
858 	for (; count != 0; count--) {
859 		/*
860 		 * Avoid releasing and reacquiring the same page lock.
861 		 */
862 		new_mtx = vm_page_lockptr(*ma);
863 		if (mtx != new_mtx) {
864 			if (mtx != NULL)
865 				mtx_unlock(mtx);
866 			mtx = new_mtx;
867 			mtx_lock(mtx);
868 		}
869 		vm_page_unhold(*ma);
870 		ma++;
871 	}
872 	if (mtx != NULL)
873 		mtx_unlock(mtx);
874 }
875 
876 vm_page_t
877 PHYS_TO_VM_PAGE(vm_paddr_t pa)
878 {
879 	vm_page_t m;
880 
881 #ifdef VM_PHYSSEG_SPARSE
882 	m = vm_phys_paddr_to_vm_page(pa);
883 	if (m == NULL)
884 		m = vm_phys_fictitious_to_vm_page(pa);
885 	return (m);
886 #elif defined(VM_PHYSSEG_DENSE)
887 	long pi;
888 
889 	pi = atop(pa);
890 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
891 		m = &vm_page_array[pi - first_page];
892 		return (m);
893 	}
894 	return (vm_phys_fictitious_to_vm_page(pa));
895 #else
896 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
897 #endif
898 }
899 
900 /*
901  *	vm_page_getfake:
902  *
903  *	Create a fictitious page with the specified physical address and
904  *	memory attribute.  The memory attribute is the only the machine-
905  *	dependent aspect of a fictitious page that must be initialized.
906  */
907 vm_page_t
908 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
909 {
910 	vm_page_t m;
911 
912 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
913 	vm_page_initfake(m, paddr, memattr);
914 	return (m);
915 }
916 
917 void
918 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
919 {
920 
921 	if ((m->flags & PG_FICTITIOUS) != 0) {
922 		/*
923 		 * The page's memattr might have changed since the
924 		 * previous initialization.  Update the pmap to the
925 		 * new memattr.
926 		 */
927 		goto memattr;
928 	}
929 	m->phys_addr = paddr;
930 	m->queue = PQ_NONE;
931 	/* Fictitious pages don't use "segind". */
932 	m->flags = PG_FICTITIOUS;
933 	/* Fictitious pages don't use "order" or "pool". */
934 	m->oflags = VPO_UNMANAGED;
935 	m->busy_lock = VPB_SINGLE_EXCLUSIVER;
936 	m->wire_count = 1;
937 	pmap_page_init(m);
938 memattr:
939 	pmap_page_set_memattr(m, memattr);
940 }
941 
942 /*
943  *	vm_page_putfake:
944  *
945  *	Release a fictitious page.
946  */
947 void
948 vm_page_putfake(vm_page_t m)
949 {
950 
951 	KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
952 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
953 	    ("vm_page_putfake: bad page %p", m));
954 	uma_zfree(fakepg_zone, m);
955 }
956 
957 /*
958  *	vm_page_updatefake:
959  *
960  *	Update the given fictitious page to the specified physical address and
961  *	memory attribute.
962  */
963 void
964 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
965 {
966 
967 	KASSERT((m->flags & PG_FICTITIOUS) != 0,
968 	    ("vm_page_updatefake: bad page %p", m));
969 	m->phys_addr = paddr;
970 	pmap_page_set_memattr(m, memattr);
971 }
972 
973 /*
974  *	vm_page_free:
975  *
976  *	Free a page.
977  */
978 void
979 vm_page_free(vm_page_t m)
980 {
981 
982 	m->flags &= ~PG_ZERO;
983 	vm_page_free_toq(m);
984 }
985 
986 /*
987  *	vm_page_free_zero:
988  *
989  *	Free a page to the zerod-pages queue
990  */
991 void
992 vm_page_free_zero(vm_page_t m)
993 {
994 
995 	m->flags |= PG_ZERO;
996 	vm_page_free_toq(m);
997 }
998 
999 /*
1000  * Unbusy and handle the page queueing for a page from the VOP_GETPAGES()
1001  * array which was optionally read ahead or behind.
1002  */
1003 void
1004 vm_page_readahead_finish(vm_page_t m)
1005 {
1006 
1007 	/* We shouldn't put invalid pages on queues. */
1008 	KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m));
1009 
1010 	/*
1011 	 * Since the page is not the actually needed one, whether it should
1012 	 * be activated or deactivated is not obvious.  Empirical results
1013 	 * have shown that deactivating the page is usually the best choice,
1014 	 * unless the page is wanted by another thread.
1015 	 */
1016 	vm_page_lock(m);
1017 	if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1018 		vm_page_activate(m);
1019 	else
1020 		vm_page_deactivate(m);
1021 	vm_page_unlock(m);
1022 	vm_page_xunbusy(m);
1023 }
1024 
1025 /*
1026  *	vm_page_sleep_if_busy:
1027  *
1028  *	Sleep and release the page queues lock if the page is busied.
1029  *	Returns TRUE if the thread slept.
1030  *
1031  *	The given page must be unlocked and object containing it must
1032  *	be locked.
1033  */
1034 int
1035 vm_page_sleep_if_busy(vm_page_t m, const char *msg)
1036 {
1037 	vm_object_t obj;
1038 
1039 	vm_page_lock_assert(m, MA_NOTOWNED);
1040 	VM_OBJECT_ASSERT_WLOCKED(m->object);
1041 
1042 	if (vm_page_busied(m)) {
1043 		/*
1044 		 * The page-specific object must be cached because page
1045 		 * identity can change during the sleep, causing the
1046 		 * re-lock of a different object.
1047 		 * It is assumed that a reference to the object is already
1048 		 * held by the callers.
1049 		 */
1050 		obj = m->object;
1051 		vm_page_lock(m);
1052 		VM_OBJECT_WUNLOCK(obj);
1053 		vm_page_busy_sleep(m, msg);
1054 		VM_OBJECT_WLOCK(obj);
1055 		return (TRUE);
1056 	}
1057 	return (FALSE);
1058 }
1059 
1060 /*
1061  *	vm_page_dirty_KBI:		[ internal use only ]
1062  *
1063  *	Set all bits in the page's dirty field.
1064  *
1065  *	The object containing the specified page must be locked if the
1066  *	call is made from the machine-independent layer.
1067  *
1068  *	See vm_page_clear_dirty_mask().
1069  *
1070  *	This function should only be called by vm_page_dirty().
1071  */
1072 void
1073 vm_page_dirty_KBI(vm_page_t m)
1074 {
1075 
1076 	/* These assertions refer to this operation by its public name. */
1077 	KASSERT((m->flags & PG_CACHED) == 0,
1078 	    ("vm_page_dirty: page in cache!"));
1079 	KASSERT(m->valid == VM_PAGE_BITS_ALL,
1080 	    ("vm_page_dirty: page is invalid!"));
1081 	m->dirty = VM_PAGE_BITS_ALL;
1082 }
1083 
1084 /*
1085  *	vm_page_insert:		[ internal use only ]
1086  *
1087  *	Inserts the given mem entry into the object and object list.
1088  *
1089  *	The object must be locked.
1090  */
1091 int
1092 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
1093 {
1094 	vm_page_t mpred;
1095 
1096 	VM_OBJECT_ASSERT_WLOCKED(object);
1097 	mpred = vm_radix_lookup_le(&object->rtree, pindex);
1098 	return (vm_page_insert_after(m, object, pindex, mpred));
1099 }
1100 
1101 /*
1102  *	vm_page_insert_after:
1103  *
1104  *	Inserts the page "m" into the specified object at offset "pindex".
1105  *
1106  *	The page "mpred" must immediately precede the offset "pindex" within
1107  *	the specified object.
1108  *
1109  *	The object must be locked.
1110  */
1111 static int
1112 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
1113     vm_page_t mpred)
1114 {
1115 	vm_page_t msucc;
1116 
1117 	VM_OBJECT_ASSERT_WLOCKED(object);
1118 	KASSERT(m->object == NULL,
1119 	    ("vm_page_insert_after: page already inserted"));
1120 	if (mpred != NULL) {
1121 		KASSERT(mpred->object == object,
1122 		    ("vm_page_insert_after: object doesn't contain mpred"));
1123 		KASSERT(mpred->pindex < pindex,
1124 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1125 		msucc = TAILQ_NEXT(mpred, listq);
1126 	} else
1127 		msucc = TAILQ_FIRST(&object->memq);
1128 	if (msucc != NULL)
1129 		KASSERT(msucc->pindex > pindex,
1130 		    ("vm_page_insert_after: msucc doesn't succeed pindex"));
1131 
1132 	/*
1133 	 * Record the object/offset pair in this page
1134 	 */
1135 	m->object = object;
1136 	m->pindex = pindex;
1137 
1138 	/*
1139 	 * Now link into the object's ordered list of backed pages.
1140 	 */
1141 	if (vm_radix_insert(&object->rtree, m)) {
1142 		m->object = NULL;
1143 		m->pindex = 0;
1144 		return (1);
1145 	}
1146 	vm_page_insert_radixdone(m, object, mpred);
1147 	return (0);
1148 }
1149 
1150 /*
1151  *	vm_page_insert_radixdone:
1152  *
1153  *	Complete page "m" insertion into the specified object after the
1154  *	radix trie hooking.
1155  *
1156  *	The page "mpred" must precede the offset "m->pindex" within the
1157  *	specified object.
1158  *
1159  *	The object must be locked.
1160  */
1161 static void
1162 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
1163 {
1164 
1165 	VM_OBJECT_ASSERT_WLOCKED(object);
1166 	KASSERT(object != NULL && m->object == object,
1167 	    ("vm_page_insert_radixdone: page %p has inconsistent object", m));
1168 	if (mpred != NULL) {
1169 		KASSERT(mpred->object == object,
1170 		    ("vm_page_insert_after: object doesn't contain mpred"));
1171 		KASSERT(mpred->pindex < m->pindex,
1172 		    ("vm_page_insert_after: mpred doesn't precede pindex"));
1173 	}
1174 
1175 	if (mpred != NULL)
1176 		TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
1177 	else
1178 		TAILQ_INSERT_HEAD(&object->memq, m, listq);
1179 
1180 	/*
1181 	 * Show that the object has one more resident page.
1182 	 */
1183 	object->resident_page_count++;
1184 
1185 	/*
1186 	 * Hold the vnode until the last page is released.
1187 	 */
1188 	if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
1189 		vhold(object->handle);
1190 
1191 	/*
1192 	 * Since we are inserting a new and possibly dirty page,
1193 	 * update the object's OBJ_MIGHTBEDIRTY flag.
1194 	 */
1195 	if (pmap_page_is_write_mapped(m))
1196 		vm_object_set_writeable_dirty(object);
1197 }
1198 
1199 /*
1200  *	vm_page_remove:
1201  *
1202  *	Removes the given mem entry from the object/offset-page
1203  *	table and the object page list, but do not invalidate/terminate
1204  *	the backing store.
1205  *
1206  *	The object must be locked.  The page must be locked if it is managed.
1207  */
1208 void
1209 vm_page_remove(vm_page_t m)
1210 {
1211 	vm_object_t object;
1212 
1213 	if ((m->oflags & VPO_UNMANAGED) == 0)
1214 		vm_page_assert_locked(m);
1215 	if ((object = m->object) == NULL)
1216 		return;
1217 	VM_OBJECT_ASSERT_WLOCKED(object);
1218 	if (vm_page_xbusied(m))
1219 		vm_page_xunbusy_maybelocked(m);
1220 
1221 	/*
1222 	 * Now remove from the object's list of backed pages.
1223 	 */
1224 	vm_radix_remove(&object->rtree, m->pindex);
1225 	TAILQ_REMOVE(&object->memq, m, listq);
1226 
1227 	/*
1228 	 * And show that the object has one fewer resident page.
1229 	 */
1230 	object->resident_page_count--;
1231 
1232 	/*
1233 	 * The vnode may now be recycled.
1234 	 */
1235 	if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
1236 		vdrop(object->handle);
1237 
1238 	m->object = NULL;
1239 }
1240 
1241 /*
1242  *	vm_page_lookup:
1243  *
1244  *	Returns the page associated with the object/offset
1245  *	pair specified; if none is found, NULL is returned.
1246  *
1247  *	The object must be locked.
1248  */
1249 vm_page_t
1250 vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
1251 {
1252 
1253 	VM_OBJECT_ASSERT_LOCKED(object);
1254 	return (vm_radix_lookup(&object->rtree, pindex));
1255 }
1256 
1257 /*
1258  *	vm_page_find_least:
1259  *
1260  *	Returns the page associated with the object with least pindex
1261  *	greater than or equal to the parameter pindex, or NULL.
1262  *
1263  *	The object must be locked.
1264  */
1265 vm_page_t
1266 vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
1267 {
1268 	vm_page_t m;
1269 
1270 	VM_OBJECT_ASSERT_LOCKED(object);
1271 	if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
1272 		m = vm_radix_lookup_ge(&object->rtree, pindex);
1273 	return (m);
1274 }
1275 
1276 /*
1277  * Returns the given page's successor (by pindex) within the object if it is
1278  * resident; if none is found, NULL is returned.
1279  *
1280  * The object must be locked.
1281  */
1282 vm_page_t
1283 vm_page_next(vm_page_t m)
1284 {
1285 	vm_page_t next;
1286 
1287 	VM_OBJECT_ASSERT_LOCKED(m->object);
1288 	if ((next = TAILQ_NEXT(m, listq)) != NULL &&
1289 	    next->pindex != m->pindex + 1)
1290 		next = NULL;
1291 	return (next);
1292 }
1293 
1294 /*
1295  * Returns the given page's predecessor (by pindex) within the object if it is
1296  * resident; if none is found, NULL is returned.
1297  *
1298  * The object must be locked.
1299  */
1300 vm_page_t
1301 vm_page_prev(vm_page_t m)
1302 {
1303 	vm_page_t prev;
1304 
1305 	VM_OBJECT_ASSERT_LOCKED(m->object);
1306 	if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL &&
1307 	    prev->pindex != m->pindex - 1)
1308 		prev = NULL;
1309 	return (prev);
1310 }
1311 
1312 /*
1313  * Uses the page mnew as a replacement for an existing page at index
1314  * pindex which must be already present in the object.
1315  *
1316  * The existing page must not be on a paging queue.
1317  */
1318 vm_page_t
1319 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex)
1320 {
1321 	vm_page_t mold;
1322 
1323 	VM_OBJECT_ASSERT_WLOCKED(object);
1324 	KASSERT(mnew->object == NULL,
1325 	    ("vm_page_replace: page already in object"));
1326 
1327 	/*
1328 	 * This function mostly follows vm_page_insert() and
1329 	 * vm_page_remove() without the radix, object count and vnode
1330 	 * dance.  Double check such functions for more comments.
1331 	 */
1332 
1333 	mnew->object = object;
1334 	mnew->pindex = pindex;
1335 	mold = vm_radix_replace(&object->rtree, mnew);
1336 	KASSERT(mold->queue == PQ_NONE,
1337 	    ("vm_page_replace: mold is on a paging queue"));
1338 
1339 	/* Keep the resident page list in sorted order. */
1340 	TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
1341 	TAILQ_REMOVE(&object->memq, mold, listq);
1342 
1343 	mold->object = NULL;
1344 	vm_page_xunbusy_maybelocked(mold);
1345 
1346 	/*
1347 	 * The object's resident_page_count does not change because we have
1348 	 * swapped one page for another, but OBJ_MIGHTBEDIRTY.
1349 	 */
1350 	if (pmap_page_is_write_mapped(mnew))
1351 		vm_object_set_writeable_dirty(object);
1352 	return (mold);
1353 }
1354 
1355 /*
1356  *	vm_page_rename:
1357  *
1358  *	Move the given memory entry from its
1359  *	current object to the specified target object/offset.
1360  *
1361  *	Note: swap associated with the page must be invalidated by the move.  We
1362  *	      have to do this for several reasons:  (1) we aren't freeing the
1363  *	      page, (2) we are dirtying the page, (3) the VM system is probably
1364  *	      moving the page from object A to B, and will then later move
1365  *	      the backing store from A to B and we can't have a conflict.
1366  *
1367  *	Note: we *always* dirty the page.  It is necessary both for the
1368  *	      fact that we moved it, and because we may be invalidating
1369  *	      swap.  If the page is on the cache, we have to deactivate it
1370  *	      or vm_page_dirty() will panic.  Dirty pages are not allowed
1371  *	      on the cache.
1372  *
1373  *	The objects must be locked.
1374  */
1375 int
1376 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
1377 {
1378 	vm_page_t mpred;
1379 	vm_pindex_t opidx;
1380 
1381 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1382 
1383 	mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex);
1384 	KASSERT(mpred == NULL || mpred->pindex != new_pindex,
1385 	    ("vm_page_rename: pindex already renamed"));
1386 
1387 	/*
1388 	 * Create a custom version of vm_page_insert() which does not depend
1389 	 * by m_prev and can cheat on the implementation aspects of the
1390 	 * function.
1391 	 */
1392 	opidx = m->pindex;
1393 	m->pindex = new_pindex;
1394 	if (vm_radix_insert(&new_object->rtree, m)) {
1395 		m->pindex = opidx;
1396 		return (1);
1397 	}
1398 
1399 	/*
1400 	 * The operation cannot fail anymore.  The removal must happen before
1401 	 * the listq iterator is tainted.
1402 	 */
1403 	m->pindex = opidx;
1404 	vm_page_lock(m);
1405 	vm_page_remove(m);
1406 
1407 	/* Return back to the new pindex to complete vm_page_insert(). */
1408 	m->pindex = new_pindex;
1409 	m->object = new_object;
1410 	vm_page_unlock(m);
1411 	vm_page_insert_radixdone(m, new_object, mpred);
1412 	vm_page_dirty(m);
1413 	return (0);
1414 }
1415 
1416 /*
1417  *	Convert all of the given object's cached pages that have a
1418  *	pindex within the given range into free pages.  If the value
1419  *	zero is given for "end", then the range's upper bound is
1420  *	infinity.  If the given object is backed by a vnode and it
1421  *	transitions from having one or more cached pages to none, the
1422  *	vnode's hold count is reduced.
1423  */
1424 void
1425 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1426 {
1427 	vm_page_t m;
1428 	boolean_t empty;
1429 
1430 	mtx_lock(&vm_page_queue_free_mtx);
1431 	if (__predict_false(vm_radix_is_empty(&object->cache))) {
1432 		mtx_unlock(&vm_page_queue_free_mtx);
1433 		return;
1434 	}
1435 	while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) {
1436 		if (end != 0 && m->pindex >= end)
1437 			break;
1438 		vm_radix_remove(&object->cache, m->pindex);
1439 		vm_page_cache_turn_free(m);
1440 	}
1441 	empty = vm_radix_is_empty(&object->cache);
1442 	mtx_unlock(&vm_page_queue_free_mtx);
1443 	if (object->type == OBJT_VNODE && empty)
1444 		vdrop(object->handle);
1445 }
1446 
1447 /*
1448  *	Returns the cached page that is associated with the given
1449  *	object and offset.  If, however, none exists, returns NULL.
1450  *
1451  *	The free page queue must be locked.
1452  */
1453 static inline vm_page_t
1454 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
1455 {
1456 
1457 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1458 	return (vm_radix_lookup(&object->cache, pindex));
1459 }
1460 
1461 /*
1462  *	Remove the given cached page from its containing object's
1463  *	collection of cached pages.
1464  *
1465  *	The free page queue must be locked.
1466  */
1467 static void
1468 vm_page_cache_remove(vm_page_t m)
1469 {
1470 
1471 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1472 	KASSERT((m->flags & PG_CACHED) != 0,
1473 	    ("vm_page_cache_remove: page %p is not cached", m));
1474 	vm_radix_remove(&m->object->cache, m->pindex);
1475 	m->object = NULL;
1476 	vm_cnt.v_cache_count--;
1477 }
1478 
1479 /*
1480  *	Transfer all of the cached pages with offset greater than or
1481  *	equal to 'offidxstart' from the original object's cache to the
1482  *	new object's cache.  However, any cached pages with offset
1483  *	greater than or equal to the new object's size are kept in the
1484  *	original object.  Initially, the new object's cache must be
1485  *	empty.  Offset 'offidxstart' in the original object must
1486  *	correspond to offset zero in the new object.
1487  *
1488  *	The new object must be locked.
1489  */
1490 void
1491 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
1492     vm_object_t new_object)
1493 {
1494 	vm_page_t m;
1495 
1496 	/*
1497 	 * Insertion into an object's collection of cached pages
1498 	 * requires the object to be locked.  In contrast, removal does
1499 	 * not.
1500 	 */
1501 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1502 	KASSERT(vm_radix_is_empty(&new_object->cache),
1503 	    ("vm_page_cache_transfer: object %p has cached pages",
1504 	    new_object));
1505 	mtx_lock(&vm_page_queue_free_mtx);
1506 	while ((m = vm_radix_lookup_ge(&orig_object->cache,
1507 	    offidxstart)) != NULL) {
1508 		/*
1509 		 * Transfer all of the pages with offset greater than or
1510 		 * equal to 'offidxstart' from the original object's
1511 		 * cache to the new object's cache.
1512 		 */
1513 		if ((m->pindex - offidxstart) >= new_object->size)
1514 			break;
1515 		vm_radix_remove(&orig_object->cache, m->pindex);
1516 		/* Update the page's object and offset. */
1517 		m->object = new_object;
1518 		m->pindex -= offidxstart;
1519 		if (vm_radix_insert(&new_object->cache, m))
1520 			vm_page_cache_turn_free(m);
1521 	}
1522 	mtx_unlock(&vm_page_queue_free_mtx);
1523 }
1524 
1525 /*
1526  *	Returns TRUE if a cached page is associated with the given object and
1527  *	offset, and FALSE otherwise.
1528  *
1529  *	The object must be locked.
1530  */
1531 boolean_t
1532 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex)
1533 {
1534 	vm_page_t m;
1535 
1536 	/*
1537 	 * Insertion into an object's collection of cached pages requires the
1538 	 * object to be locked.  Therefore, if the object is locked and the
1539 	 * object's collection is empty, there is no need to acquire the free
1540 	 * page queues lock in order to prove that the specified page doesn't
1541 	 * exist.
1542 	 */
1543 	VM_OBJECT_ASSERT_WLOCKED(object);
1544 	if (__predict_true(vm_object_cache_is_empty(object)))
1545 		return (FALSE);
1546 	mtx_lock(&vm_page_queue_free_mtx);
1547 	m = vm_page_cache_lookup(object, pindex);
1548 	mtx_unlock(&vm_page_queue_free_mtx);
1549 	return (m != NULL);
1550 }
1551 
1552 /*
1553  *	vm_page_alloc:
1554  *
1555  *	Allocate and return a page that is associated with the specified
1556  *	object and offset pair.  By default, this page is exclusive busied.
1557  *
1558  *	The caller must always specify an allocation class.
1559  *
1560  *	allocation classes:
1561  *	VM_ALLOC_NORMAL		normal process request
1562  *	VM_ALLOC_SYSTEM		system *really* needs a page
1563  *	VM_ALLOC_INTERRUPT	interrupt time request
1564  *
1565  *	optional allocation flags:
1566  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
1567  *				intends to allocate
1568  *	VM_ALLOC_IFCACHED	return page only if it is cached
1569  *	VM_ALLOC_IFNOTCACHED	return NULL, do not reactivate if the page
1570  *				is cached
1571  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1572  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1573  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1574  *				should not be exclusive busy
1575  *	VM_ALLOC_SBUSY		shared busy the allocated page
1576  *	VM_ALLOC_WIRED		wire the allocated page
1577  *	VM_ALLOC_ZERO		prefer a zeroed page
1578  *
1579  *	This routine may not sleep.
1580  */
1581 vm_page_t
1582 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
1583 {
1584 	struct vnode *vp = NULL;
1585 	vm_object_t m_object;
1586 	vm_page_t m, mpred;
1587 	int flags, req_class;
1588 
1589 	mpred = 0;	/* XXX: pacify gcc */
1590 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1591 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1592 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1593 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1594 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1595 	    req));
1596 	if (object != NULL)
1597 		VM_OBJECT_ASSERT_WLOCKED(object);
1598 
1599 	req_class = req & VM_ALLOC_CLASS_MASK;
1600 
1601 	/*
1602 	 * The page daemon is allowed to dig deeper into the free page list.
1603 	 */
1604 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1605 		req_class = VM_ALLOC_SYSTEM;
1606 
1607 	if (object != NULL) {
1608 		mpred = vm_radix_lookup_le(&object->rtree, pindex);
1609 		KASSERT(mpred == NULL || mpred->pindex != pindex,
1610 		   ("vm_page_alloc: pindex already allocated"));
1611 	}
1612 
1613 	/*
1614 	 * The page allocation request can came from consumers which already
1615 	 * hold the free page queue mutex, like vm_page_insert() in
1616 	 * vm_page_cache().
1617 	 */
1618 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
1619 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
1620 	    (req_class == VM_ALLOC_SYSTEM &&
1621 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
1622 	    (req_class == VM_ALLOC_INTERRUPT &&
1623 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) {
1624 		/*
1625 		 * Allocate from the free queue if the number of free pages
1626 		 * exceeds the minimum for the request class.
1627 		 */
1628 		if (object != NULL &&
1629 		    (m = vm_page_cache_lookup(object, pindex)) != NULL) {
1630 			if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
1631 				mtx_unlock(&vm_page_queue_free_mtx);
1632 				return (NULL);
1633 			}
1634 			if (vm_phys_unfree_page(m))
1635 				vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
1636 #if VM_NRESERVLEVEL > 0
1637 			else if (!vm_reserv_reactivate_page(m))
1638 #else
1639 			else
1640 #endif
1641 				panic("vm_page_alloc: cache page %p is missing"
1642 				    " from the free queue", m);
1643 		} else if ((req & VM_ALLOC_IFCACHED) != 0) {
1644 			mtx_unlock(&vm_page_queue_free_mtx);
1645 			return (NULL);
1646 #if VM_NRESERVLEVEL > 0
1647 		} else if (object == NULL || (object->flags & (OBJ_COLORED |
1648 		    OBJ_FICTITIOUS)) != OBJ_COLORED || (m =
1649 		    vm_reserv_alloc_page(object, pindex, mpred)) == NULL) {
1650 #else
1651 		} else {
1652 #endif
1653 			m = vm_phys_alloc_pages(object != NULL ?
1654 			    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
1655 #if VM_NRESERVLEVEL > 0
1656 			if (m == NULL && vm_reserv_reclaim_inactive()) {
1657 				m = vm_phys_alloc_pages(object != NULL ?
1658 				    VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
1659 				    0);
1660 			}
1661 #endif
1662 		}
1663 	} else {
1664 		/*
1665 		 * Not allocatable, give up.
1666 		 */
1667 		mtx_unlock(&vm_page_queue_free_mtx);
1668 		atomic_add_int(&vm_pageout_deficit,
1669 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
1670 		pagedaemon_wakeup();
1671 		return (NULL);
1672 	}
1673 
1674 	/*
1675 	 *  At this point we had better have found a good page.
1676 	 */
1677 	KASSERT(m != NULL, ("vm_page_alloc: missing page"));
1678 	KASSERT(m->queue == PQ_NONE,
1679 	    ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
1680 	KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
1681 	KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
1682 	KASSERT(!vm_page_sbusied(m),
1683 	    ("vm_page_alloc: page %p is busy", m));
1684 	KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
1685 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1686 	    ("vm_page_alloc: page %p has unexpected memattr %d", m,
1687 	    pmap_page_get_memattr(m)));
1688 	if ((m->flags & PG_CACHED) != 0) {
1689 		KASSERT((m->flags & PG_ZERO) == 0,
1690 		    ("vm_page_alloc: cached page %p is PG_ZERO", m));
1691 		KASSERT(m->valid != 0,
1692 		    ("vm_page_alloc: cached page %p is invalid", m));
1693 		if (m->object == object && m->pindex == pindex)
1694 			vm_cnt.v_reactivated++;
1695 		else
1696 			m->valid = 0;
1697 		m_object = m->object;
1698 		vm_page_cache_remove(m);
1699 		if (m_object->type == OBJT_VNODE &&
1700 		    vm_object_cache_is_empty(m_object))
1701 			vp = m_object->handle;
1702 	} else {
1703 		KASSERT(m->valid == 0,
1704 		    ("vm_page_alloc: free page %p is valid", m));
1705 		vm_phys_freecnt_adj(m, -1);
1706 		if ((m->flags & PG_ZERO) != 0)
1707 			vm_page_zero_count--;
1708 	}
1709 	mtx_unlock(&vm_page_queue_free_mtx);
1710 
1711 	/*
1712 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
1713 	 */
1714 	flags = 0;
1715 	if ((req & VM_ALLOC_ZERO) != 0)
1716 		flags = PG_ZERO;
1717 	flags &= m->flags;
1718 	if ((req & VM_ALLOC_NODUMP) != 0)
1719 		flags |= PG_NODUMP;
1720 	m->flags = flags;
1721 	m->aflags = 0;
1722 	m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ?
1723 	    VPO_UNMANAGED : 0;
1724 	m->busy_lock = VPB_UNBUSIED;
1725 	if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0)
1726 		m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1727 	if ((req & VM_ALLOC_SBUSY) != 0)
1728 		m->busy_lock = VPB_SHARERS_WORD(1);
1729 	if (req & VM_ALLOC_WIRED) {
1730 		/*
1731 		 * The page lock is not required for wiring a page until that
1732 		 * page is inserted into the object.
1733 		 */
1734 		atomic_add_int(&vm_cnt.v_wire_count, 1);
1735 		m->wire_count = 1;
1736 	}
1737 	m->act_count = 0;
1738 
1739 	if (object != NULL) {
1740 		if (vm_page_insert_after(m, object, pindex, mpred)) {
1741 			/* See the comment below about hold count. */
1742 			if (vp != NULL)
1743 				vdrop(vp);
1744 			pagedaemon_wakeup();
1745 			if (req & VM_ALLOC_WIRED) {
1746 				atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1747 				m->wire_count = 0;
1748 			}
1749 			m->object = NULL;
1750 			m->oflags = VPO_UNMANAGED;
1751 			m->busy_lock = VPB_UNBUSIED;
1752 			vm_page_free(m);
1753 			return (NULL);
1754 		}
1755 
1756 		/* Ignore device objects; the pager sets "memattr" for them. */
1757 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1758 		    (object->flags & OBJ_FICTITIOUS) == 0)
1759 			pmap_page_set_memattr(m, object->memattr);
1760 	} else
1761 		m->pindex = pindex;
1762 
1763 	/*
1764 	 * The following call to vdrop() must come after the above call
1765 	 * to vm_page_insert() in case both affect the same object and
1766 	 * vnode.  Otherwise, the affected vnode's hold count could
1767 	 * temporarily become zero.
1768 	 */
1769 	if (vp != NULL)
1770 		vdrop(vp);
1771 
1772 	/*
1773 	 * Don't wakeup too often - wakeup the pageout daemon when
1774 	 * we would be nearly out of memory.
1775 	 */
1776 	if (vm_paging_needed())
1777 		pagedaemon_wakeup();
1778 
1779 	return (m);
1780 }
1781 
1782 static void
1783 vm_page_alloc_contig_vdrop(struct spglist *lst)
1784 {
1785 
1786 	while (!SLIST_EMPTY(lst)) {
1787 		vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv);
1788 		SLIST_REMOVE_HEAD(lst, plinks.s.ss);
1789 	}
1790 }
1791 
1792 /*
1793  *	vm_page_alloc_contig:
1794  *
1795  *	Allocate a contiguous set of physical pages of the given size "npages"
1796  *	from the free lists.  All of the physical pages must be at or above
1797  *	the given physical address "low" and below the given physical address
1798  *	"high".  The given value "alignment" determines the alignment of the
1799  *	first physical page in the set.  If the given value "boundary" is
1800  *	non-zero, then the set of physical pages cannot cross any physical
1801  *	address boundary that is a multiple of that value.  Both "alignment"
1802  *	and "boundary" must be a power of two.
1803  *
1804  *	If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
1805  *	then the memory attribute setting for the physical pages is configured
1806  *	to the object's memory attribute setting.  Otherwise, the memory
1807  *	attribute setting for the physical pages is configured to "memattr",
1808  *	overriding the object's memory attribute setting.  However, if the
1809  *	object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
1810  *	memory attribute setting for the physical pages cannot be configured
1811  *	to VM_MEMATTR_DEFAULT.
1812  *
1813  *	The caller must always specify an allocation class.
1814  *
1815  *	allocation classes:
1816  *	VM_ALLOC_NORMAL		normal process request
1817  *	VM_ALLOC_SYSTEM		system *really* needs a page
1818  *	VM_ALLOC_INTERRUPT	interrupt time request
1819  *
1820  *	optional allocation flags:
1821  *	VM_ALLOC_NOBUSY		do not exclusive busy the page
1822  *	VM_ALLOC_NODUMP		do not include the page in a kernel core dump
1823  *	VM_ALLOC_NOOBJ		page is not associated with an object and
1824  *				should not be exclusive busy
1825  *	VM_ALLOC_SBUSY		shared busy the allocated page
1826  *	VM_ALLOC_WIRED		wire the allocated page
1827  *	VM_ALLOC_ZERO		prefer a zeroed page
1828  *
1829  *	This routine may not sleep.
1830  */
1831 vm_page_t
1832 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
1833     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
1834     vm_paddr_t boundary, vm_memattr_t memattr)
1835 {
1836 	struct vnode *drop;
1837 	struct spglist deferred_vdrop_list;
1838 	vm_page_t m, m_tmp, m_ret;
1839 	u_int flags;
1840 	int req_class;
1841 
1842 	KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) &&
1843 	    (object != NULL || (req & VM_ALLOC_SBUSY) == 0) &&
1844 	    ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
1845 	    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
1846 	    ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object,
1847 	    req));
1848 	if (object != NULL) {
1849 		VM_OBJECT_ASSERT_WLOCKED(object);
1850 		KASSERT(object->type == OBJT_PHYS,
1851 		    ("vm_page_alloc_contig: object %p isn't OBJT_PHYS",
1852 		    object));
1853 	}
1854 	KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
1855 	req_class = req & VM_ALLOC_CLASS_MASK;
1856 
1857 	/*
1858 	 * The page daemon is allowed to dig deeper into the free page list.
1859 	 */
1860 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
1861 		req_class = VM_ALLOC_SYSTEM;
1862 
1863 	SLIST_INIT(&deferred_vdrop_list);
1864 	mtx_lock(&vm_page_queue_free_mtx);
1865 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1866 	    vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM &&
1867 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages +
1868 	    vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT &&
1869 	    vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) {
1870 #if VM_NRESERVLEVEL > 0
1871 retry:
1872 		if (object == NULL || (object->flags & OBJ_COLORED) == 0 ||
1873 		    (m_ret = vm_reserv_alloc_contig(object, pindex, npages,
1874 		    low, high, alignment, boundary)) == NULL)
1875 #endif
1876 			m_ret = vm_phys_alloc_contig(npages, low, high,
1877 			    alignment, boundary);
1878 	} else {
1879 		mtx_unlock(&vm_page_queue_free_mtx);
1880 		atomic_add_int(&vm_pageout_deficit, npages);
1881 		pagedaemon_wakeup();
1882 		return (NULL);
1883 	}
1884 	if (m_ret != NULL)
1885 		for (m = m_ret; m < &m_ret[npages]; m++) {
1886 			drop = vm_page_alloc_init(m);
1887 			if (drop != NULL) {
1888 				/*
1889 				 * Enqueue the vnode for deferred vdrop().
1890 				 */
1891 				m->plinks.s.pv = drop;
1892 				SLIST_INSERT_HEAD(&deferred_vdrop_list, m,
1893 				    plinks.s.ss);
1894 			}
1895 		}
1896 	else {
1897 #if VM_NRESERVLEVEL > 0
1898 		if (vm_reserv_reclaim_contig(npages, low, high, alignment,
1899 		    boundary))
1900 			goto retry;
1901 #endif
1902 	}
1903 	mtx_unlock(&vm_page_queue_free_mtx);
1904 	if (m_ret == NULL)
1905 		return (NULL);
1906 
1907 	/*
1908 	 * Initialize the pages.  Only the PG_ZERO flag is inherited.
1909 	 */
1910 	flags = 0;
1911 	if ((req & VM_ALLOC_ZERO) != 0)
1912 		flags = PG_ZERO;
1913 	if ((req & VM_ALLOC_NODUMP) != 0)
1914 		flags |= PG_NODUMP;
1915 	if ((req & VM_ALLOC_WIRED) != 0)
1916 		atomic_add_int(&vm_cnt.v_wire_count, npages);
1917 	if (object != NULL) {
1918 		if (object->memattr != VM_MEMATTR_DEFAULT &&
1919 		    memattr == VM_MEMATTR_DEFAULT)
1920 			memattr = object->memattr;
1921 	}
1922 	for (m = m_ret; m < &m_ret[npages]; m++) {
1923 		m->aflags = 0;
1924 		m->flags = (m->flags | PG_NODUMP) & flags;
1925 		m->busy_lock = VPB_UNBUSIED;
1926 		if (object != NULL) {
1927 			if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
1928 				m->busy_lock = VPB_SINGLE_EXCLUSIVER;
1929 			if ((req & VM_ALLOC_SBUSY) != 0)
1930 				m->busy_lock = VPB_SHARERS_WORD(1);
1931 		}
1932 		if ((req & VM_ALLOC_WIRED) != 0)
1933 			m->wire_count = 1;
1934 		/* Unmanaged pages don't use "act_count". */
1935 		m->oflags = VPO_UNMANAGED;
1936 		if (object != NULL) {
1937 			if (vm_page_insert(m, object, pindex)) {
1938 				vm_page_alloc_contig_vdrop(
1939 				    &deferred_vdrop_list);
1940 				if (vm_paging_needed())
1941 					pagedaemon_wakeup();
1942 				if ((req & VM_ALLOC_WIRED) != 0)
1943 					atomic_subtract_int(&vm_cnt.v_wire_count,
1944 					    npages);
1945 				for (m_tmp = m, m = m_ret;
1946 				    m < &m_ret[npages]; m++) {
1947 					if ((req & VM_ALLOC_WIRED) != 0)
1948 						m->wire_count = 0;
1949 					if (m >= m_tmp) {
1950 						m->object = NULL;
1951 						m->oflags |= VPO_UNMANAGED;
1952 					}
1953 					m->busy_lock = VPB_UNBUSIED;
1954 					vm_page_free(m);
1955 				}
1956 				return (NULL);
1957 			}
1958 		} else
1959 			m->pindex = pindex;
1960 		if (memattr != VM_MEMATTR_DEFAULT)
1961 			pmap_page_set_memattr(m, memattr);
1962 		pindex++;
1963 	}
1964 	vm_page_alloc_contig_vdrop(&deferred_vdrop_list);
1965 	if (vm_paging_needed())
1966 		pagedaemon_wakeup();
1967 	return (m_ret);
1968 }
1969 
1970 /*
1971  * Initialize a page that has been freshly dequeued from a freelist.
1972  * The caller has to drop the vnode returned, if it is not NULL.
1973  *
1974  * This function may only be used to initialize unmanaged pages.
1975  *
1976  * To be called with vm_page_queue_free_mtx held.
1977  */
1978 static struct vnode *
1979 vm_page_alloc_init(vm_page_t m)
1980 {
1981 	struct vnode *drop;
1982 	vm_object_t m_object;
1983 
1984 	KASSERT(m->queue == PQ_NONE,
1985 	    ("vm_page_alloc_init: page %p has unexpected queue %d",
1986 	    m, m->queue));
1987 	KASSERT(m->wire_count == 0,
1988 	    ("vm_page_alloc_init: page %p is wired", m));
1989 	KASSERT(m->hold_count == 0,
1990 	    ("vm_page_alloc_init: page %p is held", m));
1991 	KASSERT(!vm_page_sbusied(m),
1992 	    ("vm_page_alloc_init: page %p is busy", m));
1993 	KASSERT(m->dirty == 0,
1994 	    ("vm_page_alloc_init: page %p is dirty", m));
1995 	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
1996 	    ("vm_page_alloc_init: page %p has unexpected memattr %d",
1997 	    m, pmap_page_get_memattr(m)));
1998 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1999 	drop = NULL;
2000 	if ((m->flags & PG_CACHED) != 0) {
2001 		KASSERT((m->flags & PG_ZERO) == 0,
2002 		    ("vm_page_alloc_init: cached page %p is PG_ZERO", m));
2003 		m->valid = 0;
2004 		m_object = m->object;
2005 		vm_page_cache_remove(m);
2006 		if (m_object->type == OBJT_VNODE &&
2007 		    vm_object_cache_is_empty(m_object))
2008 			drop = m_object->handle;
2009 	} else {
2010 		KASSERT(m->valid == 0,
2011 		    ("vm_page_alloc_init: free page %p is valid", m));
2012 		vm_phys_freecnt_adj(m, -1);
2013 		if ((m->flags & PG_ZERO) != 0)
2014 			vm_page_zero_count--;
2015 	}
2016 	return (drop);
2017 }
2018 
2019 /*
2020  * 	vm_page_alloc_freelist:
2021  *
2022  *	Allocate a physical page from the specified free page list.
2023  *
2024  *	The caller must always specify an allocation class.
2025  *
2026  *	allocation classes:
2027  *	VM_ALLOC_NORMAL		normal process request
2028  *	VM_ALLOC_SYSTEM		system *really* needs a page
2029  *	VM_ALLOC_INTERRUPT	interrupt time request
2030  *
2031  *	optional allocation flags:
2032  *	VM_ALLOC_COUNT(number)	the number of additional pages that the caller
2033  *				intends to allocate
2034  *	VM_ALLOC_WIRED		wire the allocated page
2035  *	VM_ALLOC_ZERO		prefer a zeroed page
2036  *
2037  *	This routine may not sleep.
2038  */
2039 vm_page_t
2040 vm_page_alloc_freelist(int flind, int req)
2041 {
2042 	struct vnode *drop;
2043 	vm_page_t m;
2044 	u_int flags;
2045 	int req_class;
2046 
2047 	req_class = req & VM_ALLOC_CLASS_MASK;
2048 
2049 	/*
2050 	 * The page daemon is allowed to dig deeper into the free page list.
2051 	 */
2052 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2053 		req_class = VM_ALLOC_SYSTEM;
2054 
2055 	/*
2056 	 * Do not allocate reserved pages unless the req has asked for it.
2057 	 */
2058 	mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE);
2059 	if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved ||
2060 	    (req_class == VM_ALLOC_SYSTEM &&
2061 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) ||
2062 	    (req_class == VM_ALLOC_INTERRUPT &&
2063 	    vm_cnt.v_free_count + vm_cnt.v_cache_count > 0))
2064 		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0);
2065 	else {
2066 		mtx_unlock(&vm_page_queue_free_mtx);
2067 		atomic_add_int(&vm_pageout_deficit,
2068 		    max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
2069 		pagedaemon_wakeup();
2070 		return (NULL);
2071 	}
2072 	if (m == NULL) {
2073 		mtx_unlock(&vm_page_queue_free_mtx);
2074 		return (NULL);
2075 	}
2076 	drop = vm_page_alloc_init(m);
2077 	mtx_unlock(&vm_page_queue_free_mtx);
2078 
2079 	/*
2080 	 * Initialize the page.  Only the PG_ZERO flag is inherited.
2081 	 */
2082 	m->aflags = 0;
2083 	flags = 0;
2084 	if ((req & VM_ALLOC_ZERO) != 0)
2085 		flags = PG_ZERO;
2086 	m->flags &= flags;
2087 	if ((req & VM_ALLOC_WIRED) != 0) {
2088 		/*
2089 		 * The page lock is not required for wiring a page that does
2090 		 * not belong to an object.
2091 		 */
2092 		atomic_add_int(&vm_cnt.v_wire_count, 1);
2093 		m->wire_count = 1;
2094 	}
2095 	/* Unmanaged pages don't use "act_count". */
2096 	m->oflags = VPO_UNMANAGED;
2097 	if (drop != NULL)
2098 		vdrop(drop);
2099 	if (vm_paging_needed())
2100 		pagedaemon_wakeup();
2101 	return (m);
2102 }
2103 
2104 #define	VPSC_ANY	0	/* No restrictions. */
2105 #define	VPSC_NORESERV	1	/* Skip reservations; implies VPSC_NOSUPER. */
2106 #define	VPSC_NOSUPER	2	/* Skip superpages. */
2107 
2108 /*
2109  *	vm_page_scan_contig:
2110  *
2111  *	Scan vm_page_array[] between the specified entries "m_start" and
2112  *	"m_end" for a run of contiguous physical pages that satisfy the
2113  *	specified conditions, and return the lowest page in the run.  The
2114  *	specified "alignment" determines the alignment of the lowest physical
2115  *	page in the run.  If the specified "boundary" is non-zero, then the
2116  *	run of physical pages cannot span a physical address that is a
2117  *	multiple of "boundary".
2118  *
2119  *	"m_end" is never dereferenced, so it need not point to a vm_page
2120  *	structure within vm_page_array[].
2121  *
2122  *	"npages" must be greater than zero.  "m_start" and "m_end" must not
2123  *	span a hole (or discontiguity) in the physical address space.  Both
2124  *	"alignment" and "boundary" must be a power of two.
2125  */
2126 vm_page_t
2127 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
2128     u_long alignment, vm_paddr_t boundary, int options)
2129 {
2130 	struct mtx *m_mtx, *new_mtx;
2131 	vm_object_t object;
2132 	vm_paddr_t pa;
2133 	vm_page_t m, m_run;
2134 #if VM_NRESERVLEVEL > 0
2135 	int level;
2136 #endif
2137 	int m_inc, order, run_ext, run_len;
2138 
2139 	KASSERT(npages > 0, ("npages is 0"));
2140 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2141 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2142 	m_run = NULL;
2143 	run_len = 0;
2144 	m_mtx = NULL;
2145 	for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
2146 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2147 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2148 
2149 		/*
2150 		 * If the current page would be the start of a run, check its
2151 		 * physical address against the end, alignment, and boundary
2152 		 * conditions.  If it doesn't satisfy these conditions, either
2153 		 * terminate the scan or advance to the next page that
2154 		 * satisfies the failed condition.
2155 		 */
2156 		if (run_len == 0) {
2157 			KASSERT(m_run == NULL, ("m_run != NULL"));
2158 			if (m + npages > m_end)
2159 				break;
2160 			pa = VM_PAGE_TO_PHYS(m);
2161 			if ((pa & (alignment - 1)) != 0) {
2162 				m_inc = atop(roundup2(pa, alignment) - pa);
2163 				continue;
2164 			}
2165 			if (rounddown2(pa ^ (pa + ptoa(npages) - 1),
2166 			    boundary) != 0) {
2167 				m_inc = atop(roundup2(pa, boundary) - pa);
2168 				continue;
2169 			}
2170 		} else
2171 			KASSERT(m_run != NULL, ("m_run == NULL"));
2172 
2173 		/*
2174 		 * Avoid releasing and reacquiring the same page lock.
2175 		 */
2176 		new_mtx = vm_page_lockptr(m);
2177 		if (m_mtx != new_mtx) {
2178 			if (m_mtx != NULL)
2179 				mtx_unlock(m_mtx);
2180 			m_mtx = new_mtx;
2181 			mtx_lock(m_mtx);
2182 		}
2183 		m_inc = 1;
2184 retry:
2185 		if (m->wire_count != 0 || m->hold_count != 0)
2186 			run_ext = 0;
2187 #if VM_NRESERVLEVEL > 0
2188 		else if ((level = vm_reserv_level(m)) >= 0 &&
2189 		    (options & VPSC_NORESERV) != 0) {
2190 			run_ext = 0;
2191 			/* Advance to the end of the reservation. */
2192 			pa = VM_PAGE_TO_PHYS(m);
2193 			m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
2194 			    pa);
2195 		}
2196 #endif
2197 		else if ((object = m->object) != NULL) {
2198 			/*
2199 			 * The page is considered eligible for relocation if
2200 			 * and only if it could be laundered or reclaimed by
2201 			 * the page daemon.
2202 			 */
2203 			if (!VM_OBJECT_TRYRLOCK(object)) {
2204 				mtx_unlock(m_mtx);
2205 				VM_OBJECT_RLOCK(object);
2206 				mtx_lock(m_mtx);
2207 				if (m->object != object) {
2208 					/*
2209 					 * The page may have been freed.
2210 					 */
2211 					VM_OBJECT_RUNLOCK(object);
2212 					goto retry;
2213 				} else if (m->wire_count != 0 ||
2214 				    m->hold_count != 0) {
2215 					run_ext = 0;
2216 					goto unlock;
2217 				}
2218 			}
2219 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2220 			    ("page %p is PG_UNHOLDFREE", m));
2221 			/* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */
2222 			if (object->type != OBJT_DEFAULT &&
2223 			    object->type != OBJT_SWAP &&
2224 			    object->type != OBJT_VNODE)
2225 				run_ext = 0;
2226 			else if ((m->flags & PG_CACHED) != 0 ||
2227 			    m != vm_page_lookup(object, m->pindex)) {
2228 				/*
2229 				 * The page is cached or recently converted
2230 				 * from cached to free.
2231 				 */
2232 #if VM_NRESERVLEVEL > 0
2233 				if (level >= 0) {
2234 					/*
2235 					 * The page is reserved.  Extend the
2236 					 * current run by one page.
2237 					 */
2238 					run_ext = 1;
2239 				} else
2240 #endif
2241 				if ((order = m->order) < VM_NFREEORDER) {
2242 					/*
2243 					 * The page is enqueued in the
2244 					 * physical memory allocator's cache/
2245 					 * free page queues.  Moreover, it is
2246 					 * the first page in a power-of-two-
2247 					 * sized run of contiguous cache/free
2248 					 * pages.  Add these pages to the end
2249 					 * of the current run, and jump
2250 					 * ahead.
2251 					 */
2252 					run_ext = 1 << order;
2253 					m_inc = 1 << order;
2254 				} else
2255 					run_ext = 0;
2256 #if VM_NRESERVLEVEL > 0
2257 			} else if ((options & VPSC_NOSUPER) != 0 &&
2258 			    (level = vm_reserv_level_iffullpop(m)) >= 0) {
2259 				run_ext = 0;
2260 				/* Advance to the end of the superpage. */
2261 				pa = VM_PAGE_TO_PHYS(m);
2262 				m_inc = atop(roundup2(pa + 1,
2263 				    vm_reserv_size(level)) - pa);
2264 #endif
2265 			} else if (object->memattr == VM_MEMATTR_DEFAULT &&
2266 			    m->queue != PQ_NONE && !vm_page_busied(m)) {
2267 				/*
2268 				 * The page is allocated but eligible for
2269 				 * relocation.  Extend the current run by one
2270 				 * page.
2271 				 */
2272 				KASSERT(pmap_page_get_memattr(m) ==
2273 				    VM_MEMATTR_DEFAULT,
2274 				    ("page %p has an unexpected memattr", m));
2275 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2276 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2277 				    ("page %p has unexpected oflags", m));
2278 				/* Don't care: VPO_NOSYNC. */
2279 				run_ext = 1;
2280 			} else
2281 				run_ext = 0;
2282 unlock:
2283 			VM_OBJECT_RUNLOCK(object);
2284 #if VM_NRESERVLEVEL > 0
2285 		} else if (level >= 0) {
2286 			/*
2287 			 * The page is reserved but not yet allocated.  In
2288 			 * other words, it is still cached or free.  Extend
2289 			 * the current run by one page.
2290 			 */
2291 			run_ext = 1;
2292 #endif
2293 		} else if ((order = m->order) < VM_NFREEORDER) {
2294 			/*
2295 			 * The page is enqueued in the physical memory
2296 			 * allocator's cache/free page queues.  Moreover, it
2297 			 * is the first page in a power-of-two-sized run of
2298 			 * contiguous cache/free pages.  Add these pages to
2299 			 * the end of the current run, and jump ahead.
2300 			 */
2301 			run_ext = 1 << order;
2302 			m_inc = 1 << order;
2303 		} else {
2304 			/*
2305 			 * Skip the page for one of the following reasons: (1)
2306 			 * It is enqueued in the physical memory allocator's
2307 			 * cache/free page queues.  However, it is not the
2308 			 * first page in a run of contiguous cache/free pages.
2309 			 * (This case rarely occurs because the scan is
2310 			 * performed in ascending order.) (2) It is not
2311 			 * reserved, and it is transitioning from free to
2312 			 * allocated.  (Conversely, the transition from
2313 			 * allocated to free for managed pages is blocked by
2314 			 * the page lock.) (3) It is allocated but not
2315 			 * contained by an object and not wired, e.g.,
2316 			 * allocated by Xen's balloon driver.
2317 			 */
2318 			run_ext = 0;
2319 		}
2320 
2321 		/*
2322 		 * Extend or reset the current run of pages.
2323 		 */
2324 		if (run_ext > 0) {
2325 			if (run_len == 0)
2326 				m_run = m;
2327 			run_len += run_ext;
2328 		} else {
2329 			if (run_len > 0) {
2330 				m_run = NULL;
2331 				run_len = 0;
2332 			}
2333 		}
2334 	}
2335 	if (m_mtx != NULL)
2336 		mtx_unlock(m_mtx);
2337 	if (run_len >= npages)
2338 		return (m_run);
2339 	return (NULL);
2340 }
2341 
2342 /*
2343  *	vm_page_reclaim_run:
2344  *
2345  *	Try to relocate each of the allocated virtual pages within the
2346  *	specified run of physical pages to a new physical address.  Free the
2347  *	physical pages underlying the relocated virtual pages.  A virtual page
2348  *	is relocatable if and only if it could be laundered or reclaimed by
2349  *	the page daemon.  Whenever possible, a virtual page is relocated to a
2350  *	physical address above "high".
2351  *
2352  *	Returns 0 if every physical page within the run was already free or
2353  *	just freed by a successful relocation.  Otherwise, returns a non-zero
2354  *	value indicating why the last attempt to relocate a virtual page was
2355  *	unsuccessful.
2356  *
2357  *	"req_class" must be an allocation class.
2358  */
2359 static int
2360 vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run,
2361     vm_paddr_t high)
2362 {
2363 	struct mtx *m_mtx, *new_mtx;
2364 	struct spglist free;
2365 	vm_object_t object;
2366 	vm_paddr_t pa;
2367 	vm_page_t m, m_end, m_new;
2368 	int error, order, req;
2369 
2370 	KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
2371 	    ("req_class is not an allocation class"));
2372 	SLIST_INIT(&free);
2373 	error = 0;
2374 	m = m_run;
2375 	m_end = m_run + npages;
2376 	m_mtx = NULL;
2377 	for (; error == 0 && m < m_end; m++) {
2378 		KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
2379 		    ("page %p is PG_FICTITIOUS or PG_MARKER", m));
2380 
2381 		/*
2382 		 * Avoid releasing and reacquiring the same page lock.
2383 		 */
2384 		new_mtx = vm_page_lockptr(m);
2385 		if (m_mtx != new_mtx) {
2386 			if (m_mtx != NULL)
2387 				mtx_unlock(m_mtx);
2388 			m_mtx = new_mtx;
2389 			mtx_lock(m_mtx);
2390 		}
2391 retry:
2392 		if (m->wire_count != 0 || m->hold_count != 0)
2393 			error = EBUSY;
2394 		else if ((object = m->object) != NULL) {
2395 			/*
2396 			 * The page is relocated if and only if it could be
2397 			 * laundered or reclaimed by the page daemon.
2398 			 */
2399 			if (!VM_OBJECT_TRYWLOCK(object)) {
2400 				mtx_unlock(m_mtx);
2401 				VM_OBJECT_WLOCK(object);
2402 				mtx_lock(m_mtx);
2403 				if (m->object != object) {
2404 					/*
2405 					 * The page may have been freed.
2406 					 */
2407 					VM_OBJECT_WUNLOCK(object);
2408 					goto retry;
2409 				} else if (m->wire_count != 0 ||
2410 				    m->hold_count != 0) {
2411 					error = EBUSY;
2412 					goto unlock;
2413 				}
2414 			}
2415 			KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2416 			    ("page %p is PG_UNHOLDFREE", m));
2417 			/* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */
2418 			if (object->type != OBJT_DEFAULT &&
2419 			    object->type != OBJT_SWAP &&
2420 			    object->type != OBJT_VNODE)
2421 				error = EINVAL;
2422 			else if ((m->flags & PG_CACHED) != 0 ||
2423 			    m != vm_page_lookup(object, m->pindex)) {
2424 				/*
2425 				 * The page is cached or recently converted
2426 				 * from cached to free.
2427 				 */
2428 				VM_OBJECT_WUNLOCK(object);
2429 				goto cached;
2430 			} else if (object->memattr != VM_MEMATTR_DEFAULT)
2431 				error = EINVAL;
2432 			else if (m->queue != PQ_NONE && !vm_page_busied(m)) {
2433 				KASSERT(pmap_page_get_memattr(m) ==
2434 				    VM_MEMATTR_DEFAULT,
2435 				    ("page %p has an unexpected memattr", m));
2436 				KASSERT((m->oflags & (VPO_SWAPINPROG |
2437 				    VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
2438 				    ("page %p has unexpected oflags", m));
2439 				/* Don't care: VPO_NOSYNC. */
2440 				if (m->valid != 0) {
2441 					/*
2442 					 * First, try to allocate a new page
2443 					 * that is above "high".  Failing
2444 					 * that, try to allocate a new page
2445 					 * that is below "m_run".  Allocate
2446 					 * the new page between the end of
2447 					 * "m_run" and "high" only as a last
2448 					 * resort.
2449 					 */
2450 					req = req_class | VM_ALLOC_NOOBJ;
2451 					if ((m->flags & PG_NODUMP) != 0)
2452 						req |= VM_ALLOC_NODUMP;
2453 					if (trunc_page(high) !=
2454 					    ~(vm_paddr_t)PAGE_MASK) {
2455 						m_new = vm_page_alloc_contig(
2456 						    NULL, 0, req, 1,
2457 						    round_page(high),
2458 						    ~(vm_paddr_t)0,
2459 						    PAGE_SIZE, 0,
2460 						    VM_MEMATTR_DEFAULT);
2461 					} else
2462 						m_new = NULL;
2463 					if (m_new == NULL) {
2464 						pa = VM_PAGE_TO_PHYS(m_run);
2465 						m_new = vm_page_alloc_contig(
2466 						    NULL, 0, req, 1,
2467 						    0, pa - 1, PAGE_SIZE, 0,
2468 						    VM_MEMATTR_DEFAULT);
2469 					}
2470 					if (m_new == NULL) {
2471 						pa += ptoa(npages);
2472 						m_new = vm_page_alloc_contig(
2473 						    NULL, 0, req, 1,
2474 						    pa, high, PAGE_SIZE, 0,
2475 						    VM_MEMATTR_DEFAULT);
2476 					}
2477 					if (m_new == NULL) {
2478 						error = ENOMEM;
2479 						goto unlock;
2480 					}
2481 					KASSERT(m_new->wire_count == 0,
2482 					    ("page %p is wired", m));
2483 
2484 					/*
2485 					 * Replace "m" with the new page.  For
2486 					 * vm_page_replace(), "m" must be busy
2487 					 * and dequeued.  Finally, change "m"
2488 					 * as if vm_page_free() was called.
2489 					 */
2490 					if (object->ref_count != 0)
2491 						pmap_remove_all(m);
2492 					m_new->aflags = m->aflags;
2493 					KASSERT(m_new->oflags == VPO_UNMANAGED,
2494 					    ("page %p is managed", m));
2495 					m_new->oflags = m->oflags & VPO_NOSYNC;
2496 					pmap_copy_page(m, m_new);
2497 					m_new->valid = m->valid;
2498 					m_new->dirty = m->dirty;
2499 					m->flags &= ~PG_ZERO;
2500 					vm_page_xbusy(m);
2501 					vm_page_remque(m);
2502 					vm_page_replace_checked(m_new, object,
2503 					    m->pindex, m);
2504 					m->valid = 0;
2505 					vm_page_undirty(m);
2506 
2507 					/*
2508 					 * The new page must be deactivated
2509 					 * before the object is unlocked.
2510 					 */
2511 					new_mtx = vm_page_lockptr(m_new);
2512 					if (m_mtx != new_mtx) {
2513 						mtx_unlock(m_mtx);
2514 						m_mtx = new_mtx;
2515 						mtx_lock(m_mtx);
2516 					}
2517 					vm_page_deactivate(m_new);
2518 				} else {
2519 					m->flags &= ~PG_ZERO;
2520 					vm_page_remque(m);
2521 					vm_page_remove(m);
2522 					KASSERT(m->dirty == 0,
2523 					    ("page %p is dirty", m));
2524 				}
2525 				SLIST_INSERT_HEAD(&free, m, plinks.s.ss);
2526 			} else
2527 				error = EBUSY;
2528 unlock:
2529 			VM_OBJECT_WUNLOCK(object);
2530 		} else {
2531 cached:
2532 			mtx_lock(&vm_page_queue_free_mtx);
2533 			order = m->order;
2534 			if (order < VM_NFREEORDER) {
2535 				/*
2536 				 * The page is enqueued in the physical memory
2537 				 * allocator's cache/free page queues.
2538 				 * Moreover, it is the first page in a power-
2539 				 * of-two-sized run of contiguous cache/free
2540 				 * pages.  Jump ahead to the last page within
2541 				 * that run, and continue from there.
2542 				 */
2543 				m += (1 << order) - 1;
2544 			}
2545 #if VM_NRESERVLEVEL > 0
2546 			else if (vm_reserv_is_page_free(m))
2547 				order = 0;
2548 #endif
2549 			mtx_unlock(&vm_page_queue_free_mtx);
2550 			if (order == VM_NFREEORDER)
2551 				error = EINVAL;
2552 		}
2553 	}
2554 	if (m_mtx != NULL)
2555 		mtx_unlock(m_mtx);
2556 	if ((m = SLIST_FIRST(&free)) != NULL) {
2557 		mtx_lock(&vm_page_queue_free_mtx);
2558 		do {
2559 			SLIST_REMOVE_HEAD(&free, plinks.s.ss);
2560 			vm_phys_freecnt_adj(m, 1);
2561 #if VM_NRESERVLEVEL > 0
2562 			if (!vm_reserv_free_page(m))
2563 #else
2564 			if (true)
2565 #endif
2566 				vm_phys_free_pages(m, 0);
2567 		} while ((m = SLIST_FIRST(&free)) != NULL);
2568 		vm_page_zero_idle_wakeup();
2569 		vm_page_free_wakeup();
2570 		mtx_unlock(&vm_page_queue_free_mtx);
2571 	}
2572 	return (error);
2573 }
2574 
2575 #define	NRUNS	16
2576 
2577 CTASSERT(powerof2(NRUNS));
2578 
2579 #define	RUN_INDEX(count)	((count) & (NRUNS - 1))
2580 
2581 #define	MIN_RECLAIM	8
2582 
2583 /*
2584  *	vm_page_reclaim_contig:
2585  *
2586  *	Reclaim allocated, contiguous physical memory satisfying the specified
2587  *	conditions by relocating the virtual pages using that physical memory.
2588  *	Returns true if reclamation is successful and false otherwise.  Since
2589  *	relocation requires the allocation of physical pages, reclamation may
2590  *	fail due to a shortage of cache/free pages.  When reclamation fails,
2591  *	callers are expected to perform VM_WAIT before retrying a failed
2592  *	allocation operation, e.g., vm_page_alloc_contig().
2593  *
2594  *	The caller must always specify an allocation class through "req".
2595  *
2596  *	allocation classes:
2597  *	VM_ALLOC_NORMAL		normal process request
2598  *	VM_ALLOC_SYSTEM		system *really* needs a page
2599  *	VM_ALLOC_INTERRUPT	interrupt time request
2600  *
2601  *	The optional allocation flags are ignored.
2602  *
2603  *	"npages" must be greater than zero.  Both "alignment" and "boundary"
2604  *	must be a power of two.
2605  */
2606 bool
2607 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
2608     u_long alignment, vm_paddr_t boundary)
2609 {
2610 	vm_paddr_t curr_low;
2611 	vm_page_t m_run, m_runs[NRUNS];
2612 	u_long count, reclaimed;
2613 	int error, i, options, req_class;
2614 
2615 	KASSERT(npages > 0, ("npages is 0"));
2616 	KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
2617 	KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
2618 	req_class = req & VM_ALLOC_CLASS_MASK;
2619 
2620 	/*
2621 	 * The page daemon is allowed to dig deeper into the free page list.
2622 	 */
2623 	if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
2624 		req_class = VM_ALLOC_SYSTEM;
2625 
2626 	/*
2627 	 * Return if the number of cached and free pages cannot satisfy the
2628 	 * requested allocation.
2629 	 */
2630 	count = vm_cnt.v_free_count + vm_cnt.v_cache_count;
2631 	if (count < npages + vm_cnt.v_free_reserved || (count < npages +
2632 	    vm_cnt.v_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
2633 	    (count < npages && req_class == VM_ALLOC_INTERRUPT))
2634 		return (false);
2635 
2636 	/*
2637 	 * Scan up to three times, relaxing the restrictions ("options") on
2638 	 * the reclamation of reservations and superpages each time.
2639 	 */
2640 	for (options = VPSC_NORESERV;;) {
2641 		/*
2642 		 * Find the highest runs that satisfy the given constraints
2643 		 * and restrictions, and record them in "m_runs".
2644 		 */
2645 		curr_low = low;
2646 		count = 0;
2647 		for (;;) {
2648 			m_run = vm_phys_scan_contig(npages, curr_low, high,
2649 			    alignment, boundary, options);
2650 			if (m_run == NULL)
2651 				break;
2652 			curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages);
2653 			m_runs[RUN_INDEX(count)] = m_run;
2654 			count++;
2655 		}
2656 
2657 		/*
2658 		 * Reclaim the highest runs in LIFO (descending) order until
2659 		 * the number of reclaimed pages, "reclaimed", is at least
2660 		 * MIN_RECLAIM.  Reset "reclaimed" each time because each
2661 		 * reclamation is idempotent, and runs will (likely) recur
2662 		 * from one scan to the next as restrictions are relaxed.
2663 		 */
2664 		reclaimed = 0;
2665 		for (i = 0; count > 0 && i < NRUNS; i++) {
2666 			count--;
2667 			m_run = m_runs[RUN_INDEX(count)];
2668 			error = vm_page_reclaim_run(req_class, npages, m_run,
2669 			    high);
2670 			if (error == 0) {
2671 				reclaimed += npages;
2672 				if (reclaimed >= MIN_RECLAIM)
2673 					return (true);
2674 			}
2675 		}
2676 
2677 		/*
2678 		 * Either relax the restrictions on the next scan or return if
2679 		 * the last scan had no restrictions.
2680 		 */
2681 		if (options == VPSC_NORESERV)
2682 			options = VPSC_NOSUPER;
2683 		else if (options == VPSC_NOSUPER)
2684 			options = VPSC_ANY;
2685 		else if (options == VPSC_ANY)
2686 			return (reclaimed != 0);
2687 	}
2688 }
2689 
2690 /*
2691  *	vm_wait:	(also see VM_WAIT macro)
2692  *
2693  *	Sleep until free pages are available for allocation.
2694  *	- Called in various places before memory allocations.
2695  */
2696 void
2697 vm_wait(void)
2698 {
2699 
2700 	mtx_lock(&vm_page_queue_free_mtx);
2701 	if (curproc == pageproc) {
2702 		vm_pageout_pages_needed = 1;
2703 		msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
2704 		    PDROP | PSWP, "VMWait", 0);
2705 	} else {
2706 		if (!vm_pageout_wanted) {
2707 			vm_pageout_wanted = true;
2708 			wakeup(&vm_pageout_wanted);
2709 		}
2710 		vm_pages_needed = true;
2711 		msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
2712 		    "vmwait", 0);
2713 	}
2714 }
2715 
2716 /*
2717  *	vm_waitpfault:	(also see VM_WAITPFAULT macro)
2718  *
2719  *	Sleep until free pages are available for allocation.
2720  *	- Called only in vm_fault so that processes page faulting
2721  *	  can be easily tracked.
2722  *	- Sleeps at a lower priority than vm_wait() so that vm_wait()ing
2723  *	  processes will be able to grab memory first.  Do not change
2724  *	  this balance without careful testing first.
2725  */
2726 void
2727 vm_waitpfault(void)
2728 {
2729 
2730 	mtx_lock(&vm_page_queue_free_mtx);
2731 	if (!vm_pageout_wanted) {
2732 		vm_pageout_wanted = true;
2733 		wakeup(&vm_pageout_wanted);
2734 	}
2735 	vm_pages_needed = true;
2736 	msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
2737 	    "pfault", 0);
2738 }
2739 
2740 struct vm_pagequeue *
2741 vm_page_pagequeue(vm_page_t m)
2742 {
2743 
2744 	return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]);
2745 }
2746 
2747 /*
2748  *	vm_page_dequeue:
2749  *
2750  *	Remove the given page from its current page queue.
2751  *
2752  *	The page must be locked.
2753  */
2754 void
2755 vm_page_dequeue(vm_page_t m)
2756 {
2757 	struct vm_pagequeue *pq;
2758 
2759 	vm_page_assert_locked(m);
2760 	KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued",
2761 	    m));
2762 	pq = vm_page_pagequeue(m);
2763 	vm_pagequeue_lock(pq);
2764 	m->queue = PQ_NONE;
2765 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2766 	vm_pagequeue_cnt_dec(pq);
2767 	vm_pagequeue_unlock(pq);
2768 }
2769 
2770 /*
2771  *	vm_page_dequeue_locked:
2772  *
2773  *	Remove the given page from its current page queue.
2774  *
2775  *	The page and page queue must be locked.
2776  */
2777 void
2778 vm_page_dequeue_locked(vm_page_t m)
2779 {
2780 	struct vm_pagequeue *pq;
2781 
2782 	vm_page_lock_assert(m, MA_OWNED);
2783 	pq = vm_page_pagequeue(m);
2784 	vm_pagequeue_assert_locked(pq);
2785 	m->queue = PQ_NONE;
2786 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2787 	vm_pagequeue_cnt_dec(pq);
2788 }
2789 
2790 /*
2791  *	vm_page_enqueue:
2792  *
2793  *	Add the given page to the specified page queue.
2794  *
2795  *	The page must be locked.
2796  */
2797 static void
2798 vm_page_enqueue(uint8_t queue, vm_page_t m)
2799 {
2800 	struct vm_pagequeue *pq;
2801 
2802 	vm_page_lock_assert(m, MA_OWNED);
2803 	KASSERT(queue < PQ_COUNT,
2804 	    ("vm_page_enqueue: invalid queue %u request for page %p",
2805 	    queue, m));
2806 	pq = &vm_phys_domain(m)->vmd_pagequeues[queue];
2807 	vm_pagequeue_lock(pq);
2808 	m->queue = queue;
2809 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2810 	vm_pagequeue_cnt_inc(pq);
2811 	vm_pagequeue_unlock(pq);
2812 }
2813 
2814 /*
2815  *	vm_page_requeue:
2816  *
2817  *	Move the given page to the tail of its current page queue.
2818  *
2819  *	The page must be locked.
2820  */
2821 void
2822 vm_page_requeue(vm_page_t m)
2823 {
2824 	struct vm_pagequeue *pq;
2825 
2826 	vm_page_lock_assert(m, MA_OWNED);
2827 	KASSERT(m->queue != PQ_NONE,
2828 	    ("vm_page_requeue: page %p is not queued", m));
2829 	pq = vm_page_pagequeue(m);
2830 	vm_pagequeue_lock(pq);
2831 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2832 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2833 	vm_pagequeue_unlock(pq);
2834 }
2835 
2836 /*
2837  *	vm_page_requeue_locked:
2838  *
2839  *	Move the given page to the tail of its current page queue.
2840  *
2841  *	The page queue must be locked.
2842  */
2843 void
2844 vm_page_requeue_locked(vm_page_t m)
2845 {
2846 	struct vm_pagequeue *pq;
2847 
2848 	KASSERT(m->queue != PQ_NONE,
2849 	    ("vm_page_requeue_locked: page %p is not queued", m));
2850 	pq = vm_page_pagequeue(m);
2851 	vm_pagequeue_assert_locked(pq);
2852 	TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
2853 	TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
2854 }
2855 
2856 /*
2857  *	vm_page_activate:
2858  *
2859  *	Put the specified page on the active list (if appropriate).
2860  *	Ensure that act_count is at least ACT_INIT but do not otherwise
2861  *	mess with it.
2862  *
2863  *	The page must be locked.
2864  */
2865 void
2866 vm_page_activate(vm_page_t m)
2867 {
2868 	int queue;
2869 
2870 	vm_page_lock_assert(m, MA_OWNED);
2871 	if ((queue = m->queue) != PQ_ACTIVE) {
2872 		if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
2873 			if (m->act_count < ACT_INIT)
2874 				m->act_count = ACT_INIT;
2875 			if (queue != PQ_NONE)
2876 				vm_page_dequeue(m);
2877 			vm_page_enqueue(PQ_ACTIVE, m);
2878 		} else
2879 			KASSERT(queue == PQ_NONE,
2880 			    ("vm_page_activate: wired page %p is queued", m));
2881 	} else {
2882 		if (m->act_count < ACT_INIT)
2883 			m->act_count = ACT_INIT;
2884 	}
2885 }
2886 
2887 /*
2888  *	vm_page_free_wakeup:
2889  *
2890  *	Helper routine for vm_page_free_toq() and vm_page_cache().  This
2891  *	routine is called when a page has been added to the cache or free
2892  *	queues.
2893  *
2894  *	The page queues must be locked.
2895  */
2896 static inline void
2897 vm_page_free_wakeup(void)
2898 {
2899 
2900 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2901 	/*
2902 	 * if pageout daemon needs pages, then tell it that there are
2903 	 * some free.
2904 	 */
2905 	if (vm_pageout_pages_needed &&
2906 	    vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) {
2907 		wakeup(&vm_pageout_pages_needed);
2908 		vm_pageout_pages_needed = 0;
2909 	}
2910 	/*
2911 	 * wakeup processes that are waiting on memory if we hit a
2912 	 * high water mark. And wakeup scheduler process if we have
2913 	 * lots of memory. this process will swapin processes.
2914 	 */
2915 	if (vm_pages_needed && !vm_page_count_min()) {
2916 		vm_pages_needed = false;
2917 		wakeup(&vm_cnt.v_free_count);
2918 	}
2919 }
2920 
2921 /*
2922  *	Turn a cached page into a free page, by changing its attributes.
2923  *	Keep the statistics up-to-date.
2924  *
2925  *	The free page queue must be locked.
2926  */
2927 static void
2928 vm_page_cache_turn_free(vm_page_t m)
2929 {
2930 
2931 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
2932 
2933 	m->object = NULL;
2934 	m->valid = 0;
2935 	KASSERT((m->flags & PG_CACHED) != 0,
2936 	    ("vm_page_cache_turn_free: page %p is not cached", m));
2937 	m->flags &= ~PG_CACHED;
2938 	vm_cnt.v_cache_count--;
2939 	vm_phys_freecnt_adj(m, 1);
2940 }
2941 
2942 /*
2943  *	vm_page_free_toq:
2944  *
2945  *	Returns the given page to the free list,
2946  *	disassociating it with any VM object.
2947  *
2948  *	The object must be locked.  The page must be locked if it is managed.
2949  */
2950 void
2951 vm_page_free_toq(vm_page_t m)
2952 {
2953 
2954 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2955 		vm_page_lock_assert(m, MA_OWNED);
2956 		KASSERT(!pmap_page_is_mapped(m),
2957 		    ("vm_page_free_toq: freeing mapped page %p", m));
2958 	} else
2959 		KASSERT(m->queue == PQ_NONE,
2960 		    ("vm_page_free_toq: unmanaged page %p is queued", m));
2961 	PCPU_INC(cnt.v_tfree);
2962 
2963 	if (vm_page_sbusied(m))
2964 		panic("vm_page_free: freeing busy page %p", m);
2965 
2966 	/*
2967 	 * Unqueue, then remove page.  Note that we cannot destroy
2968 	 * the page here because we do not want to call the pager's
2969 	 * callback routine until after we've put the page on the
2970 	 * appropriate free queue.
2971 	 */
2972 	vm_page_remque(m);
2973 	vm_page_remove(m);
2974 
2975 	/*
2976 	 * If fictitious remove object association and
2977 	 * return, otherwise delay object association removal.
2978 	 */
2979 	if ((m->flags & PG_FICTITIOUS) != 0) {
2980 		return;
2981 	}
2982 
2983 	m->valid = 0;
2984 	vm_page_undirty(m);
2985 
2986 	if (m->wire_count != 0)
2987 		panic("vm_page_free: freeing wired page %p", m);
2988 	if (m->hold_count != 0) {
2989 		m->flags &= ~PG_ZERO;
2990 		KASSERT((m->flags & PG_UNHOLDFREE) == 0,
2991 		    ("vm_page_free: freeing PG_UNHOLDFREE page %p", m));
2992 		m->flags |= PG_UNHOLDFREE;
2993 	} else {
2994 		/*
2995 		 * Restore the default memory attribute to the page.
2996 		 */
2997 		if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
2998 			pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
2999 
3000 		/*
3001 		 * Insert the page into the physical memory allocator's
3002 		 * cache/free page queues.
3003 		 */
3004 		mtx_lock(&vm_page_queue_free_mtx);
3005 		vm_phys_freecnt_adj(m, 1);
3006 #if VM_NRESERVLEVEL > 0
3007 		if (!vm_reserv_free_page(m))
3008 #else
3009 		if (TRUE)
3010 #endif
3011 			vm_phys_free_pages(m, 0);
3012 		if ((m->flags & PG_ZERO) != 0)
3013 			++vm_page_zero_count;
3014 		else
3015 			vm_page_zero_idle_wakeup();
3016 		vm_page_free_wakeup();
3017 		mtx_unlock(&vm_page_queue_free_mtx);
3018 	}
3019 }
3020 
3021 /*
3022  *	vm_page_wire:
3023  *
3024  *	Mark this page as wired down by yet
3025  *	another map, removing it from paging queues
3026  *	as necessary.
3027  *
3028  *	If the page is fictitious, then its wire count must remain one.
3029  *
3030  *	The page must be locked.
3031  */
3032 void
3033 vm_page_wire(vm_page_t m)
3034 {
3035 
3036 	/*
3037 	 * Only bump the wire statistics if the page is not already wired,
3038 	 * and only unqueue the page if it is on some queue (if it is unmanaged
3039 	 * it is already off the queues).
3040 	 */
3041 	vm_page_lock_assert(m, MA_OWNED);
3042 	if ((m->flags & PG_FICTITIOUS) != 0) {
3043 		KASSERT(m->wire_count == 1,
3044 		    ("vm_page_wire: fictitious page %p's wire count isn't one",
3045 		    m));
3046 		return;
3047 	}
3048 	if (m->wire_count == 0) {
3049 		KASSERT((m->oflags & VPO_UNMANAGED) == 0 ||
3050 		    m->queue == PQ_NONE,
3051 		    ("vm_page_wire: unmanaged page %p is queued", m));
3052 		vm_page_remque(m);
3053 		atomic_add_int(&vm_cnt.v_wire_count, 1);
3054 	}
3055 	m->wire_count++;
3056 	KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
3057 }
3058 
3059 /*
3060  * vm_page_unwire:
3061  *
3062  * Release one wiring of the specified page, potentially allowing it to be
3063  * paged out.  Returns TRUE if the number of wirings transitions to zero and
3064  * FALSE otherwise.
3065  *
3066  * Only managed pages belonging to an object can be paged out.  If the number
3067  * of wirings transitions to zero and the page is eligible for page out, then
3068  * the page is added to the specified paging queue (unless PQ_NONE is
3069  * specified).
3070  *
3071  * If a page is fictitious, then its wire count must always be one.
3072  *
3073  * A managed page must be locked.
3074  */
3075 boolean_t
3076 vm_page_unwire(vm_page_t m, uint8_t queue)
3077 {
3078 
3079 	KASSERT(queue < PQ_COUNT || queue == PQ_NONE,
3080 	    ("vm_page_unwire: invalid queue %u request for page %p",
3081 	    queue, m));
3082 	if ((m->oflags & VPO_UNMANAGED) == 0)
3083 		vm_page_assert_locked(m);
3084 	if ((m->flags & PG_FICTITIOUS) != 0) {
3085 		KASSERT(m->wire_count == 1,
3086 	    ("vm_page_unwire: fictitious page %p's wire count isn't one", m));
3087 		return (FALSE);
3088 	}
3089 	if (m->wire_count > 0) {
3090 		m->wire_count--;
3091 		if (m->wire_count == 0) {
3092 			atomic_subtract_int(&vm_cnt.v_wire_count, 1);
3093 			if ((m->oflags & VPO_UNMANAGED) == 0 &&
3094 			    m->object != NULL && queue != PQ_NONE) {
3095 				if (queue == PQ_INACTIVE)
3096 					m->flags &= ~PG_WINATCFLS;
3097 				vm_page_enqueue(queue, m);
3098 			}
3099 			return (TRUE);
3100 		} else
3101 			return (FALSE);
3102 	} else
3103 		panic("vm_page_unwire: page %p's wire count is zero", m);
3104 }
3105 
3106 /*
3107  * Move the specified page to the inactive queue.
3108  *
3109  * Many pages placed on the inactive queue should actually go
3110  * into the cache, but it is difficult to figure out which.  What
3111  * we do instead, if the inactive target is well met, is to put
3112  * clean pages at the head of the inactive queue instead of the tail.
3113  * This will cause them to be moved to the cache more quickly and
3114  * if not actively re-referenced, reclaimed more quickly.  If we just
3115  * stick these pages at the end of the inactive queue, heavy filesystem
3116  * meta-data accesses can cause an unnecessary paging load on memory bound
3117  * processes.  This optimization causes one-time-use metadata to be
3118  * reused more quickly.
3119  *
3120  * Normally noreuse is FALSE, resulting in LRU operation.  noreuse is set
3121  * to TRUE if we want this page to be 'as if it were placed in the cache',
3122  * except without unmapping it from the process address space.  In
3123  * practice this is implemented by inserting the page at the head of the
3124  * queue, using a marker page to guide FIFO insertion ordering.
3125  *
3126  * The page must be locked.
3127  */
3128 static inline void
3129 _vm_page_deactivate(vm_page_t m, boolean_t noreuse)
3130 {
3131 	struct vm_pagequeue *pq;
3132 	int queue;
3133 
3134 	vm_page_assert_locked(m);
3135 
3136 	/*
3137 	 * Ignore if the page is already inactive, unless it is unlikely to be
3138 	 * reactivated.
3139 	 */
3140 	if ((queue = m->queue) == PQ_INACTIVE && !noreuse)
3141 		return;
3142 	if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) {
3143 		pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE];
3144 		/* Avoid multiple acquisitions of the inactive queue lock. */
3145 		if (queue == PQ_INACTIVE) {
3146 			vm_pagequeue_lock(pq);
3147 			vm_page_dequeue_locked(m);
3148 		} else {
3149 			if (queue != PQ_NONE)
3150 				vm_page_dequeue(m);
3151 			m->flags &= ~PG_WINATCFLS;
3152 			vm_pagequeue_lock(pq);
3153 		}
3154 		m->queue = PQ_INACTIVE;
3155 		if (noreuse)
3156 			TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead,
3157 			    m, plinks.q);
3158 		else
3159 			TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
3160 		vm_pagequeue_cnt_inc(pq);
3161 		vm_pagequeue_unlock(pq);
3162 	}
3163 }
3164 
3165 /*
3166  * Move the specified page to the inactive queue.
3167  *
3168  * The page must be locked.
3169  */
3170 void
3171 vm_page_deactivate(vm_page_t m)
3172 {
3173 
3174 	_vm_page_deactivate(m, FALSE);
3175 }
3176 
3177 /*
3178  * Move the specified page to the inactive queue with the expectation
3179  * that it is unlikely to be reused.
3180  *
3181  * The page must be locked.
3182  */
3183 void
3184 vm_page_deactivate_noreuse(vm_page_t m)
3185 {
3186 
3187 	_vm_page_deactivate(m, TRUE);
3188 }
3189 
3190 /*
3191  * vm_page_try_to_cache:
3192  *
3193  * Returns 0 on failure, 1 on success
3194  */
3195 int
3196 vm_page_try_to_cache(vm_page_t m)
3197 {
3198 
3199 	vm_page_lock_assert(m, MA_OWNED);
3200 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3201 	if (m->dirty || m->hold_count || m->wire_count ||
3202 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
3203 		return (0);
3204 	pmap_remove_all(m);
3205 	if (m->dirty)
3206 		return (0);
3207 	vm_page_cache(m);
3208 	return (1);
3209 }
3210 
3211 /*
3212  * vm_page_try_to_free()
3213  *
3214  *	Attempt to free the page.  If we cannot free it, we do nothing.
3215  *	1 is returned on success, 0 on failure.
3216  */
3217 int
3218 vm_page_try_to_free(vm_page_t m)
3219 {
3220 
3221 	vm_page_lock_assert(m, MA_OWNED);
3222 	if (m->object != NULL)
3223 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3224 	if (m->dirty || m->hold_count || m->wire_count ||
3225 	    (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m))
3226 		return (0);
3227 	pmap_remove_all(m);
3228 	if (m->dirty)
3229 		return (0);
3230 	vm_page_free(m);
3231 	return (1);
3232 }
3233 
3234 /*
3235  * vm_page_cache
3236  *
3237  * Put the specified page onto the page cache queue (if appropriate).
3238  *
3239  * The object and page must be locked.
3240  */
3241 void
3242 vm_page_cache(vm_page_t m)
3243 {
3244 	vm_object_t object;
3245 	boolean_t cache_was_empty;
3246 
3247 	vm_page_lock_assert(m, MA_OWNED);
3248 	object = m->object;
3249 	VM_OBJECT_ASSERT_WLOCKED(object);
3250 	if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) ||
3251 	    m->hold_count || m->wire_count)
3252 		panic("vm_page_cache: attempting to cache busy page");
3253 	KASSERT(!pmap_page_is_mapped(m),
3254 	    ("vm_page_cache: page %p is mapped", m));
3255 	KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m));
3256 	if (m->valid == 0 || object->type == OBJT_DEFAULT ||
3257 	    (object->type == OBJT_SWAP &&
3258 	    !vm_pager_has_page(object, m->pindex, NULL, NULL))) {
3259 		/*
3260 		 * Hypothesis: A cache-eligible page belonging to a
3261 		 * default object or swap object but without a backing
3262 		 * store must be zero filled.
3263 		 */
3264 		vm_page_free(m);
3265 		return;
3266 	}
3267 	KASSERT((m->flags & PG_CACHED) == 0,
3268 	    ("vm_page_cache: page %p is already cached", m));
3269 
3270 	/*
3271 	 * Remove the page from the paging queues.
3272 	 */
3273 	vm_page_remque(m);
3274 
3275 	/*
3276 	 * Remove the page from the object's collection of resident
3277 	 * pages.
3278 	 */
3279 	vm_radix_remove(&object->rtree, m->pindex);
3280 	TAILQ_REMOVE(&object->memq, m, listq);
3281 	object->resident_page_count--;
3282 
3283 	/*
3284 	 * Restore the default memory attribute to the page.
3285 	 */
3286 	if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
3287 		pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
3288 
3289 	/*
3290 	 * Insert the page into the object's collection of cached pages
3291 	 * and the physical memory allocator's cache/free page queues.
3292 	 */
3293 	m->flags &= ~PG_ZERO;
3294 	mtx_lock(&vm_page_queue_free_mtx);
3295 	cache_was_empty = vm_radix_is_empty(&object->cache);
3296 	if (vm_radix_insert(&object->cache, m)) {
3297 		mtx_unlock(&vm_page_queue_free_mtx);
3298 		if (object->type == OBJT_VNODE &&
3299 		    object->resident_page_count == 0)
3300 			vdrop(object->handle);
3301 		m->object = NULL;
3302 		vm_page_free(m);
3303 		return;
3304 	}
3305 
3306 	/*
3307 	 * The above call to vm_radix_insert() could reclaim the one pre-
3308 	 * existing cached page from this object, resulting in a call to
3309 	 * vdrop().
3310 	 */
3311 	if (!cache_was_empty)
3312 		cache_was_empty = vm_radix_is_singleton(&object->cache);
3313 
3314 	m->flags |= PG_CACHED;
3315 	vm_cnt.v_cache_count++;
3316 	PCPU_INC(cnt.v_tcached);
3317 #if VM_NRESERVLEVEL > 0
3318 	if (!vm_reserv_free_page(m)) {
3319 #else
3320 	if (TRUE) {
3321 #endif
3322 		vm_phys_free_pages(m, 0);
3323 	}
3324 	vm_page_free_wakeup();
3325 	mtx_unlock(&vm_page_queue_free_mtx);
3326 
3327 	/*
3328 	 * Increment the vnode's hold count if this is the object's only
3329 	 * cached page.  Decrement the vnode's hold count if this was
3330 	 * the object's only resident page.
3331 	 */
3332 	if (object->type == OBJT_VNODE) {
3333 		if (cache_was_empty && object->resident_page_count != 0)
3334 			vhold(object->handle);
3335 		else if (!cache_was_empty && object->resident_page_count == 0)
3336 			vdrop(object->handle);
3337 	}
3338 }
3339 
3340 /*
3341  * vm_page_advise
3342  *
3343  * 	Deactivate or do nothing, as appropriate.
3344  *
3345  *	The object and page must be locked.
3346  */
3347 void
3348 vm_page_advise(vm_page_t m, int advice)
3349 {
3350 
3351 	vm_page_assert_locked(m);
3352 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3353 	if (advice == MADV_FREE)
3354 		/*
3355 		 * Mark the page clean.  This will allow the page to be freed
3356 		 * up by the system.  However, such pages are often reused
3357 		 * quickly by malloc() so we do not do anything that would
3358 		 * cause a page fault if we can help it.
3359 		 *
3360 		 * Specifically, we do not try to actually free the page now
3361 		 * nor do we try to put it in the cache (which would cause a
3362 		 * page fault on reuse).
3363 		 *
3364 		 * But we do make the page as freeable as we can without
3365 		 * actually taking the step of unmapping it.
3366 		 */
3367 		m->dirty = 0;
3368 	else if (advice != MADV_DONTNEED)
3369 		return;
3370 
3371 	/*
3372 	 * Clear any references to the page.  Otherwise, the page daemon will
3373 	 * immediately reactivate the page.
3374 	 */
3375 	vm_page_aflag_clear(m, PGA_REFERENCED);
3376 
3377 	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
3378 		vm_page_dirty(m);
3379 
3380 	/*
3381 	 * Place clean pages at the head of the inactive queue rather than the
3382 	 * tail, thus defeating the queue's LRU operation and ensuring that the
3383 	 * page will be reused quickly.
3384 	 */
3385 	_vm_page_deactivate(m, m->dirty == 0);
3386 }
3387 
3388 /*
3389  * Grab a page, waiting until we are waken up due to the page
3390  * changing state.  We keep on waiting, if the page continues
3391  * to be in the object.  If the page doesn't exist, first allocate it
3392  * and then conditionally zero it.
3393  *
3394  * This routine may sleep.
3395  *
3396  * The object must be locked on entry.  The lock will, however, be released
3397  * and reacquired if the routine sleeps.
3398  */
3399 vm_page_t
3400 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
3401 {
3402 	vm_page_t m;
3403 	int sleep;
3404 
3405 	VM_OBJECT_ASSERT_WLOCKED(object);
3406 	KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
3407 	    (allocflags & VM_ALLOC_IGN_SBUSY) != 0,
3408 	    ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
3409 retrylookup:
3410 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
3411 		sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ?
3412 		    vm_page_xbusied(m) : vm_page_busied(m);
3413 		if (sleep) {
3414 			if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3415 				return (NULL);
3416 			/*
3417 			 * Reference the page before unlocking and
3418 			 * sleeping so that the page daemon is less
3419 			 * likely to reclaim it.
3420 			 */
3421 			vm_page_aflag_set(m, PGA_REFERENCED);
3422 			vm_page_lock(m);
3423 			VM_OBJECT_WUNLOCK(object);
3424 			vm_page_busy_sleep(m, "pgrbwt");
3425 			VM_OBJECT_WLOCK(object);
3426 			goto retrylookup;
3427 		} else {
3428 			if ((allocflags & VM_ALLOC_WIRED) != 0) {
3429 				vm_page_lock(m);
3430 				vm_page_wire(m);
3431 				vm_page_unlock(m);
3432 			}
3433 			if ((allocflags &
3434 			    (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
3435 				vm_page_xbusy(m);
3436 			if ((allocflags & VM_ALLOC_SBUSY) != 0)
3437 				vm_page_sbusy(m);
3438 			return (m);
3439 		}
3440 	}
3441 	m = vm_page_alloc(object, pindex, allocflags);
3442 	if (m == NULL) {
3443 		if ((allocflags & VM_ALLOC_NOWAIT) != 0)
3444 			return (NULL);
3445 		VM_OBJECT_WUNLOCK(object);
3446 		VM_WAIT;
3447 		VM_OBJECT_WLOCK(object);
3448 		goto retrylookup;
3449 	} else if (m->valid != 0)
3450 		return (m);
3451 	if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
3452 		pmap_zero_page(m);
3453 	return (m);
3454 }
3455 
3456 /*
3457  * Mapping function for valid or dirty bits in a page.
3458  *
3459  * Inputs are required to range within a page.
3460  */
3461 vm_page_bits_t
3462 vm_page_bits(int base, int size)
3463 {
3464 	int first_bit;
3465 	int last_bit;
3466 
3467 	KASSERT(
3468 	    base + size <= PAGE_SIZE,
3469 	    ("vm_page_bits: illegal base/size %d/%d", base, size)
3470 	);
3471 
3472 	if (size == 0)		/* handle degenerate case */
3473 		return (0);
3474 
3475 	first_bit = base >> DEV_BSHIFT;
3476 	last_bit = (base + size - 1) >> DEV_BSHIFT;
3477 
3478 	return (((vm_page_bits_t)2 << last_bit) -
3479 	    ((vm_page_bits_t)1 << first_bit));
3480 }
3481 
3482 /*
3483  *	vm_page_set_valid_range:
3484  *
3485  *	Sets portions of a page valid.  The arguments are expected
3486  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3487  *	of any partial chunks touched by the range.  The invalid portion of
3488  *	such chunks will be zeroed.
3489  *
3490  *	(base + size) must be less then or equal to PAGE_SIZE.
3491  */
3492 void
3493 vm_page_set_valid_range(vm_page_t m, int base, int size)
3494 {
3495 	int endoff, frag;
3496 
3497 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3498 	if (size == 0)	/* handle degenerate case */
3499 		return;
3500 
3501 	/*
3502 	 * If the base is not DEV_BSIZE aligned and the valid
3503 	 * bit is clear, we have to zero out a portion of the
3504 	 * first block.
3505 	 */
3506 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3507 	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
3508 		pmap_zero_page_area(m, frag, base - frag);
3509 
3510 	/*
3511 	 * If the ending offset is not DEV_BSIZE aligned and the
3512 	 * valid bit is clear, we have to zero out a portion of
3513 	 * the last block.
3514 	 */
3515 	endoff = base + size;
3516 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3517 	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
3518 		pmap_zero_page_area(m, endoff,
3519 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3520 
3521 	/*
3522 	 * Assert that no previously invalid block that is now being validated
3523 	 * is already dirty.
3524 	 */
3525 	KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
3526 	    ("vm_page_set_valid_range: page %p is dirty", m));
3527 
3528 	/*
3529 	 * Set valid bits inclusive of any overlap.
3530 	 */
3531 	m->valid |= vm_page_bits(base, size);
3532 }
3533 
3534 /*
3535  * Clear the given bits from the specified page's dirty field.
3536  */
3537 static __inline void
3538 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
3539 {
3540 	uintptr_t addr;
3541 #if PAGE_SIZE < 16384
3542 	int shift;
3543 #endif
3544 
3545 	/*
3546 	 * If the object is locked and the page is neither exclusive busy nor
3547 	 * write mapped, then the page's dirty field cannot possibly be
3548 	 * set by a concurrent pmap operation.
3549 	 */
3550 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3551 	if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
3552 		m->dirty &= ~pagebits;
3553 	else {
3554 		/*
3555 		 * The pmap layer can call vm_page_dirty() without
3556 		 * holding a distinguished lock.  The combination of
3557 		 * the object's lock and an atomic operation suffice
3558 		 * to guarantee consistency of the page dirty field.
3559 		 *
3560 		 * For PAGE_SIZE == 32768 case, compiler already
3561 		 * properly aligns the dirty field, so no forcible
3562 		 * alignment is needed. Only require existence of
3563 		 * atomic_clear_64 when page size is 32768.
3564 		 */
3565 		addr = (uintptr_t)&m->dirty;
3566 #if PAGE_SIZE == 32768
3567 		atomic_clear_64((uint64_t *)addr, pagebits);
3568 #elif PAGE_SIZE == 16384
3569 		atomic_clear_32((uint32_t *)addr, pagebits);
3570 #else		/* PAGE_SIZE <= 8192 */
3571 		/*
3572 		 * Use a trick to perform a 32-bit atomic on the
3573 		 * containing aligned word, to not depend on the existence
3574 		 * of atomic_clear_{8, 16}.
3575 		 */
3576 		shift = addr & (sizeof(uint32_t) - 1);
3577 #if BYTE_ORDER == BIG_ENDIAN
3578 		shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY;
3579 #else
3580 		shift *= NBBY;
3581 #endif
3582 		addr &= ~(sizeof(uint32_t) - 1);
3583 		atomic_clear_32((uint32_t *)addr, pagebits << shift);
3584 #endif		/* PAGE_SIZE */
3585 	}
3586 }
3587 
3588 /*
3589  *	vm_page_set_validclean:
3590  *
3591  *	Sets portions of a page valid and clean.  The arguments are expected
3592  *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
3593  *	of any partial chunks touched by the range.  The invalid portion of
3594  *	such chunks will be zero'd.
3595  *
3596  *	(base + size) must be less then or equal to PAGE_SIZE.
3597  */
3598 void
3599 vm_page_set_validclean(vm_page_t m, int base, int size)
3600 {
3601 	vm_page_bits_t oldvalid, pagebits;
3602 	int endoff, frag;
3603 
3604 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3605 	if (size == 0)	/* handle degenerate case */
3606 		return;
3607 
3608 	/*
3609 	 * If the base is not DEV_BSIZE aligned and the valid
3610 	 * bit is clear, we have to zero out a portion of the
3611 	 * first block.
3612 	 */
3613 	if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
3614 	    (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
3615 		pmap_zero_page_area(m, frag, base - frag);
3616 
3617 	/*
3618 	 * If the ending offset is not DEV_BSIZE aligned and the
3619 	 * valid bit is clear, we have to zero out a portion of
3620 	 * the last block.
3621 	 */
3622 	endoff = base + size;
3623 	if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
3624 	    (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
3625 		pmap_zero_page_area(m, endoff,
3626 		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
3627 
3628 	/*
3629 	 * Set valid, clear dirty bits.  If validating the entire
3630 	 * page we can safely clear the pmap modify bit.  We also
3631 	 * use this opportunity to clear the VPO_NOSYNC flag.  If a process
3632 	 * takes a write fault on a MAP_NOSYNC memory area the flag will
3633 	 * be set again.
3634 	 *
3635 	 * We set valid bits inclusive of any overlap, but we can only
3636 	 * clear dirty bits for DEV_BSIZE chunks that are fully within
3637 	 * the range.
3638 	 */
3639 	oldvalid = m->valid;
3640 	pagebits = vm_page_bits(base, size);
3641 	m->valid |= pagebits;
3642 #if 0	/* NOT YET */
3643 	if ((frag = base & (DEV_BSIZE - 1)) != 0) {
3644 		frag = DEV_BSIZE - frag;
3645 		base += frag;
3646 		size -= frag;
3647 		if (size < 0)
3648 			size = 0;
3649 	}
3650 	pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
3651 #endif
3652 	if (base == 0 && size == PAGE_SIZE) {
3653 		/*
3654 		 * The page can only be modified within the pmap if it is
3655 		 * mapped, and it can only be mapped if it was previously
3656 		 * fully valid.
3657 		 */
3658 		if (oldvalid == VM_PAGE_BITS_ALL)
3659 			/*
3660 			 * Perform the pmap_clear_modify() first.  Otherwise,
3661 			 * a concurrent pmap operation, such as
3662 			 * pmap_protect(), could clear a modification in the
3663 			 * pmap and set the dirty field on the page before
3664 			 * pmap_clear_modify() had begun and after the dirty
3665 			 * field was cleared here.
3666 			 */
3667 			pmap_clear_modify(m);
3668 		m->dirty = 0;
3669 		m->oflags &= ~VPO_NOSYNC;
3670 	} else if (oldvalid != VM_PAGE_BITS_ALL)
3671 		m->dirty &= ~pagebits;
3672 	else
3673 		vm_page_clear_dirty_mask(m, pagebits);
3674 }
3675 
3676 void
3677 vm_page_clear_dirty(vm_page_t m, int base, int size)
3678 {
3679 
3680 	vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
3681 }
3682 
3683 /*
3684  *	vm_page_set_invalid:
3685  *
3686  *	Invalidates DEV_BSIZE'd chunks within a page.  Both the
3687  *	valid and dirty bits for the effected areas are cleared.
3688  */
3689 void
3690 vm_page_set_invalid(vm_page_t m, int base, int size)
3691 {
3692 	vm_page_bits_t bits;
3693 	vm_object_t object;
3694 
3695 	object = m->object;
3696 	VM_OBJECT_ASSERT_WLOCKED(object);
3697 	if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
3698 	    size >= object->un_pager.vnp.vnp_size)
3699 		bits = VM_PAGE_BITS_ALL;
3700 	else
3701 		bits = vm_page_bits(base, size);
3702 	if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL &&
3703 	    bits != 0)
3704 		pmap_remove_all(m);
3705 	KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) ||
3706 	    !pmap_page_is_mapped(m),
3707 	    ("vm_page_set_invalid: page %p is mapped", m));
3708 	m->valid &= ~bits;
3709 	m->dirty &= ~bits;
3710 }
3711 
3712 /*
3713  * vm_page_zero_invalid()
3714  *
3715  *	The kernel assumes that the invalid portions of a page contain
3716  *	garbage, but such pages can be mapped into memory by user code.
3717  *	When this occurs, we must zero out the non-valid portions of the
3718  *	page so user code sees what it expects.
3719  *
3720  *	Pages are most often semi-valid when the end of a file is mapped
3721  *	into memory and the file's size is not page aligned.
3722  */
3723 void
3724 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
3725 {
3726 	int b;
3727 	int i;
3728 
3729 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3730 	/*
3731 	 * Scan the valid bits looking for invalid sections that
3732 	 * must be zeroed.  Invalid sub-DEV_BSIZE'd areas ( where the
3733 	 * valid bit may be set ) have already been zeroed by
3734 	 * vm_page_set_validclean().
3735 	 */
3736 	for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
3737 		if (i == (PAGE_SIZE / DEV_BSIZE) ||
3738 		    (m->valid & ((vm_page_bits_t)1 << i))) {
3739 			if (i > b) {
3740 				pmap_zero_page_area(m,
3741 				    b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
3742 			}
3743 			b = i + 1;
3744 		}
3745 	}
3746 
3747 	/*
3748 	 * setvalid is TRUE when we can safely set the zero'd areas
3749 	 * as being valid.  We can do this if there are no cache consistancy
3750 	 * issues.  e.g. it is ok to do with UFS, but not ok to do with NFS.
3751 	 */
3752 	if (setvalid)
3753 		m->valid = VM_PAGE_BITS_ALL;
3754 }
3755 
3756 /*
3757  *	vm_page_is_valid:
3758  *
3759  *	Is (partial) page valid?  Note that the case where size == 0
3760  *	will return FALSE in the degenerate case where the page is
3761  *	entirely invalid, and TRUE otherwise.
3762  */
3763 int
3764 vm_page_is_valid(vm_page_t m, int base, int size)
3765 {
3766 	vm_page_bits_t bits;
3767 
3768 	VM_OBJECT_ASSERT_LOCKED(m->object);
3769 	bits = vm_page_bits(base, size);
3770 	return (m->valid != 0 && (m->valid & bits) == bits);
3771 }
3772 
3773 /*
3774  *	vm_page_ps_is_valid:
3775  *
3776  *	Returns TRUE if the entire (super)page is valid and FALSE otherwise.
3777  */
3778 boolean_t
3779 vm_page_ps_is_valid(vm_page_t m)
3780 {
3781 	int i, npages;
3782 
3783 	VM_OBJECT_ASSERT_LOCKED(m->object);
3784 	npages = atop(pagesizes[m->psind]);
3785 
3786 	/*
3787 	 * The physically contiguous pages that make up a superpage, i.e., a
3788 	 * page with a page size index ("psind") greater than zero, will
3789 	 * occupy adjacent entries in vm_page_array[].
3790 	 */
3791 	for (i = 0; i < npages; i++) {
3792 		if (m[i].valid != VM_PAGE_BITS_ALL)
3793 			return (FALSE);
3794 	}
3795 	return (TRUE);
3796 }
3797 
3798 /*
3799  * Set the page's dirty bits if the page is modified.
3800  */
3801 void
3802 vm_page_test_dirty(vm_page_t m)
3803 {
3804 
3805 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3806 	if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
3807 		vm_page_dirty(m);
3808 }
3809 
3810 void
3811 vm_page_lock_KBI(vm_page_t m, const char *file, int line)
3812 {
3813 
3814 	mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
3815 }
3816 
3817 void
3818 vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
3819 {
3820 
3821 	mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
3822 }
3823 
3824 int
3825 vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
3826 {
3827 
3828 	return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
3829 }
3830 
3831 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
3832 void
3833 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
3834 {
3835 
3836 	vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
3837 }
3838 
3839 void
3840 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
3841 {
3842 
3843 	mtx_assert_(vm_page_lockptr(m), a, file, line);
3844 }
3845 #endif
3846 
3847 #ifdef INVARIANTS
3848 void
3849 vm_page_object_lock_assert(vm_page_t m)
3850 {
3851 
3852 	/*
3853 	 * Certain of the page's fields may only be modified by the
3854 	 * holder of the containing object's lock or the exclusive busy.
3855 	 * holder.  Unfortunately, the holder of the write busy is
3856 	 * not recorded, and thus cannot be checked here.
3857 	 */
3858 	if (m->object != NULL && !vm_page_xbusied(m))
3859 		VM_OBJECT_ASSERT_WLOCKED(m->object);
3860 }
3861 
3862 void
3863 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits)
3864 {
3865 
3866 	if ((bits & PGA_WRITEABLE) == 0)
3867 		return;
3868 
3869 	/*
3870 	 * The PGA_WRITEABLE flag can only be set if the page is
3871 	 * managed, is exclusively busied or the object is locked.
3872 	 * Currently, this flag is only set by pmap_enter().
3873 	 */
3874 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3875 	    ("PGA_WRITEABLE on unmanaged page"));
3876 	if (!vm_page_xbusied(m))
3877 		VM_OBJECT_ASSERT_LOCKED(m->object);
3878 }
3879 #endif
3880 
3881 #include "opt_ddb.h"
3882 #ifdef DDB
3883 #include <sys/kernel.h>
3884 
3885 #include <ddb/ddb.h>
3886 
3887 DB_SHOW_COMMAND(page, vm_page_print_page_info)
3888 {
3889 	db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count);
3890 	db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count);
3891 	db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count);
3892 	db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count);
3893 	db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count);
3894 	db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
3895 	db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
3896 	db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
3897 	db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
3898 }
3899 
3900 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
3901 {
3902 	int dom;
3903 
3904 	db_printf("pq_free %d pq_cache %d\n",
3905 	    vm_cnt.v_free_count, vm_cnt.v_cache_count);
3906 	for (dom = 0; dom < vm_ndomains; dom++) {
3907 		db_printf(
3908 	"dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n",
3909 		    dom,
3910 		    vm_dom[dom].vmd_page_count,
3911 		    vm_dom[dom].vmd_free_count,
3912 		    vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
3913 		    vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
3914 		    vm_dom[dom].vmd_pass);
3915 	}
3916 }
3917 
3918 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
3919 {
3920 	vm_page_t m;
3921 	boolean_t phys;
3922 
3923 	if (!have_addr) {
3924 		db_printf("show pginfo addr\n");
3925 		return;
3926 	}
3927 
3928 	phys = strchr(modif, 'p') != NULL;
3929 	if (phys)
3930 		m = PHYS_TO_VM_PAGE(addr);
3931 	else
3932 		m = (vm_page_t)addr;
3933 	db_printf(
3934     "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n"
3935     "  af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
3936 	    m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
3937 	    m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags,
3938 	    m->flags, m->act_count, m->busy_lock, m->valid, m->dirty);
3939 }
3940 #endif /* DDB */
3941