xref: /dflybsd-src/sys/vm/vm_page.h (revision 1be4932c67b48d3aa9a9d6db1cac600d0d84a01c)
1 /*
2  * Copyright (c) 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * The Mach Operating System project at Carnegie-Mellon University.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)vm_page.h	8.2 (Berkeley) 12/13/93
37  *
38  *
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  *
64  * $FreeBSD: src/sys/vm/vm_page.h,v 1.75.2.8 2002/03/06 01:07:09 dillon Exp $
65  * $DragonFly: src/sys/vm/vm_page.h,v 1.28 2008/05/09 07:24:48 dillon Exp $
66  */
67 
68 /*
69  *	Resident memory system definitions.
70  */
71 
72 #ifndef	_VM_VM_PAGE_H_
73 #define	_VM_VM_PAGE_H_
74 
75 #if !defined(KLD_MODULE) && defined(_KERNEL)
76 #include "opt_vmpage.h"
77 #endif
78 
79 #ifndef _SYS_TYPES_H_
80 #include <sys/types.h>
81 #endif
82 #ifndef _SYS_TREE_H_
83 #include <sys/tree.h>
84 #endif
85 #ifndef _MACHINE_PMAP_H_
86 #include <machine/pmap.h>
87 #endif
88 #ifndef _VM_PMAP_H_
89 #include <vm/pmap.h>
90 #endif
91 #ifndef _MACHINE_ATOMIC_H_
92 #include <machine/atomic.h>
93 #endif
94 
95 #ifdef _KERNEL
96 
97 #ifndef _SYS_SYSTM_H_
98 #include <sys/systm.h>
99 #endif
100 #ifndef _SYS_THREAD2_H_
101 #include <sys/thread2.h>
102 #endif
103 
104 #ifdef __x86_64__
105 #include <machine/vmparam.h>
106 #endif
107 
108 #endif
109 
110 typedef enum vm_page_event { VMEVENT_NONE, VMEVENT_COW } vm_page_event_t;
111 
112 struct vm_page_action {
113 	LIST_ENTRY(vm_page_action) entry;
114 	struct vm_page		*m;
115 	vm_page_event_t		event;
116 	void			(*func)(struct vm_page *,
117 					struct vm_page_action *);
118 	void			*data;
119 };
120 
121 typedef struct vm_page_action *vm_page_action_t;
122 
123 /*
124  *	Management of resident (logical) pages.
125  *
126  *	A small structure is kept for each resident
127  *	page, indexed by page number.  Each structure
128  *	is an element of several lists:
129  *
130  *		A hash table bucket used to quickly
131  *		perform object/offset lookups
132  *
133  *		A list of all pages for a given object,
134  *		so they can be quickly deactivated at
135  *		time of deallocation.
136  *
137  *		An ordered list of pages due for pageout.
138  *
139  *	In addition, the structure contains the object
140  *	and offset to which this page belongs (for pageout),
141  *	and sundry status bits.
142  *
143  *	Fields in this structure are locked either by the lock on the
144  *	object that the page belongs to (O) or by the lock on the page
145  *	queues (P).
146  *
147  *	The 'valid' and 'dirty' fields are distinct.  A page may have dirty
148  *	bits set without having associated valid bits set.  This is used by
149  *	NFS to implement piecemeal writes.
150  */
151 
152 TAILQ_HEAD(pglist, vm_page);
153 
154 struct vm_object;
155 
156 int rb_vm_page_compare(struct vm_page *, struct vm_page *);
157 
158 struct vm_page_rb_tree;
159 RB_PROTOTYPE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare, vm_pindex_t);
160 
161 struct vm_page {
162 	TAILQ_ENTRY(vm_page) pageq;	/* vm_page_queues[] list (P)	*/
163 	RB_ENTRY(vm_page) rb_entry;	/* Red-Black tree based at object */
164 
165 	struct vm_object *object;	/* which object am I in (O,P)*/
166 	vm_pindex_t pindex;		/* offset into object (O,P) */
167 	vm_paddr_t phys_addr;		/* physical address of page */
168 	struct md_page md;		/* machine dependant stuff */
169 	u_short	queue;			/* page queue index */
170 	u_short	pc;			/* page color */
171 	u_char	act_count;		/* page usage count */
172 	u_char	busy;			/* page busy count */
173 	u_char	unused01;
174 	u_char	unused02;
175 	u_int32_t flags;		/* see below */
176 	u_int	wire_count;		/* wired down maps refs (P) */
177 	int 	hold_count;		/* page hold count */
178 
179 	/*
180 	 * NOTE that these must support one bit per DEV_BSIZE in a page!!!
181 	 * so, on normal X86 kernels, they must be at least 8 bits wide.
182 	 */
183 	u_char	valid;			/* map of valid DEV_BSIZE chunks */
184 	u_char	dirty;			/* map of dirty DEV_BSIZE chunks */
185 
186 	int	ku_pagecnt;		/* kmalloc helper */
187 };
188 
189 #ifndef __VM_PAGE_T_DEFINED__
190 #define __VM_PAGE_T_DEFINED__
191 typedef struct vm_page *vm_page_t;
192 #endif
193 
194 /*
195  * Page coloring parameters.  We default to a middle of the road optimization.
196  * Larger selections would not really hurt us but if a machine does not have
197  * a lot of memory it could cause vm_page_alloc() to eat more cpu cycles
198  * looking for free pages.
199  *
200  * Page coloring cannot be disabled.  Modules do not have access to most PQ
201  * constants because they can change between builds.
202  */
203 #if defined(_KERNEL) && !defined(KLD_MODULE)
204 
205 #if !defined(PQ_CACHESIZE)
206 #define PQ_CACHESIZE 256	/* max is 1024 (MB) */
207 #endif
208 
209 #if PQ_CACHESIZE >= 1024
210 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
211 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
212 #define PQ_L2_SIZE 256	/* A number of colors opt for 1M cache */
213 
214 #elif PQ_CACHESIZE >= 512
215 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
216 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
217 #define PQ_L2_SIZE 128	/* A number of colors opt for 512K cache */
218 
219 #elif PQ_CACHESIZE >= 256
220 #define PQ_PRIME1 13	/* Prime number somewhat less than PQ_HASH_SIZE */
221 #define PQ_PRIME2 7	/* Prime number somewhat less than PQ_HASH_SIZE */
222 #define PQ_L2_SIZE 64	/* A number of colors opt for 256K cache */
223 
224 #elif PQ_CACHESIZE >= 128
225 #define PQ_PRIME1 9	/* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
226 #define PQ_PRIME2 5	/* Prime number somewhat less than PQ_HASH_SIZE */
227 #define PQ_L2_SIZE 32	/* A number of colors opt for 128k cache */
228 
229 #else
230 #define PQ_PRIME1 5	/* Prime number somewhat less than PQ_HASH_SIZE */
231 #define PQ_PRIME2 3	/* Prime number somewhat less than PQ_HASH_SIZE */
232 #define PQ_L2_SIZE 16	/* A reasonable number of colors (opt for 64K cache) */
233 
234 #endif
235 
236 #define PQ_L2_MASK	(PQ_L2_SIZE - 1)
237 
238 #endif /* KERNEL && !KLD_MODULE */
239 
240 /*
241  *
242  * The queue array is always based on PQ_MAXL2_SIZE regardless of the actual
243  * cache size chosen in order to present a uniform interface for modules.
244  */
245 #define PQ_MAXL2_SIZE	256	/* fixed maximum (in pages) / module compat */
246 
247 #if PQ_L2_SIZE > PQ_MAXL2_SIZE
248 #error "Illegal PQ_L2_SIZE"
249 #endif
250 
251 #define PQ_NONE		0
252 #define PQ_FREE		1
253 #define PQ_INACTIVE	(1 + 1*PQ_MAXL2_SIZE)
254 #define PQ_ACTIVE	(2 + 1*PQ_MAXL2_SIZE)
255 #define PQ_CACHE	(3 + 1*PQ_MAXL2_SIZE)
256 #define PQ_HOLD		(3 + 2*PQ_MAXL2_SIZE)
257 #define PQ_COUNT	(4 + 2*PQ_MAXL2_SIZE)
258 
259 /*
260  * Scan support
261  */
262 struct vm_map;
263 
264 struct rb_vm_page_scan_info {
265 	vm_pindex_t	start_pindex;
266 	vm_pindex_t	end_pindex;
267 	int		limit;
268 	int		desired;
269 	int		error;
270 	int		pagerflags;
271 	vm_offset_t	addr;
272 	vm_pindex_t	backing_offset_index;
273 	struct vm_object *object;
274 	struct vm_object *backing_object;
275 	struct vm_page	*mpte;
276 	struct pmap	*pmap;
277 	struct vm_map	*map;
278 };
279 
280 int rb_vm_page_scancmp(struct vm_page *, void *);
281 
282 struct vpgqueues {
283 	struct pglist pl;
284 	int	*cnt;
285 	int	lcnt;
286 	int	flipflop;	/* probably not the best place */
287 };
288 
289 extern struct vpgqueues vm_page_queues[PQ_COUNT];
290 
291 /*
292  * These are the flags defined for vm_page.
293  *
294  *  PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
295  *  not under PV management but otherwise should be treated as a
296  *  normal page.  Pages not under PV management cannot be paged out
297  *  via the object/vm_page_t because there is no knowledge of their
298  *  pte mappings, nor can they be removed from their objects via
299  *  the object, and such pages are also not on any PQ queue.  The
300  *  PG_MAPPED and PG_WRITEABLE flags are not applicable.
301  *
302  *  PG_MAPPED only applies to managed pages, indicating whether the page
303  *  is mapped onto one or more pmaps.  A page might still be mapped to
304  *  special pmaps in an unmanaged fashion, for example when mapped into a
305  *  buffer cache buffer, without setting PG_MAPPED.
306  *
307  *  PG_WRITEABLE indicates that there may be a writeable managed pmap entry
308  *  somewhere, and that the page can be dirtied by hardware at any time
309  *  and may have to be tested for that.  The modified bit in unmanaged
310  *  mappings or in the special clean map is not tested.
311  *
312  *  PG_SWAPPED indicates that the page is backed by a swap block.  Any
313  *  VM object type other than OBJT_DEFAULT can have swap-backed pages now.
314  */
315 #define	PG_BUSY		0x00000001	/* page is in transit (O) */
316 #define	PG_WANTED	0x00000002	/* someone is waiting for page (O) */
317 #define PG_WINATCFLS	0x00000004	/* flush dirty page on inactive q */
318 #define	PG_FICTITIOUS	0x00000008	/* physical page doesn't exist (O) */
319 #define	PG_WRITEABLE	0x00000010	/* page is writeable */
320 #define PG_MAPPED	0x00000020	/* page is mapped (managed) */
321 #define	PG_ZERO		0x00000040	/* page is zeroed */
322 #define PG_REFERENCED	0x00000080	/* page has been referenced */
323 #define PG_CLEANCHK	0x00000100	/* page will be checked for cleaning */
324 #define PG_SWAPINPROG	0x00000200	/* swap I/O in progress on page	     */
325 #define PG_NOSYNC	0x00000400	/* do not collect for syncer */
326 #define PG_UNMANAGED	0x00000800	/* No PV management for page */
327 #define PG_MARKER	0x00001000	/* special queue marker page */
328 #define PG_RAM		0x00002000	/* read ahead mark */
329 #define PG_SWAPPED	0x00004000	/* backed by swap */
330 #define PG_NOTMETA	0x00008000	/* do not back with swap */
331 #define PG_ACTIONLIST	0x00010000	/* lookaside action list present */
332 	/* u_short, only 16 flag bits */
333 
334 /*
335  * Misc constants.
336  */
337 
338 #define ACT_DECLINE		1
339 #define ACT_ADVANCE		3
340 #define ACT_INIT		5
341 #define ACT_MAX			64
342 
343 #ifdef _KERNEL
344 /*
345  * Each pageable resident page falls into one of four lists:
346  *
347  *	free
348  *		Available for allocation now.
349  *
350  * The following are all LRU sorted:
351  *
352  *	cache
353  *		Almost available for allocation. Still in an
354  *		object, but clean and immediately freeable at
355  *		non-interrupt times.
356  *
357  *	inactive
358  *		Low activity, candidates for reclamation.
359  *		This is the list of pages that should be
360  *		paged out next.
361  *
362  *	active
363  *		Pages that are "active" i.e. they have been
364  *		recently referenced.
365  *
366  *	zero
367  *		Pages that are really free and have been pre-zeroed
368  *
369  */
370 
371 extern int vm_page_zero_count;
372 extern struct vm_page *vm_page_array;	/* First resident page in table */
373 extern int vm_page_array_size;		/* number of vm_page_t's */
374 extern long first_page;			/* first physical page number */
375 
376 #define VM_PAGE_TO_PHYS(entry)	\
377 		((entry)->phys_addr)
378 
379 #define PHYS_TO_VM_PAGE(pa)	\
380 		(&vm_page_array[atop(pa) - first_page])
381 
382 /*
383  *	Functions implemented as macros
384  */
385 
386 static __inline void
387 vm_page_flag_set(vm_page_t m, unsigned int bits)
388 {
389 	atomic_set_int(&(m)->flags, bits);
390 }
391 
392 static __inline void
393 vm_page_flag_clear(vm_page_t m, unsigned int bits)
394 {
395 	atomic_clear_int(&(m)->flags, bits);
396 }
397 
398 static __inline void
399 vm_page_busy(vm_page_t m)
400 {
401 	ASSERT_LWKT_TOKEN_HELD(&vm_token);
402 	KASSERT((m->flags & PG_BUSY) == 0,
403 		("vm_page_busy: page already busy!!!"));
404 	vm_page_flag_set(m, PG_BUSY);
405 }
406 
407 /*
408  *	vm_page_flash:
409  *
410  *	wakeup anyone waiting for the page.
411  */
412 
413 static __inline void
414 vm_page_flash(vm_page_t m)
415 {
416 	lwkt_gettoken(&vm_token);
417 	if (m->flags & PG_WANTED) {
418 		vm_page_flag_clear(m, PG_WANTED);
419 		wakeup(m);
420 	}
421 	lwkt_reltoken(&vm_token);
422 }
423 
424 /*
425  * Clear the PG_BUSY flag and wakeup anyone waiting for the page.  This
426  * is typically the last call you make on a page before moving onto
427  * other things.
428  */
429 static __inline void
430 vm_page_wakeup(vm_page_t m)
431 {
432 	KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
433 	vm_page_flag_clear(m, PG_BUSY);
434 	vm_page_flash(m);
435 }
436 
437 /*
438  * These routines manipulate the 'soft busy' count for a page.  A soft busy
439  * is almost like PG_BUSY except that it allows certain compatible operations
440  * to occur on the page while it is busy.  For example, a page undergoing a
441  * write can still be mapped read-only.
442  */
443 static __inline void
444 vm_page_io_start(vm_page_t m)
445 {
446 	atomic_add_char(&(m)->busy, 1);
447 }
448 
449 static __inline void
450 vm_page_io_finish(vm_page_t m)
451 {
452 	atomic_subtract_char(&m->busy, 1);
453 	if (m->busy == 0)
454 		vm_page_flash(m);
455 }
456 
457 
458 #if PAGE_SIZE == 4096
459 #define VM_PAGE_BITS_ALL 0xff
460 #endif
461 
462 /*
463  * Note: the code will always use nominally free pages from the free list
464  * before trying other flag-specified sources.
465  *
466  * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
467  * must be specified.  VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
468  * is also specified.
469  */
470 #define VM_ALLOC_NORMAL		0x01	/* ok to use cache pages */
471 #define VM_ALLOC_SYSTEM		0x02	/* ok to exhaust most of free list */
472 #define VM_ALLOC_INTERRUPT	0x04	/* ok to exhaust entire free list */
473 #define	VM_ALLOC_ZERO		0x08	/* req pre-zero'd memory if avail */
474 #define	VM_ALLOC_QUICK		0x10	/* like NORMAL but do not use cache */
475 #define	VM_ALLOC_RETRY		0x80	/* indefinite block (vm_page_grab()) */
476 
477 void vm_page_hold(vm_page_t);
478 void vm_page_unhold(vm_page_t);
479 void vm_page_activate (vm_page_t);
480 vm_page_t vm_page_alloc (struct vm_object *, vm_pindex_t, int);
481 vm_page_t vm_page_grab (struct vm_object *, vm_pindex_t, int);
482 void vm_page_cache (vm_page_t);
483 int vm_page_try_to_cache (vm_page_t);
484 int vm_page_try_to_free (vm_page_t);
485 void vm_page_dontneed (vm_page_t);
486 void vm_page_deactivate (vm_page_t);
487 void vm_page_insert (vm_page_t, struct vm_object *, vm_pindex_t);
488 vm_page_t vm_page_lookup (struct vm_object *, vm_pindex_t);
489 void vm_page_remove (vm_page_t);
490 void vm_page_rename (vm_page_t, struct vm_object *, vm_pindex_t);
491 void vm_page_startup (void);
492 vm_page_t vm_add_new_page (vm_paddr_t pa);
493 void vm_page_unmanage (vm_page_t);
494 void vm_page_unwire (vm_page_t, int);
495 void vm_page_wire (vm_page_t);
496 void vm_page_unqueue (vm_page_t);
497 void vm_page_unqueue_nowakeup (vm_page_t);
498 void vm_page_set_validclean (vm_page_t, int, int);
499 void vm_page_set_validdirty (vm_page_t, int, int);
500 void vm_page_set_valid (vm_page_t, int, int);
501 void vm_page_set_dirty (vm_page_t, int, int);
502 void vm_page_clear_dirty (vm_page_t, int, int);
503 void vm_page_set_invalid (vm_page_t, int, int);
504 int vm_page_is_valid (vm_page_t, int, int);
505 void vm_page_test_dirty (vm_page_t);
506 int vm_page_bits (int, int);
507 vm_page_t vm_page_list_find(int basequeue, int index, boolean_t prefer_zero);
508 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
509 void vm_page_free_toq(vm_page_t m);
510 vm_page_t vm_page_free_fromq_fast(void);
511 vm_offset_t vm_contig_pg_kmap(int, u_long, vm_map_t, int);
512 void vm_contig_pg_free(int, u_long);
513 void vm_page_event_internal(vm_page_t, vm_page_event_t);
514 void vm_page_dirty(vm_page_t m);
515 void vm_page_register_action(vm_page_action_t action, vm_page_event_t event);
516 void vm_page_unregister_action(vm_page_action_t action);
517 
518 /*
519  * Reduce the protection of a page.  This routine never raises the
520  * protection and therefore can be safely called if the page is already
521  * at VM_PROT_NONE (it will be a NOP effectively ).
522  *
523  * VM_PROT_NONE will remove all user mappings of a page.  This is often
524  * necessary when a page changes state (for example, turns into a copy-on-write
525  * page or needs to be frozen for write I/O) in order to force a fault, or
526  * to force a page's dirty bits to be synchronized and avoid hardware
527  * (modified/accessed) bit update races with pmap changes.
528  *
529  * Since 'prot' is usually a constant, this inline usually winds up optimizing
530  * out the primary conditional.
531  *
532  * WARNING: VM_PROT_NONE can block, but will loop until all mappings have
533  * been cleared.  Callers should be aware that other page related elements
534  * might have changed, however.
535  */
536 static __inline void
537 vm_page_protect(vm_page_t mem, int prot)
538 {
539 	if (prot == VM_PROT_NONE) {
540 		if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
541 			pmap_page_protect(mem, VM_PROT_NONE);
542 			/* PG_WRITEABLE & PG_MAPPED cleared by call */
543 		}
544 	} else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
545 		pmap_page_protect(mem, VM_PROT_READ);
546 		/* PG_WRITEABLE cleared by call */
547 	}
548 }
549 
550 /*
551  * Zero-fill the specified page.  The entire contents of the page will be
552  * zero'd out.
553  */
554 static __inline boolean_t
555 vm_page_zero_fill(vm_page_t m)
556 {
557 	pmap_zero_page(VM_PAGE_TO_PHYS(m));
558 	return (TRUE);
559 }
560 
561 /*
562  * Copy the contents of src_m to dest_m.  The pages must be stable but spl
563  * and other protections depend on context.
564  */
565 static __inline void
566 vm_page_copy(vm_page_t src_m, vm_page_t dest_m)
567 {
568 	pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
569 	dest_m->valid = VM_PAGE_BITS_ALL;
570 	dest_m->dirty = VM_PAGE_BITS_ALL;
571 }
572 
573 /*
574  * Free a page.  The page must be marked BUSY.
575  *
576  * The clearing of PG_ZERO is a temporary safety until the code can be
577  * reviewed to determine that PG_ZERO is being properly cleared on
578  * write faults or maps.  PG_ZERO was previously cleared in
579  * vm_page_alloc().
580  */
581 static __inline void
582 vm_page_free(vm_page_t m)
583 {
584 	vm_page_flag_clear(m, PG_ZERO);
585 	vm_page_free_toq(m);
586 }
587 
588 /*
589  * Free a page to the zerod-pages queue
590  */
591 static __inline void
592 vm_page_free_zero(vm_page_t m)
593 {
594 #ifdef __x86_64__
595 	/* JG DEBUG64 We check if the page is really zeroed. */
596 	char *p = (char *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
597 	int i;
598 
599 	for (i = 0; i < PAGE_SIZE; i++) {
600 		if (p[i] != 0) {
601 			panic("non-zero page in vm_page_free_zero()");
602 		}
603 	}
604 
605 #endif
606 	vm_page_flag_set(m, PG_ZERO);
607 	vm_page_free_toq(m);
608 }
609 
610 /*
611  * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
612  * m->busy is zero.  Returns TRUE if it had to sleep ( including if
613  * it almost had to sleep and made temporary spl*() mods), FALSE
614  * otherwise.
615  *
616  * This routine assumes that interrupts can only remove the busy
617  * status from a page, not set the busy status or change it from
618  * PG_BUSY to m->busy or vise versa (which would create a timing
619  * window).
620  *
621  * Note: as an inline, 'also_m_busy' is usually a constant and well
622  * optimized.
623  */
624 static __inline int
625 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
626 {
627 	if ((m->flags & PG_BUSY) || (also_m_busy && m->busy))  {
628 		lwkt_gettoken(&vm_token);
629 		if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
630 			/*
631 			 * Page is busy. Wait and retry.
632 			 */
633 			vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
634 			tsleep(m, 0, msg, 0);
635 		}
636 		lwkt_reltoken(&vm_token);
637 		return(TRUE);
638 		/* not reached */
639 	}
640 	return(FALSE);
641 }
642 
643 /*
644  * Set page to not be dirty.  Note: does not clear pmap modify bits .
645  */
646 static __inline void
647 vm_page_undirty(vm_page_t m)
648 {
649 	m->dirty = 0;
650 }
651 
652 #endif				/* _KERNEL */
653 #endif				/* !_VM_VM_PAGE_H_ */
654