xref: /dflybsd-src/sys/vm/vm_page.h (revision d83c779ab2c938232fa7b53777cd18cc9c4fc8e4)
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 	vm_page_event_t		event;
115 	void			(*func)(struct vm_page *,
116 					struct vm_page_action *);
117 	void			*data;
118 };
119 
120 typedef struct vm_page_action *vm_page_action_t;
121 
122 /*
123  *	Management of resident (logical) pages.
124  *
125  *	A small structure is kept for each resident
126  *	page, indexed by page number.  Each structure
127  *	is an element of several lists:
128  *
129  *		A hash table bucket used to quickly
130  *		perform object/offset lookups
131  *
132  *		A list of all pages for a given object,
133  *		so they can be quickly deactivated at
134  *		time of deallocation.
135  *
136  *		An ordered list of pages due for pageout.
137  *
138  *	In addition, the structure contains the object
139  *	and offset to which this page belongs (for pageout),
140  *	and sundry status bits.
141  *
142  *	Fields in this structure are locked either by the lock on the
143  *	object that the page belongs to (O) or by the lock on the page
144  *	queues (P).
145  *
146  *	The 'valid' and 'dirty' fields are distinct.  A page may have dirty
147  *	bits set without having associated valid bits set.  This is used by
148  *	NFS to implement piecemeal writes.
149  */
150 
151 TAILQ_HEAD(pglist, vm_page);
152 
153 struct vm_object;
154 
155 int rb_vm_page_compare(struct vm_page *, struct vm_page *);
156 
157 struct vm_page_rb_tree;
158 RB_PROTOTYPE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare, vm_pindex_t);
159 
160 struct vm_page {
161 	TAILQ_ENTRY(vm_page) pageq;	/* vm_page_queues[] list (P)	*/
162 	RB_ENTRY(vm_page) rb_entry;	/* Red-Black tree based at object */
163 
164 	struct vm_object *object;	/* which object am I in (O,P)*/
165 	vm_pindex_t pindex;		/* offset into object (O,P) */
166 	vm_paddr_t phys_addr;		/* physical address of page */
167 	struct md_page md;		/* machine dependant stuff */
168 	u_short	queue;			/* page queue index */
169 	u_short	flags;			/* see below */
170 	u_short	pc;			/* page color */
171 	u_char	act_count;		/* page usage count */
172 	u_char	busy;			/* page busy count */
173 	u_int	wire_count;		/* wired down maps refs (P) */
174 	int 	hold_count;		/* page hold count */
175 
176 	/*
177 	 * NOTE that these must support one bit per DEV_BSIZE in a page!!!
178 	 * so, on normal X86 kernels, they must be at least 8 bits wide.
179 	 */
180 #if PAGE_SIZE == 4096
181 	u_char	valid;			/* map of valid DEV_BSIZE chunks */
182 	u_char	dirty;			/* map of dirty DEV_BSIZE chunks */
183 #elif PAGE_SIZE == 8192
184 	u_short	valid;			/* map of valid DEV_BSIZE chunks */
185 	u_short	dirty;			/* map of dirty DEV_BSIZE chunks */
186 #endif
187 	LIST_HEAD(,vm_page_action) action_list;
188 };
189 
190 #ifndef __VM_PAGE_T_DEFINED__
191 #define __VM_PAGE_T_DEFINED__
192 typedef struct vm_page *vm_page_t;
193 #endif
194 
195 /*
196  * note: currently use SWAPBLK_NONE as an absolute value rather then
197  * a flag bit.
198  */
199 #define SWAPBLK_MASK	((daddr_t)((u_daddr_t)-1 >> 1))		/* mask */
200 #define SWAPBLK_NONE	((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */
201 
202 /*
203  * Page coloring parameters.  We default to a middle of the road optimization.
204  * Larger selections would not really hurt us but if a machine does not have
205  * a lot of memory it could cause vm_page_alloc() to eat more cpu cycles
206  * looking for free pages.
207  *
208  * Page coloring cannot be disabled.  Modules do not have access to most PQ
209  * constants because they can change between builds.
210  */
211 #if defined(_KERNEL) && !defined(KLD_MODULE)
212 
213 #if !defined(PQ_CACHESIZE)
214 #define PQ_CACHESIZE 256	/* max is 1024 (MB) */
215 #endif
216 
217 #if PQ_CACHESIZE >= 1024
218 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
219 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
220 #define PQ_L2_SIZE 256	/* A number of colors opt for 1M cache */
221 
222 #elif PQ_CACHESIZE >= 512
223 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
224 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
225 #define PQ_L2_SIZE 128	/* A number of colors opt for 512K cache */
226 
227 #elif PQ_CACHESIZE >= 256
228 #define PQ_PRIME1 13	/* Prime number somewhat less than PQ_HASH_SIZE */
229 #define PQ_PRIME2 7	/* Prime number somewhat less than PQ_HASH_SIZE */
230 #define PQ_L2_SIZE 64	/* A number of colors opt for 256K cache */
231 
232 #elif PQ_CACHESIZE >= 128
233 #define PQ_PRIME1 9	/* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
234 #define PQ_PRIME2 5	/* Prime number somewhat less than PQ_HASH_SIZE */
235 #define PQ_L2_SIZE 32	/* A number of colors opt for 128k cache */
236 
237 #else
238 #define PQ_PRIME1 5	/* Prime number somewhat less than PQ_HASH_SIZE */
239 #define PQ_PRIME2 3	/* Prime number somewhat less than PQ_HASH_SIZE */
240 #define PQ_L2_SIZE 16	/* A reasonable number of colors (opt for 64K cache) */
241 
242 #endif
243 
244 #define PQ_L2_MASK	(PQ_L2_SIZE - 1)
245 
246 #endif /* KERNEL && !KLD_MODULE */
247 
248 /*
249  *
250  * The queue array is always based on PQ_MAXL2_SIZE regardless of the actual
251  * cache size chosen in order to present a uniform interface for modules.
252  */
253 #define PQ_MAXL2_SIZE	256	/* fixed maximum (in pages) / module compat */
254 
255 #if PQ_L2_SIZE > PQ_MAXL2_SIZE
256 #error "Illegal PQ_L2_SIZE"
257 #endif
258 
259 #define PQ_NONE		0
260 #define PQ_FREE		1
261 #define PQ_INACTIVE	(1 + 1*PQ_MAXL2_SIZE)
262 #define PQ_ACTIVE	(2 + 1*PQ_MAXL2_SIZE)
263 #define PQ_CACHE	(3 + 1*PQ_MAXL2_SIZE)
264 #define PQ_HOLD		(3 + 2*PQ_MAXL2_SIZE)
265 #define PQ_COUNT	(4 + 2*PQ_MAXL2_SIZE)
266 
267 /*
268  * Scan support
269  */
270 struct vm_map;
271 
272 struct rb_vm_page_scan_info {
273 	vm_pindex_t	start_pindex;
274 	vm_pindex_t	end_pindex;
275 	int		limit;
276 	int		desired;
277 	int		error;
278 	int		pagerflags;
279 	vm_offset_t	addr;
280 	vm_pindex_t	backing_offset_index;
281 	struct vm_object *object;
282 	struct vm_object *backing_object;
283 	struct vm_page	*mpte;
284 	struct pmap	*pmap;
285 	struct vm_map	*map;
286 };
287 
288 int rb_vm_page_scancmp(struct vm_page *, void *);
289 
290 struct vpgqueues {
291 	struct pglist pl;
292 	int	*cnt;
293 	int	lcnt;
294 	int	flipflop;	/* probably not the best place */
295 };
296 
297 extern struct vpgqueues vm_page_queues[PQ_COUNT];
298 
299 /*
300  * These are the flags defined for vm_page.
301  *
302  *  PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
303  *  not under PV management but otherwise should be treated as a
304  *  normal page.  Pages not under PV management cannot be paged out
305  *  via the object/vm_page_t because there is no knowledge of their
306  *  pte mappings, nor can they be removed from their objects via
307  *  the object, and such pages are also not on any PQ queue.  The
308  *  PG_MAPPED and PG_WRITEABLE flags are not applicable.
309  *
310  *  PG_MAPPED only applies to managed pages, indicating whether the page
311  *  is mapped onto one or more pmaps.  A page might still be mapped to
312  *  special pmaps in an unmanaged fashion, for example when mapped into a
313  *  buffer cache buffer, without setting PG_MAPPED.
314  *
315  *  PG_WRITEABLE indicates that there may be a writeable managed pmap entry
316  *  somewhere, and that the page can be dirtied by hardware at any time
317  *  and may have to be tested for that.  The modified bit in unmanaged
318  *  mappings or in the special clean map is not tested.
319  *
320  *  PG_SWAPPED indicates that the page is backed by a swap block.  Any
321  *  VM object type other than OBJT_DEFAULT can have swap-backed pages now.
322  */
323 #define	PG_BUSY		0x0001		/* page is in transit (O) */
324 #define	PG_WANTED	0x0002		/* someone is waiting for page (O) */
325 #define PG_WINATCFLS	0x0004		/* flush dirty page on inactive q */
326 #define	PG_FICTITIOUS	0x0008		/* physical page doesn't exist (O) */
327 #define	PG_WRITEABLE	0x0010		/* page is writeable */
328 #define PG_MAPPED	0x0020		/* page is mapped (managed) */
329 #define	PG_ZERO		0x0040		/* page is zeroed */
330 #define PG_REFERENCED	0x0080		/* page has been referenced */
331 #define PG_CLEANCHK	0x0100		/* page will be checked for cleaning */
332 #define PG_SWAPINPROG	0x0200		/* swap I/O in progress on page	     */
333 #define PG_NOSYNC	0x0400		/* do not collect for syncer */
334 #define PG_UNMANAGED	0x0800		/* No PV management for page */
335 #define PG_MARKER	0x1000		/* special queue marker page */
336 #define PG_RAM		0x2000		/* read ahead mark */
337 #define PG_SWAPPED	0x4000		/* backed by swap */
338 #define PG_NOTMETA	0x8000		/* do not back with swap */
339 	/* u_short, only 16 flag bits */
340 
341 /*
342  * Misc constants.
343  */
344 
345 #define ACT_DECLINE		1
346 #define ACT_ADVANCE		3
347 #define ACT_INIT		5
348 #define ACT_MAX			64
349 
350 #ifdef _KERNEL
351 /*
352  * Each pageable resident page falls into one of four lists:
353  *
354  *	free
355  *		Available for allocation now.
356  *
357  * The following are all LRU sorted:
358  *
359  *	cache
360  *		Almost available for allocation. Still in an
361  *		object, but clean and immediately freeable at
362  *		non-interrupt times.
363  *
364  *	inactive
365  *		Low activity, candidates for reclamation.
366  *		This is the list of pages that should be
367  *		paged out next.
368  *
369  *	active
370  *		Pages that are "active" i.e. they have been
371  *		recently referenced.
372  *
373  *	zero
374  *		Pages that are really free and have been pre-zeroed
375  *
376  */
377 
378 extern int vm_page_zero_count;
379 extern struct vm_page *vm_page_array;	/* First resident page in table */
380 extern int vm_page_array_size;		/* number of vm_page_t's */
381 extern long first_page;			/* first physical page number */
382 
383 #define VM_PAGE_TO_PHYS(entry)	\
384 		((entry)->phys_addr)
385 
386 #define PHYS_TO_VM_PAGE(pa)	\
387 		(&vm_page_array[atop(pa) - first_page])
388 
389 /*
390  *	Functions implemented as macros
391  */
392 
393 static __inline void
394 vm_page_flag_set(vm_page_t m, unsigned int bits)
395 {
396 	atomic_set_short(&(m)->flags, bits);
397 }
398 
399 static __inline void
400 vm_page_flag_clear(vm_page_t m, unsigned int bits)
401 {
402 	atomic_clear_short(&(m)->flags, bits);
403 }
404 
405 static __inline void
406 vm_page_busy(vm_page_t m)
407 {
408 	KASSERT((m->flags & PG_BUSY) == 0,
409 		("vm_page_busy: page already busy!!!"));
410 	vm_page_flag_set(m, PG_BUSY);
411 }
412 
413 /*
414  *	vm_page_flash:
415  *
416  *	wakeup anyone waiting for the page.
417  */
418 
419 static __inline void
420 vm_page_flash(vm_page_t m)
421 {
422 	if (m->flags & PG_WANTED) {
423 		vm_page_flag_clear(m, PG_WANTED);
424 		wakeup(m);
425 	}
426 }
427 
428 /*
429  * Clear the PG_BUSY flag and wakeup anyone waiting for the page.  This
430  * is typically the last call you make on a page before moving onto
431  * other things.
432  */
433 static __inline void
434 vm_page_wakeup(vm_page_t m)
435 {
436 	KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
437 	vm_page_flag_clear(m, PG_BUSY);
438 	vm_page_flash(m);
439 }
440 
441 /*
442  * These routines manipulate the 'soft busy' count for a page.  A soft busy
443  * is almost like PG_BUSY except that it allows certain compatible operations
444  * to occur on the page while it is busy.  For example, a page undergoing a
445  * write can still be mapped read-only.
446  */
447 static __inline void
448 vm_page_io_start(vm_page_t m)
449 {
450 	atomic_add_char(&(m)->busy, 1);
451 }
452 
453 static __inline void
454 vm_page_io_finish(vm_page_t m)
455 {
456 	atomic_subtract_char(&m->busy, 1);
457 	if (m->busy == 0)
458 		vm_page_flash(m);
459 }
460 
461 
462 #if PAGE_SIZE == 4096
463 #define VM_PAGE_BITS_ALL 0xff
464 #endif
465 
466 #if PAGE_SIZE == 8192
467 #define VM_PAGE_BITS_ALL 0xffff
468 #endif
469 
470 /*
471  * Note: the code will always use nominally free pages from the free list
472  * before trying other flag-specified sources.
473  *
474  * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
475  * must be specified.  VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
476  * is also specified.
477  */
478 #define VM_ALLOC_NORMAL		0x01	/* ok to use cache pages */
479 #define VM_ALLOC_SYSTEM		0x02	/* ok to exhaust most of free list */
480 #define VM_ALLOC_INTERRUPT	0x04	/* ok to exhaust entire free list */
481 #define	VM_ALLOC_ZERO		0x08	/* req pre-zero'd memory if avail */
482 #define	VM_ALLOC_QUICK		0x10	/* like NORMAL but do not use cache */
483 #define	VM_ALLOC_RETRY		0x80	/* indefinite block (vm_page_grab()) */
484 
485 void vm_page_unhold(vm_page_t mem);
486 void vm_page_activate (vm_page_t);
487 vm_page_t vm_page_alloc (struct vm_object *, vm_pindex_t, int);
488 vm_page_t vm_page_grab (struct vm_object *, vm_pindex_t, int);
489 void vm_page_cache (vm_page_t);
490 int vm_page_try_to_cache (vm_page_t);
491 int vm_page_try_to_free (vm_page_t);
492 void vm_page_dontneed (vm_page_t);
493 void vm_page_deactivate (vm_page_t);
494 void vm_page_insert (vm_page_t, struct vm_object *, vm_pindex_t);
495 vm_page_t vm_page_lookup (struct vm_object *, vm_pindex_t);
496 void vm_page_remove (vm_page_t);
497 void vm_page_rename (vm_page_t, struct vm_object *, vm_pindex_t);
498 vm_offset_t vm_page_startup (vm_offset_t);
499 vm_page_t vm_add_new_page (vm_paddr_t pa);
500 void vm_page_unmanage (vm_page_t);
501 void vm_page_unwire (vm_page_t, int);
502 void vm_page_wire (vm_page_t);
503 void vm_page_unqueue (vm_page_t);
504 void vm_page_unqueue_nowakeup (vm_page_t);
505 void vm_page_set_validclean (vm_page_t, int, int);
506 void vm_page_set_validdirty (vm_page_t, int, int);
507 void vm_page_set_valid (vm_page_t, int, int);
508 void vm_page_set_dirty (vm_page_t, int, int);
509 void vm_page_clear_dirty (vm_page_t, int, int);
510 void vm_page_set_invalid (vm_page_t, int, int);
511 int vm_page_is_valid (vm_page_t, int, int);
512 void vm_page_test_dirty (vm_page_t);
513 int vm_page_bits (int, int);
514 vm_page_t vm_page_list_find(int basequeue, int index, boolean_t prefer_zero);
515 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
516 void vm_page_free_toq(vm_page_t m);
517 vm_offset_t vm_contig_pg_kmap(int, u_long, vm_map_t, int);
518 void vm_contig_pg_free(int, u_long);
519 void vm_page_event_internal(vm_page_t, vm_page_event_t);
520 void vm_page_dirty(vm_page_t m);
521 
522 /*
523  * Holding a page keeps it from being reused.  Other parts of the system
524  * can still disassociate the page from its current object and free it, or
525  * perform read or write I/O on it and/or otherwise manipulate the page,
526  * but if the page is held the VM system will leave the page and its data
527  * intact and not reuse the page for other purposes until the last hold
528  * reference is released.  (see vm_page_wire() if you want to prevent the
529  * page from being disassociated from its object too).
530  *
531  * This routine must be called while at splvm() or better.
532  *
533  * The caller must still validate the contents of the page and, if necessary,
534  * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
535  * before manipulating the page.
536  */
537 static __inline void
538 vm_page_hold(vm_page_t mem)
539 {
540 	mem->hold_count++;
541 }
542 
543 /*
544  * Reduce the protection of a page.  This routine never raises the
545  * protection and therefore can be safely called if the page is already
546  * at VM_PROT_NONE (it will be a NOP effectively ).
547  *
548  * VM_PROT_NONE will remove all user mappings of a page.  This is often
549  * necessary when a page changes state (for example, turns into a copy-on-write
550  * page or needs to be frozen for write I/O) in order to force a fault, or
551  * to force a page's dirty bits to be synchronized and avoid hardware
552  * (modified/accessed) bit update races with pmap changes.
553  *
554  * Since 'prot' is usually a constant, this inline usually winds up optimizing
555  * out the primary conditional.
556  *
557  * WARNING: VM_PROT_NONE can block, but will loop until all mappings have
558  * been cleared.  Callers should be aware that other page related elements
559  * might have changed, however.
560  */
561 static __inline void
562 vm_page_protect(vm_page_t mem, int prot)
563 {
564 	if (prot == VM_PROT_NONE) {
565 		if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
566 			pmap_page_protect(mem, VM_PROT_NONE);
567 			/* PG_WRITEABLE & PG_MAPPED cleared by call */
568 		}
569 	} else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
570 		pmap_page_protect(mem, VM_PROT_READ);
571 		/* PG_WRITEABLE cleared by call */
572 	}
573 }
574 
575 /*
576  * Zero-fill the specified page.  The entire contents of the page will be
577  * zero'd out.
578  */
579 static __inline boolean_t
580 vm_page_zero_fill(vm_page_t m)
581 {
582 	pmap_zero_page(VM_PAGE_TO_PHYS(m));
583 	return (TRUE);
584 }
585 
586 /*
587  * Copy the contents of src_m to dest_m.  The pages must be stable but spl
588  * and other protections depend on context.
589  */
590 static __inline void
591 vm_page_copy(vm_page_t src_m, vm_page_t dest_m)
592 {
593 	pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
594 	dest_m->valid = VM_PAGE_BITS_ALL;
595 	dest_m->dirty = VM_PAGE_BITS_ALL;
596 }
597 
598 /*
599  * Free a page.  The page must be marked BUSY.
600  *
601  * The clearing of PG_ZERO is a temporary safety until the code can be
602  * reviewed to determine that PG_ZERO is being properly cleared on
603  * write faults or maps.  PG_ZERO was previously cleared in
604  * vm_page_alloc().
605  */
606 static __inline void
607 vm_page_free(vm_page_t m)
608 {
609 	vm_page_flag_clear(m, PG_ZERO);
610 	vm_page_free_toq(m);
611 }
612 
613 /*
614  * Free a page to the zerod-pages queue
615  */
616 static __inline void
617 vm_page_free_zero(vm_page_t m)
618 {
619 #ifdef __x86_64__
620 	/* JG DEBUG64 We check if the page is really zeroed. */
621 	char *p = (char *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
622 	int i;
623 
624 	for (i = 0; i < PAGE_SIZE; i++) {
625 		if (p[i] != 0) {
626 			panic("non-zero page in vm_page_free_zero()");
627 		}
628 	}
629 
630 #endif
631 	vm_page_flag_set(m, PG_ZERO);
632 	vm_page_free_toq(m);
633 }
634 
635 /*
636  * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
637  * m->busy is zero.  Returns TRUE if it had to sleep ( including if
638  * it almost had to sleep and made temporary spl*() mods), FALSE
639  * otherwise.
640  *
641  * This routine assumes that interrupts can only remove the busy
642  * status from a page, not set the busy status or change it from
643  * PG_BUSY to m->busy or vise versa (which would create a timing
644  * window).
645  *
646  * Note: as an inline, 'also_m_busy' is usually a constant and well
647  * optimized.
648  */
649 static __inline int
650 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
651 {
652 	if ((m->flags & PG_BUSY) || (also_m_busy && m->busy))  {
653 		crit_enter();
654 		if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
655 			/*
656 			 * Page is busy. Wait and retry.
657 			 */
658 			vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
659 			tsleep(m, 0, msg, 0);
660 		}
661 		crit_exit();
662 		return(TRUE);
663 		/* not reached */
664 	}
665 	return(FALSE);
666 }
667 
668 /*
669  * Set page to not be dirty.  Note: does not clear pmap modify bits .
670  */
671 static __inline void
672 vm_page_undirty(vm_page_t m)
673 {
674 	m->dirty = 0;
675 }
676 
677 #endif				/* _KERNEL */
678 #endif				/* !_VM_VM_PAGE_H_ */
679