xref: /freebsd/sys/vm/vm_page.h (revision bfe691b2f75de2224c7ceb304ebcdef2b42d4179)
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from: @(#)vm_page.h	8.2 (Berkeley) 12/13/93
33  *
34  *
35  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36  * All rights reserved.
37  *
38  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39  *
40  * Permission to use, copy, modify and distribute this software and
41  * its documentation is hereby granted, provided that both the copyright
42  * notice and this permission notice appear in all copies of the
43  * software, derivative works or modified versions, and any portions
44  * thereof, and that both notices appear in supporting documentation.
45  *
46  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49  *
50  * Carnegie Mellon requests users of this software to return to
51  *
52  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
53  *  School of Computer Science
54  *  Carnegie Mellon University
55  *  Pittsburgh PA 15213-3890
56  *
57  * any improvements or extensions that they make and grant Carnegie the
58  * rights to redistribute these changes.
59  *
60  * $FreeBSD$
61  */
62 
63 /*
64  *	Resident memory system definitions.
65  */
66 
67 #ifndef	_VM_PAGE_
68 #define	_VM_PAGE_
69 
70 #include <vm/pmap.h>
71 
72 /*
73  *	Management of resident (logical) pages.
74  *
75  *	A small structure is kept for each resident
76  *	page, indexed by page number.  Each structure
77  *	is an element of several lists:
78  *
79  *		A hash table bucket used to quickly
80  *		perform object/offset lookups
81  *
82  *		A list of all pages for a given object,
83  *		so they can be quickly deactivated at
84  *		time of deallocation.
85  *
86  *		An ordered list of pages due for pageout.
87  *
88  *	In addition, the structure contains the object
89  *	and offset to which this page belongs (for pageout),
90  *	and sundry status bits.
91  *
92  *	Fields in this structure are locked either by the lock on the
93  *	object that the page belongs to (O) or by the lock on the page
94  *	queues (P).
95  *
96  *	The 'valid' and 'dirty' fields are distinct.  A page may have dirty
97  *	bits set without having associated valid bits set.  This is used by
98  *	NFS to implement piecemeal writes.
99  */
100 
101 TAILQ_HEAD(pglist, vm_page);
102 
103 struct vm_page {
104 	TAILQ_ENTRY(vm_page) pageq;	/* queue info for FIFO queue or free list (P) */
105 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) 	*/
106 	struct vm_page *left;		/* splay tree link (O)		*/
107 	struct vm_page *right;		/* splay tree link (O)		*/
108 
109 	vm_object_t object;		/* which object am I in (O,P)*/
110 	vm_pindex_t pindex;		/* offset into object (O,P) */
111 	vm_paddr_t phys_addr;		/* physical address of page */
112 	struct md_page md;		/* machine dependant stuff */
113 	u_short	queue;			/* page queue index */
114 	u_short	flags,			/* see below */
115 		pc;			/* page color */
116 	u_short wire_count;		/* wired down maps refs (P) */
117 	u_int cow;			/* page cow mapping count */
118 	short hold_count;		/* page hold count */
119 	u_short oflags;			/* page flags (O) */
120 	u_char	act_count;		/* page usage count */
121 	u_char	busy;			/* page busy count (O) */
122 	/* NOTE that these must support one bit per DEV_BSIZE in a page!!! */
123 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
124 #if PAGE_SIZE == 4096
125 	u_char	valid;			/* map of valid DEV_BSIZE chunks (O) */
126 	u_char	dirty;			/* map of dirty DEV_BSIZE chunks */
127 #elif PAGE_SIZE == 8192
128 	u_short	valid;			/* map of valid DEV_BSIZE chunks (O) */
129 	u_short	dirty;			/* map of dirty DEV_BSIZE chunks */
130 #elif PAGE_SIZE == 16384
131 	u_int valid;			/* map of valid DEV_BSIZE chunks (O) */
132 	u_int dirty;			/* map of dirty DEV_BSIZE chunks */
133 #elif PAGE_SIZE == 32768
134 	u_long valid;			/* map of valid DEV_BSIZE chunks (O) */
135 	u_long dirty;			/* map of dirty DEV_BSIZE chunks */
136 #endif
137 };
138 
139 /*
140  * Page flags stored in oflags:
141  *
142  * Access to these page flags is synchronized by the lock on the object
143  * containing the page (O).
144  */
145 #define	VPO_BUSY	0x0001	/* page is in transit */
146 #define	VPO_WANTED	0x0002	/* someone is waiting for page */
147 #define	VPO_CLEANCHK	0x0100	/* page will be checked for cleaning */
148 #define	VPO_SWAPINPROG	0x0200	/* swap I/O in progress on page */
149 #define	VPO_NOSYNC	0x0400	/* do not collect for syncer */
150 
151 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
152 #if PAGE_SIZE == 32768
153 #ifdef CTASSERT
154 CTASSERT(sizeof(u_long) >= 8);
155 #endif
156 #endif
157 
158 /* PQ_CACHE and PQ_FREE represents a PQ_NUMCOLORS consecutive queue. */
159 #define PQ_NONE		0
160 #define PQ_FREE		1
161 #define PQ_INACTIVE	(page_queue_coloring.inactive)
162 #define PQ_ACTIVE	(page_queue_coloring.active)
163 #define PQ_CACHE	(page_queue_coloring.cache)
164 #define PQ_HOLD		(page_queue_coloring.hold)
165 #define PQ_COUNT	(page_queue_coloring.count)
166 #define PQ_MAXCOLORS	1024
167 #define PQ_MAXCOUNT	(4 + 2 * PQ_MAXCOLORS)
168 #define PQ_NUMCOLORS	(page_queue_coloring.numcolors)
169 #define PQ_PRIME1	(page_queue_coloring.prime1)
170 #define PQ_PRIME2	(page_queue_coloring.prime2)
171 #define PQ_COLORMASK	(page_queue_coloring.colormask)
172 #define PQ_MAXLENGTH	(page_queue_coloring.maxlength)
173 
174 /* Returns the real queue a page is on. */
175 #define VM_PAGE_GETQUEUE(m)	((m)->queue)
176 
177 /* Returns the well known queue a page is on. */
178 #define VM_PAGE_GETKNOWNQUEUE1(m)	((m)->queue - (m)->pc)
179 #define VM_PAGE_GETKNOWNQUEUE2(m)	VM_PAGE_GETQUEUE(m)
180 
181 /* Given the real queue number and a page color return the well know queue. */
182 #define VM_PAGE_RESOLVEQUEUE(m, q)	((q) - (m)->pc)
183 
184 /* Returns true if the page is in the named well known queue. */
185 #define VM_PAGE_INQUEUE1(m, q)	(VM_PAGE_GETKNOWNQUEUE1(m) == (q))
186 #define VM_PAGE_INQUEUE2(m, q)	(VM_PAGE_GETKNOWNQUEUE2(m) == (q))
187 
188 /* Sets the queue a page is on. */
189 #define VM_PAGE_SETQUEUE1(m, q)	(VM_PAGE_GETQUEUE(m) = (q) + (m)->pc)
190 #define VM_PAGE_SETQUEUE2(m, q)	(VM_PAGE_GETQUEUE(m) = (q))
191 
192 struct vpgqueues {
193 	struct pglist pl;
194 	int	*cnt;
195 	int	lcnt;
196 };
197 
198 struct pq_coloring {
199 	int numcolors;
200 	int colormask;
201 	int prime1;
202 	int prime2;
203 	int inactive;
204 	int active;
205 	int cache;
206 	int hold;
207 	int count;
208 	int maxlength;
209 };
210 
211 extern struct vpgqueues vm_page_queues[PQ_MAXCOUNT];
212 extern struct mtx vm_page_queue_free_mtx;
213 extern struct pq_coloring page_queue_coloring;
214 
215 /*
216  * These are the flags defined for vm_page.
217  *
218  * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
219  * 	 not under PV management but otherwise should be treated as a
220  *	 normal page.  Pages not under PV management cannot be paged out
221  *	 via the object/vm_page_t because there is no knowledge of their
222  *	 pte mappings, nor can they be removed from their objects via
223  *	 the object, and such pages are also not on any PQ queue.
224  */
225 #define PG_WINATCFLS	0x0004		/* flush dirty page on inactive q */
226 #define	PG_FICTITIOUS	0x0008		/* physical page doesn't exist (O) */
227 #define	PG_WRITEABLE	0x0010		/* page is mapped writeable */
228 #define	PG_ZERO		0x0040		/* page is zeroed */
229 #define PG_REFERENCED	0x0080		/* page has been referenced */
230 #define PG_UNMANAGED	0x0800		/* No PV management for page */
231 #define PG_MARKER	0x1000		/* special queue marker page */
232 #define	PG_SLAB		0x2000		/* object pointer is actually a slab */
233 
234 /*
235  * Misc constants.
236  */
237 #define ACT_DECLINE		1
238 #define ACT_ADVANCE		3
239 #define ACT_INIT		5
240 #define ACT_MAX			64
241 
242 #ifdef _KERNEL
243 /*
244  * Each pageable resident page falls into one of four lists:
245  *
246  *	free
247  *		Available for allocation now.
248  *
249  * The following are all LRU sorted:
250  *
251  *	cache
252  *		Almost available for allocation. Still in an
253  *		object, but clean and immediately freeable at
254  *		non-interrupt times.
255  *
256  *	inactive
257  *		Low activity, candidates for reclamation.
258  *		This is the list of pages that should be
259  *		paged out next.
260  *
261  *	active
262  *		Pages that are "active" i.e. they have been
263  *		recently referenced.
264  *
265  *	zero
266  *		Pages that are really free and have been pre-zeroed
267  *
268  */
269 
270 extern int vm_page_zero_count;
271 
272 extern vm_page_t vm_page_array;		/* First resident page in table */
273 extern int vm_page_array_size;		/* number of vm_page_t's */
274 extern long first_page;			/* first physical page number */
275 
276 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
277 
278 #define PHYS_TO_VM_PAGE(pa) \
279 		(&vm_page_array[atop(pa) - first_page ])
280 
281 extern struct mtx vm_page_queue_mtx;
282 #define vm_page_lock_queues()   mtx_lock(&vm_page_queue_mtx)
283 #define vm_page_unlock_queues() mtx_unlock(&vm_page_queue_mtx)
284 
285 #if PAGE_SIZE == 4096
286 #define VM_PAGE_BITS_ALL 0xffu
287 #elif PAGE_SIZE == 8192
288 #define VM_PAGE_BITS_ALL 0xffffu
289 #elif PAGE_SIZE == 16384
290 #define VM_PAGE_BITS_ALL 0xffffffffu
291 #elif PAGE_SIZE == 32768
292 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
293 #endif
294 
295 /* page allocation classes: */
296 #define VM_ALLOC_NORMAL		0
297 #define VM_ALLOC_INTERRUPT	1
298 #define VM_ALLOC_SYSTEM		2
299 #define	VM_ALLOC_CLASS_MASK	3
300 /* page allocation flags: */
301 #define	VM_ALLOC_WIRED		0x0020	/* non pageable */
302 #define	VM_ALLOC_ZERO		0x0040	/* Try to obtain a zeroed page */
303 #define	VM_ALLOC_RETRY		0x0080	/* vm_page_grab() only */
304 #define	VM_ALLOC_NOOBJ		0x0100	/* No associated object */
305 #define	VM_ALLOC_NOBUSY		0x0200	/* Do not busy the page */
306 
307 void vm_page_flag_set(vm_page_t m, unsigned short bits);
308 void vm_page_flag_clear(vm_page_t m, unsigned short bits);
309 void vm_page_busy(vm_page_t m);
310 void vm_page_flash(vm_page_t m);
311 void vm_page_io_start(vm_page_t m);
312 void vm_page_io_finish(vm_page_t m);
313 void vm_page_hold(vm_page_t mem);
314 void vm_page_unhold(vm_page_t mem);
315 void vm_page_free(vm_page_t m);
316 void vm_page_free_zero(vm_page_t m);
317 void vm_page_dirty(vm_page_t m);
318 void vm_page_wakeup(vm_page_t m);
319 
320 void vm_pageq_init(void);
321 void vm_pageq_add_new_page(vm_paddr_t pa);
322 void vm_pageq_enqueue(int queue, vm_page_t m);
323 void vm_pageq_remove_nowakeup(vm_page_t m);
324 void vm_pageq_remove(vm_page_t m);
325 vm_page_t vm_pageq_find(int basequeue, int index, boolean_t prefer_zero);
326 void vm_pageq_requeue(vm_page_t m);
327 
328 void vm_page_activate (vm_page_t);
329 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
330 vm_page_t vm_page_alloc_contig (vm_pindex_t, vm_paddr_t, vm_paddr_t,
331 	    vm_offset_t, vm_offset_t);
332 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
333 void vm_page_cache (register vm_page_t);
334 int vm_page_try_to_cache (vm_page_t);
335 int vm_page_try_to_free (vm_page_t);
336 void vm_page_dontneed (register vm_page_t);
337 void vm_page_deactivate (vm_page_t);
338 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
339 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
340 void vm_page_remove (vm_page_t);
341 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
342 vm_page_t vm_page_select_cache(int);
343 void vm_page_sleep(vm_page_t m, const char *msg);
344 vm_page_t vm_page_splay(vm_pindex_t, vm_page_t);
345 vm_offset_t vm_page_startup(vm_offset_t vaddr);
346 void vm_page_unwire (vm_page_t, int);
347 void vm_page_wire (vm_page_t);
348 void vm_page_set_validclean (vm_page_t, int, int);
349 void vm_page_clear_dirty (vm_page_t, int, int);
350 void vm_page_set_invalid (vm_page_t, int, int);
351 int vm_page_is_valid (vm_page_t, int, int);
352 void vm_page_test_dirty (vm_page_t);
353 int vm_page_bits (int, int);
354 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
355 void vm_page_free_toq(vm_page_t m);
356 void vm_page_zero_idle_wakeup(void);
357 void vm_page_cowfault (vm_page_t);
358 void vm_page_cowsetup (vm_page_t);
359 void vm_page_cowclear (vm_page_t);
360 
361 /*
362  *	vm_page_sleep_if_busy:
363  *
364  *	Sleep and release the page queues lock if VPO_BUSY is set or,
365  *	if also_m_busy is TRUE, busy is non-zero.  Returns TRUE if the
366  *	thread slept and the page queues lock was released.
367  *	Otherwise, retains the page queues lock and returns FALSE.
368  *
369  *	The object containing the given page must be locked.
370  */
371 static __inline int
372 vm_page_sleep_if_busy(vm_page_t m, int also_m_busy, const char *msg)
373 {
374 
375 	if ((m->oflags & VPO_BUSY) || (also_m_busy && m->busy)) {
376 		vm_page_sleep(m, msg);
377 		return (TRUE);
378 	}
379 	return (FALSE);
380 }
381 
382 /*
383  *	vm_page_undirty:
384  *
385  *	Set page to not be dirty.  Note: does not clear pmap modify bits
386  */
387 static __inline void
388 vm_page_undirty(vm_page_t m)
389 {
390 	m->dirty = 0;
391 }
392 
393 #endif				/* _KERNEL */
394 #endif				/* !_VM_PAGE_ */
395