xref: /freebsd/sys/vm/vm_page.h (revision 43764a7ffa9ad6eba3275410bc2397d3d398f75f)
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), its corresponding page lock (P),
94  *	or by the lock on the page queues (Q).
95  *
96  */
97 
98 TAILQ_HEAD(pglist, vm_page);
99 
100 struct vm_page {
101 	TAILQ_ENTRY(vm_page) pageq;	/* queue info for FIFO queue or free list (Q) */
102 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) 	*/
103 	struct vm_page *left;		/* splay tree link (O)		*/
104 	struct vm_page *right;		/* splay tree link (O)		*/
105 
106 	vm_object_t object;		/* which object am I in (O,P)*/
107 	vm_pindex_t pindex;		/* offset into object (O,P) */
108 	vm_paddr_t phys_addr;		/* physical address of page */
109 	struct md_page md;		/* machine dependant stuff */
110 	uint8_t	queue;			/* page queue index (P,Q) */
111 	int8_t segind;
112 	u_short	flags;			/* see below */
113 	uint8_t	order;			/* index of the buddy queue */
114 	uint8_t pool;
115 	u_short cow;			/* page cow mapping count (P) */
116 	u_int wire_count;		/* wired down maps refs (P) */
117 	short hold_count;		/* page hold count (P) */
118 	u_short oflags;			/* page flags (O) */
119 	u_char	act_count;		/* page usage count (O) */
120 	u_char	busy;			/* page busy count (O) */
121 	/* NOTE that these must support one bit per DEV_BSIZE in a page!!! */
122 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
123 #if PAGE_SIZE == 4096
124 	u_char	valid;			/* map of valid DEV_BSIZE chunks (O) */
125 	u_char	dirty;			/* map of dirty DEV_BSIZE chunks (O) */
126 #elif PAGE_SIZE == 8192
127 	u_short	valid;			/* map of valid DEV_BSIZE chunks (O) */
128 	u_short	dirty;			/* map of dirty DEV_BSIZE chunks (O) */
129 #elif PAGE_SIZE == 16384
130 	u_int valid;			/* map of valid DEV_BSIZE chunks (O) */
131 	u_int dirty;			/* map of dirty DEV_BSIZE chunks (O) */
132 #elif PAGE_SIZE == 32768
133 	u_long valid;			/* map of valid DEV_BSIZE chunks (O) */
134 	u_long dirty;			/* map of dirty DEV_BSIZE chunks (O) */
135 #endif
136 };
137 
138 /*
139  * Page flags stored in oflags:
140  *
141  * Access to these page flags is synchronized by the lock on the object
142  * containing the page (O).
143  */
144 #define	VPO_BUSY	0x0001	/* page is in transit */
145 #define	VPO_WANTED	0x0002	/* someone is waiting for page */
146 #define	VPO_SWAPINPROG	0x0200	/* swap I/O in progress on page */
147 #define	VPO_NOSYNC	0x0400	/* do not collect for syncer */
148 
149 #define	PQ_NONE		255
150 #define	PQ_INACTIVE	0
151 #define	PQ_ACTIVE	1
152 #define	PQ_HOLD		2
153 #define	PQ_COUNT	3
154 
155 struct vpgqueues {
156 	struct pglist pl;
157 	int	*cnt;
158 };
159 
160 extern struct vpgqueues vm_page_queues[PQ_COUNT];
161 
162 struct vpglocks {
163 	struct mtx	data;
164 	char		pad[CACHE_LINE_SIZE - sizeof(struct mtx)];
165 } __aligned(CACHE_LINE_SIZE);
166 
167 extern struct vpglocks vm_page_queue_free_lock;
168 extern struct vpglocks pa_lock[];
169 
170 #if defined(__arm__)
171 #define	PDRSHIFT	PDR_SHIFT
172 #elif !defined(PDRSHIFT)
173 #define PDRSHIFT	21
174 #endif
175 
176 #define	pa_index(pa)	((pa) >> PDRSHIFT)
177 #define	PA_LOCKPTR(pa)	&pa_lock[pa_index((pa)) % PA_LOCK_COUNT].data
178 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
179 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
180 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
181 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
182 #define	PA_UNLOCK_COND(pa) 			\
183 	do {		   			\
184 		if ((pa) != 0) {		\
185 			PA_UNLOCK((pa));	\
186 			(pa) = 0;		\
187 		}				\
188 	} while (0)
189 
190 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
191 
192 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
193 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
194 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
195 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
196 #define	vm_page_lock_assert(m, a)	mtx_assert(vm_page_lockptr((m)), (a))
197 
198 #define	vm_page_queue_free_mtx	vm_page_queue_free_lock.data
199 /*
200  * These are the flags defined for vm_page.
201  *
202  * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
203  * 	 not under PV management but otherwise should be treated as a
204  *	 normal page.  Pages not under PV management cannot be paged out
205  *	 via the object/vm_page_t because there is no knowledge of their
206  *	 pte mappings, nor can they be removed from their objects via
207  *	 the object, and such pages are also not on any PQ queue.
208  *
209  * PG_REFERENCED may be cleared only if the object containing the page is
210  * locked.
211  *
212  * PG_WRITEABLE is set exclusively on managed pages by pmap_enter().  When it
213  * does so, the page must be VPO_BUSY.
214  */
215 #define	PG_CACHED	0x0001		/* page is cached */
216 #define	PG_FREE		0x0002		/* page is free */
217 #define PG_WINATCFLS	0x0004		/* flush dirty page on inactive q */
218 #define	PG_FICTITIOUS	0x0008		/* physical page doesn't exist (O) */
219 #define	PG_WRITEABLE	0x0010		/* page is mapped writeable */
220 #define	PG_ZERO		0x0040		/* page is zeroed */
221 #define PG_REFERENCED	0x0080		/* page has been referenced */
222 #define PG_UNMANAGED	0x0800		/* No PV management for page */
223 #define PG_MARKER	0x1000		/* special queue marker page */
224 #define	PG_SLAB		0x2000		/* object pointer is actually a slab */
225 
226 /*
227  * Misc constants.
228  */
229 #define ACT_DECLINE		1
230 #define ACT_ADVANCE		3
231 #define ACT_INIT		5
232 #define ACT_MAX			64
233 
234 #ifdef _KERNEL
235 
236 #include <vm/vm_param.h>
237 
238 /*
239  * Each pageable resident page falls into one of five lists:
240  *
241  *	free
242  *		Available for allocation now.
243  *
244  *	cache
245  *		Almost available for allocation. Still associated with
246  *		an object, but clean and immediately freeable.
247  *
248  *	hold
249  *		Will become free after a pending I/O operation
250  *		completes.
251  *
252  * The following lists are LRU sorted:
253  *
254  *	inactive
255  *		Low activity, candidates for reclamation.
256  *		This is the list of pages that should be
257  *		paged out next.
258  *
259  *	active
260  *		Pages that are "active" i.e. they have been
261  *		recently referenced.
262  *
263  */
264 
265 struct vnode;
266 extern int vm_page_zero_count;
267 
268 extern vm_page_t vm_page_array;		/* First resident page in table */
269 extern int vm_page_array_size;		/* number of vm_page_t's */
270 extern long first_page;			/* first physical page number */
271 
272 #define	VM_PAGE_IS_FREE(m)	(((m)->flags & PG_FREE) != 0)
273 
274 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
275 
276 vm_page_t vm_phys_paddr_to_vm_page(vm_paddr_t pa);
277 
278 static __inline vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
279 
280 static __inline vm_page_t
281 PHYS_TO_VM_PAGE(vm_paddr_t pa)
282 {
283 #ifdef VM_PHYSSEG_SPARSE
284 	return (vm_phys_paddr_to_vm_page(pa));
285 #elif defined(VM_PHYSSEG_DENSE)
286 	return (&vm_page_array[atop(pa) - first_page]);
287 #else
288 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
289 #endif
290 }
291 
292 extern struct vpglocks vm_page_queue_lock;
293 
294 #define	vm_page_queue_mtx	vm_page_queue_lock.data
295 #define vm_page_lock_queues()   mtx_lock(&vm_page_queue_mtx)
296 #define vm_page_unlock_queues() mtx_unlock(&vm_page_queue_mtx)
297 
298 #if PAGE_SIZE == 4096
299 #define VM_PAGE_BITS_ALL 0xffu
300 #elif PAGE_SIZE == 8192
301 #define VM_PAGE_BITS_ALL 0xffffu
302 #elif PAGE_SIZE == 16384
303 #define VM_PAGE_BITS_ALL 0xffffffffu
304 #elif PAGE_SIZE == 32768
305 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
306 #endif
307 
308 /* page allocation classes: */
309 #define VM_ALLOC_NORMAL		0
310 #define VM_ALLOC_INTERRUPT	1
311 #define VM_ALLOC_SYSTEM		2
312 #define	VM_ALLOC_CLASS_MASK	3
313 /* page allocation flags: */
314 #define	VM_ALLOC_WIRED		0x0020	/* non pageable */
315 #define	VM_ALLOC_ZERO		0x0040	/* Try to obtain a zeroed page */
316 #define	VM_ALLOC_RETRY		0x0080	/* Mandatory with vm_page_grab() */
317 #define	VM_ALLOC_NOOBJ		0x0100	/* No associated object */
318 #define	VM_ALLOC_NOBUSY		0x0200	/* Do not busy the page */
319 #define	VM_ALLOC_IFCACHED	0x0400	/* Fail if the page is not cached */
320 #define	VM_ALLOC_IFNOTCACHED	0x0800	/* Fail if the page is cached */
321 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* vm_page_grab() only */
322 
323 #define	VM_ALLOC_COUNT_SHIFT	16
324 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
325 
326 void vm_page_flag_set(vm_page_t m, unsigned short bits);
327 void vm_page_flag_clear(vm_page_t m, unsigned short bits);
328 void vm_page_busy(vm_page_t m);
329 void vm_page_flash(vm_page_t m);
330 void vm_page_io_start(vm_page_t m);
331 void vm_page_io_finish(vm_page_t m);
332 void vm_page_hold(vm_page_t mem);
333 void vm_page_unhold(vm_page_t mem);
334 void vm_page_free(vm_page_t m);
335 void vm_page_free_zero(vm_page_t m);
336 void vm_page_dirty(vm_page_t m);
337 void vm_page_wakeup(vm_page_t m);
338 
339 void vm_pageq_remove(vm_page_t m);
340 
341 void vm_page_activate (vm_page_t);
342 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
343 vm_page_t vm_page_alloc_freelist(int, int);
344 struct vnode *vm_page_alloc_init(vm_page_t);
345 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
346 void vm_page_cache(vm_page_t);
347 void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t);
348 void vm_page_cache_remove(vm_page_t);
349 void vm_page_cache_transfer(vm_object_t, vm_pindex_t, vm_object_t);
350 int vm_page_try_to_cache (vm_page_t);
351 int vm_page_try_to_free (vm_page_t);
352 void vm_page_dontneed(vm_page_t);
353 void vm_page_deactivate (vm_page_t);
354 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
355 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
356 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
357 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
358 vm_page_t vm_page_next(vm_page_t m);
359 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
360 vm_page_t vm_page_prev(vm_page_t m);
361 void vm_page_putfake(vm_page_t m);
362 void vm_page_remove (vm_page_t);
363 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
364 void vm_page_requeue(vm_page_t m);
365 void vm_page_set_valid(vm_page_t m, int base, int size);
366 void vm_page_sleep(vm_page_t m, const char *msg);
367 vm_page_t vm_page_splay(vm_pindex_t, vm_page_t);
368 vm_offset_t vm_page_startup(vm_offset_t vaddr);
369 void vm_page_unhold_pages(vm_page_t *ma, int count);
370 void vm_page_unwire (vm_page_t, int);
371 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
372 void vm_page_wire (vm_page_t);
373 void vm_page_set_validclean (vm_page_t, int, int);
374 void vm_page_clear_dirty (vm_page_t, int, int);
375 void vm_page_set_invalid (vm_page_t, int, int);
376 int vm_page_is_valid (vm_page_t, int, int);
377 void vm_page_test_dirty (vm_page_t);
378 int vm_page_bits (int, int);
379 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
380 void vm_page_free_toq(vm_page_t m);
381 void vm_page_zero_idle_wakeup(void);
382 void vm_page_cowfault (vm_page_t);
383 int vm_page_cowsetup(vm_page_t);
384 void vm_page_cowclear (vm_page_t);
385 
386 #ifdef INVARIANTS
387 void vm_page_object_lock_assert(vm_page_t m);
388 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
389 #else
390 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
391 #endif
392 
393 /*
394  *	vm_page_sleep_if_busy:
395  *
396  *	Sleep and release the page queues lock if VPO_BUSY is set or,
397  *	if also_m_busy is TRUE, busy is non-zero.  Returns TRUE if the
398  *	thread slept and the page queues lock was released.
399  *	Otherwise, retains the page queues lock and returns FALSE.
400  *
401  *	The object containing the given page must be locked.
402  */
403 static __inline int
404 vm_page_sleep_if_busy(vm_page_t m, int also_m_busy, const char *msg)
405 {
406 
407 	if ((m->oflags & VPO_BUSY) || (also_m_busy && m->busy)) {
408 		vm_page_sleep(m, msg);
409 		return (TRUE);
410 	}
411 	return (FALSE);
412 }
413 
414 /*
415  *	vm_page_undirty:
416  *
417  *	Set page to not be dirty.  Note: does not clear pmap modify bits
418  */
419 static __inline void
420 vm_page_undirty(vm_page_t m)
421 {
422 
423 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
424 	m->dirty = 0;
425 }
426 
427 #endif				/* _KERNEL */
428 #endif				/* !_VM_PAGE_ */
429