xref: /freebsd/sys/vm/vm_page.h (revision 682c9e0fed0115eb6f283e755901c0aac90e86e8)
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  *	In general, operations on this structure's mutable fields are
93  *	synchronized using either one of or a combination of the lock on the
94  *	object that the page belongs to (O), the pool lock for the page (P),
95  *	or the lock for either the free or paging queues (Q).  If a field is
96  *	annotated below with two of these locks, then holding either lock is
97  *	sufficient for read access, but both locks are required for write
98  *	access.
99  *
100  *	In contrast, the synchronization of accesses to the page's dirty field
101  *	is machine dependent (M).  In the machine-independent layer, the lock
102  *	on the object that the page belongs to must be held in order to
103  *	operate on the field.  However, the pmap layer is permitted to set
104  *	all bits within the field without holding that lock.  Therefore, if
105  *	the underlying architecture does not support atomic read-modify-write
106  *	operations on the field's type, then the machine-independent layer
107  *	must also hold the page queues lock when performing read-modify-write
108  *	operations and the pmap layer must hold the page queues lock when
109  *	setting the field.  In the machine-independent layer, the
110  *	implementation of read-modify-write operations on the field is
111  *	encapsulated in vm_page_clear_dirty_mask().
112  */
113 
114 TAILQ_HEAD(pglist, vm_page);
115 
116 struct vm_page {
117 	TAILQ_ENTRY(vm_page) pageq;	/* queue info for FIFO queue or free list (Q) */
118 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) 	*/
119 	struct vm_page *left;		/* splay tree link (O)		*/
120 	struct vm_page *right;		/* splay tree link (O)		*/
121 
122 	vm_object_t object;		/* which object am I in (O,P)*/
123 	vm_pindex_t pindex;		/* offset into object (O,P) */
124 	vm_paddr_t phys_addr;		/* physical address of page */
125 	struct md_page md;		/* machine dependant stuff */
126 	uint8_t	queue;			/* page queue index (P,Q) */
127 	int8_t segind;
128 	u_short	flags;			/* see below */
129 	uint8_t	order;			/* index of the buddy queue */
130 	uint8_t pool;
131 	u_short cow;			/* page cow mapping count (P) */
132 	u_int wire_count;		/* wired down maps refs (P) */
133 	short hold_count;		/* page hold count (P) */
134 	u_short oflags;			/* page flags (O) */
135 	u_char	act_count;		/* page usage count (O) */
136 	u_char	busy;			/* page busy count (O) */
137 	/* NOTE that these must support one bit per DEV_BSIZE in a page!!! */
138 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
139 	/* In reality, support for 32KB pages is not fully implemented. */
140 #if PAGE_SIZE == 4096
141 	u_char	valid;			/* map of valid DEV_BSIZE chunks (O) */
142 	u_char	dirty;			/* map of dirty DEV_BSIZE chunks (M) */
143 #elif PAGE_SIZE == 8192
144 	u_short	valid;			/* map of valid DEV_BSIZE chunks (O) */
145 	u_short	dirty;			/* map of dirty DEV_BSIZE chunks (M) */
146 #elif PAGE_SIZE == 16384
147 	u_int valid;			/* map of valid DEV_BSIZE chunks (O) */
148 	u_int dirty;			/* map of dirty DEV_BSIZE chunks (M) */
149 #elif PAGE_SIZE == 32768
150 	u_long valid;			/* map of valid DEV_BSIZE chunks (O) */
151 	u_long dirty;			/* map of dirty DEV_BSIZE chunks (M) */
152 #endif
153 };
154 
155 /*
156  * Page flags stored in oflags:
157  *
158  * Access to these page flags is synchronized by the lock on the object
159  * containing the page (O).
160  */
161 #define	VPO_BUSY	0x0001	/* page is in transit */
162 #define	VPO_WANTED	0x0002	/* someone is waiting for page */
163 #define	VPO_SWAPINPROG	0x0200	/* swap I/O in progress on page */
164 #define	VPO_NOSYNC	0x0400	/* do not collect for syncer */
165 
166 #define	PQ_NONE		255
167 #define	PQ_INACTIVE	0
168 #define	PQ_ACTIVE	1
169 #define	PQ_HOLD		2
170 #define	PQ_COUNT	3
171 
172 struct vpgqueues {
173 	struct pglist pl;
174 	int	*cnt;
175 };
176 
177 extern struct vpgqueues vm_page_queues[PQ_COUNT];
178 
179 struct vpglocks {
180 	struct mtx	data;
181 	char		pad[CACHE_LINE_SIZE - sizeof(struct mtx)];
182 } __aligned(CACHE_LINE_SIZE);
183 
184 extern struct vpglocks vm_page_queue_free_lock;
185 extern struct vpglocks pa_lock[];
186 
187 #if defined(__arm__)
188 #define	PDRSHIFT	PDR_SHIFT
189 #elif !defined(PDRSHIFT)
190 #define PDRSHIFT	21
191 #endif
192 
193 #define	pa_index(pa)	((pa) >> PDRSHIFT)
194 #define	PA_LOCKPTR(pa)	&pa_lock[pa_index((pa)) % PA_LOCK_COUNT].data
195 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
196 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
197 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
198 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
199 #define	PA_UNLOCK_COND(pa) 			\
200 	do {		   			\
201 		if ((pa) != 0) {		\
202 			PA_UNLOCK((pa));	\
203 			(pa) = 0;		\
204 		}				\
205 	} while (0)
206 
207 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
208 
209 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
210 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
211 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
212 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
213 #define	vm_page_lock_assert(m, a)	mtx_assert(vm_page_lockptr((m)), (a))
214 
215 #define	vm_page_queue_free_mtx	vm_page_queue_free_lock.data
216 /*
217  * These are the flags defined for vm_page.
218  *
219  * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
220  * 	 not under PV management but otherwise should be treated as a
221  *	 normal page.  Pages not under PV management cannot be paged out
222  *	 via the object/vm_page_t because there is no knowledge of their
223  *	 pte mappings, nor can they be removed from their objects via
224  *	 the object, and such pages are also not on any PQ queue.
225  *
226  * PG_REFERENCED may be cleared only if the object containing the page is
227  * locked.
228  *
229  * PG_WRITEABLE is set exclusively on managed pages by pmap_enter().  When it
230  * does so, the page must be VPO_BUSY.
231  */
232 #define	PG_CACHED	0x0001		/* page is cached */
233 #define	PG_FREE		0x0002		/* page is free */
234 #define PG_WINATCFLS	0x0004		/* flush dirty page on inactive q */
235 #define	PG_FICTITIOUS	0x0008		/* physical page doesn't exist (O) */
236 #define	PG_WRITEABLE	0x0010		/* page is mapped writeable */
237 #define	PG_ZERO		0x0040		/* page is zeroed */
238 #define PG_REFERENCED	0x0080		/* page has been referenced */
239 #define PG_UNMANAGED	0x0800		/* No PV management for page */
240 #define PG_MARKER	0x1000		/* special queue marker page */
241 #define	PG_SLAB		0x2000		/* object pointer is actually a slab */
242 
243 /*
244  * Misc constants.
245  */
246 #define ACT_DECLINE		1
247 #define ACT_ADVANCE		3
248 #define ACT_INIT		5
249 #define ACT_MAX			64
250 
251 #ifdef _KERNEL
252 
253 #include <vm/vm_param.h>
254 
255 /*
256  * Each pageable resident page falls into one of five lists:
257  *
258  *	free
259  *		Available for allocation now.
260  *
261  *	cache
262  *		Almost available for allocation. Still associated with
263  *		an object, but clean and immediately freeable.
264  *
265  *	hold
266  *		Will become free after a pending I/O operation
267  *		completes.
268  *
269  * The following lists are LRU sorted:
270  *
271  *	inactive
272  *		Low activity, candidates for reclamation.
273  *		This is the list of pages that should be
274  *		paged out next.
275  *
276  *	active
277  *		Pages that are "active" i.e. they have been
278  *		recently referenced.
279  *
280  */
281 
282 struct vnode;
283 extern int vm_page_zero_count;
284 
285 extern vm_page_t vm_page_array;		/* First resident page in table */
286 extern int vm_page_array_size;		/* number of vm_page_t's */
287 extern long first_page;			/* first physical page number */
288 
289 #define	VM_PAGE_IS_FREE(m)	(((m)->flags & PG_FREE) != 0)
290 
291 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
292 
293 vm_page_t vm_phys_paddr_to_vm_page(vm_paddr_t pa);
294 
295 static __inline vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
296 
297 static __inline vm_page_t
298 PHYS_TO_VM_PAGE(vm_paddr_t pa)
299 {
300 #ifdef VM_PHYSSEG_SPARSE
301 	return (vm_phys_paddr_to_vm_page(pa));
302 #elif defined(VM_PHYSSEG_DENSE)
303 	return (&vm_page_array[atop(pa) - first_page]);
304 #else
305 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
306 #endif
307 }
308 
309 extern struct vpglocks vm_page_queue_lock;
310 
311 #define	vm_page_queue_mtx	vm_page_queue_lock.data
312 #define vm_page_lock_queues()   mtx_lock(&vm_page_queue_mtx)
313 #define vm_page_unlock_queues() mtx_unlock(&vm_page_queue_mtx)
314 
315 #if PAGE_SIZE == 4096
316 #define VM_PAGE_BITS_ALL 0xffu
317 #elif PAGE_SIZE == 8192
318 #define VM_PAGE_BITS_ALL 0xffffu
319 #elif PAGE_SIZE == 16384
320 #define VM_PAGE_BITS_ALL 0xffffffffu
321 #elif PAGE_SIZE == 32768
322 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
323 #endif
324 
325 /* page allocation classes: */
326 #define VM_ALLOC_NORMAL		0
327 #define VM_ALLOC_INTERRUPT	1
328 #define VM_ALLOC_SYSTEM		2
329 #define	VM_ALLOC_CLASS_MASK	3
330 /* page allocation flags: */
331 #define	VM_ALLOC_WIRED		0x0020	/* non pageable */
332 #define	VM_ALLOC_ZERO		0x0040	/* Try to obtain a zeroed page */
333 #define	VM_ALLOC_RETRY		0x0080	/* Mandatory with vm_page_grab() */
334 #define	VM_ALLOC_NOOBJ		0x0100	/* No associated object */
335 #define	VM_ALLOC_NOBUSY		0x0200	/* Do not busy the page */
336 #define	VM_ALLOC_IFCACHED	0x0400	/* Fail if the page is not cached */
337 #define	VM_ALLOC_IFNOTCACHED	0x0800	/* Fail if the page is cached */
338 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* vm_page_grab() only */
339 
340 #define	VM_ALLOC_COUNT_SHIFT	16
341 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
342 
343 void vm_page_flag_set(vm_page_t m, unsigned short bits);
344 void vm_page_flag_clear(vm_page_t m, unsigned short bits);
345 void vm_page_busy(vm_page_t m);
346 void vm_page_flash(vm_page_t m);
347 void vm_page_io_start(vm_page_t m);
348 void vm_page_io_finish(vm_page_t m);
349 void vm_page_hold(vm_page_t mem);
350 void vm_page_unhold(vm_page_t mem);
351 void vm_page_free(vm_page_t m);
352 void vm_page_free_zero(vm_page_t m);
353 void vm_page_dirty(vm_page_t m);
354 void vm_page_wakeup(vm_page_t m);
355 
356 void vm_pageq_remove(vm_page_t m);
357 
358 void vm_page_activate (vm_page_t);
359 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
360 vm_page_t vm_page_alloc_freelist(int, int);
361 struct vnode *vm_page_alloc_init(vm_page_t);
362 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
363 void vm_page_cache(vm_page_t);
364 void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t);
365 void vm_page_cache_remove(vm_page_t);
366 void vm_page_cache_transfer(vm_object_t, vm_pindex_t, vm_object_t);
367 int vm_page_try_to_cache (vm_page_t);
368 int vm_page_try_to_free (vm_page_t);
369 void vm_page_dontneed(vm_page_t);
370 void vm_page_deactivate (vm_page_t);
371 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
372 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
373 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
374 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
375 vm_page_t vm_page_next(vm_page_t m);
376 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
377 vm_page_t vm_page_prev(vm_page_t m);
378 void vm_page_putfake(vm_page_t m);
379 void vm_page_remove (vm_page_t);
380 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
381 void vm_page_requeue(vm_page_t m);
382 void vm_page_set_valid(vm_page_t m, int base, int size);
383 void vm_page_sleep(vm_page_t m, const char *msg);
384 vm_page_t vm_page_splay(vm_pindex_t, vm_page_t);
385 vm_offset_t vm_page_startup(vm_offset_t vaddr);
386 void vm_page_unhold_pages(vm_page_t *ma, int count);
387 void vm_page_unwire (vm_page_t, int);
388 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
389 void vm_page_wire (vm_page_t);
390 void vm_page_set_validclean (vm_page_t, int, int);
391 void vm_page_clear_dirty (vm_page_t, int, int);
392 void vm_page_set_invalid (vm_page_t, int, int);
393 int vm_page_is_valid (vm_page_t, int, int);
394 void vm_page_test_dirty (vm_page_t);
395 int vm_page_bits (int, int);
396 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
397 void vm_page_free_toq(vm_page_t m);
398 void vm_page_zero_idle_wakeup(void);
399 void vm_page_cowfault (vm_page_t);
400 int vm_page_cowsetup(vm_page_t);
401 void vm_page_cowclear (vm_page_t);
402 
403 #ifdef INVARIANTS
404 void vm_page_object_lock_assert(vm_page_t m);
405 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
406 #else
407 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
408 #endif
409 
410 /*
411  *	vm_page_sleep_if_busy:
412  *
413  *	Sleep and release the page queues lock if VPO_BUSY is set or,
414  *	if also_m_busy is TRUE, busy is non-zero.  Returns TRUE if the
415  *	thread slept and the page queues lock was released.
416  *	Otherwise, retains the page queues lock and returns FALSE.
417  *
418  *	The object containing the given page must be locked.
419  */
420 static __inline int
421 vm_page_sleep_if_busy(vm_page_t m, int also_m_busy, const char *msg)
422 {
423 
424 	if ((m->oflags & VPO_BUSY) || (also_m_busy && m->busy)) {
425 		vm_page_sleep(m, msg);
426 		return (TRUE);
427 	}
428 	return (FALSE);
429 }
430 
431 /*
432  *	vm_page_undirty:
433  *
434  *	Set page to not be dirty.  Note: does not clear pmap modify bits
435  */
436 static __inline void
437 vm_page_undirty(vm_page_t m)
438 {
439 
440 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
441 	m->dirty = 0;
442 }
443 
444 #endif				/* _KERNEL */
445 #endif				/* !_VM_PAGE_ */
446