xref: /freebsd/sys/vm/vm_page.h (revision d9e232109f0de1a30cffdabe17140ca35016eeec)
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 collections:
78  *
79  *		A radix tree 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 queue (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
101  *	dirty field is machine dependent (M).  In the
102  *	machine-independent layer, the lock on the object that the
103  *	page belongs to must be held in order to operate on the field.
104  *	However, the pmap layer is permitted to set all bits within
105  *	the field without holding that lock.  If the underlying
106  *	architecture does not support atomic read-modify-write
107  *	operations on the field's type, then the machine-independent
108  *	layer uses a 32-bit atomic on the aligned 32-bit word that
109  *	contains the dirty field.  In the machine-independent layer,
110  *	the implementation of read-modify-write operations on the
111  *	field is encapsulated in vm_page_clear_dirty_mask().
112  */
113 
114 #if PAGE_SIZE == 4096
115 #define VM_PAGE_BITS_ALL 0xffu
116 typedef uint8_t vm_page_bits_t;
117 #elif PAGE_SIZE == 8192
118 #define VM_PAGE_BITS_ALL 0xffffu
119 typedef uint16_t vm_page_bits_t;
120 #elif PAGE_SIZE == 16384
121 #define VM_PAGE_BITS_ALL 0xffffffffu
122 typedef uint32_t vm_page_bits_t;
123 #elif PAGE_SIZE == 32768
124 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
125 typedef uint64_t vm_page_bits_t;
126 #endif
127 
128 struct vm_page {
129 	union {
130 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
131 		struct {
132 			SLIST_ENTRY(vm_page) ss; /* private slists */
133 			void *pv;
134 		} s;
135 		struct {
136 			u_long p;
137 			u_long v;
138 		} memguard;
139 	} plinks;
140 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
141 	vm_object_t object;		/* which object am I in (O,P) */
142 	vm_pindex_t pindex;		/* offset into object (O,P) */
143 	vm_paddr_t phys_addr;		/* physical address of page */
144 	struct md_page md;		/* machine dependant stuff */
145 	uint8_t	queue;			/* page queue index (P,Q) */
146 	int8_t segind;
147 	short hold_count;		/* page hold count (P) */
148 	uint8_t	order;			/* index of the buddy queue */
149 	uint8_t pool;
150 	u_short cow;			/* page cow mapping count (P) */
151 	u_int wire_count;		/* wired down maps refs (P) */
152 	uint8_t aflags;			/* access is atomic */
153 	uint8_t oflags;			/* page VPO_* flags (O) */
154 	uint16_t flags;			/* page PG_* flags (P) */
155 	u_char	act_count;		/* page usage count (P) */
156 	u_char __pad0;			/* unused padding */
157 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
158 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
159 	vm_page_bits_t valid;		/* map of valid DEV_BSIZE chunks (O) */
160 	vm_page_bits_t dirty;		/* map of dirty DEV_BSIZE chunks (M) */
161 	volatile u_int busy_lock;	/* busy owners lock */
162 };
163 
164 /*
165  * Page flags stored in oflags:
166  *
167  * Access to these page flags is synchronized by the lock on the object
168  * containing the page (O).
169  *
170  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
171  * 	 indicates that the page is not under PV management but
172  * 	 otherwise should be treated as a normal page.  Pages not
173  * 	 under PV management cannot be paged out via the
174  * 	 object/vm_page_t because there is no knowledge of their pte
175  * 	 mappings, and such pages are also not on any PQ queue.
176  *
177  */
178 #define	VPO_UNUSED01	0x01		/* --available-- */
179 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
180 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
181 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
182 #define	VPO_NOSYNC	0x10		/* do not collect for syncer */
183 
184 /*
185  * Busy page implementation details.
186  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
187  * even if the support for owner identity is removed because of size
188  * constraints.  Checks on lock recursion are then not possible, while the
189  * lock assertions effectiveness is someway reduced.
190  */
191 #define	VPB_BIT_SHARED		0x01
192 #define	VPB_BIT_EXCLUSIVE	0x02
193 #define	VPB_BIT_WAITERS		0x04
194 #define	VPB_BIT_FLAGMASK						\
195 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
196 
197 #define	VPB_SHARERS_SHIFT	3
198 #define	VPB_SHARERS(x)							\
199 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
200 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
201 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
202 
203 #define	VPB_SINGLE_EXCLUSIVER	VPB_BIT_EXCLUSIVE
204 
205 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
206 
207 #define	PQ_NONE		255
208 #define	PQ_INACTIVE	0
209 #define	PQ_ACTIVE	1
210 #define	PQ_COUNT	2
211 
212 TAILQ_HEAD(pglist, vm_page);
213 SLIST_HEAD(spglist, vm_page);
214 
215 struct vm_pagequeue {
216 	struct mtx	pq_mutex;
217 	struct pglist	pq_pl;
218 	int		pq_cnt;
219 	int		* const pq_vcnt;
220 	const char	* const pq_name;
221 } __aligned(CACHE_LINE_SIZE);
222 
223 
224 struct vm_domain {
225 	struct vm_pagequeue vmd_pagequeues[PQ_COUNT];
226 	u_int vmd_page_count;
227 	u_int vmd_free_count;
228 	long vmd_segs;	/* bitmask of the segments */
229 	boolean_t vmd_oom;
230 	int vmd_pass;	/* local pagedaemon pass */
231 	struct vm_page vmd_marker; /* marker for pagedaemon private use */
232 };
233 
234 extern struct vm_domain vm_dom[MAXMEMDOM];
235 
236 #define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
237 #define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
238 #define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
239 
240 #ifdef _KERNEL
241 static __inline void
242 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
243 {
244 
245 #ifdef notyet
246 	vm_pagequeue_assert_locked(pq);
247 #endif
248 	pq->pq_cnt += addend;
249 	atomic_add_int(pq->pq_vcnt, addend);
250 }
251 #define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
252 #define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
253 #endif	/* _KERNEL */
254 
255 extern struct mtx_padalign vm_page_queue_free_mtx;
256 extern struct mtx_padalign pa_lock[];
257 
258 #if defined(__arm__)
259 #define	PDRSHIFT	PDR_SHIFT
260 #elif !defined(PDRSHIFT)
261 #define PDRSHIFT	21
262 #endif
263 
264 #define	pa_index(pa)	((pa) >> PDRSHIFT)
265 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
266 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
267 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
268 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
269 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
270 #define	PA_UNLOCK_COND(pa) 			\
271 	do {		   			\
272 		if ((pa) != 0) {		\
273 			PA_UNLOCK((pa));	\
274 			(pa) = 0;		\
275 		}				\
276 	} while (0)
277 
278 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
279 
280 #ifdef KLD_MODULE
281 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
282 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
283 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
284 #else	/* !KLD_MODULE */
285 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
286 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
287 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
288 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
289 #endif
290 #if defined(INVARIANTS)
291 #define	vm_page_assert_locked(m)		\
292     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
293 #define	vm_page_lock_assert(m, a)		\
294     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
295 #else
296 #define	vm_page_assert_locked(m)
297 #define	vm_page_lock_assert(m, a)
298 #endif
299 
300 /*
301  * The vm_page's aflags are updated using atomic operations.  To set or clear
302  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
303  * must be used.  Neither these flags nor these functions are part of the KBI.
304  *
305  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
306  * both the MI and MD VM layers.  However, kernel loadable modules should not
307  * directly set this flag.  They should call vm_page_reference() instead.
308  *
309  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().  When it
310  * does so, the page must be exclusive busied.  The MI VM layer must never
311  * access this flag directly.  Instead, it should call
312  * pmap_page_is_write_mapped().
313  *
314  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
315  * at least one executable mapping.  It is not consumed by the MI VM layer.
316  */
317 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
318 #define	PGA_REFERENCED	0x02		/* page has been referenced */
319 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
320 
321 /*
322  * Page flags.  If changed at any other time than page allocation or
323  * freeing, the modification must be protected by the vm_page lock.
324  */
325 #define	PG_CACHED	0x0001		/* page is cached */
326 #define	PG_FREE		0x0002		/* page is free */
327 #define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
328 #define	PG_ZERO		0x0008		/* page is zeroed */
329 #define	PG_MARKER	0x0010		/* special queue marker page */
330 #define	PG_SLAB		0x0020		/* object pointer is actually a slab */
331 #define	PG_WINATCFLS	0x0040		/* flush dirty page on inactive q */
332 #define	PG_NODUMP	0x0080		/* don't include this page in a dump */
333 #define	PG_UNHOLDFREE	0x0100		/* delayed free of a held page */
334 
335 /*
336  * Misc constants.
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 #include <sys/systm.h>
346 
347 #include <machine/atomic.h>
348 
349 /*
350  * Each pageable resident page falls into one of four lists:
351  *
352  *	free
353  *		Available for allocation now.
354  *
355  *	cache
356  *		Almost available for allocation. Still associated with
357  *		an object, but clean and immediately freeable.
358  *
359  * The following lists are LRU sorted:
360  *
361  *	inactive
362  *		Low activity, candidates for reclamation.
363  *		This is the list of pages that should be
364  *		paged out next.
365  *
366  *	active
367  *		Pages that are "active" i.e. they have been
368  *		recently referenced.
369  *
370  */
371 
372 extern int vm_page_zero_count;
373 
374 extern vm_page_t vm_page_array;		/* First resident page in table */
375 extern long vm_page_array_size;		/* number of vm_page_t's */
376 extern long first_page;			/* first physical page number */
377 
378 #define	VM_PAGE_IS_FREE(m)	(((m)->flags & PG_FREE) != 0)
379 
380 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
381 
382 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
383 
384 /* page allocation classes: */
385 #define VM_ALLOC_NORMAL		0
386 #define VM_ALLOC_INTERRUPT	1
387 #define VM_ALLOC_SYSTEM		2
388 #define	VM_ALLOC_CLASS_MASK	3
389 /* page allocation flags: */
390 #define	VM_ALLOC_WIRED		0x0020	/* non pageable */
391 #define	VM_ALLOC_ZERO		0x0040	/* Try to obtain a zeroed page */
392 #define	VM_ALLOC_RETRY		0x0080	/* Mandatory with vm_page_grab() */
393 #define	VM_ALLOC_NOOBJ		0x0100	/* No associated object */
394 #define	VM_ALLOC_NOBUSY		0x0200	/* Do not busy the page */
395 #define	VM_ALLOC_IFCACHED	0x0400	/* Fail if the page is not cached */
396 #define	VM_ALLOC_IFNOTCACHED	0x0800	/* Fail if the page is cached */
397 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* vm_page_grab() only */
398 #define	VM_ALLOC_NODUMP		0x2000	/* don't include in dump */
399 #define	VM_ALLOC_SBUSY		0x4000	/* Shared busy the page */
400 
401 #define	VM_ALLOC_COUNT_SHIFT	16
402 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
403 
404 #ifdef M_NOWAIT
405 static inline int
406 malloc2vm_flags(int malloc_flags)
407 {
408 	int pflags;
409 
410 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
411 	    (malloc_flags & M_NOWAIT) != 0,
412 	    ("M_USE_RESERVE requires M_NOWAIT"));
413 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
414 	    VM_ALLOC_SYSTEM;
415 	if ((malloc_flags & M_ZERO) != 0)
416 		pflags |= VM_ALLOC_ZERO;
417 	if ((malloc_flags & M_NODUMP) != 0)
418 		pflags |= VM_ALLOC_NODUMP;
419 	return (pflags);
420 }
421 #endif
422 
423 void vm_page_busy_downgrade(vm_page_t m);
424 void vm_page_busy_sleep(vm_page_t m, const char *msg);
425 void vm_page_flash(vm_page_t m);
426 void vm_page_hold(vm_page_t mem);
427 void vm_page_unhold(vm_page_t mem);
428 void vm_page_free(vm_page_t m);
429 void vm_page_free_zero(vm_page_t m);
430 
431 void vm_page_activate (vm_page_t);
432 void vm_page_advise(vm_page_t m, int advice);
433 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
434 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
435     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
436     vm_paddr_t boundary, vm_memattr_t memattr);
437 vm_page_t vm_page_alloc_freelist(int, int);
438 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
439 void vm_page_cache(vm_page_t);
440 void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t);
441 void vm_page_cache_transfer(vm_object_t, vm_pindex_t, vm_object_t);
442 int vm_page_try_to_cache (vm_page_t);
443 int vm_page_try_to_free (vm_page_t);
444 void vm_page_deactivate (vm_page_t);
445 void vm_page_dequeue(vm_page_t m);
446 void vm_page_dequeue_locked(vm_page_t m);
447 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
448 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
449 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
450 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
451 boolean_t vm_page_is_cached(vm_object_t object, vm_pindex_t pindex);
452 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
453 vm_page_t vm_page_next(vm_page_t m);
454 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
455 struct vm_pagequeue *vm_page_pagequeue(vm_page_t m);
456 vm_page_t vm_page_prev(vm_page_t m);
457 void vm_page_putfake(vm_page_t m);
458 void vm_page_readahead_finish(vm_page_t m);
459 void vm_page_reference(vm_page_t m);
460 void vm_page_remove (vm_page_t);
461 int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
462 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
463     vm_pindex_t pindex);
464 void vm_page_requeue(vm_page_t m);
465 void vm_page_requeue_locked(vm_page_t m);
466 int vm_page_sbusied(vm_page_t m);
467 void vm_page_set_valid_range(vm_page_t m, int base, int size);
468 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
469 vm_offset_t vm_page_startup(vm_offset_t vaddr);
470 void vm_page_sunbusy(vm_page_t m);
471 int vm_page_trysbusy(vm_page_t m);
472 void vm_page_unhold_pages(vm_page_t *ma, int count);
473 void vm_page_unwire (vm_page_t, int);
474 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
475 void vm_page_wire (vm_page_t);
476 void vm_page_xunbusy_hard(vm_page_t m);
477 void vm_page_set_validclean (vm_page_t, int, int);
478 void vm_page_clear_dirty (vm_page_t, int, int);
479 void vm_page_set_invalid (vm_page_t, int, int);
480 int vm_page_is_valid (vm_page_t, int, int);
481 void vm_page_test_dirty (vm_page_t);
482 vm_page_bits_t vm_page_bits(int base, int size);
483 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
484 void vm_page_free_toq(vm_page_t m);
485 void vm_page_zero_idle_wakeup(void);
486 void vm_page_cowfault (vm_page_t);
487 int vm_page_cowsetup(vm_page_t);
488 void vm_page_cowclear (vm_page_t);
489 
490 void vm_page_dirty_KBI(vm_page_t m);
491 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
492 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
493 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
494 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
495 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
496 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
497 #endif
498 
499 #define	vm_page_assert_sbusied(m)					\
500 	KASSERT(vm_page_sbusied(m),					\
501 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
502 	    (void *)m, __FILE__, __LINE__));
503 
504 #define	vm_page_assert_unbusied(m)					\
505 	KASSERT(!vm_page_busied(m),					\
506 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
507 	    (void *)m, __FILE__, __LINE__));
508 
509 #define	vm_page_assert_xbusied(m)					\
510 	KASSERT(vm_page_xbusied(m),					\
511 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
512 	    (void *)m, __FILE__, __LINE__));
513 
514 #define	vm_page_busied(m)						\
515 	((m)->busy_lock != VPB_UNBUSIED)
516 
517 #define	vm_page_sbusy(m) do {						\
518 	if (!vm_page_trysbusy(m))					\
519 		panic("%s: page %p failed shared busing", __func__, m);	\
520 } while (0)
521 
522 #define	vm_page_tryxbusy(m)						\
523 	(atomic_cmpset_acq_int(&m->busy_lock, VPB_UNBUSIED,		\
524 	    VPB_SINGLE_EXCLUSIVER))
525 
526 #define	vm_page_xbusied(m)						\
527 	((m->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
528 
529 #define	vm_page_xbusy(m) do {						\
530 	if (!vm_page_tryxbusy(m))					\
531 		panic("%s: page %p failed exclusive busing", __func__,	\
532 		    m);							\
533 } while (0)
534 
535 #define	vm_page_xunbusy(m) do {						\
536 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
537 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
538 		vm_page_xunbusy_hard(m);				\
539 } while (0)
540 
541 #ifdef INVARIANTS
542 void vm_page_object_lock_assert(vm_page_t m);
543 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
544 #else
545 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
546 #endif
547 
548 /*
549  * We want to use atomic updates for the aflags field, which is 8 bits wide.
550  * However, not all architectures support atomic operations on 8-bit
551  * destinations.  In order that we can easily use a 32-bit operation, we
552  * require that the aflags field be 32-bit aligned.
553  */
554 CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0);
555 
556 /*
557  *	Clear the given bits in the specified page.
558  */
559 static inline void
560 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
561 {
562 	uint32_t *addr, val;
563 
564 	/*
565 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
566 	 */
567 	if ((bits & PGA_REFERENCED) != 0)
568 		vm_page_assert_locked(m);
569 
570 	/*
571 	 * Access the whole 32-bit word containing the aflags field with an
572 	 * atomic update.  Parallel non-atomic updates to the other fields
573 	 * within this word are handled properly by the atomic update.
574 	 */
575 	addr = (void *)&m->aflags;
576 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
577 	    ("vm_page_aflag_clear: aflags is misaligned"));
578 	val = bits;
579 #if BYTE_ORDER == BIG_ENDIAN
580 	val <<= 24;
581 #endif
582 	atomic_clear_32(addr, val);
583 }
584 
585 /*
586  *	Set the given bits in the specified page.
587  */
588 static inline void
589 vm_page_aflag_set(vm_page_t m, uint8_t bits)
590 {
591 	uint32_t *addr, val;
592 
593 	/*
594 	 * The PGA_WRITEABLE flag can only be set if the page is managed and
595 	 * exclusive busied.  Currently, this flag is only set by pmap_enter().
596 	 */
597 	KASSERT((bits & PGA_WRITEABLE) == 0 ||
598 	    ((m->oflags & VPO_UNMANAGED) == 0 && vm_page_xbusied(m)),
599 	    ("vm_page_aflag_set: PGA_WRITEABLE and not exclusive busy"));
600 
601 	/*
602 	 * Access the whole 32-bit word containing the aflags field with an
603 	 * atomic update.  Parallel non-atomic updates to the other fields
604 	 * within this word are handled properly by the atomic update.
605 	 */
606 	addr = (void *)&m->aflags;
607 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
608 	    ("vm_page_aflag_set: aflags is misaligned"));
609 	val = bits;
610 #if BYTE_ORDER == BIG_ENDIAN
611 	val <<= 24;
612 #endif
613 	atomic_set_32(addr, val);
614 }
615 
616 /*
617  *	vm_page_dirty:
618  *
619  *	Set all bits in the page's dirty field.
620  *
621  *	The object containing the specified page must be locked if the
622  *	call is made from the machine-independent layer.
623  *
624  *	See vm_page_clear_dirty_mask().
625  */
626 static __inline void
627 vm_page_dirty(vm_page_t m)
628 {
629 
630 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
631 #if defined(KLD_MODULE) || defined(INVARIANTS)
632 	vm_page_dirty_KBI(m);
633 #else
634 	m->dirty = VM_PAGE_BITS_ALL;
635 #endif
636 }
637 
638 /*
639  *	vm_page_remque:
640  *
641  *	If the given page is in a page queue, then remove it from that page
642  *	queue.
643  *
644  *	The page must be locked.
645  */
646 static inline void
647 vm_page_remque(vm_page_t m)
648 {
649 
650 	if (m->queue != PQ_NONE)
651 		vm_page_dequeue(m);
652 }
653 
654 /*
655  *	vm_page_undirty:
656  *
657  *	Set page to not be dirty.  Note: does not clear pmap modify bits
658  */
659 static __inline void
660 vm_page_undirty(vm_page_t m)
661 {
662 
663 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
664 	m->dirty = 0;
665 }
666 
667 #endif				/* _KERNEL */
668 #endif				/* !_VM_PAGE_ */
669