xref: /freebsd/sys/vm/vm_page.h (revision febdb468801f35e51c6c5c22221cfce9197c6f3b)
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 	u_int wire_count;		/* wired down maps refs (P) */
146 	volatile u_int busy_lock;	/* busy owners lock */
147 	uint16_t hold_count;		/* page hold count (P) */
148 	uint16_t flags;			/* page PG_* flags (P) */
149 	uint8_t aflags;			/* access is atomic */
150 	uint8_t oflags;			/* page VPO_* flags (O) */
151 	uint8_t	queue;			/* page queue index (P,Q) */
152 	int8_t segind;
153 	uint8_t	order;			/* index of the buddy queue */
154 	uint8_t pool;
155 	u_char	act_count;		/* page usage count (P) */
156 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
157 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
158 	vm_page_bits_t valid;		/* map of valid DEV_BSIZE chunks (O) */
159 	vm_page_bits_t dirty;		/* map of dirty DEV_BSIZE chunks (M) */
160 };
161 
162 /*
163  * Page flags stored in oflags:
164  *
165  * Access to these page flags is synchronized by the lock on the object
166  * containing the page (O).
167  *
168  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
169  * 	 indicates that the page is not under PV management but
170  * 	 otherwise should be treated as a normal page.  Pages not
171  * 	 under PV management cannot be paged out via the
172  * 	 object/vm_page_t because there is no knowledge of their pte
173  * 	 mappings, and such pages are also not on any PQ queue.
174  *
175  */
176 #define	VPO_UNUSED01	0x01		/* --available-- */
177 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
178 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
179 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
180 #define	VPO_NOSYNC	0x10		/* do not collect for syncer */
181 
182 /*
183  * Busy page implementation details.
184  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
185  * even if the support for owner identity is removed because of size
186  * constraints.  Checks on lock recursion are then not possible, while the
187  * lock assertions effectiveness is someway reduced.
188  */
189 #define	VPB_BIT_SHARED		0x01
190 #define	VPB_BIT_EXCLUSIVE	0x02
191 #define	VPB_BIT_WAITERS		0x04
192 #define	VPB_BIT_FLAGMASK						\
193 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
194 
195 #define	VPB_SHARERS_SHIFT	3
196 #define	VPB_SHARERS(x)							\
197 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
198 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
199 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
200 
201 #define	VPB_SINGLE_EXCLUSIVER	VPB_BIT_EXCLUSIVE
202 
203 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
204 
205 #define	PQ_NONE		255
206 #define	PQ_INACTIVE	0
207 #define	PQ_ACTIVE	1
208 #define	PQ_COUNT	2
209 
210 TAILQ_HEAD(pglist, vm_page);
211 SLIST_HEAD(spglist, vm_page);
212 
213 struct vm_pagequeue {
214 	struct mtx	pq_mutex;
215 	struct pglist	pq_pl;
216 	int		pq_cnt;
217 	int		* const pq_vcnt;
218 	const char	* const pq_name;
219 } __aligned(CACHE_LINE_SIZE);
220 
221 
222 struct vm_domain {
223 	struct vm_pagequeue vmd_pagequeues[PQ_COUNT];
224 	u_int vmd_page_count;
225 	u_int vmd_free_count;
226 	long vmd_segs;	/* bitmask of the segments */
227 	boolean_t vmd_oom;
228 	int vmd_pass;	/* local pagedaemon pass */
229 	struct vm_page vmd_marker; /* marker for pagedaemon private use */
230 };
231 
232 extern struct vm_domain vm_dom[MAXMEMDOM];
233 
234 #define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
235 #define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
236 #define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
237 
238 #ifdef _KERNEL
239 static __inline void
240 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
241 {
242 
243 #ifdef notyet
244 	vm_pagequeue_assert_locked(pq);
245 #endif
246 	pq->pq_cnt += addend;
247 	atomic_add_int(pq->pq_vcnt, addend);
248 }
249 #define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
250 #define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
251 #endif	/* _KERNEL */
252 
253 extern struct mtx_padalign vm_page_queue_free_mtx;
254 extern struct mtx_padalign pa_lock[];
255 
256 #if defined(__arm__)
257 #define	PDRSHIFT	PDR_SHIFT
258 #elif !defined(PDRSHIFT)
259 #define PDRSHIFT	21
260 #endif
261 
262 #define	pa_index(pa)	((pa) >> PDRSHIFT)
263 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
264 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
265 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
266 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
267 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
268 #define	PA_UNLOCK_COND(pa) 			\
269 	do {		   			\
270 		if ((pa) != 0) {		\
271 			PA_UNLOCK((pa));	\
272 			(pa) = 0;		\
273 		}				\
274 	} while (0)
275 
276 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
277 
278 #ifdef KLD_MODULE
279 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
280 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
281 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
282 #else	/* !KLD_MODULE */
283 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
284 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
285 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
286 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
287 #endif
288 #if defined(INVARIANTS)
289 #define	vm_page_assert_locked(m)		\
290     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
291 #define	vm_page_lock_assert(m, a)		\
292     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
293 #else
294 #define	vm_page_assert_locked(m)
295 #define	vm_page_lock_assert(m, a)
296 #endif
297 
298 /*
299  * The vm_page's aflags are updated using atomic operations.  To set or clear
300  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
301  * must be used.  Neither these flags nor these functions are part of the KBI.
302  *
303  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
304  * both the MI and MD VM layers.  However, kernel loadable modules should not
305  * directly set this flag.  They should call vm_page_reference() instead.
306  *
307  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().  When it
308  * does so, the page must be exclusive busied.  The MI VM layer must never
309  * access this flag directly.  Instead, it should call
310  * pmap_page_is_write_mapped().
311  *
312  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
313  * at least one executable mapping.  It is not consumed by the MI VM layer.
314  */
315 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
316 #define	PGA_REFERENCED	0x02		/* page has been referenced */
317 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
318 
319 /*
320  * Page flags.  If changed at any other time than page allocation or
321  * freeing, the modification must be protected by the vm_page lock.
322  */
323 #define	PG_CACHED	0x0001		/* page is cached */
324 #define	PG_FREE		0x0002		/* page is free */
325 #define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
326 #define	PG_ZERO		0x0008		/* page is zeroed */
327 #define	PG_MARKER	0x0010		/* special queue marker page */
328 #define	PG_WINATCFLS	0x0040		/* flush dirty page on inactive q */
329 #define	PG_NODUMP	0x0080		/* don't include this page in a dump */
330 #define	PG_UNHOLDFREE	0x0100		/* delayed free of a held page */
331 
332 /*
333  * Misc constants.
334  */
335 #define ACT_DECLINE		1
336 #define ACT_ADVANCE		3
337 #define ACT_INIT		5
338 #define ACT_MAX			64
339 
340 #ifdef _KERNEL
341 
342 #include <sys/systm.h>
343 
344 #include <machine/atomic.h>
345 
346 /*
347  * Each pageable resident page falls into one of four lists:
348  *
349  *	free
350  *		Available for allocation now.
351  *
352  *	cache
353  *		Almost available for allocation. Still associated with
354  *		an object, but clean and immediately freeable.
355  *
356  * The following lists are LRU sorted:
357  *
358  *	inactive
359  *		Low activity, candidates for reclamation.
360  *		This is the list of pages that should be
361  *		paged out next.
362  *
363  *	active
364  *		Pages that are "active" i.e. they have been
365  *		recently referenced.
366  *
367  */
368 
369 extern int vm_page_zero_count;
370 
371 extern vm_page_t vm_page_array;		/* First resident page in table */
372 extern long vm_page_array_size;		/* number of vm_page_t's */
373 extern long first_page;			/* first physical page number */
374 
375 #define	VM_PAGE_IS_FREE(m)	(((m)->flags & PG_FREE) != 0)
376 
377 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
378 
379 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
380 
381 /* page allocation classes: */
382 #define VM_ALLOC_NORMAL		0
383 #define VM_ALLOC_INTERRUPT	1
384 #define VM_ALLOC_SYSTEM		2
385 #define	VM_ALLOC_CLASS_MASK	3
386 /* page allocation flags: */
387 #define	VM_ALLOC_WIRED		0x0020	/* non pageable */
388 #define	VM_ALLOC_ZERO		0x0040	/* Try to obtain a zeroed page */
389 #define	VM_ALLOC_NOOBJ		0x0100	/* No associated object */
390 #define	VM_ALLOC_NOBUSY		0x0200	/* Do not busy the page */
391 #define	VM_ALLOC_IFCACHED	0x0400	/* Fail if the page is not cached */
392 #define	VM_ALLOC_IFNOTCACHED	0x0800	/* Fail if the page is cached */
393 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* vm_page_grab() only */
394 #define	VM_ALLOC_NODUMP		0x2000	/* don't include in dump */
395 #define	VM_ALLOC_SBUSY		0x4000	/* Shared busy the page */
396 
397 #define	VM_ALLOC_COUNT_SHIFT	16
398 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
399 
400 #ifdef M_NOWAIT
401 static inline int
402 malloc2vm_flags(int malloc_flags)
403 {
404 	int pflags;
405 
406 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
407 	    (malloc_flags & M_NOWAIT) != 0,
408 	    ("M_USE_RESERVE requires M_NOWAIT"));
409 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
410 	    VM_ALLOC_SYSTEM;
411 	if ((malloc_flags & M_ZERO) != 0)
412 		pflags |= VM_ALLOC_ZERO;
413 	if ((malloc_flags & M_NODUMP) != 0)
414 		pflags |= VM_ALLOC_NODUMP;
415 	return (pflags);
416 }
417 #endif
418 
419 void vm_page_busy_downgrade(vm_page_t m);
420 void vm_page_busy_sleep(vm_page_t m, const char *msg);
421 void vm_page_flash(vm_page_t m);
422 void vm_page_hold(vm_page_t mem);
423 void vm_page_unhold(vm_page_t mem);
424 void vm_page_free(vm_page_t m);
425 void vm_page_free_zero(vm_page_t m);
426 
427 void vm_page_activate (vm_page_t);
428 void vm_page_advise(vm_page_t m, int advice);
429 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
430 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
431     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
432     vm_paddr_t boundary, vm_memattr_t memattr);
433 vm_page_t vm_page_alloc_freelist(int, int);
434 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
435 void vm_page_cache(vm_page_t);
436 void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t);
437 void vm_page_cache_transfer(vm_object_t, vm_pindex_t, vm_object_t);
438 int vm_page_try_to_cache (vm_page_t);
439 int vm_page_try_to_free (vm_page_t);
440 void vm_page_deactivate (vm_page_t);
441 void vm_page_dequeue(vm_page_t m);
442 void vm_page_dequeue_locked(vm_page_t m);
443 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
444 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
445 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
446 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
447 boolean_t vm_page_is_cached(vm_object_t object, vm_pindex_t pindex);
448 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
449 vm_page_t vm_page_next(vm_page_t m);
450 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
451 struct vm_pagequeue *vm_page_pagequeue(vm_page_t m);
452 vm_page_t vm_page_prev(vm_page_t m);
453 void vm_page_putfake(vm_page_t m);
454 void vm_page_readahead_finish(vm_page_t m);
455 void vm_page_reference(vm_page_t m);
456 void vm_page_remove (vm_page_t);
457 int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
458 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
459     vm_pindex_t pindex);
460 void vm_page_requeue(vm_page_t m);
461 void vm_page_requeue_locked(vm_page_t m);
462 int vm_page_sbusied(vm_page_t m);
463 void vm_page_set_valid_range(vm_page_t m, int base, int size);
464 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
465 vm_offset_t vm_page_startup(vm_offset_t vaddr);
466 void vm_page_sunbusy(vm_page_t m);
467 int vm_page_trysbusy(vm_page_t m);
468 void vm_page_unhold_pages(vm_page_t *ma, int count);
469 void vm_page_unwire (vm_page_t, int);
470 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
471 void vm_page_wire (vm_page_t);
472 void vm_page_xunbusy_hard(vm_page_t m);
473 void vm_page_set_validclean (vm_page_t, int, int);
474 void vm_page_clear_dirty (vm_page_t, int, int);
475 void vm_page_set_invalid (vm_page_t, int, int);
476 int vm_page_is_valid (vm_page_t, int, int);
477 void vm_page_test_dirty (vm_page_t);
478 vm_page_bits_t vm_page_bits(int base, int size);
479 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
480 void vm_page_free_toq(vm_page_t m);
481 void vm_page_zero_idle_wakeup(void);
482 
483 void vm_page_dirty_KBI(vm_page_t m);
484 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
485 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
486 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
487 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
488 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
489 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
490 #endif
491 
492 #define	vm_page_assert_sbusied(m)					\
493 	KASSERT(vm_page_sbusied(m),					\
494 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
495 	    (void *)m, __FILE__, __LINE__));
496 
497 #define	vm_page_assert_unbusied(m)					\
498 	KASSERT(!vm_page_busied(m),					\
499 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
500 	    (void *)m, __FILE__, __LINE__));
501 
502 #define	vm_page_assert_xbusied(m)					\
503 	KASSERT(vm_page_xbusied(m),					\
504 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
505 	    (void *)m, __FILE__, __LINE__));
506 
507 #define	vm_page_busied(m)						\
508 	((m)->busy_lock != VPB_UNBUSIED)
509 
510 #define	vm_page_sbusy(m) do {						\
511 	if (!vm_page_trysbusy(m))					\
512 		panic("%s: page %p failed shared busing", __func__, m);	\
513 } while (0)
514 
515 #define	vm_page_tryxbusy(m)						\
516 	(atomic_cmpset_acq_int(&m->busy_lock, VPB_UNBUSIED,		\
517 	    VPB_SINGLE_EXCLUSIVER))
518 
519 #define	vm_page_xbusied(m)						\
520 	((m->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
521 
522 #define	vm_page_xbusy(m) do {						\
523 	if (!vm_page_tryxbusy(m))					\
524 		panic("%s: page %p failed exclusive busing", __func__,	\
525 		    m);							\
526 } while (0)
527 
528 #define	vm_page_xunbusy(m) do {						\
529 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
530 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
531 		vm_page_xunbusy_hard(m);				\
532 } while (0)
533 
534 #ifdef INVARIANTS
535 void vm_page_object_lock_assert(vm_page_t m);
536 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
537 #else
538 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
539 #endif
540 
541 /*
542  * We want to use atomic updates for the aflags field, which is 8 bits wide.
543  * However, not all architectures support atomic operations on 8-bit
544  * destinations.  In order that we can easily use a 32-bit operation, we
545  * require that the aflags field be 32-bit aligned.
546  */
547 CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0);
548 
549 /*
550  *	Clear the given bits in the specified page.
551  */
552 static inline void
553 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
554 {
555 	uint32_t *addr, val;
556 
557 	/*
558 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
559 	 */
560 	if ((bits & PGA_REFERENCED) != 0)
561 		vm_page_assert_locked(m);
562 
563 	/*
564 	 * Access the whole 32-bit word containing the aflags field with an
565 	 * atomic update.  Parallel non-atomic updates to the other fields
566 	 * within this word are handled properly by the atomic update.
567 	 */
568 	addr = (void *)&m->aflags;
569 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
570 	    ("vm_page_aflag_clear: aflags is misaligned"));
571 	val = bits;
572 #if BYTE_ORDER == BIG_ENDIAN
573 	val <<= 24;
574 #endif
575 	atomic_clear_32(addr, val);
576 }
577 
578 /*
579  *	Set the given bits in the specified page.
580  */
581 static inline void
582 vm_page_aflag_set(vm_page_t m, uint8_t bits)
583 {
584 	uint32_t *addr, val;
585 
586 	/*
587 	 * The PGA_WRITEABLE flag can only be set if the page is managed and
588 	 * exclusive busied.  Currently, this flag is only set by pmap_enter().
589 	 */
590 	KASSERT((bits & PGA_WRITEABLE) == 0 ||
591 	    ((m->oflags & VPO_UNMANAGED) == 0 && vm_page_xbusied(m)),
592 	    ("vm_page_aflag_set: PGA_WRITEABLE and not exclusive busy"));
593 
594 	/*
595 	 * Access the whole 32-bit word containing the aflags field with an
596 	 * atomic update.  Parallel non-atomic updates to the other fields
597 	 * within this word are handled properly by the atomic update.
598 	 */
599 	addr = (void *)&m->aflags;
600 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
601 	    ("vm_page_aflag_set: aflags is misaligned"));
602 	val = bits;
603 #if BYTE_ORDER == BIG_ENDIAN
604 	val <<= 24;
605 #endif
606 	atomic_set_32(addr, val);
607 }
608 
609 /*
610  *	vm_page_dirty:
611  *
612  *	Set all bits in the page's dirty field.
613  *
614  *	The object containing the specified page must be locked if the
615  *	call is made from the machine-independent layer.
616  *
617  *	See vm_page_clear_dirty_mask().
618  */
619 static __inline void
620 vm_page_dirty(vm_page_t m)
621 {
622 
623 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
624 #if defined(KLD_MODULE) || defined(INVARIANTS)
625 	vm_page_dirty_KBI(m);
626 #else
627 	m->dirty = VM_PAGE_BITS_ALL;
628 #endif
629 }
630 
631 /*
632  *	vm_page_remque:
633  *
634  *	If the given page is in a page queue, then remove it from that page
635  *	queue.
636  *
637  *	The page must be locked.
638  */
639 static inline void
640 vm_page_remque(vm_page_t m)
641 {
642 
643 	if (m->queue != PQ_NONE)
644 		vm_page_dequeue(m);
645 }
646 
647 /*
648  *	vm_page_undirty:
649  *
650  *	Set page to not be dirty.  Note: does not clear pmap modify bits
651  */
652 static __inline void
653 vm_page_undirty(vm_page_t m)
654 {
655 
656 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
657 	m->dirty = 0;
658 }
659 
660 #endif				/* _KERNEL */
661 #endif				/* !_VM_PAGE_ */
662