vm_page.h revision 307672
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: stable/10/sys/vm/vm_page.h 307672 2016-10-20 13:12:19Z kib $
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
116typedef uint8_t vm_page_bits_t;
117#elif PAGE_SIZE == 8192
118#define VM_PAGE_BITS_ALL 0xffffu
119typedef uint16_t vm_page_bits_t;
120#elif PAGE_SIZE == 16384
121#define VM_PAGE_BITS_ALL 0xffffffffu
122typedef uint32_t vm_page_bits_t;
123#elif PAGE_SIZE == 32768
124#define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
125typedef uint64_t vm_page_bits_t;
126#endif
127
128struct 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	int8_t psind;			/* pagesizes[] index (O) */
161};
162
163/*
164 * Page flags stored in oflags:
165 *
166 * Access to these page flags is synchronized by the lock on the object
167 * containing the page (O).
168 *
169 * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
170 * 	 indicates that the page is not under PV management but
171 * 	 otherwise should be treated as a normal page.  Pages not
172 * 	 under PV management cannot be paged out via the
173 * 	 object/vm_page_t because there is no knowledge of their pte
174 * 	 mappings, and such pages are also not on any PQ queue.
175 *
176 */
177#define	VPO_UNUSED01	0x01		/* --available-- */
178#define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
179#define	VPO_UNMANAGED	0x04		/* no PV management for page */
180#define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
181#define	VPO_NOSYNC	0x10		/* do not collect for syncer */
182
183/*
184 * Busy page implementation details.
185 * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
186 * even if the support for owner identity is removed because of size
187 * constraints.  Checks on lock recursion are then not possible, while the
188 * lock assertions effectiveness is someway reduced.
189 */
190#define	VPB_BIT_SHARED		0x01
191#define	VPB_BIT_EXCLUSIVE	0x02
192#define	VPB_BIT_WAITERS		0x04
193#define	VPB_BIT_FLAGMASK						\
194	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
195
196#define	VPB_SHARERS_SHIFT	3
197#define	VPB_SHARERS(x)							\
198	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
199#define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
200#define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
201
202#define	VPB_SINGLE_EXCLUSIVER	VPB_BIT_EXCLUSIVE
203
204#define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
205
206#define	PQ_NONE		255
207#define	PQ_INACTIVE	0
208#define	PQ_ACTIVE	1
209#define	PQ_COUNT	2
210
211TAILQ_HEAD(pglist, vm_page);
212SLIST_HEAD(spglist, vm_page);
213
214struct vm_pagequeue {
215	struct mtx	pq_mutex;
216	struct pglist	pq_pl;
217	int		pq_cnt;
218	u_int		* const pq_vcnt;
219	const char	* const pq_name;
220} __aligned(CACHE_LINE_SIZE);
221
222
223struct vm_domain {
224	struct vm_pagequeue vmd_pagequeues[PQ_COUNT];
225	u_int vmd_page_count;
226	u_int vmd_free_count;
227	long vmd_segs;	/* bitmask of the segments */
228	boolean_t vmd_oom;
229	int vmd_pass;	/* local pagedaemon pass */
230	int vmd_oom_seq;
231	int vmd_last_active_scan;
232	struct vm_page vmd_marker; /* marker for pagedaemon private use */
233};
234
235extern struct vm_domain vm_dom[MAXMEMDOM];
236
237#define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
238#define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
239#define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
240
241#ifdef _KERNEL
242static __inline void
243vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
244{
245
246#ifdef notyet
247	vm_pagequeue_assert_locked(pq);
248#endif
249	pq->pq_cnt += addend;
250	atomic_add_int(pq->pq_vcnt, addend);
251}
252#define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
253#define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
254#endif	/* _KERNEL */
255
256extern struct mtx_padalign vm_page_queue_free_mtx;
257extern struct mtx_padalign pa_lock[];
258
259#if defined(__arm__)
260#define	PDRSHIFT	PDR_SHIFT
261#elif !defined(PDRSHIFT)
262#define PDRSHIFT	21
263#endif
264
265#define	pa_index(pa)	((pa) >> PDRSHIFT)
266#define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
267#define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
268#define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
269#define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
270#define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
271#define	PA_UNLOCK_COND(pa) 			\
272	do {		   			\
273		if ((pa) != 0) {		\
274			PA_UNLOCK((pa));	\
275			(pa) = 0;		\
276		}				\
277	} while (0)
278
279#define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
280
281#ifdef KLD_MODULE
282#define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
283#define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
284#define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
285#else	/* !KLD_MODULE */
286#define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
287#define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
288#define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
289#define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
290#endif
291#if defined(INVARIANTS)
292#define	vm_page_assert_locked(m)		\
293    vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
294#define	vm_page_lock_assert(m, a)		\
295    vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
296#else
297#define	vm_page_assert_locked(m)
298#define	vm_page_lock_assert(m, a)
299#endif
300
301/*
302 * The vm_page's aflags are updated using atomic operations.  To set or clear
303 * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
304 * must be used.  Neither these flags nor these functions are part of the KBI.
305 *
306 * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
307 * both the MI and MD VM layers.  However, kernel loadable modules should not
308 * directly set this flag.  They should call vm_page_reference() instead.
309 *
310 * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
311 * When it does so, the object must be locked, or the page must be
312 * exclusive busied.  The MI VM layer must never access this flag
313 * directly.  Instead, it should call pmap_page_is_write_mapped().
314 *
315 * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
316 * at least one executable mapping.  It is not consumed by the MI VM layer.
317 */
318#define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
319#define	PGA_REFERENCED	0x02		/* page has been referenced */
320#define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
321
322/*
323 * Page flags.  If changed at any other time than page allocation or
324 * freeing, the modification must be protected by the vm_page lock.
325 */
326#define	PG_CACHED	0x0001		/* page is cached */
327#define	PG_FREE		0x0002		/* page is free */
328#define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
329#define	PG_ZERO		0x0008		/* page is zeroed */
330#define	PG_MARKER	0x0010		/* special queue marker page */
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
372extern int vm_page_zero_count;
373
374extern vm_page_t vm_page_array;		/* First resident page in table */
375extern long vm_page_array_size;		/* number of vm_page_t's */
376extern 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
382vm_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_NOOBJ		0x0100	/* No associated object */
393#define	VM_ALLOC_NOBUSY		0x0200	/* Do not busy the page */
394#define	VM_ALLOC_IFCACHED	0x0400	/* Fail if the page is not cached */
395#define	VM_ALLOC_IFNOTCACHED	0x0800	/* Fail if the page is cached */
396#define	VM_ALLOC_IGN_SBUSY	0x1000	/* vm_page_grab() only */
397#define	VM_ALLOC_NODUMP		0x2000	/* don't include in dump */
398#define	VM_ALLOC_SBUSY		0x4000	/* Shared busy the page */
399
400#define	VM_ALLOC_COUNT_SHIFT	16
401#define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
402
403#ifdef M_NOWAIT
404static inline int
405malloc2vm_flags(int malloc_flags)
406{
407	int pflags;
408
409	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
410	    (malloc_flags & M_NOWAIT) != 0,
411	    ("M_USE_RESERVE requires M_NOWAIT"));
412	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
413	    VM_ALLOC_SYSTEM;
414	if ((malloc_flags & M_ZERO) != 0)
415		pflags |= VM_ALLOC_ZERO;
416	if ((malloc_flags & M_NODUMP) != 0)
417		pflags |= VM_ALLOC_NODUMP;
418	return (pflags);
419}
420#endif
421
422void vm_page_busy_downgrade(vm_page_t m);
423void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
424void vm_page_flash(vm_page_t m);
425void vm_page_hold(vm_page_t mem);
426void vm_page_unhold(vm_page_t mem);
427void vm_page_free(vm_page_t m);
428void vm_page_free_zero(vm_page_t m);
429
430void vm_page_activate (vm_page_t);
431void vm_page_advise(vm_page_t m, int advice);
432vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
433vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
434    u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
435    vm_paddr_t boundary, vm_memattr_t memattr);
436vm_page_t vm_page_alloc_freelist(int, int);
437vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
438void vm_page_cache(vm_page_t);
439void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t);
440void vm_page_cache_transfer(vm_object_t, vm_pindex_t, vm_object_t);
441int vm_page_try_to_cache (vm_page_t);
442int vm_page_try_to_free (vm_page_t);
443void vm_page_deactivate (vm_page_t);
444void vm_page_dequeue(vm_page_t m);
445void vm_page_dequeue_locked(vm_page_t m);
446vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
447vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
448void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
449int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
450boolean_t vm_page_is_cached(vm_object_t object, vm_pindex_t pindex);
451vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
452vm_page_t vm_page_next(vm_page_t m);
453int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
454struct vm_pagequeue *vm_page_pagequeue(vm_page_t m);
455vm_page_t vm_page_prev(vm_page_t m);
456boolean_t vm_page_ps_is_valid(vm_page_t m);
457void vm_page_putfake(vm_page_t m);
458void vm_page_readahead_finish(vm_page_t m);
459void vm_page_reference(vm_page_t m);
460void vm_page_remove (vm_page_t);
461int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
462vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
463    vm_pindex_t pindex);
464void vm_page_requeue(vm_page_t m);
465void vm_page_requeue_locked(vm_page_t m);
466int vm_page_sbusied(vm_page_t m);
467void vm_page_set_valid_range(vm_page_t m, int base, int size);
468int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
469vm_offset_t vm_page_startup(vm_offset_t vaddr);
470void vm_page_sunbusy(vm_page_t m);
471int vm_page_trysbusy(vm_page_t m);
472void vm_page_unhold_pages(vm_page_t *ma, int count);
473void vm_page_unwire (vm_page_t, int);
474void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
475void vm_page_wire (vm_page_t);
476void vm_page_xunbusy_hard(vm_page_t m);
477void vm_page_set_validclean (vm_page_t, int, int);
478void vm_page_clear_dirty (vm_page_t, int, int);
479void vm_page_set_invalid (vm_page_t, int, int);
480int vm_page_is_valid (vm_page_t, int, int);
481void vm_page_test_dirty (vm_page_t);
482vm_page_bits_t vm_page_bits(int base, int size);
483void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
484void vm_page_free_toq(vm_page_t m);
485void vm_page_zero_idle_wakeup(void);
486
487void vm_page_dirty_KBI(vm_page_t m);
488void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
489void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
490int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
491#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
492void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
493void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
494#endif
495
496#define	vm_page_assert_sbusied(m)					\
497	KASSERT(vm_page_sbusied(m),					\
498	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
499	    (void *)m, __FILE__, __LINE__));
500
501#define	vm_page_assert_unbusied(m)					\
502	KASSERT(!vm_page_busied(m),					\
503	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
504	    (void *)m, __FILE__, __LINE__));
505
506#define	vm_page_assert_xbusied(m)					\
507	KASSERT(vm_page_xbusied(m),					\
508	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
509	    (void *)m, __FILE__, __LINE__));
510
511#define	vm_page_busied(m)						\
512	((m)->busy_lock != VPB_UNBUSIED)
513
514#define	vm_page_sbusy(m) do {						\
515	if (!vm_page_trysbusy(m))					\
516		panic("%s: page %p failed shared busing", __func__, m);	\
517} while (0)
518
519#define	vm_page_tryxbusy(m)						\
520	(atomic_cmpset_acq_int(&m->busy_lock, VPB_UNBUSIED,		\
521	    VPB_SINGLE_EXCLUSIVER))
522
523#define	vm_page_xbusied(m)						\
524	((m->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
525
526#define	vm_page_xbusy(m) do {						\
527	if (!vm_page_tryxbusy(m))					\
528		panic("%s: page %p failed exclusive busing", __func__,	\
529		    m);							\
530} while (0)
531
532#define	vm_page_xunbusy(m) do {						\
533	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
534	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
535		vm_page_xunbusy_hard(m);				\
536} while (0)
537
538#ifdef INVARIANTS
539void vm_page_object_lock_assert(vm_page_t m);
540#define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
541void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits);
542#define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
543	vm_page_assert_pga_writeable(m, bits)
544#else
545#define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
546#define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
547#endif
548
549/*
550 * We want to use atomic updates for the aflags field, which is 8 bits wide.
551 * However, not all architectures support atomic operations on 8-bit
552 * destinations.  In order that we can easily use a 32-bit operation, we
553 * require that the aflags field be 32-bit aligned.
554 */
555CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0);
556
557/*
558 *	Clear the given bits in the specified page.
559 */
560static inline void
561vm_page_aflag_clear(vm_page_t m, uint8_t bits)
562{
563	uint32_t *addr, val;
564
565	/*
566	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
567	 */
568	if ((bits & PGA_REFERENCED) != 0)
569		vm_page_assert_locked(m);
570
571	/*
572	 * Access the whole 32-bit word containing the aflags field with an
573	 * atomic update.  Parallel non-atomic updates to the other fields
574	 * within this word are handled properly by the atomic update.
575	 */
576	addr = (void *)&m->aflags;
577	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
578	    ("vm_page_aflag_clear: aflags is misaligned"));
579	val = bits;
580#if BYTE_ORDER == BIG_ENDIAN
581	val <<= 24;
582#endif
583	atomic_clear_32(addr, val);
584}
585
586/*
587 *	Set the given bits in the specified page.
588 */
589static inline void
590vm_page_aflag_set(vm_page_t m, uint8_t bits)
591{
592	uint32_t *addr, val;
593
594	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
595
596	/*
597	 * Access the whole 32-bit word containing the aflags field with an
598	 * atomic update.  Parallel non-atomic updates to the other fields
599	 * within this word are handled properly by the atomic update.
600	 */
601	addr = (void *)&m->aflags;
602	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
603	    ("vm_page_aflag_set: aflags is misaligned"));
604	val = bits;
605#if BYTE_ORDER == BIG_ENDIAN
606	val <<= 24;
607#endif
608	atomic_set_32(addr, val);
609}
610
611/*
612 *	vm_page_dirty:
613 *
614 *	Set all bits in the page's dirty field.
615 *
616 *	The object containing the specified page must be locked if the
617 *	call is made from the machine-independent layer.
618 *
619 *	See vm_page_clear_dirty_mask().
620 */
621static __inline void
622vm_page_dirty(vm_page_t m)
623{
624
625	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
626#if defined(KLD_MODULE) || defined(INVARIANTS)
627	vm_page_dirty_KBI(m);
628#else
629	m->dirty = VM_PAGE_BITS_ALL;
630#endif
631}
632
633/*
634 *	vm_page_remque:
635 *
636 *	If the given page is in a page queue, then remove it from that page
637 *	queue.
638 *
639 *	The page must be locked.
640 */
641static inline void
642vm_page_remque(vm_page_t m)
643{
644
645	if (m->queue != PQ_NONE)
646		vm_page_dequeue(m);
647}
648
649/*
650 *	vm_page_undirty:
651 *
652 *	Set page to not be dirty.  Note: does not clear pmap modify bits
653 */
654static __inline void
655vm_page_undirty(vm_page_t m)
656{
657
658	VM_PAGE_OBJECT_LOCK_ASSERT(m);
659	m->dirty = 0;
660}
661
662#endif				/* _KERNEL */
663#endif				/* !_VM_PAGE_ */
664