xref: /freebsd/sys/vm/vm_page.h (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
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 psind;			/* pagesizes[] index (O) */
153 	int8_t segind;
154 	uint8_t	order;			/* index of the buddy queue */
155 	uint8_t pool;
156 	u_char	act_count;		/* page usage count (P) */
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 };
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 
211 TAILQ_HEAD(pglist, vm_page);
212 SLIST_HEAD(spglist, vm_page);
213 
214 struct vm_pagequeue {
215 	struct mtx	pq_mutex;
216 	struct pglist	pq_pl;
217 	int		pq_cnt;
218 	int		* const pq_vcnt;
219 	const char	* const pq_name;
220 } __aligned(CACHE_LINE_SIZE);
221 
222 
223 struct 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 	struct vm_page vmd_inacthead; /* marker for LRU-defeating insertions */
234 };
235 
236 extern struct vm_domain vm_dom[MAXMEMDOM];
237 
238 #define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
239 #define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
240 #define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
241 
242 #ifdef _KERNEL
243 static __inline void
244 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
245 {
246 
247 #ifdef notyet
248 	vm_pagequeue_assert_locked(pq);
249 #endif
250 	pq->pq_cnt += addend;
251 	atomic_add_int(pq->pq_vcnt, addend);
252 }
253 #define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
254 #define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
255 #endif	/* _KERNEL */
256 
257 extern struct mtx_padalign vm_page_queue_free_mtx;
258 extern struct mtx_padalign pa_lock[];
259 
260 #if defined(__arm__)
261 #define	PDRSHIFT	PDR_SHIFT
262 #elif !defined(PDRSHIFT)
263 #define PDRSHIFT	21
264 #endif
265 
266 #define	pa_index(pa)	((pa) >> PDRSHIFT)
267 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
268 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
269 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
270 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
271 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
272 #define	PA_UNLOCK_COND(pa) 			\
273 	do {		   			\
274 		if ((pa) != 0) {		\
275 			PA_UNLOCK((pa));	\
276 			(pa) = 0;		\
277 		}				\
278 	} while (0)
279 
280 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
281 
282 #ifdef KLD_MODULE
283 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
284 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
285 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
286 #else	/* !KLD_MODULE */
287 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
288 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
289 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
290 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
291 #endif
292 #if defined(INVARIANTS)
293 #define	vm_page_assert_locked(m)		\
294     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
295 #define	vm_page_lock_assert(m, a)		\
296     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
297 #else
298 #define	vm_page_assert_locked(m)
299 #define	vm_page_lock_assert(m, a)
300 #endif
301 
302 /*
303  * The vm_page's aflags are updated using atomic operations.  To set or clear
304  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
305  * must be used.  Neither these flags nor these functions are part of the KBI.
306  *
307  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
308  * both the MI and MD VM layers.  However, kernel loadable modules should not
309  * directly set this flag.  They should call vm_page_reference() instead.
310  *
311  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
312  * When it does so, the object must be locked, or the page must be
313  * exclusive busied.  The MI VM layer must never access this flag
314  * directly.  Instead, it should call pmap_page_is_write_mapped().
315  *
316  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
317  * at least one executable mapping.  It is not consumed by the MI VM layer.
318  */
319 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
320 #define	PGA_REFERENCED	0x02		/* page has been referenced */
321 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
322 
323 /*
324  * Page flags.  If changed at any other time than page allocation or
325  * freeing, the modification must be protected by the vm_page lock.
326  */
327 #define	PG_CACHED	0x0001		/* page is cached */
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 
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_TO_PHYS(entry)	((entry)->phys_addr)
379 
380 /*
381  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
382  * page to which the given physical address belongs. The correct vm_page_t
383  * object is returned for addresses that are not page-aligned.
384  */
385 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
386 
387 /*
388  * Page allocation parameters for vm_page for the functions
389  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
390  * vm_page_alloc_freelist().  Some functions support only a subset
391  * of the flags, and ignore others, see the flags legend.
392  *
393  * Bits 0 - 1 define class.
394  * Bits 2 - 15 dedicated for flags.
395  * Legend:
396  * (a) - vm_page_alloc() supports the flag.
397  * (c) - vm_page_alloc_contig() supports the flag.
398  * (f) - vm_page_alloc_freelist() supports the flag.
399  * (g) - vm_page_grab() supports the flag.
400  * Bits above 15 define the count of additional pages that the caller
401  * intends to allocate.
402  */
403 #define VM_ALLOC_NORMAL		0
404 #define VM_ALLOC_INTERRUPT	1
405 #define VM_ALLOC_SYSTEM		2
406 #define	VM_ALLOC_CLASS_MASK	3
407 #define	VM_ALLOC_WIRED		0x0020	/* (acfg) Allocate non pageable page */
408 #define	VM_ALLOC_ZERO		0x0040	/* (acfg) Try to obtain a zeroed page */
409 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
410 #define	VM_ALLOC_NOBUSY		0x0200	/* (acg) Do not busy the page */
411 #define	VM_ALLOC_IFCACHED	0x0400	/* (ag) Fail if page is not cached */
412 #define	VM_ALLOC_IFNOTCACHED	0x0800	/* (ag) Fail if page is cached */
413 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (g) Ignore shared busy flag */
414 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
415 #define	VM_ALLOC_SBUSY		0x4000	/* (acg) Shared busy the page */
416 #define	VM_ALLOC_NOWAIT		0x8000	/* (g) Do not sleep, return NULL */
417 #define	VM_ALLOC_COUNT_SHIFT	16
418 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
419 
420 #ifdef M_NOWAIT
421 static inline int
422 malloc2vm_flags(int malloc_flags)
423 {
424 	int pflags;
425 
426 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
427 	    (malloc_flags & M_NOWAIT) != 0,
428 	    ("M_USE_RESERVE requires M_NOWAIT"));
429 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
430 	    VM_ALLOC_SYSTEM;
431 	if ((malloc_flags & M_ZERO) != 0)
432 		pflags |= VM_ALLOC_ZERO;
433 	if ((malloc_flags & M_NODUMP) != 0)
434 		pflags |= VM_ALLOC_NODUMP;
435 	return (pflags);
436 }
437 #endif
438 
439 void vm_page_busy_downgrade(vm_page_t m);
440 void vm_page_busy_sleep(vm_page_t m, const char *msg);
441 void vm_page_flash(vm_page_t m);
442 void vm_page_hold(vm_page_t mem);
443 void vm_page_unhold(vm_page_t mem);
444 void vm_page_free(vm_page_t m);
445 void vm_page_free_zero(vm_page_t m);
446 
447 void vm_page_activate (vm_page_t);
448 void vm_page_advise(vm_page_t m, int advice);
449 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
450 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
451     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
452     vm_paddr_t boundary, vm_memattr_t memattr);
453 vm_page_t vm_page_alloc_freelist(int, int);
454 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
455 void vm_page_cache(vm_page_t);
456 void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t);
457 void vm_page_cache_transfer(vm_object_t, vm_pindex_t, vm_object_t);
458 int vm_page_try_to_cache (vm_page_t);
459 int vm_page_try_to_free (vm_page_t);
460 void vm_page_deactivate (vm_page_t);
461 void vm_page_deactivate_noreuse(vm_page_t);
462 void vm_page_dequeue(vm_page_t m);
463 void vm_page_dequeue_locked(vm_page_t m);
464 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
465 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
466 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
467 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
468 boolean_t vm_page_is_cached(vm_object_t object, vm_pindex_t pindex);
469 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
470 vm_page_t vm_page_next(vm_page_t m);
471 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
472 struct vm_pagequeue *vm_page_pagequeue(vm_page_t m);
473 vm_page_t vm_page_prev(vm_page_t m);
474 boolean_t vm_page_ps_is_valid(vm_page_t m);
475 void vm_page_putfake(vm_page_t m);
476 void vm_page_readahead_finish(vm_page_t m);
477 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
478     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
479 void vm_page_reference(vm_page_t m);
480 void vm_page_remove (vm_page_t);
481 int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
482 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
483     vm_pindex_t pindex);
484 void vm_page_requeue(vm_page_t m);
485 void vm_page_requeue_locked(vm_page_t m);
486 int vm_page_sbusied(vm_page_t m);
487 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
488     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
489 void vm_page_set_valid_range(vm_page_t m, int base, int size);
490 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
491 vm_offset_t vm_page_startup(vm_offset_t vaddr);
492 void vm_page_sunbusy(vm_page_t m);
493 int vm_page_trysbusy(vm_page_t m);
494 void vm_page_unhold_pages(vm_page_t *ma, int count);
495 boolean_t vm_page_unwire(vm_page_t m, uint8_t queue);
496 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
497 void vm_page_wire (vm_page_t);
498 void vm_page_xunbusy_hard(vm_page_t m);
499 void vm_page_set_validclean (vm_page_t, int, int);
500 void vm_page_clear_dirty (vm_page_t, int, int);
501 void vm_page_set_invalid (vm_page_t, int, int);
502 int vm_page_is_valid (vm_page_t, int, int);
503 void vm_page_test_dirty (vm_page_t);
504 vm_page_bits_t vm_page_bits(int base, int size);
505 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
506 void vm_page_free_toq(vm_page_t m);
507 void vm_page_zero_idle_wakeup(void);
508 
509 void vm_page_dirty_KBI(vm_page_t m);
510 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
511 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
512 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
513 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
514 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
515 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
516 #endif
517 
518 #define	vm_page_assert_sbusied(m)					\
519 	KASSERT(vm_page_sbusied(m),					\
520 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
521 	    (m), __FILE__, __LINE__))
522 
523 #define	vm_page_assert_unbusied(m)					\
524 	KASSERT(!vm_page_busied(m),					\
525 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
526 	    (m), __FILE__, __LINE__))
527 
528 #define	vm_page_assert_xbusied(m)					\
529 	KASSERT(vm_page_xbusied(m),					\
530 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
531 	    (m), __FILE__, __LINE__))
532 
533 #define	vm_page_busied(m)						\
534 	((m)->busy_lock != VPB_UNBUSIED)
535 
536 #define	vm_page_sbusy(m) do {						\
537 	if (!vm_page_trysbusy(m))					\
538 		panic("%s: page %p failed shared busying", __func__,	\
539 		    (m));						\
540 } while (0)
541 
542 #define	vm_page_tryxbusy(m)						\
543 	(atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED,		\
544 	    VPB_SINGLE_EXCLUSIVER))
545 
546 #define	vm_page_xbusied(m)						\
547 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
548 
549 #define	vm_page_xbusy(m) do {						\
550 	if (!vm_page_tryxbusy(m))					\
551 		panic("%s: page %p failed exclusive busying", __func__,	\
552 		    (m));						\
553 } while (0)
554 
555 #define	vm_page_xunbusy(m) do {						\
556 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
557 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
558 		vm_page_xunbusy_hard(m);				\
559 } while (0)
560 
561 #ifdef INVARIANTS
562 void vm_page_object_lock_assert(vm_page_t m);
563 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
564 void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits);
565 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
566 	vm_page_assert_pga_writeable(m, bits)
567 #else
568 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
569 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
570 #endif
571 
572 /*
573  * We want to use atomic updates for the aflags field, which is 8 bits wide.
574  * However, not all architectures support atomic operations on 8-bit
575  * destinations.  In order that we can easily use a 32-bit operation, we
576  * require that the aflags field be 32-bit aligned.
577  */
578 CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0);
579 
580 /*
581  *	Clear the given bits in the specified page.
582  */
583 static inline void
584 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
585 {
586 	uint32_t *addr, val;
587 
588 	/*
589 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
590 	 */
591 	if ((bits & PGA_REFERENCED) != 0)
592 		vm_page_assert_locked(m);
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_clear: aflags is misaligned"));
602 	val = bits;
603 #if BYTE_ORDER == BIG_ENDIAN
604 	val <<= 24;
605 #endif
606 	atomic_clear_32(addr, val);
607 }
608 
609 /*
610  *	Set the given bits in the specified page.
611  */
612 static inline void
613 vm_page_aflag_set(vm_page_t m, uint8_t bits)
614 {
615 	uint32_t *addr, val;
616 
617 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
618 
619 	/*
620 	 * Access the whole 32-bit word containing the aflags field with an
621 	 * atomic update.  Parallel non-atomic updates to the other fields
622 	 * within this word are handled properly by the atomic update.
623 	 */
624 	addr = (void *)&m->aflags;
625 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
626 	    ("vm_page_aflag_set: aflags is misaligned"));
627 	val = bits;
628 #if BYTE_ORDER == BIG_ENDIAN
629 	val <<= 24;
630 #endif
631 	atomic_set_32(addr, val);
632 }
633 
634 /*
635  *	vm_page_dirty:
636  *
637  *	Set all bits in the page's dirty field.
638  *
639  *	The object containing the specified page must be locked if the
640  *	call is made from the machine-independent layer.
641  *
642  *	See vm_page_clear_dirty_mask().
643  */
644 static __inline void
645 vm_page_dirty(vm_page_t m)
646 {
647 
648 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
649 #if defined(KLD_MODULE) || defined(INVARIANTS)
650 	vm_page_dirty_KBI(m);
651 #else
652 	m->dirty = VM_PAGE_BITS_ALL;
653 #endif
654 }
655 
656 /*
657  *	vm_page_remque:
658  *
659  *	If the given page is in a page queue, then remove it from that page
660  *	queue.
661  *
662  *	The page must be locked.
663  */
664 static inline void
665 vm_page_remque(vm_page_t m)
666 {
667 
668 	if (m->queue != PQ_NONE)
669 		vm_page_dequeue(m);
670 }
671 
672 /*
673  *	vm_page_undirty:
674  *
675  *	Set page to not be dirty.  Note: does not clear pmap modify bits
676  */
677 static __inline void
678 vm_page_undirty(vm_page_t m)
679 {
680 
681 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
682 	m->dirty = 0;
683 }
684 
685 static inline void
686 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
687     vm_page_t mold)
688 {
689 	vm_page_t mret;
690 
691 	mret = vm_page_replace(mnew, object, pindex);
692 	KASSERT(mret == mold,
693 	    ("invalid page replacement, mold=%p, mret=%p", mold, mret));
694 
695 	/* Unused if !INVARIANTS. */
696 	(void)mold;
697 	(void)mret;
698 }
699 
700 #endif				/* _KERNEL */
701 #endif				/* !_VM_PAGE_ */
702