xref: /freebsd/sys/vm/vm_page.h (revision 38069501)
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  * 3. 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 dependent 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_LAUNDRY	2
210 #define	PQ_UNSWAPPABLE	3
211 #define	PQ_COUNT	4
212 
213 #ifndef VM_PAGE_HAVE_PGLIST
214 TAILQ_HEAD(pglist, vm_page);
215 #define VM_PAGE_HAVE_PGLIST
216 #endif
217 SLIST_HEAD(spglist, vm_page);
218 
219 struct vm_pagequeue {
220 	struct mtx	pq_mutex;
221 	struct pglist	pq_pl;
222 	int		pq_cnt;
223 	u_int		* const pq_vcnt;
224 	const char	* const pq_name;
225 } __aligned(CACHE_LINE_SIZE);
226 
227 
228 struct vm_domain {
229 	struct vm_pagequeue vmd_pagequeues[PQ_COUNT];
230 	u_int vmd_page_count;
231 	u_int vmd_free_count;
232 	long vmd_segs;	/* bitmask of the segments */
233 	boolean_t vmd_oom;
234 	int vmd_oom_seq;
235 	int vmd_last_active_scan;
236 	struct vm_page vmd_laundry_marker;
237 	struct vm_page vmd_marker; /* marker for pagedaemon private use */
238 	struct vm_page vmd_inacthead; /* marker for LRU-defeating insertions */
239 };
240 
241 extern struct vm_domain vm_dom[MAXMEMDOM];
242 
243 #define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
244 #define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
245 #define	vm_pagequeue_lockptr(pq)	(&(pq)->pq_mutex)
246 #define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
247 
248 #ifdef _KERNEL
249 extern vm_page_t bogus_page;
250 
251 static __inline void
252 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
253 {
254 
255 #ifdef notyet
256 	vm_pagequeue_assert_locked(pq);
257 #endif
258 	pq->pq_cnt += addend;
259 	atomic_add_int(pq->pq_vcnt, addend);
260 }
261 #define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
262 #define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
263 #endif	/* _KERNEL */
264 
265 extern struct mtx_padalign vm_page_queue_free_mtx;
266 extern struct mtx_padalign pa_lock[];
267 
268 #if defined(__arm__)
269 #define	PDRSHIFT	PDR_SHIFT
270 #elif !defined(PDRSHIFT)
271 #define PDRSHIFT	21
272 #endif
273 
274 #define	pa_index(pa)	((pa) >> PDRSHIFT)
275 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
276 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
277 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
278 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
279 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
280 #define	PA_UNLOCK_COND(pa) 			\
281 	do {		   			\
282 		if ((pa) != 0) {		\
283 			PA_UNLOCK((pa));	\
284 			(pa) = 0;		\
285 		}				\
286 	} while (0)
287 
288 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
289 
290 #ifdef KLD_MODULE
291 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
292 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
293 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
294 #else	/* !KLD_MODULE */
295 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
296 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
297 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
298 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
299 #endif
300 #if defined(INVARIANTS)
301 #define	vm_page_assert_locked(m)		\
302     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
303 #define	vm_page_lock_assert(m, a)		\
304     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
305 #else
306 #define	vm_page_assert_locked(m)
307 #define	vm_page_lock_assert(m, a)
308 #endif
309 
310 /*
311  * The vm_page's aflags are updated using atomic operations.  To set or clear
312  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
313  * must be used.  Neither these flags nor these functions are part of the KBI.
314  *
315  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
316  * both the MI and MD VM layers.  However, kernel loadable modules should not
317  * directly set this flag.  They should call vm_page_reference() instead.
318  *
319  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
320  * When it does so, the object must be locked, or the page must be
321  * exclusive busied.  The MI VM layer must never access this flag
322  * directly.  Instead, it should call pmap_page_is_write_mapped().
323  *
324  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
325  * at least one executable mapping.  It is not consumed by the MI VM layer.
326  */
327 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
328 #define	PGA_REFERENCED	0x02		/* page has been referenced */
329 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
330 
331 /*
332  * Page flags.  If changed at any other time than page allocation or
333  * freeing, the modification must be protected by the vm_page lock.
334  */
335 #define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
336 #define	PG_ZERO		0x0008		/* page is zeroed */
337 #define	PG_MARKER	0x0010		/* special queue marker page */
338 #define	PG_NODUMP	0x0080		/* don't include this page in a dump */
339 #define	PG_UNHOLDFREE	0x0100		/* delayed free of a held page */
340 
341 /*
342  * Misc constants.
343  */
344 #define ACT_DECLINE		1
345 #define ACT_ADVANCE		3
346 #define ACT_INIT		5
347 #define ACT_MAX			64
348 
349 #ifdef _KERNEL
350 
351 #include <sys/systm.h>
352 
353 #include <machine/atomic.h>
354 
355 /*
356  * Each pageable resident page falls into one of five lists:
357  *
358  *	free
359  *		Available for allocation now.
360  *
361  *	inactive
362  *		Low activity, candidates for reclamation.
363  *		This list is approximately LRU ordered.
364  *
365  *	laundry
366  *		This is the list of pages that should be
367  *		paged out next.
368  *
369  *	unswappable
370  *		Dirty anonymous pages that cannot be paged
371  *		out because no swap device is configured.
372  *
373  *	active
374  *		Pages that are "active", i.e., they have been
375  *		recently referenced.
376  *
377  */
378 
379 extern int vm_page_zero_count;
380 
381 extern vm_page_t vm_page_array;		/* First resident page in table */
382 extern long vm_page_array_size;		/* number of vm_page_t's */
383 extern long first_page;			/* first physical page number */
384 
385 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
386 
387 /*
388  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
389  * page to which the given physical address belongs. The correct vm_page_t
390  * object is returned for addresses that are not page-aligned.
391  */
392 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
393 
394 /*
395  * Page allocation parameters for vm_page for the functions
396  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
397  * vm_page_alloc_freelist().  Some functions support only a subset
398  * of the flags, and ignore others, see the flags legend.
399  *
400  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
401  * and the vm_page_grab*() functions.  See these functions for details.
402  *
403  * Bits 0 - 1 define class.
404  * Bits 2 - 15 dedicated for flags.
405  * Legend:
406  * (a) - vm_page_alloc() supports the flag.
407  * (c) - vm_page_alloc_contig() supports the flag.
408  * (f) - vm_page_alloc_freelist() supports the flag.
409  * (g) - vm_page_grab() supports the flag.
410  * (p) - vm_page_grab_pages() supports the flag.
411  * Bits above 15 define the count of additional pages that the caller
412  * intends to allocate.
413  */
414 #define VM_ALLOC_NORMAL		0
415 #define VM_ALLOC_INTERRUPT	1
416 #define VM_ALLOC_SYSTEM		2
417 #define	VM_ALLOC_CLASS_MASK	3
418 #define	VM_ALLOC_WIRED		0x0020	/* (acfgp) Allocate a wired page */
419 #define	VM_ALLOC_ZERO		0x0040	/* (acfgp) Allocate a prezeroed page */
420 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
421 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
422 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
423 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
424 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
425 #define	VM_ALLOC_NOWAIT		0x8000	/* (gp) Do not sleep */
426 #define	VM_ALLOC_COUNT_SHIFT	16
427 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
428 
429 #ifdef M_NOWAIT
430 static inline int
431 malloc2vm_flags(int malloc_flags)
432 {
433 	int pflags;
434 
435 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
436 	    (malloc_flags & M_NOWAIT) != 0,
437 	    ("M_USE_RESERVE requires M_NOWAIT"));
438 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
439 	    VM_ALLOC_SYSTEM;
440 	if ((malloc_flags & M_ZERO) != 0)
441 		pflags |= VM_ALLOC_ZERO;
442 	if ((malloc_flags & M_NODUMP) != 0)
443 		pflags |= VM_ALLOC_NODUMP;
444 	return (pflags);
445 }
446 #endif
447 
448 /*
449  * Predicates supported by vm_page_ps_test():
450  *
451  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
452  *	However, it can be spuriously false when the (super)page has become
453  *	dirty in the pmap but that information has not been propagated to the
454  *	machine-independent layer.
455  */
456 #define	PS_ALL_DIRTY	0x1
457 #define	PS_ALL_VALID	0x2
458 #define	PS_NONE_BUSY	0x4
459 
460 void vm_page_busy_downgrade(vm_page_t m);
461 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
462 void vm_page_flash(vm_page_t m);
463 void vm_page_hold(vm_page_t mem);
464 void vm_page_unhold(vm_page_t mem);
465 void vm_page_free(vm_page_t m);
466 void vm_page_free_zero(vm_page_t m);
467 
468 void vm_page_activate (vm_page_t);
469 void vm_page_advise(vm_page_t m, int advice);
470 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
471 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
472 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
473     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
474     vm_paddr_t boundary, vm_memattr_t memattr);
475 vm_page_t vm_page_alloc_freelist(int, int);
476 void vm_page_change_lock(vm_page_t m, struct mtx **mtx);
477 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
478 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
479     vm_page_t *ma, int count);
480 void vm_page_deactivate (vm_page_t);
481 void vm_page_deactivate_noreuse(vm_page_t);
482 void vm_page_dequeue(vm_page_t m);
483 void vm_page_dequeue_locked(vm_page_t m);
484 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
485 void vm_page_free_phys_pglist(struct pglist *tq);
486 bool vm_page_free_prep(vm_page_t m, bool pagequeue_locked);
487 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
488 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
489 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
490 void vm_page_launder(vm_page_t m);
491 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
492 vm_page_t vm_page_next(vm_page_t m);
493 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
494 struct vm_pagequeue *vm_page_pagequeue(vm_page_t m);
495 vm_page_t vm_page_prev(vm_page_t m);
496 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
497 void vm_page_putfake(vm_page_t m);
498 void vm_page_readahead_finish(vm_page_t m);
499 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
500     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
501 void vm_page_reference(vm_page_t m);
502 void vm_page_remove (vm_page_t);
503 int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
504 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
505     vm_pindex_t pindex);
506 void vm_page_requeue(vm_page_t m);
507 void vm_page_requeue_locked(vm_page_t m);
508 int vm_page_sbusied(vm_page_t m);
509 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
510     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
511 void vm_page_set_valid_range(vm_page_t m, int base, int size);
512 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
513 vm_offset_t vm_page_startup(vm_offset_t vaddr);
514 void vm_page_sunbusy(vm_page_t m);
515 bool vm_page_try_to_free(vm_page_t m);
516 int vm_page_trysbusy(vm_page_t m);
517 void vm_page_unhold_pages(vm_page_t *ma, int count);
518 void vm_page_unswappable(vm_page_t m);
519 boolean_t vm_page_unwire(vm_page_t m, uint8_t queue);
520 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
521 void vm_page_wire (vm_page_t);
522 void vm_page_xunbusy_hard(vm_page_t m);
523 void vm_page_xunbusy_maybelocked(vm_page_t m);
524 void vm_page_set_validclean (vm_page_t, int, int);
525 void vm_page_clear_dirty (vm_page_t, int, int);
526 void vm_page_set_invalid (vm_page_t, int, int);
527 int vm_page_is_valid (vm_page_t, int, int);
528 void vm_page_test_dirty (vm_page_t);
529 vm_page_bits_t vm_page_bits(int base, int size);
530 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
531 void vm_page_free_toq(vm_page_t m);
532 
533 void vm_page_dirty_KBI(vm_page_t m);
534 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
535 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
536 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
537 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
538 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
539 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
540 #endif
541 
542 #define	vm_page_assert_sbusied(m)					\
543 	KASSERT(vm_page_sbusied(m),					\
544 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
545 	    (m), __FILE__, __LINE__))
546 
547 #define	vm_page_assert_unbusied(m)					\
548 	KASSERT(!vm_page_busied(m),					\
549 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
550 	    (m), __FILE__, __LINE__))
551 
552 #define	vm_page_assert_xbusied(m)					\
553 	KASSERT(vm_page_xbusied(m),					\
554 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
555 	    (m), __FILE__, __LINE__))
556 
557 #define	vm_page_busied(m)						\
558 	((m)->busy_lock != VPB_UNBUSIED)
559 
560 #define	vm_page_sbusy(m) do {						\
561 	if (!vm_page_trysbusy(m))					\
562 		panic("%s: page %p failed shared busying", __func__,	\
563 		    (m));						\
564 } while (0)
565 
566 #define	vm_page_tryxbusy(m)						\
567 	(atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED,		\
568 	    VPB_SINGLE_EXCLUSIVER))
569 
570 #define	vm_page_xbusied(m)						\
571 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
572 
573 #define	vm_page_xbusy(m) do {						\
574 	if (!vm_page_tryxbusy(m))					\
575 		panic("%s: page %p failed exclusive busying", __func__,	\
576 		    (m));						\
577 } while (0)
578 
579 /* Note: page m's lock must not be owned by the caller. */
580 #define	vm_page_xunbusy(m) do {						\
581 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
582 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
583 		vm_page_xunbusy_hard(m);				\
584 } while (0)
585 
586 #ifdef INVARIANTS
587 void vm_page_object_lock_assert(vm_page_t m);
588 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
589 void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits);
590 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
591 	vm_page_assert_pga_writeable(m, bits)
592 #else
593 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
594 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
595 #endif
596 
597 /*
598  * We want to use atomic updates for the aflags field, which is 8 bits wide.
599  * However, not all architectures support atomic operations on 8-bit
600  * destinations.  In order that we can easily use a 32-bit operation, we
601  * require that the aflags field be 32-bit aligned.
602  */
603 CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0);
604 
605 /*
606  *	Clear the given bits in the specified page.
607  */
608 static inline void
609 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
610 {
611 	uint32_t *addr, val;
612 
613 	/*
614 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
615 	 */
616 	if ((bits & PGA_REFERENCED) != 0)
617 		vm_page_assert_locked(m);
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_clear: aflags is misaligned"));
627 	val = bits;
628 #if BYTE_ORDER == BIG_ENDIAN
629 	val <<= 24;
630 #endif
631 	atomic_clear_32(addr, val);
632 }
633 
634 /*
635  *	Set the given bits in the specified page.
636  */
637 static inline void
638 vm_page_aflag_set(vm_page_t m, uint8_t bits)
639 {
640 	uint32_t *addr, val;
641 
642 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
643 
644 	/*
645 	 * Access the whole 32-bit word containing the aflags field with an
646 	 * atomic update.  Parallel non-atomic updates to the other fields
647 	 * within this word are handled properly by the atomic update.
648 	 */
649 	addr = (void *)&m->aflags;
650 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
651 	    ("vm_page_aflag_set: aflags is misaligned"));
652 	val = bits;
653 #if BYTE_ORDER == BIG_ENDIAN
654 	val <<= 24;
655 #endif
656 	atomic_set_32(addr, val);
657 }
658 
659 /*
660  *	vm_page_dirty:
661  *
662  *	Set all bits in the page's dirty field.
663  *
664  *	The object containing the specified page must be locked if the
665  *	call is made from the machine-independent layer.
666  *
667  *	See vm_page_clear_dirty_mask().
668  */
669 static __inline void
670 vm_page_dirty(vm_page_t m)
671 {
672 
673 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
674 #if defined(KLD_MODULE) || defined(INVARIANTS)
675 	vm_page_dirty_KBI(m);
676 #else
677 	m->dirty = VM_PAGE_BITS_ALL;
678 #endif
679 }
680 
681 /*
682  *	vm_page_remque:
683  *
684  *	If the given page is in a page queue, then remove it from that page
685  *	queue.
686  *
687  *	The page must be locked.
688  */
689 static inline void
690 vm_page_remque(vm_page_t m)
691 {
692 
693 	if (m->queue != PQ_NONE)
694 		vm_page_dequeue(m);
695 }
696 
697 /*
698  *	vm_page_undirty:
699  *
700  *	Set page to not be dirty.  Note: does not clear pmap modify bits
701  */
702 static __inline void
703 vm_page_undirty(vm_page_t m)
704 {
705 
706 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
707 	m->dirty = 0;
708 }
709 
710 static inline void
711 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
712     vm_page_t mold)
713 {
714 	vm_page_t mret;
715 
716 	mret = vm_page_replace(mnew, object, pindex);
717 	KASSERT(mret == mold,
718 	    ("invalid page replacement, mold=%p, mret=%p", mold, mret));
719 
720 	/* Unused if !INVARIANTS. */
721 	(void)mold;
722 	(void)mret;
723 }
724 
725 static inline bool
726 vm_page_active(vm_page_t m)
727 {
728 
729 	return (m->queue == PQ_ACTIVE);
730 }
731 
732 static inline bool
733 vm_page_inactive(vm_page_t m)
734 {
735 
736 	return (m->queue == PQ_INACTIVE);
737 }
738 
739 static inline bool
740 vm_page_in_laundry(vm_page_t m)
741 {
742 
743 	return (m->queue == PQ_LAUNDRY || m->queue == PQ_UNSWAPPABLE);
744 }
745 
746 #endif				/* _KERNEL */
747 #endif				/* !_VM_PAGE_ */
748