xref: /freebsd/sys/vm/vm_page.h (revision 0440b3d2cbf83afb55209a9938b31a48912f222c)
1 /*-
2  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * The Mach Operating System project at Carnegie-Mellon University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
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 
61 /*
62  *	Resident memory system definitions.
63  */
64 
65 #ifndef	_VM_PAGE_
66 #define	_VM_PAGE_
67 
68 #include <vm/pmap.h>
69 #include <vm/_vm_phys.h>
70 
71 /*
72  *	Management of resident (logical) pages.
73  *
74  *	A small structure is kept for each resident
75  *	page, indexed by page number.  Each structure
76  *	is an element of several collections:
77  *
78  *		A radix tree used to quickly
79  *		perform object/offset lookups
80  *
81  *		An ordered list of pages due for pageout.
82  *
83  *	In addition, the structure contains the object
84  *	and offset to which this page belongs (for pageout),
85  *	and sundry status bits.
86  *
87  *	In general, operations on this structure's mutable fields are
88  *	synchronized using either one of or a combination of locks.  If a
89  *	field is annotated with two of these locks then holding either is
90  *	sufficient for read access but both are required for write access.
91  *	The queue lock for a page depends on the value of its queue field and is
92  *	described in detail below.
93  *
94  *	The following annotations are possible:
95  *	(A) the field must be accessed using atomic(9) and may require
96  *	    additional synchronization.
97  *	(B) the page busy lock.
98  *	(C) the field is immutable.
99  *	(F) the per-domain lock for the free queues.
100  *	(M) Machine dependent, defined by pmap layer.
101  *	(O) the object that the page belongs to.
102  *	(Q) the page's queue lock.
103  *
104  *	The busy lock is an embedded reader-writer lock that protects the
105  *	page's contents and identity (i.e., its <object, pindex> tuple) as
106  *	well as certain valid/dirty modifications.  To avoid bloating the
107  *	the page structure, the busy lock lacks some of the features available
108  *	the kernel's general-purpose synchronization primitives.  As a result,
109  *	busy lock ordering rules are not verified, lock recursion is not
110  *	detected, and an attempt to xbusy a busy page or sbusy an xbusy page
111  *	results will trigger a panic rather than causing the thread to block.
112  *	vm_page_sleep_if_busy() can be used to sleep until the page's busy
113  *	state changes, after which the caller must re-lookup the page and
114  *	re-evaluate its state.  vm_page_busy_acquire() will block until
115  *	the lock is acquired.
116  *
117  *	The valid field is protected by the page busy lock (B) and object
118  *	lock (O).  Transitions from invalid to valid are generally done
119  *	via I/O or zero filling and do not require the object lock.
120  *	These must be protected with the busy lock to prevent page-in or
121  *	creation races.  Page invalidation generally happens as a result
122  *	of truncate or msync.  When invalidated, pages must not be present
123  *	in pmap and must hold the object lock to prevent concurrent
124  *	speculative read-only mappings that do not require busy.  I/O
125  *	routines may check for validity without a lock if they are prepared
126  *	to handle invalidation races with higher level locks (vnode) or are
127  *	unconcerned with races so long as they hold a reference to prevent
128  *	recycling.  When a valid bit is set while holding a shared busy
129  *	lock (A) atomic operations are used to protect against concurrent
130  *	modification.
131  *
132  *	In contrast, the synchronization of accesses to the page's
133  *	dirty field is a mix of machine dependent (M) and busy (B).  In
134  *	the machine-independent layer, the page busy must be held to
135  *	operate on the field.  However, the pmap layer is permitted to
136  *	set all bits within the field without holding that lock.  If the
137  *	underlying architecture does not support atomic read-modify-write
138  *	operations on the field's type, then the machine-independent
139  *	layer uses a 32-bit atomic on the aligned 32-bit word that
140  *	contains the dirty field.  In the machine-independent layer,
141  *	the implementation of read-modify-write operations on the
142  *	field is encapsulated in vm_page_clear_dirty_mask().  An
143  *	exclusive busy lock combined with pmap_remove_{write/all}() is the
144  *	only way to ensure a page can not become dirty.  I/O generally
145  *	removes the page from pmap to ensure exclusive access and atomic
146  *	writes.
147  *
148  *	The ref_count field tracks references to the page.  References that
149  *	prevent the page from being reclaimable are called wirings and are
150  *	counted in the low bits of ref_count.  The containing object's
151  *	reference, if one exists, is counted using the VPRC_OBJREF bit in the
152  *	ref_count field.  Additionally, the VPRC_BLOCKED bit is used to
153  *	atomically check for wirings and prevent new wirings via
154  *	pmap_extract_and_hold().  When a page belongs to an object, it may be
155  *	wired only when the object is locked, or the page is busy, or by
156  *	pmap_extract_and_hold().  As a result, if the object is locked and the
157  *	page is not busy (or is exclusively busied by the current thread), and
158  *	the page is unmapped, its wire count will not increase.  The ref_count
159  *	field is updated using atomic operations in most cases, except when it
160  *	is known that no other references to the page exist, such as in the page
161  *	allocator.  A page may be present in the page queues, or even actively
162  *	scanned by the page daemon, without an explicitly counted referenced.
163  *	The page daemon must therefore handle the possibility of a concurrent
164  *	free of the page.
165  *
166  *	The queue state of a page consists of the queue and act_count fields of
167  *	its atomically updated state, and the subset of atomic flags specified
168  *	by PGA_QUEUE_STATE_MASK.  The queue field contains the page's page queue
169  *	index, or PQ_NONE if it does not belong to a page queue.  To modify the
170  *	queue field, the page queue lock corresponding to the old value must be
171  *	held, unless that value is PQ_NONE, in which case the queue index must
172  *	be updated using an atomic RMW operation.  There is one exception to
173  *	this rule: the page daemon may transition the queue field from
174  *	PQ_INACTIVE to PQ_NONE immediately prior to freeing the page during an
175  *	inactive queue scan.  At that point the page is already dequeued and no
176  *	other references to that vm_page structure can exist.  The PGA_ENQUEUED
177  *	flag, when set, indicates that the page structure is physically inserted
178  *	into the queue corresponding to the page's queue index, and may only be
179  *	set or cleared with the corresponding page queue lock held.
180  *
181  *	To avoid contention on page queue locks, page queue operations (enqueue,
182  *	dequeue, requeue) are batched using fixed-size per-CPU queues.  A
183  *	deferred operation is requested by setting one of the flags in
184  *	PGA_QUEUE_OP_MASK and inserting an entry into a batch queue.  When a
185  *	queue is full, an attempt to insert a new entry will lock the page
186  *	queues and trigger processing of the pending entries.  The
187  *	type-stability of vm_page structures is crucial to this scheme since the
188  *	processing of entries in a given batch queue may be deferred
189  *	indefinitely.  In particular, a page may be freed with pending batch
190  *	queue entries.  The page queue operation flags must be set using atomic
191  *	RWM operations.
192  */
193 
194 #if PAGE_SIZE == 4096
195 #define VM_PAGE_BITS_ALL 0xffu
196 typedef uint8_t vm_page_bits_t;
197 #elif PAGE_SIZE == 8192
198 #define VM_PAGE_BITS_ALL 0xffffu
199 typedef uint16_t vm_page_bits_t;
200 #elif PAGE_SIZE == 16384
201 #define VM_PAGE_BITS_ALL 0xffffffffu
202 typedef uint32_t vm_page_bits_t;
203 #elif PAGE_SIZE == 32768
204 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
205 typedef uint64_t vm_page_bits_t;
206 #endif
207 
208 typedef union vm_page_astate {
209 	struct {
210 		uint16_t flags;
211 		uint8_t	queue;
212 		uint8_t act_count;
213 	};
214 	uint32_t _bits;
215 } vm_page_astate_t;
216 
217 struct vm_page {
218 	union {
219 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
220 		struct {
221 			SLIST_ENTRY(vm_page) ss; /* private slists */
222 		} s;
223 		struct {
224 			u_long p;
225 			u_long v;
226 		} memguard;
227 		struct {
228 			void *slab;
229 			void *zone;
230 		} uma;
231 	} plinks;
232 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
233 	vm_object_t object;		/* which object am I in (O) */
234 	vm_pindex_t pindex;		/* offset into object (O,P) */
235 	vm_paddr_t phys_addr;		/* physical address of page (C) */
236 	struct md_page md;		/* machine dependent stuff */
237 	u_int ref_count;		/* page references (A) */
238 	u_int busy_lock;		/* busy owners lock (A) */
239 	union vm_page_astate a;		/* state accessed atomically (A) */
240 	uint8_t order;			/* index of the buddy queue (F) */
241 	uint8_t pool;			/* vm_phys freepool index (F) */
242 	uint8_t flags;			/* page PG_* flags (P) */
243 	uint8_t oflags;			/* page VPO_* flags (O) */
244 	int8_t psind;			/* pagesizes[] index (O) */
245 	int8_t segind;			/* vm_phys segment index (C) */
246 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
247 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
248 	vm_page_bits_t valid;		/* valid DEV_BSIZE chunk map (O,B) */
249 	vm_page_bits_t dirty;		/* dirty DEV_BSIZE chunk map (M,B) */
250 };
251 
252 /*
253  * Special bits used in the ref_count field.
254  *
255  * ref_count is normally used to count wirings that prevent the page from being
256  * reclaimed, but also supports several special types of references that do not
257  * prevent reclamation.  Accesses to the ref_count field must be atomic unless
258  * the page is unallocated.
259  *
260  * VPRC_OBJREF is the reference held by the containing object.  It can set or
261  * cleared only when the corresponding object's write lock is held.
262  *
263  * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while
264  * attempting to tear down all mappings of a given page.  The page busy lock and
265  * object write lock must both be held in order to set or clear this bit.
266  */
267 #define	VPRC_BLOCKED	0x40000000u	/* mappings are being removed */
268 #define	VPRC_OBJREF	0x80000000u	/* object reference, cleared with (O) */
269 #define	VPRC_WIRE_COUNT(c)	((c) & ~(VPRC_BLOCKED | VPRC_OBJREF))
270 #define	VPRC_WIRE_COUNT_MAX	(~(VPRC_BLOCKED | VPRC_OBJREF))
271 
272 /*
273  * Page flags stored in oflags:
274  *
275  * Access to these page flags is synchronized by the lock on the object
276  * containing the page (O).
277  *
278  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
279  * 	 indicates that the page is not under PV management but
280  * 	 otherwise should be treated as a normal page.  Pages not
281  * 	 under PV management cannot be paged out via the
282  * 	 object/vm_page_t because there is no knowledge of their pte
283  * 	 mappings, and such pages are also not on any PQ queue.
284  *
285  */
286 #define	VPO_KMEM_EXEC	0x01		/* kmem mapping allows execution */
287 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
288 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
289 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
290 
291 /*
292  * Busy page implementation details.
293  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
294  * even if the support for owner identity is removed because of size
295  * constraints.  Checks on lock recursion are then not possible, while the
296  * lock assertions effectiveness is someway reduced.
297  */
298 #define	VPB_BIT_SHARED		0x01
299 #define	VPB_BIT_EXCLUSIVE	0x02
300 #define	VPB_BIT_WAITERS		0x04
301 #define	VPB_BIT_FLAGMASK						\
302 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
303 
304 #define	VPB_SHARERS_SHIFT	3
305 #define	VPB_SHARERS(x)							\
306 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
307 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
308 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
309 
310 #define	VPB_SINGLE_EXCLUSIVE	VPB_BIT_EXCLUSIVE
311 #ifdef INVARIANTS
312 #define	VPB_CURTHREAD_EXCLUSIVE						\
313 	(VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK))
314 #else
315 #define	VPB_CURTHREAD_EXCLUSIVE	VPB_SINGLE_EXCLUSIVE
316 #endif
317 
318 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
319 
320 /* Freed lock blocks both shared and exclusive. */
321 #define	VPB_FREED		(0xffffffff - VPB_BIT_SHARED)
322 
323 #define	PQ_NONE		255
324 #define	PQ_INACTIVE	0
325 #define	PQ_ACTIVE	1
326 #define	PQ_LAUNDRY	2
327 #define	PQ_UNSWAPPABLE	3
328 #define	PQ_COUNT	4
329 
330 #ifndef VM_PAGE_HAVE_PGLIST
331 TAILQ_HEAD(pglist, vm_page);
332 #define VM_PAGE_HAVE_PGLIST
333 #endif
334 SLIST_HEAD(spglist, vm_page);
335 
336 #ifdef _KERNEL
337 extern vm_page_t bogus_page;
338 #endif	/* _KERNEL */
339 
340 extern struct mtx_padalign pa_lock[];
341 
342 #if defined(__arm__)
343 #define	PDRSHIFT	PDR_SHIFT
344 #elif !defined(PDRSHIFT)
345 #define PDRSHIFT	21
346 #endif
347 
348 #define	pa_index(pa)	((pa) >> PDRSHIFT)
349 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
350 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
351 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
352 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
353 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
354 #define	PA_UNLOCK_COND(pa) 			\
355 	do {		   			\
356 		if ((pa) != 0) {		\
357 			PA_UNLOCK((pa));	\
358 			(pa) = 0;		\
359 		}				\
360 	} while (0)
361 
362 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
363 
364 #if defined(KLD_MODULE) && !defined(KLD_TIED)
365 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
366 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
367 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
368 #else	/* !KLD_MODULE */
369 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
370 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
371 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
372 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
373 #endif
374 #if defined(INVARIANTS)
375 #define	vm_page_assert_locked(m)		\
376     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
377 #define	vm_page_lock_assert(m, a)		\
378     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
379 #else
380 #define	vm_page_assert_locked(m)
381 #define	vm_page_lock_assert(m, a)
382 #endif
383 
384 /*
385  * The vm_page's aflags are updated using atomic operations.  To set or clear
386  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
387  * must be used.  Neither these flags nor these functions are part of the KBI.
388  *
389  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
390  * both the MI and MD VM layers.  However, kernel loadable modules should not
391  * directly set this flag.  They should call vm_page_reference() instead.
392  *
393  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
394  * When it does so, the object must be locked, or the page must be
395  * exclusive busied.  The MI VM layer must never access this flag
396  * directly.  Instead, it should call pmap_page_is_write_mapped().
397  *
398  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
399  * at least one executable mapping.  It is not consumed by the MI VM layer.
400  *
401  * PGA_NOSYNC must be set and cleared with the page busy lock held.
402  *
403  * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
404  * from a page queue, respectively.  It determines whether the plinks.q field
405  * of the page is valid.  To set or clear this flag, page's "queue" field must
406  * be a valid queue index, and the corresponding page queue lock must be held.
407  *
408  * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
409  * queue, and cleared when the dequeue request is processed.  A page may
410  * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
411  * is requested after the page is scheduled to be enqueued but before it is
412  * actually inserted into the page queue.
413  *
414  * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
415  * in its page queue.
416  *
417  * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
418  * the inactive queue, thus bypassing LRU.
419  *
420  * The PGA_DEQUEUE, PGA_REQUEUE and PGA_REQUEUE_HEAD flags must be set using an
421  * atomic RMW operation to ensure that the "queue" field is a valid queue index,
422  * and the corresponding page queue lock must be held when clearing any of the
423  * flags.
424  *
425  * PGA_SWAP_FREE is used to defer freeing swap space to the pageout daemon
426  * when the context that dirties the page does not have the object write lock
427  * held.
428  */
429 #define	PGA_WRITEABLE	0x0001		/* page may be mapped writeable */
430 #define	PGA_REFERENCED	0x0002		/* page has been referenced */
431 #define	PGA_EXECUTABLE	0x0004		/* page may be mapped executable */
432 #define	PGA_ENQUEUED	0x0008		/* page is enqueued in a page queue */
433 #define	PGA_DEQUEUE	0x0010		/* page is due to be dequeued */
434 #define	PGA_REQUEUE	0x0020		/* page is due to be requeued */
435 #define	PGA_REQUEUE_HEAD 0x0040		/* page requeue should bypass LRU */
436 #define	PGA_NOSYNC	0x0080		/* do not collect for syncer */
437 #define	PGA_SWAP_FREE	0x0100		/* page with swap space was dirtied */
438 #define	PGA_SWAP_SPACE	0x0200		/* page has allocated swap space */
439 
440 #define	PGA_QUEUE_OP_MASK	(PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD)
441 #define	PGA_QUEUE_STATE_MASK	(PGA_ENQUEUED | PGA_QUEUE_OP_MASK)
442 
443 /*
444  * Page flags.  Updates to these flags are not synchronized, and thus they must
445  * be set during page allocation or free to avoid races.
446  *
447  * The PG_PCPU_CACHE flag is set at allocation time if the page was
448  * allocated from a per-CPU cache.  It is cleared the next time that the
449  * page is allocated from the physical memory allocator.
450  */
451 #define	PG_PCPU_CACHE	0x01		/* was allocated from per-CPU caches */
452 #define	PG_FICTITIOUS	0x02		/* physical page doesn't exist */
453 #define	PG_ZERO		0x04		/* page is zeroed */
454 #define	PG_MARKER	0x08		/* special queue marker page */
455 #define	PG_NODUMP	0x10		/* don't include this page in a dump */
456 #define	PG_NOFREE	0x20		/* page should never be freed. */
457 
458 /*
459  * Misc constants.
460  */
461 #define ACT_DECLINE		1
462 #define ACT_ADVANCE		3
463 #define ACT_INIT		5
464 #define ACT_MAX			64
465 
466 #ifdef _KERNEL
467 
468 #include <sys/kassert.h>
469 #include <machine/atomic.h>
470 struct pctrie_iter;
471 
472 /*
473  * Each pageable resident page falls into one of five lists:
474  *
475  *	free
476  *		Available for allocation now.
477  *
478  *	inactive
479  *		Low activity, candidates for reclamation.
480  *		This list is approximately LRU ordered.
481  *
482  *	laundry
483  *		This is the list of pages that should be
484  *		paged out next.
485  *
486  *	unswappable
487  *		Dirty anonymous pages that cannot be paged
488  *		out because no swap device is configured.
489  *
490  *	active
491  *		Pages that are "active", i.e., they have been
492  *		recently referenced.
493  *
494  */
495 
496 extern vm_page_t vm_page_array;		/* First resident page in table */
497 extern long vm_page_array_size;		/* number of vm_page_t's */
498 extern long first_page;			/* first physical page number */
499 
500 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
501 
502 /*
503  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
504  * page to which the given physical address belongs. The correct vm_page_t
505  * object is returned for addresses that are not page-aligned.
506  */
507 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
508 
509 /*
510  * Page allocation parameters for vm_page for the functions
511  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
512  * vm_page_alloc_freelist().  Some functions support only a subset
513  * of the flags, and ignore others, see the flags legend.
514  *
515  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
516  * and the vm_page_grab*() functions.  See these functions for details.
517  *
518  * Bits 0 - 1 define class.
519  * Bits 2 - 15 dedicated for flags.
520  * Legend:
521  * (a) - vm_page_alloc() supports the flag.
522  * (c) - vm_page_alloc_contig() supports the flag.
523  * (g) - vm_page_grab() supports the flag.
524  * (n) - vm_page_alloc_noobj() and vm_page_alloc_freelist() support the flag.
525  * (p) - vm_page_grab_pages() supports the flag.
526  * Bits above 15 define the count of additional pages that the caller
527  * intends to allocate.
528  */
529 #define VM_ALLOC_NORMAL		0
530 #define VM_ALLOC_INTERRUPT	1
531 #define VM_ALLOC_SYSTEM		2
532 #define	VM_ALLOC_CLASS_MASK	3
533 #define	VM_ALLOC_WAITOK		0x0008	/* (acn) Sleep and retry */
534 #define	VM_ALLOC_WAITFAIL	0x0010	/* (acn) Sleep and return error */
535 #define	VM_ALLOC_WIRED		0x0020	/* (acgnp) Allocate a wired page */
536 #define	VM_ALLOC_ZERO		0x0040	/* (acgnp) Allocate a zeroed page */
537 #define	VM_ALLOC_NORECLAIM	0x0080	/* (c) Do not reclaim after failure */
538 #define	VM_ALLOC_NOFREE		0x0100	/* (an) Page will never be released */
539 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
540 #define	VM_ALLOC_NOCREAT	0x0400	/* (gp) Don't create a page */
541 #define	VM_ALLOC_AVAIL1		0x0800
542 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
543 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
544 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
545 #define	VM_ALLOC_NOWAIT		0x8000	/* (acgnp) Do not sleep */
546 #define	VM_ALLOC_COUNT_MAX	0xffff
547 #define	VM_ALLOC_COUNT_SHIFT	16
548 #define	VM_ALLOC_COUNT_MASK	(VM_ALLOC_COUNT(VM_ALLOC_COUNT_MAX))
549 #define	VM_ALLOC_COUNT(count)	({				\
550 	KASSERT((count) <= VM_ALLOC_COUNT_MAX,			\
551 	    ("%s: invalid VM_ALLOC_COUNT value", __func__));	\
552 	(count) << VM_ALLOC_COUNT_SHIFT;			\
553 })
554 
555 #ifdef M_NOWAIT
556 static inline int
557 malloc2vm_flags(int malloc_flags)
558 {
559 	int pflags;
560 
561 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
562 	    (malloc_flags & M_NOWAIT) != 0,
563 	    ("M_USE_RESERVE requires M_NOWAIT"));
564 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
565 	    VM_ALLOC_SYSTEM;
566 	if ((malloc_flags & M_ZERO) != 0)
567 		pflags |= VM_ALLOC_ZERO;
568 	if ((malloc_flags & M_NODUMP) != 0)
569 		pflags |= VM_ALLOC_NODUMP;
570 	if ((malloc_flags & M_NOWAIT))
571 		pflags |= VM_ALLOC_NOWAIT;
572 	if ((malloc_flags & M_WAITOK))
573 		pflags |= VM_ALLOC_WAITOK;
574 	if ((malloc_flags & M_NORECLAIM))
575 		pflags |= VM_ALLOC_NORECLAIM;
576 	if ((malloc_flags & M_NEVERFREED))
577 		pflags |= VM_ALLOC_NOFREE;
578 	return (pflags);
579 }
580 #endif
581 
582 /*
583  * Predicates supported by vm_page_ps_test():
584  *
585  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
586  *	However, it can be spuriously false when the (super)page has become
587  *	dirty in the pmap but that information has not been propagated to the
588  *	machine-independent layer.
589  */
590 #define	PS_ALL_DIRTY	0x1
591 #define	PS_ALL_VALID	0x2
592 #define	PS_NONE_BUSY	0x4
593 
594 void vm_page_activate (vm_page_t);
595 void vm_page_advise(vm_page_t m, int advice);
596 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
597 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
598     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
599     vm_paddr_t boundary, vm_memattr_t memattr);
600 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
601     vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
602     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
603     vm_memattr_t memattr);
604 vm_page_t vm_page_alloc_domain_iter(vm_object_t object, vm_pindex_t pindex,
605     int domain, int req, struct pctrie_iter *pages);
606 vm_page_t vm_page_alloc_iter(vm_object_t object, vm_pindex_t pindex, int req,
607     struct pctrie_iter *pages);
608 vm_page_t vm_page_alloc_noobj(int);
609 vm_page_t vm_page_alloc_noobj_domain(int, int);
610 vm_page_t vm_page_alloc_noobj_contig(int req, u_long npages, vm_paddr_t low,
611     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
612     vm_memattr_t memattr);
613 vm_page_t vm_page_alloc_noobj_contig_domain(int domain, int req, u_long npages,
614     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
615     vm_memattr_t memattr);
616 void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set);
617 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
618 bool vm_page_busy_acquire(vm_page_t m, int allocflags);
619 void vm_page_busy_downgrade(vm_page_t m);
620 int vm_page_busy_tryupgrade(vm_page_t m);
621 bool vm_page_busy_sleep(vm_page_t m, const char *msg, int allocflags);
622 void vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m,
623     vm_pindex_t pindex, const char *wmesg, int allocflags);
624 void vm_page_deactivate(vm_page_t m);
625 void vm_page_deactivate_noreuse(vm_page_t m);
626 void vm_page_dequeue(vm_page_t m);
627 void vm_page_dequeue_deferred(vm_page_t m);
628 void vm_page_free(vm_page_t m);
629 void vm_page_free_invalid(vm_page_t m);
630 int vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
631 void vm_page_free_zero(vm_page_t m);
632 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
633 int vm_page_grab_zero_partial(vm_object_t object, vm_pindex_t pindex, int base,
634     int end);
635 vm_page_t vm_page_grab(vm_object_t, vm_pindex_t, int);
636 vm_page_t vm_page_grab_iter(vm_object_t object, vm_pindex_t pindex,
637     int allocflags, struct pctrie_iter *pages);
638 vm_page_t vm_page_grab_unlocked(vm_object_t, vm_pindex_t, int);
639 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
640     vm_page_t *ma, int count);
641 int vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex,
642     int allocflags, vm_page_t *ma, int count);
643 int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex,
644     int allocflags);
645 int vm_page_grab_valid_iter(vm_page_t *mp, vm_object_t object,
646     vm_pindex_t pindex, int allocflags, struct pctrie_iter *pages);
647 int vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object,
648     vm_pindex_t pindex, int allocflags);
649 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
650 void vm_page_init_marker(vm_page_t marker, int queue, uint16_t aflags);
651 void vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind, int pool);
652 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
653 void vm_page_invalid(vm_page_t m);
654 void vm_page_iter_free(struct pctrie_iter *pages, vm_page_t m);
655 void vm_page_iter_init(struct pctrie_iter *, vm_object_t);
656 int vm_page_iter_insert(vm_page_t m, vm_object_t, vm_pindex_t,
657     struct pctrie_iter *);
658 void vm_page_iter_limit_init(struct pctrie_iter *, vm_object_t, vm_pindex_t);
659 bool vm_page_iter_remove(struct pctrie_iter *pages, vm_page_t m);
660 bool vm_page_iter_rename(struct pctrie_iter *old_pages, vm_page_t m,
661     vm_object_t new_object, vm_pindex_t new_pindex);
662 void vm_page_launder(vm_page_t m);
663 vm_page_t vm_page_lookup(vm_object_t, vm_pindex_t);
664 vm_page_t vm_page_lookup_unlocked(vm_object_t, vm_pindex_t);
665 void vm_page_pqbatch_drain(void);
666 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
667 bool vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old,
668     vm_page_astate_t new);
669 bool vm_page_ps_test(vm_page_t m, int psind, int flags, vm_page_t skip_m);
670 void vm_page_putfake(vm_page_t m);
671 void vm_page_readahead_finish(vm_page_t m);
672 int vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
673     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
674 int vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
675     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
676 int vm_page_reclaim_contig_domain_ext(int domain, int req, u_long npages,
677     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
678     int desired_runs);
679 void vm_page_reference(vm_page_t m);
680 #define	VPR_TRYFREE	0x01
681 #define	VPR_NOREUSE	0x02
682 void vm_page_release(vm_page_t m, int flags);
683 void vm_page_release_locked(vm_page_t m, int flags);
684 vm_page_t vm_page_relookup(vm_object_t, vm_pindex_t);
685 bool vm_page_remove(vm_page_t);
686 bool vm_page_remove_xbusy(vm_page_t);
687 void vm_page_replace(vm_page_t mnew, vm_object_t object,
688     vm_pindex_t pindex, vm_page_t mold);
689 int vm_page_sbusied(vm_page_t m);
690 vm_page_bits_t vm_page_set_dirty(vm_page_t m);
691 void vm_page_set_valid_range(vm_page_t m, int base, int size);
692 vm_offset_t vm_page_startup(vm_offset_t vaddr);
693 void vm_page_sunbusy(vm_page_t m);
694 bool vm_page_try_remove_all(vm_page_t m);
695 bool vm_page_try_remove_write(vm_page_t m);
696 int vm_page_trysbusy(vm_page_t m);
697 int vm_page_tryxbusy(vm_page_t m);
698 void vm_page_unhold_pages(vm_page_t *ma, int count);
699 void vm_page_unswappable(vm_page_t m);
700 void vm_page_unwire(vm_page_t m, uint8_t queue);
701 bool vm_page_unwire_noq(vm_page_t m);
702 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
703 void vm_page_wire(vm_page_t);
704 bool vm_page_wire_mapped(vm_page_t m);
705 void vm_page_xunbusy_hard(vm_page_t m);
706 void vm_page_xunbusy_hard_unchecked(vm_page_t m);
707 void vm_page_set_validclean (vm_page_t, int, int);
708 void vm_page_clear_dirty(vm_page_t, int, int);
709 void vm_page_set_invalid(vm_page_t, int, int);
710 void vm_page_valid(vm_page_t m);
711 int vm_page_is_valid(vm_page_t, int, int);
712 void vm_page_test_dirty(vm_page_t);
713 vm_page_bits_t vm_page_bits(int base, int size);
714 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
715 
716 void vm_page_dirty_KBI(vm_page_t m);
717 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
718 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
719 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
720 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
721 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
722 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
723 #endif
724 
725 #define	vm_page_busy_fetch(m)	atomic_load_int(&(m)->busy_lock)
726 
727 #define	vm_page_assert_busied(m)					\
728 	KASSERT(vm_page_busied(m),					\
729 	    ("vm_page_assert_busied: page %p not busy @ %s:%d", \
730 	    (m), __FILE__, __LINE__))
731 
732 #define	vm_page_assert_sbusied(m)					\
733 	KASSERT(vm_page_sbusied(m),					\
734 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
735 	    (m), __FILE__, __LINE__))
736 
737 #define	vm_page_assert_unbusied(m)					\
738 	KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) !=		\
739 	    VPB_CURTHREAD_EXCLUSIVE,					\
740 	    ("vm_page_assert_unbusied: page %p busy_lock %#x owned"	\
741 	     " by me (%p) @ %s:%d",					\
742 	    (m), (m)->busy_lock, curthread, __FILE__, __LINE__));	\
743 
744 #define	vm_page_assert_xbusied_unchecked(m) do {			\
745 	KASSERT(vm_page_xbusied(m),					\
746 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
747 	    (m), __FILE__, __LINE__));					\
748 } while (0)
749 #define	vm_page_assert_xbusied(m) do {					\
750 	vm_page_assert_xbusied_unchecked(m);				\
751 	KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) ==		\
752 	    VPB_CURTHREAD_EXCLUSIVE,					\
753 	    ("vm_page_assert_xbusied: page %p busy_lock %#x not owned"	\
754 	     " by me (%p) @ %s:%d",					\
755 	    (m), (m)->busy_lock, curthread, __FILE__, __LINE__));	\
756 } while (0)
757 
758 #define	vm_page_busied(m)						\
759 	(vm_page_busy_fetch(m) != VPB_UNBUSIED)
760 
761 #define	vm_page_xbusied(m)						\
762 	((vm_page_busy_fetch(m) & VPB_SINGLE_EXCLUSIVE) != 0)
763 
764 #define	vm_page_busy_freed(m)						\
765 	(vm_page_busy_fetch(m) == VPB_FREED)
766 
767 /* Note: page m's lock must not be owned by the caller. */
768 #define	vm_page_xunbusy(m) do {						\
769 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
770 	    VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED))			\
771 		vm_page_xunbusy_hard(m);				\
772 } while (0)
773 #define	vm_page_xunbusy_unchecked(m) do {				\
774 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
775 	    VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED))			\
776 		vm_page_xunbusy_hard_unchecked(m);			\
777 } while (0)
778 
779 #ifdef INVARIANTS
780 void vm_page_object_busy_assert(vm_page_t m);
781 #define	VM_PAGE_OBJECT_BUSY_ASSERT(m)	vm_page_object_busy_assert(m)
782 void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits);
783 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
784 	vm_page_assert_pga_writeable(m, bits)
785 /*
786  * Claim ownership of a page's xbusy state.  In non-INVARIANTS kernels this
787  * operation is a no-op since ownership is not tracked.  In particular
788  * this macro does not provide any synchronization with the previous owner.
789  */
790 #define	vm_page_xbusy_claim(m) do {					\
791 	u_int _busy_lock;						\
792 									\
793 	vm_page_assert_xbusied_unchecked((m));				\
794 	do {								\
795 		_busy_lock = vm_page_busy_fetch(m);			\
796 	} while (!atomic_cmpset_int(&(m)->busy_lock, _busy_lock,	\
797 	    (_busy_lock & VPB_BIT_FLAGMASK) | VPB_CURTHREAD_EXCLUSIVE)); \
798 } while (0)
799 #else
800 #define	VM_PAGE_OBJECT_BUSY_ASSERT(m)	(void)0
801 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
802 #define	vm_page_xbusy_claim(m)
803 #endif
804 
805 #if BYTE_ORDER == BIG_ENDIAN
806 #define	VM_PAGE_AFLAG_SHIFT	16
807 #else
808 #define	VM_PAGE_AFLAG_SHIFT	0
809 #endif
810 
811 /*
812  *	Load a snapshot of a page's 32-bit atomic state.
813  */
814 static inline vm_page_astate_t
815 vm_page_astate_load(vm_page_t m)
816 {
817 	vm_page_astate_t a;
818 
819 	a._bits = atomic_load_32(&m->a._bits);
820 	return (a);
821 }
822 
823 /*
824  *	Atomically compare and set a page's atomic state.
825  */
826 static inline bool
827 vm_page_astate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new)
828 {
829 
830 	KASSERT(new.queue == PQ_INACTIVE || (new.flags & PGA_REQUEUE_HEAD) == 0,
831 	    ("%s: invalid head requeue request for page %p", __func__, m));
832 	KASSERT((new.flags & PGA_ENQUEUED) == 0 || new.queue != PQ_NONE,
833 	    ("%s: setting PGA_ENQUEUED with PQ_NONE in page %p", __func__, m));
834 	KASSERT(new._bits != old->_bits,
835 	    ("%s: bits are unchanged", __func__));
836 
837 	return (atomic_fcmpset_32(&m->a._bits, &old->_bits, new._bits) != 0);
838 }
839 
840 /*
841  *	Clear the given bits in the specified page.
842  */
843 static inline void
844 vm_page_aflag_clear(vm_page_t m, uint16_t bits)
845 {
846 	uint32_t *addr, val;
847 
848 	/*
849 	 * Access the whole 32-bit word containing the aflags field with an
850 	 * atomic update.  Parallel non-atomic updates to the other fields
851 	 * within this word are handled properly by the atomic update.
852 	 */
853 	addr = (void *)&m->a;
854 	val = bits << VM_PAGE_AFLAG_SHIFT;
855 	atomic_clear_32(addr, val);
856 }
857 
858 /*
859  *	Set the given bits in the specified page.
860  */
861 static inline void
862 vm_page_aflag_set(vm_page_t m, uint16_t bits)
863 {
864 	uint32_t *addr, val;
865 
866 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
867 
868 	/*
869 	 * Access the whole 32-bit word containing the aflags field with an
870 	 * atomic update.  Parallel non-atomic updates to the other fields
871 	 * within this word are handled properly by the atomic update.
872 	 */
873 	addr = (void *)&m->a;
874 	val = bits << VM_PAGE_AFLAG_SHIFT;
875 	atomic_set_32(addr, val);
876 }
877 
878 /*
879  *	vm_page_dirty:
880  *
881  *	Set all bits in the page's dirty field.
882  *
883  *	The object containing the specified page must be locked if the
884  *	call is made from the machine-independent layer.
885  *
886  *	See vm_page_clear_dirty_mask().
887  */
888 static __inline void
889 vm_page_dirty(vm_page_t m)
890 {
891 
892 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
893 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
894 	vm_page_dirty_KBI(m);
895 #else
896 	m->dirty = VM_PAGE_BITS_ALL;
897 #endif
898 }
899 
900 /*
901  *	vm_page_undirty:
902  *
903  *	Set page to not be dirty.  Note: does not clear pmap modify bits
904  */
905 static __inline void
906 vm_page_undirty(vm_page_t m)
907 {
908 
909 	VM_PAGE_OBJECT_BUSY_ASSERT(m);
910 	m->dirty = 0;
911 }
912 
913 static inline uint8_t
914 _vm_page_queue(vm_page_astate_t as)
915 {
916 
917 	if ((as.flags & PGA_DEQUEUE) != 0)
918 		return (PQ_NONE);
919 	return (as.queue);
920 }
921 
922 /*
923  *	vm_page_queue:
924  *
925  *	Return the index of the queue containing m.
926  */
927 static inline uint8_t
928 vm_page_queue(vm_page_t m)
929 {
930 
931 	return (_vm_page_queue(vm_page_astate_load(m)));
932 }
933 
934 static inline bool
935 vm_page_active(vm_page_t m)
936 {
937 
938 	return (vm_page_queue(m) == PQ_ACTIVE);
939 }
940 
941 static inline bool
942 vm_page_inactive(vm_page_t m)
943 {
944 
945 	return (vm_page_queue(m) == PQ_INACTIVE);
946 }
947 
948 static inline bool
949 vm_page_in_laundry(vm_page_t m)
950 {
951 	uint8_t queue;
952 
953 	queue = vm_page_queue(m);
954 	return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
955 }
956 
957 static inline void
958 vm_page_clearref(vm_page_t m)
959 {
960 	u_int r;
961 
962 	r = m->ref_count;
963 	while (atomic_fcmpset_int(&m->ref_count, &r, r & (VPRC_BLOCKED |
964 	    VPRC_OBJREF)) == 0)
965 		;
966 }
967 
968 /*
969  *	vm_page_drop:
970  *
971  *	Release a reference to a page and return the old reference count.
972  */
973 static inline u_int
974 vm_page_drop(vm_page_t m, u_int val)
975 {
976 	u_int old;
977 
978 	/*
979 	 * Synchronize with vm_page_free_prep(): ensure that all updates to the
980 	 * page structure are visible before it is freed.
981 	 */
982 	atomic_thread_fence_rel();
983 	old = atomic_fetchadd_int(&m->ref_count, -val);
984 	KASSERT(old != VPRC_BLOCKED,
985 	    ("vm_page_drop: page %p has an invalid refcount value", m));
986 	return (old);
987 }
988 
989 /*
990  *	vm_page_wired:
991  *
992  *	Perform a racy check to determine whether a reference prevents the page
993  *	from being reclaimable.  If the page's object is locked, and the page is
994  *	unmapped and exclusively busied by the current thread, no new wirings
995  *	may be created.
996  */
997 static inline bool
998 vm_page_wired(vm_page_t m)
999 {
1000 
1001 	return (VPRC_WIRE_COUNT(m->ref_count) > 0);
1002 }
1003 
1004 static inline bool
1005 vm_page_all_valid(vm_page_t m)
1006 {
1007 
1008 	return (m->valid == VM_PAGE_BITS_ALL);
1009 }
1010 
1011 static inline bool
1012 vm_page_any_valid(vm_page_t m)
1013 {
1014 
1015 	return (m->valid != 0);
1016 }
1017 
1018 static inline bool
1019 vm_page_none_valid(vm_page_t m)
1020 {
1021 
1022 	return (m->valid == 0);
1023 }
1024 
1025 static inline int
1026 vm_page_domain(vm_page_t m __numa_used)
1027 {
1028 #ifdef NUMA
1029 	int domn, segind;
1030 
1031 	segind = m->segind;
1032 	KASSERT(segind < vm_phys_nsegs, ("segind %d m %p", segind, m));
1033 	domn = vm_phys_segs[segind].domain;
1034 	KASSERT(domn >= 0 && domn < vm_ndomains, ("domain %d m %p", domn, m));
1035 	return (domn);
1036 #else
1037 	return (0);
1038 #endif
1039 }
1040 
1041 #endif				/* _KERNEL */
1042 #endif				/* !_VM_PAGE_ */
1043