xref: /freebsd/sys/vm/vm_page.h (revision d34048812292b714a0bf99967270d18fe3097c62)
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  *	from: @(#)vm_page.h	8.2 (Berkeley) 12/13/93
35  *
36  *
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  *
62  * $FreeBSD$
63  */
64 
65 /*
66  *	Resident memory system definitions.
67  */
68 
69 #ifndef	_VM_PAGE_
70 #define	_VM_PAGE_
71 
72 #include <vm/pmap.h>
73 
74 /*
75  *	Management of resident (logical) pages.
76  *
77  *	A small structure is kept for each resident
78  *	page, indexed by page number.  Each structure
79  *	is an element of several collections:
80  *
81  *		A radix tree used to quickly
82  *		perform object/offset lookups
83  *
84  *		A list of all pages for a given object,
85  *		so they can be quickly deactivated at
86  *		time of deallocation.
87  *
88  *		An ordered list of pages due for pageout.
89  *
90  *	In addition, the structure contains the object
91  *	and offset to which this page belongs (for pageout),
92  *	and sundry status bits.
93  *
94  *	In general, operations on this structure's mutable fields are
95  *	synchronized using either one of or a combination of the lock on the
96  *	object that the page belongs to (O), the page lock (P),
97  *	the per-domain lock for the free queues (F), or the page's queue
98  *	lock (Q).  The physical address of a page is used to select its page
99  *	lock from a pool.  The queue lock for a page depends on the value of
100  *	its queue field and described in detail below.  If a field is
101  *	annotated below with two of these locks, then holding either lock is
102  *	sufficient for read access, but both locks are required for write
103  *	access.  An annotation of (C) indicates that the field is immutable.
104  *
105  *	In contrast, the synchronization of accesses to the page's
106  *	dirty field is machine dependent (M).  In the
107  *	machine-independent layer, the lock on the object that the
108  *	page belongs to must be held in order to operate on the field.
109  *	However, the pmap layer is permitted to set all bits within
110  *	the field without holding that lock.  If the underlying
111  *	architecture does not support atomic read-modify-write
112  *	operations on the field's type, then the machine-independent
113  *	layer uses a 32-bit atomic on the aligned 32-bit word that
114  *	contains the dirty field.  In the machine-independent layer,
115  *	the implementation of read-modify-write operations on the
116  *	field is encapsulated in vm_page_clear_dirty_mask().
117  *
118  *	The page structure contains two counters which prevent page reuse.
119  *	Both counters are protected by the page lock (P).  The hold
120  *	counter counts transient references obtained via a pmap lookup, and
121  *	is also used to prevent page reclamation in situations where it is
122  *	undesirable to block other accesses to the page.  The wire counter
123  *	is used to implement mlock(2) and is non-zero for pages containing
124  *	kernel memory.  Pages that are wired or held will not be reclaimed
125  *	or laundered by the page daemon, but are treated differently during
126  *	a page queue scan: held pages remain at their position in the queue,
127  *	while wired pages are removed from the queue and must later be
128  *	re-enqueued appropriately by the unwiring thread.  It is legal to
129  *	call vm_page_free() on a held page; doing so causes it to be removed
130  *	from its object and page queue, and the page is released to the
131  *	allocator once the last hold reference is dropped.  In contrast,
132  *	wired pages may not be freed.
133  *
134  *	In some pmap implementations, the wire count of a page table page is
135  *	used to track the number of populated entries.
136  *
137  *	The busy lock is an embedded reader-writer lock which protects the
138  *	page's contents and identity (i.e., its <object, pindex> tuple) and
139  *	interlocks with the object lock (O).  In particular, a page may be
140  *	busied or unbusied only with the object write lock held.  To avoid
141  *	bloating the page structure, the busy lock lacks some of the
142  *	features available to the kernel's general-purpose synchronization
143  *	primitives.  As a result, busy lock ordering rules are not verified,
144  *	lock recursion is not detected, and an attempt to xbusy a busy page
145  *	or sbusy an xbusy page results will trigger a panic rather than
146  *	causing the thread to block.  vm_page_sleep_if_busy() can be used to
147  *	sleep until the page's busy state changes, after which the caller
148  *	must re-lookup the page and re-evaluate its state.
149  *
150  *	The queue field is the index of the page queue containing the
151  *	page, or PQ_NONE if the page is not enqueued.  The queue lock of a
152  *	page is the page queue lock corresponding to the page queue index,
153  *	or the page lock (P) for the page if it is not enqueued.  To modify
154  *	the queue field, the queue lock for the old value of the field must
155  *	be held.  It is invalid for a page's queue field to transition
156  *	between two distinct page queue indices.  That is, when updating
157  *	the queue field, either the new value or the old value must be
158  *	PQ_NONE.
159  *
160  *	To avoid contention on page queue locks, page queue operations
161  *	(enqueue, dequeue, requeue) are batched using per-CPU queues.
162  *	A deferred operation is requested by inserting an entry into a
163  *	batch queue; the entry is simply a pointer to the page, and the
164  *	request type is encoded in the page's aflags field using the values
165  *	in PGA_QUEUE_STATE_MASK.  The type-stability of struct vm_pages is
166  *	crucial to this scheme since the processing of entries in a given
167  *	batch queue may be deferred indefinitely.  In particular, a page
168  *	may be freed before its pending batch queue entries have been
169  *	processed.  The page lock (P) must be held to schedule a batched
170  *	queue operation, and the page queue lock must be held in order to
171  *	process batch queue entries for the page queue.
172  */
173 
174 #if PAGE_SIZE == 4096
175 #define VM_PAGE_BITS_ALL 0xffu
176 typedef uint8_t vm_page_bits_t;
177 #elif PAGE_SIZE == 8192
178 #define VM_PAGE_BITS_ALL 0xffffu
179 typedef uint16_t vm_page_bits_t;
180 #elif PAGE_SIZE == 16384
181 #define VM_PAGE_BITS_ALL 0xffffffffu
182 typedef uint32_t vm_page_bits_t;
183 #elif PAGE_SIZE == 32768
184 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
185 typedef uint64_t vm_page_bits_t;
186 #endif
187 
188 struct vm_page {
189 	union {
190 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
191 		struct {
192 			SLIST_ENTRY(vm_page) ss; /* private slists */
193 			void *pv;
194 		} s;
195 		struct {
196 			u_long p;
197 			u_long v;
198 		} memguard;
199 	} plinks;
200 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
201 	vm_object_t object;		/* which object am I in (O,P) */
202 	vm_pindex_t pindex;		/* offset into object (O,P) */
203 	vm_paddr_t phys_addr;		/* physical address of page (C) */
204 	struct md_page md;		/* machine dependent stuff */
205 	u_int wire_count;		/* wired down maps refs (P) */
206 	volatile u_int busy_lock;	/* busy owners lock */
207 	uint16_t hold_count;		/* page hold count (P) */
208 	uint16_t flags;			/* page PG_* flags (P) */
209 	uint8_t aflags;			/* access is atomic */
210 	uint8_t oflags;			/* page VPO_* flags (O) */
211 	uint8_t queue;			/* page queue index (Q) */
212 	int8_t psind;			/* pagesizes[] index (O) */
213 	int8_t segind;			/* vm_phys segment index (C) */
214 	uint8_t	order;			/* index of the buddy queue (F) */
215 	uint8_t pool;			/* vm_phys freepool index (F) */
216 	u_char	act_count;		/* page usage count (P) */
217 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
218 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
219 	vm_page_bits_t valid;		/* map of valid DEV_BSIZE chunks (O) */
220 	vm_page_bits_t dirty;		/* map of dirty DEV_BSIZE chunks (M) */
221 };
222 
223 /*
224  * Page flags stored in oflags:
225  *
226  * Access to these page flags is synchronized by the lock on the object
227  * containing the page (O).
228  *
229  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
230  * 	 indicates that the page is not under PV management but
231  * 	 otherwise should be treated as a normal page.  Pages not
232  * 	 under PV management cannot be paged out via the
233  * 	 object/vm_page_t because there is no knowledge of their pte
234  * 	 mappings, and such pages are also not on any PQ queue.
235  *
236  */
237 #define	VPO_KMEM_EXEC	0x01		/* kmem mapping allows execution */
238 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
239 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
240 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
241 #define	VPO_NOSYNC	0x10		/* do not collect for syncer */
242 
243 /*
244  * Busy page implementation details.
245  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
246  * even if the support for owner identity is removed because of size
247  * constraints.  Checks on lock recursion are then not possible, while the
248  * lock assertions effectiveness is someway reduced.
249  */
250 #define	VPB_BIT_SHARED		0x01
251 #define	VPB_BIT_EXCLUSIVE	0x02
252 #define	VPB_BIT_WAITERS		0x04
253 #define	VPB_BIT_FLAGMASK						\
254 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
255 
256 #define	VPB_SHARERS_SHIFT	3
257 #define	VPB_SHARERS(x)							\
258 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
259 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
260 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
261 
262 #define	VPB_SINGLE_EXCLUSIVER	VPB_BIT_EXCLUSIVE
263 
264 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
265 
266 #define	PQ_NONE		255
267 #define	PQ_INACTIVE	0
268 #define	PQ_ACTIVE	1
269 #define	PQ_LAUNDRY	2
270 #define	PQ_UNSWAPPABLE	3
271 #define	PQ_COUNT	4
272 
273 #ifndef VM_PAGE_HAVE_PGLIST
274 TAILQ_HEAD(pglist, vm_page);
275 #define VM_PAGE_HAVE_PGLIST
276 #endif
277 SLIST_HEAD(spglist, vm_page);
278 
279 #ifdef _KERNEL
280 extern vm_page_t bogus_page;
281 #endif	/* _KERNEL */
282 
283 extern struct mtx_padalign pa_lock[];
284 
285 #if defined(__arm__)
286 #define	PDRSHIFT	PDR_SHIFT
287 #elif !defined(PDRSHIFT)
288 #define PDRSHIFT	21
289 #endif
290 
291 #define	pa_index(pa)	((pa) >> PDRSHIFT)
292 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
293 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
294 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
295 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
296 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
297 #define	PA_UNLOCK_COND(pa) 			\
298 	do {		   			\
299 		if ((pa) != 0) {		\
300 			PA_UNLOCK((pa));	\
301 			(pa) = 0;		\
302 		}				\
303 	} while (0)
304 
305 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
306 
307 #if defined(KLD_MODULE) && !defined(KLD_TIED)
308 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
309 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
310 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
311 #else	/* !KLD_MODULE */
312 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
313 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
314 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
315 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
316 #endif
317 #if defined(INVARIANTS)
318 #define	vm_page_assert_locked(m)		\
319     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
320 #define	vm_page_lock_assert(m, a)		\
321     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
322 #else
323 #define	vm_page_assert_locked(m)
324 #define	vm_page_lock_assert(m, a)
325 #endif
326 
327 /*
328  * The vm_page's aflags are updated using atomic operations.  To set or clear
329  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
330  * must be used.  Neither these flags nor these functions are part of the KBI.
331  *
332  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
333  * both the MI and MD VM layers.  However, kernel loadable modules should not
334  * directly set this flag.  They should call vm_page_reference() instead.
335  *
336  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
337  * When it does so, the object must be locked, or the page must be
338  * exclusive busied.  The MI VM layer must never access this flag
339  * directly.  Instead, it should call pmap_page_is_write_mapped().
340  *
341  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
342  * at least one executable mapping.  It is not consumed by the MI VM layer.
343  *
344  * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
345  * from a page queue, respectively.  It determines whether the plinks.q field
346  * of the page is valid.  To set or clear this flag, the queue lock for the
347  * page must be held: the page queue lock corresponding to the page's "queue"
348  * field if its value is not PQ_NONE, and the page lock otherwise.
349  *
350  * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
351  * queue, and cleared when the dequeue request is processed.  A page may
352  * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
353  * is requested after the page is scheduled to be enqueued but before it is
354  * actually inserted into the page queue.  The page lock must be held to set
355  * this flag, and the queue lock for the page must be held to clear it.
356  *
357  * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
358  * in its page queue.  The page lock must be held to set this flag, and the
359  * queue lock for the page must be held to clear it.
360  *
361  * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
362  * the inactive queue, thus bypassing LRU.  The page lock must be held to
363  * set this flag, and the queue lock for the page must be held to clear it.
364  */
365 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
366 #define	PGA_REFERENCED	0x02		/* page has been referenced */
367 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
368 #define	PGA_ENQUEUED	0x08		/* page is enqueued in a page queue */
369 #define	PGA_DEQUEUE	0x10		/* page is due to be dequeued */
370 #define	PGA_REQUEUE	0x20		/* page is due to be requeued */
371 #define	PGA_REQUEUE_HEAD 0x40		/* page requeue should bypass LRU */
372 
373 #define	PGA_QUEUE_STATE_MASK	(PGA_ENQUEUED | PGA_DEQUEUE | PGA_REQUEUE | \
374 				PGA_REQUEUE_HEAD)
375 
376 /*
377  * Page flags.  If changed at any other time than page allocation or
378  * freeing, the modification must be protected by the vm_page lock.
379  */
380 #define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
381 #define	PG_ZERO		0x0008		/* page is zeroed */
382 #define	PG_MARKER	0x0010		/* special queue marker page */
383 #define	PG_NODUMP	0x0080		/* don't include this page in a dump */
384 #define	PG_UNHOLDFREE	0x0100		/* delayed free of a held page */
385 
386 /*
387  * Misc constants.
388  */
389 #define ACT_DECLINE		1
390 #define ACT_ADVANCE		3
391 #define ACT_INIT		5
392 #define ACT_MAX			64
393 
394 #ifdef _KERNEL
395 
396 #include <sys/systm.h>
397 
398 #include <machine/atomic.h>
399 
400 /*
401  * Each pageable resident page falls into one of five lists:
402  *
403  *	free
404  *		Available for allocation now.
405  *
406  *	inactive
407  *		Low activity, candidates for reclamation.
408  *		This list is approximately LRU ordered.
409  *
410  *	laundry
411  *		This is the list of pages that should be
412  *		paged out next.
413  *
414  *	unswappable
415  *		Dirty anonymous pages that cannot be paged
416  *		out because no swap device is configured.
417  *
418  *	active
419  *		Pages that are "active", i.e., they have been
420  *		recently referenced.
421  *
422  */
423 
424 extern vm_page_t vm_page_array;		/* First resident page in table */
425 extern long vm_page_array_size;		/* number of vm_page_t's */
426 extern long first_page;			/* first physical page number */
427 
428 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
429 
430 /*
431  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
432  * page to which the given physical address belongs. The correct vm_page_t
433  * object is returned for addresses that are not page-aligned.
434  */
435 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
436 
437 /*
438  * Page allocation parameters for vm_page for the functions
439  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
440  * vm_page_alloc_freelist().  Some functions support only a subset
441  * of the flags, and ignore others, see the flags legend.
442  *
443  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
444  * and the vm_page_grab*() functions.  See these functions for details.
445  *
446  * Bits 0 - 1 define class.
447  * Bits 2 - 15 dedicated for flags.
448  * Legend:
449  * (a) - vm_page_alloc() supports the flag.
450  * (c) - vm_page_alloc_contig() supports the flag.
451  * (f) - vm_page_alloc_freelist() supports the flag.
452  * (g) - vm_page_grab() supports the flag.
453  * (p) - vm_page_grab_pages() supports the flag.
454  * Bits above 15 define the count of additional pages that the caller
455  * intends to allocate.
456  */
457 #define VM_ALLOC_NORMAL		0
458 #define VM_ALLOC_INTERRUPT	1
459 #define VM_ALLOC_SYSTEM		2
460 #define	VM_ALLOC_CLASS_MASK	3
461 #define	VM_ALLOC_WAITOK		0x0008	/* (acf) Sleep and retry */
462 #define	VM_ALLOC_WAITFAIL	0x0010	/* (acf) Sleep and return error */
463 #define	VM_ALLOC_WIRED		0x0020	/* (acfgp) Allocate a wired page */
464 #define	VM_ALLOC_ZERO		0x0040	/* (acfgp) Allocate a prezeroed page */
465 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
466 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
467 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
468 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
469 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
470 #define	VM_ALLOC_NOWAIT		0x8000	/* (acfgp) Do not sleep */
471 #define	VM_ALLOC_COUNT_SHIFT	16
472 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
473 
474 #ifdef M_NOWAIT
475 static inline int
476 malloc2vm_flags(int malloc_flags)
477 {
478 	int pflags;
479 
480 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
481 	    (malloc_flags & M_NOWAIT) != 0,
482 	    ("M_USE_RESERVE requires M_NOWAIT"));
483 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
484 	    VM_ALLOC_SYSTEM;
485 	if ((malloc_flags & M_ZERO) != 0)
486 		pflags |= VM_ALLOC_ZERO;
487 	if ((malloc_flags & M_NODUMP) != 0)
488 		pflags |= VM_ALLOC_NODUMP;
489 	if ((malloc_flags & M_NOWAIT))
490 		pflags |= VM_ALLOC_NOWAIT;
491 	if ((malloc_flags & M_WAITOK))
492 		pflags |= VM_ALLOC_WAITOK;
493 	return (pflags);
494 }
495 #endif
496 
497 /*
498  * Predicates supported by vm_page_ps_test():
499  *
500  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
501  *	However, it can be spuriously false when the (super)page has become
502  *	dirty in the pmap but that information has not been propagated to the
503  *	machine-independent layer.
504  */
505 #define	PS_ALL_DIRTY	0x1
506 #define	PS_ALL_VALID	0x2
507 #define	PS_NONE_BUSY	0x4
508 
509 void vm_page_busy_downgrade(vm_page_t m);
510 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
511 void vm_page_flash(vm_page_t m);
512 void vm_page_hold(vm_page_t mem);
513 void vm_page_unhold(vm_page_t mem);
514 void vm_page_free(vm_page_t m);
515 void vm_page_free_zero(vm_page_t m);
516 
517 void vm_page_activate (vm_page_t);
518 void vm_page_advise(vm_page_t m, int advice);
519 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
520 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int);
521 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
522 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
523     vm_page_t);
524 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
525     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
526     vm_paddr_t boundary, vm_memattr_t memattr);
527 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
528     vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
529     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
530     vm_memattr_t memattr);
531 vm_page_t vm_page_alloc_freelist(int, int);
532 vm_page_t vm_page_alloc_freelist_domain(int, int, int);
533 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
534 void vm_page_change_lock(vm_page_t m, struct mtx **mtx);
535 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
536 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
537     vm_page_t *ma, int count);
538 void vm_page_deactivate(vm_page_t);
539 void vm_page_deactivate_noreuse(vm_page_t);
540 void vm_page_dequeue(vm_page_t m);
541 void vm_page_dequeue_deferred(vm_page_t m);
542 void vm_page_drain_pqbatch(void);
543 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
544 bool vm_page_free_prep(vm_page_t m);
545 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
546 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
547 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
548 void vm_page_launder(vm_page_t m);
549 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
550 vm_page_t vm_page_next(vm_page_t m);
551 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
552 struct vm_pagequeue *vm_page_pagequeue(vm_page_t m);
553 vm_page_t vm_page_prev(vm_page_t m);
554 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
555 void vm_page_putfake(vm_page_t m);
556 void vm_page_readahead_finish(vm_page_t m);
557 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
558     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
559 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
560     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
561 void vm_page_reference(vm_page_t m);
562 void vm_page_remove (vm_page_t);
563 int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
564 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
565     vm_pindex_t pindex);
566 void vm_page_requeue(vm_page_t m);
567 int vm_page_sbusied(vm_page_t m);
568 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
569     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
570 void vm_page_set_valid_range(vm_page_t m, int base, int size);
571 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
572 vm_offset_t vm_page_startup(vm_offset_t vaddr);
573 void vm_page_sunbusy(vm_page_t m);
574 bool vm_page_try_to_free(vm_page_t m);
575 int vm_page_trysbusy(vm_page_t m);
576 void vm_page_unhold_pages(vm_page_t *ma, int count);
577 void vm_page_unswappable(vm_page_t m);
578 bool vm_page_unwire(vm_page_t m, uint8_t queue);
579 bool vm_page_unwire_noq(vm_page_t m);
580 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
581 void vm_page_wire (vm_page_t);
582 void vm_page_xunbusy_hard(vm_page_t m);
583 void vm_page_xunbusy_maybelocked(vm_page_t m);
584 void vm_page_set_validclean (vm_page_t, int, int);
585 void vm_page_clear_dirty (vm_page_t, int, int);
586 void vm_page_set_invalid (vm_page_t, int, int);
587 int vm_page_is_valid (vm_page_t, int, int);
588 void vm_page_test_dirty (vm_page_t);
589 vm_page_bits_t vm_page_bits(int base, int size);
590 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
591 void vm_page_free_toq(vm_page_t m);
592 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
593 
594 void vm_page_dirty_KBI(vm_page_t m);
595 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
596 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
597 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
598 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
599 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
600 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
601 #endif
602 
603 #define	vm_page_assert_sbusied(m)					\
604 	KASSERT(vm_page_sbusied(m),					\
605 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
606 	    (m), __FILE__, __LINE__))
607 
608 #define	vm_page_assert_unbusied(m)					\
609 	KASSERT(!vm_page_busied(m),					\
610 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
611 	    (m), __FILE__, __LINE__))
612 
613 #define	vm_page_assert_xbusied(m)					\
614 	KASSERT(vm_page_xbusied(m),					\
615 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
616 	    (m), __FILE__, __LINE__))
617 
618 #define	vm_page_busied(m)						\
619 	((m)->busy_lock != VPB_UNBUSIED)
620 
621 #define	vm_page_sbusy(m) do {						\
622 	if (!vm_page_trysbusy(m))					\
623 		panic("%s: page %p failed shared busying", __func__,	\
624 		    (m));						\
625 } while (0)
626 
627 #define	vm_page_tryxbusy(m)						\
628 	(atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED,		\
629 	    VPB_SINGLE_EXCLUSIVER))
630 
631 #define	vm_page_xbusied(m)						\
632 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
633 
634 #define	vm_page_xbusy(m) do {						\
635 	if (!vm_page_tryxbusy(m))					\
636 		panic("%s: page %p failed exclusive busying", __func__,	\
637 		    (m));						\
638 } while (0)
639 
640 /* Note: page m's lock must not be owned by the caller. */
641 #define	vm_page_xunbusy(m) do {						\
642 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
643 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
644 		vm_page_xunbusy_hard(m);				\
645 } while (0)
646 
647 #ifdef INVARIANTS
648 void vm_page_object_lock_assert(vm_page_t m);
649 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
650 void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits);
651 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
652 	vm_page_assert_pga_writeable(m, bits)
653 #else
654 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
655 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
656 #endif
657 
658 /*
659  * We want to use atomic updates for the aflags field, which is 8 bits wide.
660  * However, not all architectures support atomic operations on 8-bit
661  * destinations.  In order that we can easily use a 32-bit operation, we
662  * require that the aflags field be 32-bit aligned.
663  */
664 CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0);
665 
666 /*
667  *	Clear the given bits in the specified page.
668  */
669 static inline void
670 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
671 {
672 	uint32_t *addr, val;
673 
674 	/*
675 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
676 	 */
677 	if ((bits & PGA_REFERENCED) != 0)
678 		vm_page_assert_locked(m);
679 
680 	/*
681 	 * Access the whole 32-bit word containing the aflags field with an
682 	 * atomic update.  Parallel non-atomic updates to the other fields
683 	 * within this word are handled properly by the atomic update.
684 	 */
685 	addr = (void *)&m->aflags;
686 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
687 	    ("vm_page_aflag_clear: aflags is misaligned"));
688 	val = bits;
689 #if BYTE_ORDER == BIG_ENDIAN
690 	val <<= 24;
691 #endif
692 	atomic_clear_32(addr, val);
693 }
694 
695 /*
696  *	Set the given bits in the specified page.
697  */
698 static inline void
699 vm_page_aflag_set(vm_page_t m, uint8_t bits)
700 {
701 	uint32_t *addr, val;
702 
703 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
704 
705 	/*
706 	 * Access the whole 32-bit word containing the aflags field with an
707 	 * atomic update.  Parallel non-atomic updates to the other fields
708 	 * within this word are handled properly by the atomic update.
709 	 */
710 	addr = (void *)&m->aflags;
711 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
712 	    ("vm_page_aflag_set: aflags is misaligned"));
713 	val = bits;
714 #if BYTE_ORDER == BIG_ENDIAN
715 	val <<= 24;
716 #endif
717 	atomic_set_32(addr, val);
718 }
719 
720 /*
721  *	vm_page_dirty:
722  *
723  *	Set all bits in the page's dirty field.
724  *
725  *	The object containing the specified page must be locked if the
726  *	call is made from the machine-independent layer.
727  *
728  *	See vm_page_clear_dirty_mask().
729  */
730 static __inline void
731 vm_page_dirty(vm_page_t m)
732 {
733 
734 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
735 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
736 	vm_page_dirty_KBI(m);
737 #else
738 	m->dirty = VM_PAGE_BITS_ALL;
739 #endif
740 }
741 
742 /*
743  *	vm_page_undirty:
744  *
745  *	Set page to not be dirty.  Note: does not clear pmap modify bits
746  */
747 static __inline void
748 vm_page_undirty(vm_page_t m)
749 {
750 
751 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
752 	m->dirty = 0;
753 }
754 
755 static inline void
756 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
757     vm_page_t mold)
758 {
759 	vm_page_t mret;
760 
761 	mret = vm_page_replace(mnew, object, pindex);
762 	KASSERT(mret == mold,
763 	    ("invalid page replacement, mold=%p, mret=%p", mold, mret));
764 
765 	/* Unused if !INVARIANTS. */
766 	(void)mold;
767 	(void)mret;
768 }
769 
770 /*
771  *	vm_page_queue:
772  *
773  *	Return the index of the queue containing m.  This index is guaranteed
774  *	not to change while the page lock is held.
775  */
776 static inline uint8_t
777 vm_page_queue(vm_page_t m)
778 {
779 
780 	vm_page_assert_locked(m);
781 
782 	if ((m->aflags & PGA_DEQUEUE) != 0)
783 		return (PQ_NONE);
784 	atomic_thread_fence_acq();
785 	return (m->queue);
786 }
787 
788 static inline bool
789 vm_page_active(vm_page_t m)
790 {
791 
792 	return (vm_page_queue(m) == PQ_ACTIVE);
793 }
794 
795 static inline bool
796 vm_page_inactive(vm_page_t m)
797 {
798 
799 	return (vm_page_queue(m) == PQ_INACTIVE);
800 }
801 
802 static inline bool
803 vm_page_in_laundry(vm_page_t m)
804 {
805 	uint8_t queue;
806 
807 	queue = vm_page_queue(m);
808 	return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
809 }
810 
811 /*
812  *	vm_page_held:
813  *
814  *	Return true if a reference prevents the page from being reclaimable.
815  */
816 static inline bool
817 vm_page_held(vm_page_t m)
818 {
819 
820 	return (m->hold_count > 0 || m->wire_count > 0);
821 }
822 
823 #endif				/* _KERNEL */
824 #endif				/* !_VM_PAGE_ */
825