xref: /freebsd/sys/vm/vm_page.h (revision f6a3b357e9be4c6423c85eff9a847163a0d307c8)
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 page,
151  *	or PQ_NONE if the page is not enqueued.  The queue lock of a page is
152  *	the page queue lock corresponding to the page queue index, or the
153  *	page lock (P) for the page if it is not enqueued.  To modify the
154  *	queue field, the queue lock for the old value of the field must be
155  *	held.  There is one exception to this rule: the page daemon may
156  *	transition the queue field from PQ_INACTIVE to PQ_NONE immediately
157  *	prior to freeing a page during an inactive queue scan.  At that
158  *	point the page has already been physically dequeued and no other
159  *	references to that vm_page structure exist.
160  *
161  *	To avoid contention on page queue locks, page queue operations
162  *	(enqueue, dequeue, requeue) are batched using per-CPU queues.  A
163  *	deferred operation is requested by inserting an entry into a batch
164  *	queue; the entry is simply a pointer to the page, and the request
165  *	type is encoded in the page's aflags field using the values in
166  *	PGA_QUEUE_STATE_MASK.  The type-stability of struct vm_pages is
167  *	crucial to this scheme since the processing of entries in a given
168  *	batch queue may be deferred indefinitely.  In particular, a page may
169  *	be freed before its pending batch queue entries have been processed.
170  *	The page lock (P) must be held to schedule a batched queue
171  *	operation, and the page queue lock must be held in order to process
172  *	batch queue entries for the page queue.  There is one exception to
173  *	this rule: the thread freeing a page may schedule a dequeue without
174  *	holding the page lock.  In this scenario the only other thread which
175  *	may hold a reference to the page is the page daemon, which is
176  *	careful to avoid modifying the page's queue state once the dequeue
177  *	has been requested by setting PGA_DEQUEUE.
178  */
179 
180 #if PAGE_SIZE == 4096
181 #define VM_PAGE_BITS_ALL 0xffu
182 typedef uint8_t vm_page_bits_t;
183 #elif PAGE_SIZE == 8192
184 #define VM_PAGE_BITS_ALL 0xffffu
185 typedef uint16_t vm_page_bits_t;
186 #elif PAGE_SIZE == 16384
187 #define VM_PAGE_BITS_ALL 0xffffffffu
188 typedef uint32_t vm_page_bits_t;
189 #elif PAGE_SIZE == 32768
190 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
191 typedef uint64_t vm_page_bits_t;
192 #endif
193 
194 struct vm_page {
195 	union {
196 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
197 		struct {
198 			SLIST_ENTRY(vm_page) ss; /* private slists */
199 			void *pv;
200 		} s;
201 		struct {
202 			u_long p;
203 			u_long v;
204 		} memguard;
205 	} plinks;
206 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
207 	vm_object_t object;		/* which object am I in (O,P) */
208 	vm_pindex_t pindex;		/* offset into object (O,P) */
209 	vm_paddr_t phys_addr;		/* physical address of page (C) */
210 	struct md_page md;		/* machine dependent stuff */
211 	u_int wire_count;		/* wired down maps refs (P) */
212 	volatile u_int busy_lock;	/* busy owners lock */
213 	uint16_t flags;			/* page PG_* flags (P) */
214 	uint8_t	order;			/* index of the buddy queue (F) */
215 	uint8_t pool;			/* vm_phys freepool index (F) */
216 	uint8_t aflags;			/* access is atomic */
217 	uint8_t oflags;			/* page VPO_* flags (O) */
218 	uint8_t queue;			/* page queue index (Q) */
219 	int8_t psind;			/* pagesizes[] index (O) */
220 	int8_t segind;			/* vm_phys segment index (C) */
221 	u_char	act_count;		/* page usage count (P) */
222 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
223 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
224 	vm_page_bits_t valid;		/* map of valid DEV_BSIZE chunks (O) */
225 	vm_page_bits_t dirty;		/* map of dirty DEV_BSIZE chunks (M) */
226 };
227 
228 /*
229  * Page flags stored in oflags:
230  *
231  * Access to these page flags is synchronized by the lock on the object
232  * containing the page (O).
233  *
234  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
235  * 	 indicates that the page is not under PV management but
236  * 	 otherwise should be treated as a normal page.  Pages not
237  * 	 under PV management cannot be paged out via the
238  * 	 object/vm_page_t because there is no knowledge of their pte
239  * 	 mappings, and such pages are also not on any PQ queue.
240  *
241  */
242 #define	VPO_KMEM_EXEC	0x01		/* kmem mapping allows execution */
243 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
244 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
245 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
246 #define	VPO_NOSYNC	0x10		/* do not collect for syncer */
247 
248 /*
249  * Busy page implementation details.
250  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
251  * even if the support for owner identity is removed because of size
252  * constraints.  Checks on lock recursion are then not possible, while the
253  * lock assertions effectiveness is someway reduced.
254  */
255 #define	VPB_BIT_SHARED		0x01
256 #define	VPB_BIT_EXCLUSIVE	0x02
257 #define	VPB_BIT_WAITERS		0x04
258 #define	VPB_BIT_FLAGMASK						\
259 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
260 
261 #define	VPB_SHARERS_SHIFT	3
262 #define	VPB_SHARERS(x)							\
263 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
264 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
265 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
266 
267 #define	VPB_SINGLE_EXCLUSIVER	VPB_BIT_EXCLUSIVE
268 
269 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
270 
271 #define	PQ_NONE		255
272 #define	PQ_INACTIVE	0
273 #define	PQ_ACTIVE	1
274 #define	PQ_LAUNDRY	2
275 #define	PQ_UNSWAPPABLE	3
276 #define	PQ_COUNT	4
277 
278 #ifndef VM_PAGE_HAVE_PGLIST
279 TAILQ_HEAD(pglist, vm_page);
280 #define VM_PAGE_HAVE_PGLIST
281 #endif
282 SLIST_HEAD(spglist, vm_page);
283 
284 #ifdef _KERNEL
285 extern vm_page_t bogus_page;
286 #endif	/* _KERNEL */
287 
288 extern struct mtx_padalign pa_lock[];
289 
290 #if defined(__arm__)
291 #define	PDRSHIFT	PDR_SHIFT
292 #elif !defined(PDRSHIFT)
293 #define PDRSHIFT	21
294 #endif
295 
296 #define	pa_index(pa)	((pa) >> PDRSHIFT)
297 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
298 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
299 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
300 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
301 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
302 #define	PA_UNLOCK_COND(pa) 			\
303 	do {		   			\
304 		if ((pa) != 0) {		\
305 			PA_UNLOCK((pa));	\
306 			(pa) = 0;		\
307 		}				\
308 	} while (0)
309 
310 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
311 
312 #if defined(KLD_MODULE) && !defined(KLD_TIED)
313 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
314 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
315 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
316 #else	/* !KLD_MODULE */
317 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
318 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
319 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
320 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
321 #endif
322 #if defined(INVARIANTS)
323 #define	vm_page_assert_locked(m)		\
324     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
325 #define	vm_page_lock_assert(m, a)		\
326     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
327 #else
328 #define	vm_page_assert_locked(m)
329 #define	vm_page_lock_assert(m, a)
330 #endif
331 
332 /*
333  * The vm_page's aflags are updated using atomic operations.  To set or clear
334  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
335  * must be used.  Neither these flags nor these functions are part of the KBI.
336  *
337  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
338  * both the MI and MD VM layers.  However, kernel loadable modules should not
339  * directly set this flag.  They should call vm_page_reference() instead.
340  *
341  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
342  * When it does so, the object must be locked, or the page must be
343  * exclusive busied.  The MI VM layer must never access this flag
344  * directly.  Instead, it should call pmap_page_is_write_mapped().
345  *
346  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
347  * at least one executable mapping.  It is not consumed by the MI VM layer.
348  *
349  * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
350  * from a page queue, respectively.  It determines whether the plinks.q field
351  * of the page is valid.  To set or clear this flag, the queue lock for the
352  * page must be held: the page queue lock corresponding to the page's "queue"
353  * field if its value is not PQ_NONE, and the page lock otherwise.
354  *
355  * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
356  * queue, and cleared when the dequeue request is processed.  A page may
357  * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
358  * is requested after the page is scheduled to be enqueued but before it is
359  * actually inserted into the page queue.  For allocated pages, the page lock
360  * must be held to set this flag, but it may be set by vm_page_free_prep()
361  * without the page lock held.  The page queue lock must be held to clear the
362  * PGA_DEQUEUE flag.
363  *
364  * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
365  * in its page queue.  The page lock must be held to set this flag, and the
366  * queue lock for the page must be held to clear it.
367  *
368  * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
369  * the inactive queue, thus bypassing LRU.  The page lock must be held to
370  * set this flag, and the queue lock for the page must be held to clear it.
371  */
372 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
373 #define	PGA_REFERENCED	0x02		/* page has been referenced */
374 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
375 #define	PGA_ENQUEUED	0x08		/* page is enqueued in a page queue */
376 #define	PGA_DEQUEUE	0x10		/* page is due to be dequeued */
377 #define	PGA_REQUEUE	0x20		/* page is due to be requeued */
378 #define	PGA_REQUEUE_HEAD 0x40		/* page requeue should bypass LRU */
379 
380 #define	PGA_QUEUE_STATE_MASK	(PGA_ENQUEUED | PGA_DEQUEUE | PGA_REQUEUE | \
381 				PGA_REQUEUE_HEAD)
382 
383 /*
384  * Page flags.  If changed at any other time than page allocation or
385  * freeing, the modification must be protected by the vm_page lock.
386  *
387  * The PG_PCPU_CACHE flag is set at allocation time if the page was
388  * allocated from a per-CPU cache.  It is cleared the next time that the
389  * page is allocated from the physical memory allocator.
390  */
391 #define	PG_PCPU_CACHE	0x0001		/* was allocated from per-CPU caches */
392 #define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
393 #define	PG_ZERO		0x0008		/* page is zeroed */
394 #define	PG_MARKER	0x0010		/* special queue marker page */
395 #define	PG_NODUMP	0x0080		/* don't include this page in a dump */
396 
397 /*
398  * Misc constants.
399  */
400 #define ACT_DECLINE		1
401 #define ACT_ADVANCE		3
402 #define ACT_INIT		5
403 #define ACT_MAX			64
404 
405 #ifdef _KERNEL
406 
407 #include <sys/systm.h>
408 
409 #include <machine/atomic.h>
410 
411 /*
412  * Each pageable resident page falls into one of five lists:
413  *
414  *	free
415  *		Available for allocation now.
416  *
417  *	inactive
418  *		Low activity, candidates for reclamation.
419  *		This list is approximately LRU ordered.
420  *
421  *	laundry
422  *		This is the list of pages that should be
423  *		paged out next.
424  *
425  *	unswappable
426  *		Dirty anonymous pages that cannot be paged
427  *		out because no swap device is configured.
428  *
429  *	active
430  *		Pages that are "active", i.e., they have been
431  *		recently referenced.
432  *
433  */
434 
435 extern vm_page_t vm_page_array;		/* First resident page in table */
436 extern long vm_page_array_size;		/* number of vm_page_t's */
437 extern long first_page;			/* first physical page number */
438 
439 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
440 
441 /*
442  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
443  * page to which the given physical address belongs. The correct vm_page_t
444  * object is returned for addresses that are not page-aligned.
445  */
446 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
447 
448 /*
449  * Page allocation parameters for vm_page for the functions
450  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
451  * vm_page_alloc_freelist().  Some functions support only a subset
452  * of the flags, and ignore others, see the flags legend.
453  *
454  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
455  * and the vm_page_grab*() functions.  See these functions for details.
456  *
457  * Bits 0 - 1 define class.
458  * Bits 2 - 15 dedicated for flags.
459  * Legend:
460  * (a) - vm_page_alloc() supports the flag.
461  * (c) - vm_page_alloc_contig() supports the flag.
462  * (f) - vm_page_alloc_freelist() supports the flag.
463  * (g) - vm_page_grab() supports the flag.
464  * (p) - vm_page_grab_pages() supports the flag.
465  * Bits above 15 define the count of additional pages that the caller
466  * intends to allocate.
467  */
468 #define VM_ALLOC_NORMAL		0
469 #define VM_ALLOC_INTERRUPT	1
470 #define VM_ALLOC_SYSTEM		2
471 #define	VM_ALLOC_CLASS_MASK	3
472 #define	VM_ALLOC_WAITOK		0x0008	/* (acf) Sleep and retry */
473 #define	VM_ALLOC_WAITFAIL	0x0010	/* (acf) Sleep and return error */
474 #define	VM_ALLOC_WIRED		0x0020	/* (acfgp) Allocate a wired page */
475 #define	VM_ALLOC_ZERO		0x0040	/* (acfgp) Allocate a prezeroed page */
476 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
477 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
478 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
479 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
480 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
481 #define	VM_ALLOC_NOWAIT		0x8000	/* (acfgp) Do not sleep */
482 #define	VM_ALLOC_COUNT_SHIFT	16
483 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
484 
485 #ifdef M_NOWAIT
486 static inline int
487 malloc2vm_flags(int malloc_flags)
488 {
489 	int pflags;
490 
491 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
492 	    (malloc_flags & M_NOWAIT) != 0,
493 	    ("M_USE_RESERVE requires M_NOWAIT"));
494 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
495 	    VM_ALLOC_SYSTEM;
496 	if ((malloc_flags & M_ZERO) != 0)
497 		pflags |= VM_ALLOC_ZERO;
498 	if ((malloc_flags & M_NODUMP) != 0)
499 		pflags |= VM_ALLOC_NODUMP;
500 	if ((malloc_flags & M_NOWAIT))
501 		pflags |= VM_ALLOC_NOWAIT;
502 	if ((malloc_flags & M_WAITOK))
503 		pflags |= VM_ALLOC_WAITOK;
504 	return (pflags);
505 }
506 #endif
507 
508 /*
509  * Predicates supported by vm_page_ps_test():
510  *
511  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
512  *	However, it can be spuriously false when the (super)page has become
513  *	dirty in the pmap but that information has not been propagated to the
514  *	machine-independent layer.
515  */
516 #define	PS_ALL_DIRTY	0x1
517 #define	PS_ALL_VALID	0x2
518 #define	PS_NONE_BUSY	0x4
519 
520 void vm_page_busy_downgrade(vm_page_t m);
521 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
522 void vm_page_flash(vm_page_t m);
523 void vm_page_free(vm_page_t m);
524 void vm_page_free_zero(vm_page_t m);
525 
526 void vm_page_activate (vm_page_t);
527 void vm_page_advise(vm_page_t m, int advice);
528 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
529 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int);
530 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
531 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
532     vm_page_t);
533 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
534     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
535     vm_paddr_t boundary, vm_memattr_t memattr);
536 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
537     vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
538     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
539     vm_memattr_t memattr);
540 vm_page_t vm_page_alloc_freelist(int, int);
541 vm_page_t vm_page_alloc_freelist_domain(int, int, int);
542 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
543 void vm_page_change_lock(vm_page_t m, struct mtx **mtx);
544 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
545 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
546     vm_page_t *ma, int count);
547 void vm_page_deactivate(vm_page_t);
548 void vm_page_deactivate_noreuse(vm_page_t);
549 void vm_page_dequeue(vm_page_t m);
550 void vm_page_dequeue_deferred(vm_page_t m);
551 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
552 bool vm_page_free_prep(vm_page_t m);
553 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
554 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
555 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
556 void vm_page_launder(vm_page_t m);
557 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
558 vm_page_t vm_page_next(vm_page_t m);
559 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
560 void vm_page_pqbatch_drain(void);
561 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
562 vm_page_t vm_page_prev(vm_page_t m);
563 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
564 void vm_page_putfake(vm_page_t m);
565 void vm_page_readahead_finish(vm_page_t m);
566 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
567     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
568 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
569     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
570 void vm_page_reference(vm_page_t m);
571 #define	VPR_TRYFREE	0x01
572 #define	VPR_NOREUSE	0x02
573 void vm_page_release(vm_page_t m, int flags);
574 void vm_page_release_locked(vm_page_t m, int flags);
575 bool vm_page_remove(vm_page_t);
576 int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t);
577 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
578     vm_pindex_t pindex);
579 void vm_page_requeue(vm_page_t m);
580 int vm_page_sbusied(vm_page_t m);
581 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
582     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
583 void vm_page_set_valid_range(vm_page_t m, int base, int size);
584 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
585 vm_offset_t vm_page_startup(vm_offset_t vaddr);
586 void vm_page_sunbusy(vm_page_t m);
587 void vm_page_swapqueue(vm_page_t m, uint8_t oldq, uint8_t newq);
588 int vm_page_trysbusy(vm_page_t m);
589 void vm_page_unhold_pages(vm_page_t *ma, int count);
590 void vm_page_unswappable(vm_page_t m);
591 bool vm_page_unwire(vm_page_t m, uint8_t queue);
592 bool vm_page_unwire_noq(vm_page_t m);
593 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
594 void vm_page_wire (vm_page_t);
595 void vm_page_xunbusy_hard(vm_page_t m);
596 void vm_page_xunbusy_maybelocked(vm_page_t m);
597 void vm_page_set_validclean (vm_page_t, int, int);
598 void vm_page_clear_dirty (vm_page_t, int, int);
599 void vm_page_set_invalid (vm_page_t, int, int);
600 int vm_page_is_valid (vm_page_t, int, int);
601 void vm_page_test_dirty (vm_page_t);
602 vm_page_bits_t vm_page_bits(int base, int size);
603 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
604 void vm_page_free_toq(vm_page_t m);
605 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
606 
607 void vm_page_dirty_KBI(vm_page_t m);
608 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
609 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
610 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
611 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
612 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
613 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
614 #endif
615 
616 #define	vm_page_assert_sbusied(m)					\
617 	KASSERT(vm_page_sbusied(m),					\
618 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
619 	    (m), __FILE__, __LINE__))
620 
621 #define	vm_page_assert_unbusied(m)					\
622 	KASSERT(!vm_page_busied(m),					\
623 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
624 	    (m), __FILE__, __LINE__))
625 
626 #define	vm_page_assert_xbusied(m)					\
627 	KASSERT(vm_page_xbusied(m),					\
628 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
629 	    (m), __FILE__, __LINE__))
630 
631 #define	vm_page_busied(m)						\
632 	((m)->busy_lock != VPB_UNBUSIED)
633 
634 #define	vm_page_sbusy(m) do {						\
635 	if (!vm_page_trysbusy(m))					\
636 		panic("%s: page %p failed shared busying", __func__,	\
637 		    (m));						\
638 } while (0)
639 
640 #define	vm_page_tryxbusy(m)						\
641 	(atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED,		\
642 	    VPB_SINGLE_EXCLUSIVER))
643 
644 #define	vm_page_xbusied(m)						\
645 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
646 
647 #define	vm_page_xbusy(m) do {						\
648 	if (!vm_page_tryxbusy(m))					\
649 		panic("%s: page %p failed exclusive busying", __func__,	\
650 		    (m));						\
651 } while (0)
652 
653 /* Note: page m's lock must not be owned by the caller. */
654 #define	vm_page_xunbusy(m) do {						\
655 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
656 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
657 		vm_page_xunbusy_hard(m);				\
658 } while (0)
659 
660 #ifdef INVARIANTS
661 void vm_page_object_lock_assert(vm_page_t m);
662 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
663 void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits);
664 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
665 	vm_page_assert_pga_writeable(m, bits)
666 #else
667 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
668 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
669 #endif
670 
671 /*
672  * We want to use atomic updates for the aflags field, which is 8 bits wide.
673  * However, not all architectures support atomic operations on 8-bit
674  * destinations.  In order that we can easily use a 32-bit operation, we
675  * require that the aflags field be 32-bit aligned.
676  */
677 _Static_assert(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0,
678     "aflags field is not 32-bit aligned");
679 
680 /*
681  * We want to be able to update the aflags and queue fields atomically in
682  * the same operation.
683  */
684 _Static_assert(offsetof(struct vm_page, aflags) / sizeof(uint32_t) ==
685     offsetof(struct vm_page, queue) / sizeof(uint32_t),
686     "aflags and queue fields do not belong to the same 32-bit word");
687 _Static_assert(offsetof(struct vm_page, queue) % sizeof(uint32_t) == 2,
688     "queue field is at an unexpected offset");
689 _Static_assert(sizeof(((struct vm_page *)NULL)->queue) == 1,
690     "queue field has an unexpected size");
691 
692 #if BYTE_ORDER == LITTLE_ENDIAN
693 #define	VM_PAGE_AFLAG_SHIFT	0
694 #define	VM_PAGE_QUEUE_SHIFT	16
695 #else
696 #define	VM_PAGE_AFLAG_SHIFT	24
697 #define	VM_PAGE_QUEUE_SHIFT	8
698 #endif
699 #define	VM_PAGE_QUEUE_MASK	(0xff << VM_PAGE_QUEUE_SHIFT)
700 
701 /*
702  *	Clear the given bits in the specified page.
703  */
704 static inline void
705 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
706 {
707 	uint32_t *addr, val;
708 
709 	/*
710 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
711 	 */
712 	if ((bits & PGA_REFERENCED) != 0)
713 		vm_page_assert_locked(m);
714 
715 	/*
716 	 * Access the whole 32-bit word containing the aflags field with an
717 	 * atomic update.  Parallel non-atomic updates to the other fields
718 	 * within this word are handled properly by the atomic update.
719 	 */
720 	addr = (void *)&m->aflags;
721 	val = bits << VM_PAGE_AFLAG_SHIFT;
722 	atomic_clear_32(addr, val);
723 }
724 
725 /*
726  *	Set the given bits in the specified page.
727  */
728 static inline void
729 vm_page_aflag_set(vm_page_t m, uint8_t bits)
730 {
731 	uint32_t *addr, val;
732 
733 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
734 
735 	/*
736 	 * Access the whole 32-bit word containing the aflags field with an
737 	 * atomic update.  Parallel non-atomic updates to the other fields
738 	 * within this word are handled properly by the atomic update.
739 	 */
740 	addr = (void *)&m->aflags;
741 	val = bits << VM_PAGE_AFLAG_SHIFT;
742 	atomic_set_32(addr, val);
743 }
744 
745 /*
746  *	Atomically update the queue state of the page.  The operation fails if
747  *	any of the queue flags in "fflags" are set or if the "queue" field of
748  *	the page does not match the expected value; if the operation is
749  *	successful, the flags in "nflags" are set and all other queue state
750  *	flags are cleared.
751  */
752 static inline bool
753 vm_page_pqstate_cmpset(vm_page_t m, uint32_t oldq, uint32_t newq,
754     uint32_t fflags, uint32_t nflags)
755 {
756 	uint32_t *addr, nval, oval, qsmask;
757 
758 	vm_page_assert_locked(m);
759 
760 	fflags <<= VM_PAGE_AFLAG_SHIFT;
761 	nflags <<= VM_PAGE_AFLAG_SHIFT;
762 	newq <<= VM_PAGE_QUEUE_SHIFT;
763 	oldq <<= VM_PAGE_QUEUE_SHIFT;
764 	qsmask = ((PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD) <<
765 	    VM_PAGE_AFLAG_SHIFT) | VM_PAGE_QUEUE_MASK;
766 
767 	addr = (void *)&m->aflags;
768 	oval = atomic_load_32(addr);
769 	do {
770 		if ((oval & fflags) != 0)
771 			return (false);
772 		if ((oval & VM_PAGE_QUEUE_MASK) != oldq)
773 			return (false);
774 		nval = (oval & ~qsmask) | nflags | newq;
775 	} while (!atomic_fcmpset_32(addr, &oval, nval));
776 
777 	return (true);
778 }
779 
780 /*
781  *	vm_page_dirty:
782  *
783  *	Set all bits in the page's dirty field.
784  *
785  *	The object containing the specified page must be locked if the
786  *	call is made from the machine-independent layer.
787  *
788  *	See vm_page_clear_dirty_mask().
789  */
790 static __inline void
791 vm_page_dirty(vm_page_t m)
792 {
793 
794 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
795 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
796 	vm_page_dirty_KBI(m);
797 #else
798 	m->dirty = VM_PAGE_BITS_ALL;
799 #endif
800 }
801 
802 /*
803  *	vm_page_undirty:
804  *
805  *	Set page to not be dirty.  Note: does not clear pmap modify bits
806  */
807 static __inline void
808 vm_page_undirty(vm_page_t m)
809 {
810 
811 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
812 	m->dirty = 0;
813 }
814 
815 static inline void
816 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
817     vm_page_t mold)
818 {
819 	vm_page_t mret;
820 
821 	mret = vm_page_replace(mnew, object, pindex);
822 	KASSERT(mret == mold,
823 	    ("invalid page replacement, mold=%p, mret=%p", mold, mret));
824 
825 	/* Unused if !INVARIANTS. */
826 	(void)mold;
827 	(void)mret;
828 }
829 
830 /*
831  *	vm_page_queue:
832  *
833  *	Return the index of the queue containing m.  This index is guaranteed
834  *	not to change while the page lock is held.
835  */
836 static inline uint8_t
837 vm_page_queue(vm_page_t m)
838 {
839 
840 	vm_page_assert_locked(m);
841 
842 	if ((m->aflags & PGA_DEQUEUE) != 0)
843 		return (PQ_NONE);
844 	atomic_thread_fence_acq();
845 	return (m->queue);
846 }
847 
848 static inline bool
849 vm_page_active(vm_page_t m)
850 {
851 
852 	return (vm_page_queue(m) == PQ_ACTIVE);
853 }
854 
855 static inline bool
856 vm_page_inactive(vm_page_t m)
857 {
858 
859 	return (vm_page_queue(m) == PQ_INACTIVE);
860 }
861 
862 static inline bool
863 vm_page_in_laundry(vm_page_t m)
864 {
865 	uint8_t queue;
866 
867 	queue = vm_page_queue(m);
868 	return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
869 }
870 
871 /*
872  *	vm_page_wired:
873  *
874  *	Return true if a reference prevents the page from being reclaimable.
875  */
876 static inline bool
877 vm_page_wired(vm_page_t m)
878 {
879 
880 	return (m->wire_count > 0);
881 }
882 
883 #endif				/* _KERNEL */
884 #endif				/* !_VM_PAGE_ */
885