xref: /freebsd/sys/vm/vm_page.h (revision 1781ad707e1278d05f66038a44652a76a02f8567)
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 ref_count field tracks references to the page.  References that
119  *	prevent the page from being reclaimable are called wirings and are
120  *	counted in the low bits of ref_count.  The containing object's
121  *	reference, if one exists, is counted using the VPRC_OBJREF bit in the
122  *	ref_count field.  Additionally, the VPRC_BLOCKED bit is used to
123  *	atomically check for wirings and prevent new wirings via
124  *	pmap_extract_and_hold().  When a page belongs to an object, it may be
125  *	wired only when the object is locked, or the page is busy, or by
126  *	pmap_extract_and_hold().  As a result, if the object is locked and the
127  *	page is not busy (or is exclusively busied by the current thread), and
128  *	the page is unmapped, its wire count will not increase.  The ref_count
129  *	field is updated using atomic operations in most cases, except when it
130  *	is known that no other references to the page exist, such as in the page
131  *	allocator.  A page may be present in the page queues, or even actively
132  *	scanned by the page daemon, without an explicitly counted referenced.
133  *	The page daemon must therefore handle the possibility of a concurrent
134  *	free of the page.
135  *
136  *	The busy lock is an embedded reader-writer lock which protects the
137  *	page's contents and identity (i.e., its <object, pindex> tuple) and
138  *	interlocks with the object lock (O).  In particular, a page may be
139  *	busied or unbusied only with the object write lock held.  To avoid
140  *	bloating the page structure, the busy lock lacks some of the
141  *	features available to the kernel's general-purpose synchronization
142  *	primitives.  As a result, busy lock ordering rules are not verified,
143  *	lock recursion is not detected, and an attempt to xbusy a busy page
144  *	or sbusy an xbusy page results will trigger a panic rather than
145  *	causing the thread to block.  vm_page_sleep_if_busy() can be used to
146  *	sleep until the page's busy state changes, after which the caller
147  *	must re-lookup the page and re-evaluate its state.
148  *
149  *	The queue field is the index of the page queue containing the page,
150  *	or PQ_NONE if the page is not enqueued.  The queue lock of a page is
151  *	the page queue lock corresponding to the page queue index, or the
152  *	page lock (P) for the page if it is not enqueued.  To modify the
153  *	queue field, the queue lock for the old value of the field must be
154  *	held.  There is one exception to this rule: the page daemon may
155  *	transition the queue field from PQ_INACTIVE to PQ_NONE immediately
156  *	prior to freeing a page during an inactive queue scan.  At that
157  *	point the page has already been physically dequeued and no other
158  *	references to that vm_page structure exist.
159  *
160  *	To avoid contention on page queue locks, page queue operations
161  *	(enqueue, dequeue, requeue) are batched using per-CPU queues.  A
162  *	deferred operation is requested by inserting an entry into a batch
163  *	queue; the entry is simply a pointer to the page, and the request
164  *	type is encoded in the page's aflags field using the values in
165  *	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 may
168  *	be freed before its pending batch queue entries have been processed.
169  *	The page lock (P) must be held to schedule a batched queue
170  *	operation, and the page queue lock must be held in order to process
171  *	batch queue entries for the page queue.  There is one exception to
172  *	this rule: the thread freeing a page may schedule a dequeue without
173  *	holding the page lock.  In this scenario the only other thread which
174  *	may hold a reference to the page is the page daemon, which is
175  *	careful to avoid modifying the page's queue state once the dequeue
176  *	has been requested by setting PGA_DEQUEUE.
177  */
178 
179 #if PAGE_SIZE == 4096
180 #define VM_PAGE_BITS_ALL 0xffu
181 typedef uint8_t vm_page_bits_t;
182 #elif PAGE_SIZE == 8192
183 #define VM_PAGE_BITS_ALL 0xffffu
184 typedef uint16_t vm_page_bits_t;
185 #elif PAGE_SIZE == 16384
186 #define VM_PAGE_BITS_ALL 0xffffffffu
187 typedef uint32_t vm_page_bits_t;
188 #elif PAGE_SIZE == 32768
189 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
190 typedef uint64_t vm_page_bits_t;
191 #endif
192 
193 struct vm_page {
194 	union {
195 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
196 		struct {
197 			SLIST_ENTRY(vm_page) ss; /* private slists */
198 			void *pv;
199 		} s;
200 		struct {
201 			u_long p;
202 			u_long v;
203 		} memguard;
204 	} plinks;
205 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
206 	vm_object_t object;		/* which object am I in (O) */
207 	vm_pindex_t pindex;		/* offset into object (O,P) */
208 	vm_paddr_t phys_addr;		/* physical address of page (C) */
209 	struct md_page md;		/* machine dependent stuff */
210 	union {
211 		u_int wire_count;
212 		u_int ref_count;	/* page references */
213 	};
214 	volatile u_int busy_lock;	/* busy owners lock */
215 	uint16_t flags;			/* page PG_* flags (P) */
216 	uint8_t	order;			/* index of the buddy queue (F) */
217 	uint8_t pool;			/* vm_phys freepool index (F) */
218 	uint8_t aflags;			/* access is atomic */
219 	uint8_t oflags;			/* page VPO_* flags (O) */
220 	uint8_t queue;			/* page queue index (Q) */
221 	int8_t psind;			/* pagesizes[] index (O) */
222 	int8_t segind;			/* vm_phys segment index (C) */
223 	u_char	act_count;		/* page usage count (P) */
224 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
225 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
226 	vm_page_bits_t valid;		/* map of valid DEV_BSIZE chunks (O) */
227 	vm_page_bits_t dirty;		/* map of dirty DEV_BSIZE chunks (M) */
228 };
229 
230 /*
231  * Special bits used in the ref_count field.
232  *
233  * ref_count is normally used to count wirings that prevent the page from being
234  * reclaimed, but also supports several special types of references that do not
235  * prevent reclamation.  Accesses to the ref_count field must be atomic unless
236  * the page is unallocated.
237  *
238  * VPRC_OBJREF is the reference held by the containing object.  It can set or
239  * cleared only when the corresponding object's write lock is held.
240  *
241  * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while
242  * attempting to tear down all mappings of a given page.  The page lock and
243  * object write lock must both be held in order to set or clear this bit.
244  */
245 #define	VPRC_BLOCKED	0x40000000u	/* mappings are being removed */
246 #define	VPRC_OBJREF	0x80000000u	/* object reference, cleared with (O) */
247 #define	VPRC_WIRE_COUNT(c)	((c) & ~(VPRC_BLOCKED | VPRC_OBJREF))
248 #define	VPRC_WIRE_COUNT_MAX	(~(VPRC_BLOCKED | VPRC_OBJREF))
249 
250 /*
251  * Page flags stored in oflags:
252  *
253  * Access to these page flags is synchronized by the lock on the object
254  * containing the page (O).
255  *
256  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
257  * 	 indicates that the page is not under PV management but
258  * 	 otherwise should be treated as a normal page.  Pages not
259  * 	 under PV management cannot be paged out via the
260  * 	 object/vm_page_t because there is no knowledge of their pte
261  * 	 mappings, and such pages are also not on any PQ queue.
262  *
263  */
264 #define	VPO_KMEM_EXEC	0x01		/* kmem mapping allows execution */
265 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
266 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
267 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
268 #define	VPO_NOSYNC	0x10		/* do not collect for syncer */
269 
270 /*
271  * Busy page implementation details.
272  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
273  * even if the support for owner identity is removed because of size
274  * constraints.  Checks on lock recursion are then not possible, while the
275  * lock assertions effectiveness is someway reduced.
276  */
277 #define	VPB_BIT_SHARED		0x01
278 #define	VPB_BIT_EXCLUSIVE	0x02
279 #define	VPB_BIT_WAITERS		0x04
280 #define	VPB_BIT_FLAGMASK						\
281 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
282 
283 #define	VPB_SHARERS_SHIFT	3
284 #define	VPB_SHARERS(x)							\
285 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
286 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
287 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
288 
289 #define	VPB_SINGLE_EXCLUSIVER	VPB_BIT_EXCLUSIVE
290 
291 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
292 
293 #define	PQ_NONE		255
294 #define	PQ_INACTIVE	0
295 #define	PQ_ACTIVE	1
296 #define	PQ_LAUNDRY	2
297 #define	PQ_UNSWAPPABLE	3
298 #define	PQ_COUNT	4
299 
300 #ifndef VM_PAGE_HAVE_PGLIST
301 TAILQ_HEAD(pglist, vm_page);
302 #define VM_PAGE_HAVE_PGLIST
303 #endif
304 SLIST_HEAD(spglist, vm_page);
305 
306 #ifdef _KERNEL
307 extern vm_page_t bogus_page;
308 #endif	/* _KERNEL */
309 
310 extern struct mtx_padalign pa_lock[];
311 
312 #if defined(__arm__)
313 #define	PDRSHIFT	PDR_SHIFT
314 #elif !defined(PDRSHIFT)
315 #define PDRSHIFT	21
316 #endif
317 
318 #define	pa_index(pa)	((pa) >> PDRSHIFT)
319 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
320 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
321 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
322 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
323 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
324 #define	PA_UNLOCK_COND(pa) 			\
325 	do {		   			\
326 		if ((pa) != 0) {		\
327 			PA_UNLOCK((pa));	\
328 			(pa) = 0;		\
329 		}				\
330 	} while (0)
331 
332 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
333 
334 #if defined(KLD_MODULE) && !defined(KLD_TIED)
335 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
336 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
337 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
338 #else	/* !KLD_MODULE */
339 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
340 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
341 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
342 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
343 #endif
344 #if defined(INVARIANTS)
345 #define	vm_page_assert_locked(m)		\
346     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
347 #define	vm_page_lock_assert(m, a)		\
348     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
349 #else
350 #define	vm_page_assert_locked(m)
351 #define	vm_page_lock_assert(m, a)
352 #endif
353 
354 /*
355  * The vm_page's aflags are updated using atomic operations.  To set or clear
356  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
357  * must be used.  Neither these flags nor these functions are part of the KBI.
358  *
359  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
360  * both the MI and MD VM layers.  However, kernel loadable modules should not
361  * directly set this flag.  They should call vm_page_reference() instead.
362  *
363  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
364  * When it does so, the object must be locked, or the page must be
365  * exclusive busied.  The MI VM layer must never access this flag
366  * directly.  Instead, it should call pmap_page_is_write_mapped().
367  *
368  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
369  * at least one executable mapping.  It is not consumed by the MI VM layer.
370  *
371  * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
372  * from a page queue, respectively.  It determines whether the plinks.q field
373  * of the page is valid.  To set or clear this flag, the queue lock for the
374  * page must be held: the page queue lock corresponding to the page's "queue"
375  * field if its value is not PQ_NONE, and the page lock otherwise.
376  *
377  * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
378  * queue, and cleared when the dequeue request is processed.  A page may
379  * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
380  * is requested after the page is scheduled to be enqueued but before it is
381  * actually inserted into the page queue.  For allocated pages, the page lock
382  * must be held to set this flag, but it may be set by vm_page_free_prep()
383  * without the page lock held.  The page queue lock must be held to clear the
384  * PGA_DEQUEUE flag.
385  *
386  * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
387  * in its page queue.  The page lock must be held to set this flag, and the
388  * queue lock for the page must be held to clear it.
389  *
390  * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
391  * the inactive queue, thus bypassing LRU.  The page lock must be held to
392  * set this flag, and the queue lock for the page must be held to clear it.
393  */
394 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
395 #define	PGA_REFERENCED	0x02		/* page has been referenced */
396 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
397 #define	PGA_ENQUEUED	0x08		/* page is enqueued in a page queue */
398 #define	PGA_DEQUEUE	0x10		/* page is due to be dequeued */
399 #define	PGA_REQUEUE	0x20		/* page is due to be requeued */
400 #define	PGA_REQUEUE_HEAD 0x40		/* page requeue should bypass LRU */
401 
402 #define	PGA_QUEUE_STATE_MASK	(PGA_ENQUEUED | PGA_DEQUEUE | PGA_REQUEUE | \
403 				PGA_REQUEUE_HEAD)
404 
405 /*
406  * Page flags.  If changed at any other time than page allocation or
407  * freeing, the modification must be protected by the vm_page lock.
408  *
409  * The PG_PCPU_CACHE flag is set at allocation time if the page was
410  * allocated from a per-CPU cache.  It is cleared the next time that the
411  * page is allocated from the physical memory allocator.
412  */
413 #define	PG_PCPU_CACHE	0x0001		/* was allocated from per-CPU caches */
414 #define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
415 #define	PG_ZERO		0x0008		/* page is zeroed */
416 #define	PG_MARKER	0x0010		/* special queue marker page */
417 #define	PG_NODUMP	0x0080		/* don't include this page in a dump */
418 
419 /*
420  * Misc constants.
421  */
422 #define ACT_DECLINE		1
423 #define ACT_ADVANCE		3
424 #define ACT_INIT		5
425 #define ACT_MAX			64
426 
427 #ifdef _KERNEL
428 
429 #include <sys/systm.h>
430 
431 #include <machine/atomic.h>
432 
433 /*
434  * Each pageable resident page falls into one of five lists:
435  *
436  *	free
437  *		Available for allocation now.
438  *
439  *	inactive
440  *		Low activity, candidates for reclamation.
441  *		This list is approximately LRU ordered.
442  *
443  *	laundry
444  *		This is the list of pages that should be
445  *		paged out next.
446  *
447  *	unswappable
448  *		Dirty anonymous pages that cannot be paged
449  *		out because no swap device is configured.
450  *
451  *	active
452  *		Pages that are "active", i.e., they have been
453  *		recently referenced.
454  *
455  */
456 
457 extern vm_page_t vm_page_array;		/* First resident page in table */
458 extern long vm_page_array_size;		/* number of vm_page_t's */
459 extern long first_page;			/* first physical page number */
460 
461 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
462 
463 /*
464  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
465  * page to which the given physical address belongs. The correct vm_page_t
466  * object is returned for addresses that are not page-aligned.
467  */
468 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
469 
470 /*
471  * Page allocation parameters for vm_page for the functions
472  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
473  * vm_page_alloc_freelist().  Some functions support only a subset
474  * of the flags, and ignore others, see the flags legend.
475  *
476  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
477  * and the vm_page_grab*() functions.  See these functions for details.
478  *
479  * Bits 0 - 1 define class.
480  * Bits 2 - 15 dedicated for flags.
481  * Legend:
482  * (a) - vm_page_alloc() supports the flag.
483  * (c) - vm_page_alloc_contig() supports the flag.
484  * (f) - vm_page_alloc_freelist() supports the flag.
485  * (g) - vm_page_grab() supports the flag.
486  * (p) - vm_page_grab_pages() supports the flag.
487  * Bits above 15 define the count of additional pages that the caller
488  * intends to allocate.
489  */
490 #define VM_ALLOC_NORMAL		0
491 #define VM_ALLOC_INTERRUPT	1
492 #define VM_ALLOC_SYSTEM		2
493 #define	VM_ALLOC_CLASS_MASK	3
494 #define	VM_ALLOC_WAITOK		0x0008	/* (acf) Sleep and retry */
495 #define	VM_ALLOC_WAITFAIL	0x0010	/* (acf) Sleep and return error */
496 #define	VM_ALLOC_WIRED		0x0020	/* (acfgp) Allocate a wired page */
497 #define	VM_ALLOC_ZERO		0x0040	/* (acfgp) Allocate a prezeroed page */
498 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
499 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
500 #define	VM_ALLOC_NOCREAT	0x0400	/* (gp) Don't create a page */
501 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
502 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
503 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
504 #define	VM_ALLOC_NOWAIT		0x8000	/* (acfgp) Do not sleep */
505 #define	VM_ALLOC_COUNT_SHIFT	16
506 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
507 
508 #ifdef M_NOWAIT
509 static inline int
510 malloc2vm_flags(int malloc_flags)
511 {
512 	int pflags;
513 
514 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
515 	    (malloc_flags & M_NOWAIT) != 0,
516 	    ("M_USE_RESERVE requires M_NOWAIT"));
517 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
518 	    VM_ALLOC_SYSTEM;
519 	if ((malloc_flags & M_ZERO) != 0)
520 		pflags |= VM_ALLOC_ZERO;
521 	if ((malloc_flags & M_NODUMP) != 0)
522 		pflags |= VM_ALLOC_NODUMP;
523 	if ((malloc_flags & M_NOWAIT))
524 		pflags |= VM_ALLOC_NOWAIT;
525 	if ((malloc_flags & M_WAITOK))
526 		pflags |= VM_ALLOC_WAITOK;
527 	return (pflags);
528 }
529 #endif
530 
531 /*
532  * Predicates supported by vm_page_ps_test():
533  *
534  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
535  *	However, it can be spuriously false when the (super)page has become
536  *	dirty in the pmap but that information has not been propagated to the
537  *	machine-independent layer.
538  */
539 #define	PS_ALL_DIRTY	0x1
540 #define	PS_ALL_VALID	0x2
541 #define	PS_NONE_BUSY	0x4
542 
543 int vm_page_busy_acquire(vm_page_t m, int allocflags);
544 void vm_page_busy_downgrade(vm_page_t m);
545 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
546 void vm_page_free(vm_page_t m);
547 void vm_page_free_zero(vm_page_t m);
548 
549 void vm_page_activate (vm_page_t);
550 void vm_page_advise(vm_page_t m, int advice);
551 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
552 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int);
553 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
554 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
555     vm_page_t);
556 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
557     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
558     vm_paddr_t boundary, vm_memattr_t memattr);
559 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
560     vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
561     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
562     vm_memattr_t memattr);
563 vm_page_t vm_page_alloc_freelist(int, int);
564 vm_page_t vm_page_alloc_freelist_domain(int, int, int);
565 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
566 void vm_page_change_lock(vm_page_t m, struct mtx **mtx);
567 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
568 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
569     vm_page_t *ma, int count);
570 int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex,
571     int allocflags);
572 void vm_page_deactivate(vm_page_t);
573 void vm_page_deactivate_noreuse(vm_page_t);
574 void vm_page_dequeue(vm_page_t m);
575 void vm_page_dequeue_deferred(vm_page_t m);
576 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
577 bool vm_page_free_prep(vm_page_t m);
578 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
579 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
580 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
581 void vm_page_launder(vm_page_t m);
582 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
583 vm_page_t vm_page_next(vm_page_t m);
584 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
585 void vm_page_pqbatch_drain(void);
586 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
587 vm_page_t vm_page_prev(vm_page_t m);
588 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
589 void vm_page_putfake(vm_page_t m);
590 void vm_page_readahead_finish(vm_page_t m);
591 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
592     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
593 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
594     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
595 void vm_page_reference(vm_page_t m);
596 #define	VPR_TRYFREE	0x01
597 #define	VPR_NOREUSE	0x02
598 void vm_page_release(vm_page_t m, int flags);
599 void vm_page_release_locked(vm_page_t m, int flags);
600 bool vm_page_remove(vm_page_t);
601 int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t);
602 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
603     vm_pindex_t pindex);
604 void vm_page_requeue(vm_page_t m);
605 int vm_page_sbusied(vm_page_t m);
606 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
607     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
608 void vm_page_set_valid_range(vm_page_t m, int base, int size);
609 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
610 int vm_page_sleep_if_xbusy(vm_page_t m, const char *msg);
611 vm_offset_t vm_page_startup(vm_offset_t vaddr);
612 void vm_page_sunbusy(vm_page_t m);
613 void vm_page_swapqueue(vm_page_t m, uint8_t oldq, uint8_t newq);
614 bool vm_page_try_remove_all(vm_page_t m);
615 bool vm_page_try_remove_write(vm_page_t m);
616 int vm_page_trysbusy(vm_page_t m);
617 void vm_page_unhold_pages(vm_page_t *ma, int count);
618 void vm_page_unswappable(vm_page_t m);
619 void vm_page_unwire(vm_page_t m, uint8_t queue);
620 bool vm_page_unwire_noq(vm_page_t m);
621 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
622 void vm_page_wire(vm_page_t);
623 bool vm_page_wire_mapped(vm_page_t m);
624 void vm_page_xunbusy_hard(vm_page_t m);
625 void vm_page_set_validclean (vm_page_t, int, int);
626 void vm_page_clear_dirty (vm_page_t, int, int);
627 void vm_page_set_invalid (vm_page_t, int, int);
628 int vm_page_is_valid (vm_page_t, int, int);
629 void vm_page_test_dirty (vm_page_t);
630 vm_page_bits_t vm_page_bits(int base, int size);
631 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
632 void vm_page_free_toq(vm_page_t m);
633 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
634 
635 void vm_page_dirty_KBI(vm_page_t m);
636 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
637 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
638 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
639 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
640 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
641 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
642 #endif
643 
644 #define	vm_page_assert_sbusied(m)					\
645 	KASSERT(vm_page_sbusied(m),					\
646 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
647 	    (m), __FILE__, __LINE__))
648 
649 #define	vm_page_assert_unbusied(m)					\
650 	KASSERT(!vm_page_busied(m),					\
651 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
652 	    (m), __FILE__, __LINE__))
653 
654 #define	vm_page_assert_xbusied(m)					\
655 	KASSERT(vm_page_xbusied(m),					\
656 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
657 	    (m), __FILE__, __LINE__))
658 
659 #define	vm_page_busied(m)						\
660 	((m)->busy_lock != VPB_UNBUSIED)
661 
662 #define	vm_page_sbusy(m) do {						\
663 	if (!vm_page_trysbusy(m))					\
664 		panic("%s: page %p failed shared busying", __func__,	\
665 		    (m));						\
666 } while (0)
667 
668 #define	vm_page_tryxbusy(m)						\
669 	(atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED,		\
670 	    VPB_SINGLE_EXCLUSIVER))
671 
672 #define	vm_page_xbusied(m)						\
673 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
674 
675 #define	vm_page_xbusy(m) do {						\
676 	if (!vm_page_tryxbusy(m))					\
677 		panic("%s: page %p failed exclusive busying", __func__,	\
678 		    (m));						\
679 } while (0)
680 
681 /* Note: page m's lock must not be owned by the caller. */
682 #define	vm_page_xunbusy(m) do {						\
683 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
684 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
685 		vm_page_xunbusy_hard(m);				\
686 } while (0)
687 
688 #ifdef INVARIANTS
689 void vm_page_object_lock_assert(vm_page_t m);
690 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
691 void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits);
692 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
693 	vm_page_assert_pga_writeable(m, bits)
694 #else
695 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
696 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
697 #endif
698 
699 /*
700  * We want to use atomic updates for the aflags field, which is 8 bits wide.
701  * However, not all architectures support atomic operations on 8-bit
702  * destinations.  In order that we can easily use a 32-bit operation, we
703  * require that the aflags field be 32-bit aligned.
704  */
705 _Static_assert(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0,
706     "aflags field is not 32-bit aligned");
707 
708 /*
709  * We want to be able to update the aflags and queue fields atomically in
710  * the same operation.
711  */
712 _Static_assert(offsetof(struct vm_page, aflags) / sizeof(uint32_t) ==
713     offsetof(struct vm_page, queue) / sizeof(uint32_t),
714     "aflags and queue fields do not belong to the same 32-bit word");
715 _Static_assert(offsetof(struct vm_page, queue) % sizeof(uint32_t) == 2,
716     "queue field is at an unexpected offset");
717 _Static_assert(sizeof(((struct vm_page *)NULL)->queue) == 1,
718     "queue field has an unexpected size");
719 
720 #if BYTE_ORDER == LITTLE_ENDIAN
721 #define	VM_PAGE_AFLAG_SHIFT	0
722 #define	VM_PAGE_QUEUE_SHIFT	16
723 #else
724 #define	VM_PAGE_AFLAG_SHIFT	24
725 #define	VM_PAGE_QUEUE_SHIFT	8
726 #endif
727 #define	VM_PAGE_QUEUE_MASK	(0xff << VM_PAGE_QUEUE_SHIFT)
728 
729 /*
730  *	Clear the given bits in the specified page.
731  */
732 static inline void
733 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
734 {
735 	uint32_t *addr, val;
736 
737 	/*
738 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
739 	 */
740 	if ((bits & PGA_REFERENCED) != 0)
741 		vm_page_assert_locked(m);
742 
743 	/*
744 	 * Access the whole 32-bit word containing the aflags field with an
745 	 * atomic update.  Parallel non-atomic updates to the other fields
746 	 * within this word are handled properly by the atomic update.
747 	 */
748 	addr = (void *)&m->aflags;
749 	val = bits << VM_PAGE_AFLAG_SHIFT;
750 	atomic_clear_32(addr, val);
751 }
752 
753 /*
754  *	Set the given bits in the specified page.
755  */
756 static inline void
757 vm_page_aflag_set(vm_page_t m, uint8_t bits)
758 {
759 	uint32_t *addr, val;
760 
761 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
762 
763 	/*
764 	 * Access the whole 32-bit word containing the aflags field with an
765 	 * atomic update.  Parallel non-atomic updates to the other fields
766 	 * within this word are handled properly by the atomic update.
767 	 */
768 	addr = (void *)&m->aflags;
769 	val = bits << VM_PAGE_AFLAG_SHIFT;
770 	atomic_set_32(addr, val);
771 }
772 
773 /*
774  *	Atomically update the queue state of the page.  The operation fails if
775  *	any of the queue flags in "fflags" are set or if the "queue" field of
776  *	the page does not match the expected value; if the operation is
777  *	successful, the flags in "nflags" are set and all other queue state
778  *	flags are cleared.
779  */
780 static inline bool
781 vm_page_pqstate_cmpset(vm_page_t m, uint32_t oldq, uint32_t newq,
782     uint32_t fflags, uint32_t nflags)
783 {
784 	uint32_t *addr, nval, oval, qsmask;
785 
786 	fflags <<= VM_PAGE_AFLAG_SHIFT;
787 	nflags <<= VM_PAGE_AFLAG_SHIFT;
788 	newq <<= VM_PAGE_QUEUE_SHIFT;
789 	oldq <<= VM_PAGE_QUEUE_SHIFT;
790 	qsmask = ((PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD) <<
791 	    VM_PAGE_AFLAG_SHIFT) | VM_PAGE_QUEUE_MASK;
792 
793 	addr = (void *)&m->aflags;
794 	oval = atomic_load_32(addr);
795 	do {
796 		if ((oval & fflags) != 0)
797 			return (false);
798 		if ((oval & VM_PAGE_QUEUE_MASK) != oldq)
799 			return (false);
800 		nval = (oval & ~qsmask) | nflags | newq;
801 	} while (!atomic_fcmpset_32(addr, &oval, nval));
802 
803 	return (true);
804 }
805 
806 /*
807  *	vm_page_dirty:
808  *
809  *	Set all bits in the page's dirty field.
810  *
811  *	The object containing the specified page must be locked if the
812  *	call is made from the machine-independent layer.
813  *
814  *	See vm_page_clear_dirty_mask().
815  */
816 static __inline void
817 vm_page_dirty(vm_page_t m)
818 {
819 
820 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
821 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
822 	vm_page_dirty_KBI(m);
823 #else
824 	m->dirty = VM_PAGE_BITS_ALL;
825 #endif
826 }
827 
828 /*
829  *	vm_page_undirty:
830  *
831  *	Set page to not be dirty.  Note: does not clear pmap modify bits
832  */
833 static __inline void
834 vm_page_undirty(vm_page_t m)
835 {
836 
837 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
838 	m->dirty = 0;
839 }
840 
841 static inline void
842 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
843     vm_page_t mold)
844 {
845 	vm_page_t mret;
846 
847 	mret = vm_page_replace(mnew, object, pindex);
848 	KASSERT(mret == mold,
849 	    ("invalid page replacement, mold=%p, mret=%p", mold, mret));
850 
851 	/* Unused if !INVARIANTS. */
852 	(void)mold;
853 	(void)mret;
854 }
855 
856 /*
857  *	vm_page_queue:
858  *
859  *	Return the index of the queue containing m.  This index is guaranteed
860  *	not to change while the page lock is held.
861  */
862 static inline uint8_t
863 vm_page_queue(vm_page_t m)
864 {
865 
866 	vm_page_assert_locked(m);
867 
868 	if ((m->aflags & PGA_DEQUEUE) != 0)
869 		return (PQ_NONE);
870 	atomic_thread_fence_acq();
871 	return (m->queue);
872 }
873 
874 static inline bool
875 vm_page_active(vm_page_t m)
876 {
877 
878 	return (vm_page_queue(m) == PQ_ACTIVE);
879 }
880 
881 static inline bool
882 vm_page_inactive(vm_page_t m)
883 {
884 
885 	return (vm_page_queue(m) == PQ_INACTIVE);
886 }
887 
888 static inline bool
889 vm_page_in_laundry(vm_page_t m)
890 {
891 	uint8_t queue;
892 
893 	queue = vm_page_queue(m);
894 	return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
895 }
896 
897 /*
898  *	vm_page_drop:
899  *
900  *	Release a reference to a page and return the old reference count.
901  */
902 static inline u_int
903 vm_page_drop(vm_page_t m, u_int val)
904 {
905 	u_int old;
906 
907 	/*
908 	 * Synchronize with vm_page_free_prep(): ensure that all updates to the
909 	 * page structure are visible before it is freed.
910 	 */
911 	atomic_thread_fence_rel();
912 	old = atomic_fetchadd_int(&m->ref_count, -val);
913 	KASSERT(old != VPRC_BLOCKED,
914 	    ("vm_page_drop: page %p has an invalid refcount value", m));
915 	return (old);
916 }
917 
918 /*
919  *	vm_page_wired:
920  *
921  *	Perform a racy check to determine whether a reference prevents the page
922  *	from being reclaimable.  If the page's object is locked, and the page is
923  *	unmapped and unbusied or exclusively busied by the current thread, no
924  *	new wirings may be created.
925  */
926 static inline bool
927 vm_page_wired(vm_page_t m)
928 {
929 
930 	return (VPRC_WIRE_COUNT(m->ref_count) > 0);
931 }
932 
933 #endif				/* _KERNEL */
934 #endif				/* !_VM_PAGE_ */
935