xref: /freebsd/sys/vm/vm_page.h (revision ce6a89e27cd190313be39bb479880aeda4778436)
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 locks.  If a
96  *	field is annotated with two of these locks then holding either is
97  *	sufficient for read access but both are required for write access.
98  *	The physical address of a page is used to select its page lock from
99  *	a pool.  The queue lock for a page depends on the value of its queue
100  *	field and is described in detail below.
101  *
102  *	The following annotations are possible:
103  *	(A) the field is atomic and may require additional synchronization.
104  *	(B) the page busy lock.
105  *	(C) the field is immutable.
106  *	(F) the per-domain lock for the free queues
107  *	(M) Machine dependent, defined by pmap layer.
108  *	(O) the object that the page belongs to.
109  *	(P) the page lock.
110  *	(Q) the page's queue lock.
111  *
112  *	The busy lock is an embedded reader-writer lock that protects the
113  *	page's contents and identity (i.e., its <object, pindex> tuple) as
114  *	well as certain valid/dirty modifications.  To avoid bloating the
115  *	the page structure, the busy lock lacks some of the features available
116  *	the kernel's general-purpose synchronization primitives.  As a result,
117  *	busy lock ordering rules are not verified, lock recursion is not
118  *	detected, and an attempt to xbusy a busy page or sbusy an xbusy page
119  *	results will trigger a panic rather than causing the thread to block.
120  *	vm_page_sleep_if_busy() can be used to sleep until the page's busy
121  *	state changes, after which the caller must re-lookup the page and
122  *	re-evaluate its state.  vm_page_busy_acquire() will block until
123  *	the lock is acquired.
124  *
125  *	The valid field is protected by the page busy lock (B) and object
126  *	lock (O).  Transitions from invalid to valid are generally done
127  *	via I/O or zero filling and do not require the object lock.
128  *	These must be protected with the busy lock to prevent page-in or
129  *	creation races.  Page invalidation generally happens as a result
130  *	of truncate or msync.  When invalidated, pages must not be present
131  *	in pmap and must hold the object lock to prevent concurrent
132  *	speculative read-only mappings that do not require busy.  I/O
133  *	routines may check for validity without a lock if they are prepared
134  *	to handle invalidation races with higher level locks (vnode) or are
135  *	unconcerned with races so long as they hold a reference to prevent
136  *	recycling.  When a valid bit is set while holding a shared busy
137  *	lock (A) atomic operations are used to protect against concurrent
138  *	modification.
139  *
140  *	In contrast, the synchronization of accesses to the page's
141  *	dirty field is a mix of machine dependent (M) and busy (B).  In
142  *	the machine-independent layer, the page busy must be held to
143  *	operate on the field.  However, the pmap layer is permitted to
144  *	set all bits within the field without holding that lock.  If the
145  *	underlying architecture does not support atomic read-modify-write
146  *	operations on the field's type, then the machine-independent
147  *	layer uses a 32-bit atomic on the aligned 32-bit word that
148  *	contains the dirty field.  In the machine-independent layer,
149  *	the implementation of read-modify-write operations on the
150  *	field is encapsulated in vm_page_clear_dirty_mask().  An
151  *	exclusive busy lock combined with pmap_remove_{write/all}() is the
152  *	only way to ensure a page can not become dirty.  I/O generally
153  *	removes the page from pmap to ensure exclusive access and atomic
154  *	writes.
155  *
156  *	The ref_count field tracks references to the page.  References that
157  *	prevent the page from being reclaimable are called wirings and are
158  *	counted in the low bits of ref_count.  The containing object's
159  *	reference, if one exists, is counted using the VPRC_OBJREF bit in the
160  *	ref_count field.  Additionally, the VPRC_BLOCKED bit is used to
161  *	atomically check for wirings and prevent new wirings via
162  *	pmap_extract_and_hold().  When a page belongs to an object, it may be
163  *	wired only when the object is locked, or the page is busy, or by
164  *	pmap_extract_and_hold().  As a result, if the object is locked and the
165  *	page is not busy (or is exclusively busied by the current thread), and
166  *	the page is unmapped, its wire count will not increase.  The ref_count
167  *	field is updated using atomic operations in most cases, except when it
168  *	is known that no other references to the page exist, such as in the page
169  *	allocator.  A page may be present in the page queues, or even actively
170  *	scanned by the page daemon, without an explicitly counted referenced.
171  *	The page daemon must therefore handle the possibility of a concurrent
172  *	free of the page.
173  *
174  *	The queue state of a page consists of the queue and act_count fields of
175  *	its atomically updated state, and the subset of atomic flags specified
176  *	by PGA_QUEUE_STATE_MASK.  The queue field contains the page's page queue
177  *	index, or PQ_NONE if it does not belong to a page queue.  To modify the
178  *	queue field, the page queue lock corresponding to the old value must be
179  *	held, unless that value is PQ_NONE, in which case the queue index must
180  *	be updated using an atomic RMW operation.  There is one exception to
181  *	this rule: the page daemon may transition the queue field from
182  *	PQ_INACTIVE to PQ_NONE immediately prior to freeing the page during an
183  *	inactive queue scan.  At that point the page is already dequeued and no
184  *	other references to that vm_page structure can exist.  The PGA_ENQUEUED
185  *	flag, when set, indicates that the page structure is physically inserted
186  *	into the queue corresponding to the page's queue index, and may only be
187  *	set or cleared with the corresponding page queue lock held.
188  *
189  *	To avoid contention on page queue locks, page queue operations (enqueue,
190  *	dequeue, requeue) are batched using fixed-size per-CPU queues.  A
191  *	deferred operation is requested by setting one of the flags in
192  *	PGA_QUEUE_OP_MASK and inserting an entry into a batch queue.  When a
193  *	queue is full, an attempt to insert a new entry will lock the page
194  *	queues and trigger processing of the pending entries.  The
195  *	type-stability of vm_page structures is crucial to this scheme since the
196  *	processing of entries in a given batch queue may be deferred
197  *	indefinitely.  In particular, a page may be freed with pending batch
198  *	queue entries.  The page queue operation flags must be set using atomic
199  *	RWM operations.
200  */
201 
202 #if PAGE_SIZE == 4096
203 #define VM_PAGE_BITS_ALL 0xffu
204 typedef uint8_t vm_page_bits_t;
205 #elif PAGE_SIZE == 8192
206 #define VM_PAGE_BITS_ALL 0xffffu
207 typedef uint16_t vm_page_bits_t;
208 #elif PAGE_SIZE == 16384
209 #define VM_PAGE_BITS_ALL 0xffffffffu
210 typedef uint32_t vm_page_bits_t;
211 #elif PAGE_SIZE == 32768
212 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
213 typedef uint64_t vm_page_bits_t;
214 #endif
215 
216 typedef union vm_page_astate {
217 	struct {
218 		uint16_t flags;
219 		uint8_t	queue;
220 		uint8_t act_count;
221 	};
222 	uint32_t _bits;
223 } vm_page_astate_t;
224 
225 struct vm_page {
226 	union {
227 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
228 		struct {
229 			SLIST_ENTRY(vm_page) ss; /* private slists */
230 		} s;
231 		struct {
232 			u_long p;
233 			u_long v;
234 		} memguard;
235 		struct {
236 			void *slab;
237 			void *zone;
238 		} uma;
239 	} plinks;
240 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
241 	vm_object_t object;		/* which object am I in (O) */
242 	vm_pindex_t pindex;		/* offset into object (O,P) */
243 	vm_paddr_t phys_addr;		/* physical address of page (C) */
244 	struct md_page md;		/* machine dependent stuff */
245 	u_int ref_count;		/* page references (A) */
246 	volatile u_int busy_lock;	/* busy owners lock */
247 	union vm_page_astate a;		/* state accessed atomically */
248 	uint8_t order;			/* index of the buddy queue (F) */
249 	uint8_t pool;			/* vm_phys freepool index (F) */
250 	uint8_t flags;			/* page PG_* flags (P) */
251 	uint8_t oflags;			/* page VPO_* flags (O) */
252 	int8_t psind;			/* pagesizes[] index (O) */
253 	int8_t segind;			/* vm_phys segment index (C) */
254 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
255 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
256 	vm_page_bits_t valid;		/* valid DEV_BSIZE chunk map (O,B) */
257 	vm_page_bits_t dirty;		/* dirty DEV_BSIZE chunk map (M,B) */
258 };
259 
260 /*
261  * Special bits used in the ref_count field.
262  *
263  * ref_count is normally used to count wirings that prevent the page from being
264  * reclaimed, but also supports several special types of references that do not
265  * prevent reclamation.  Accesses to the ref_count field must be atomic unless
266  * the page is unallocated.
267  *
268  * VPRC_OBJREF is the reference held by the containing object.  It can set or
269  * cleared only when the corresponding object's write lock is held.
270  *
271  * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while
272  * attempting to tear down all mappings of a given page.  The page lock and
273  * object write lock must both be held in order to set or clear this bit.
274  */
275 #define	VPRC_BLOCKED	0x40000000u	/* mappings are being removed */
276 #define	VPRC_OBJREF	0x80000000u	/* object reference, cleared with (O) */
277 #define	VPRC_WIRE_COUNT(c)	((c) & ~(VPRC_BLOCKED | VPRC_OBJREF))
278 #define	VPRC_WIRE_COUNT_MAX	(~(VPRC_BLOCKED | VPRC_OBJREF))
279 
280 /*
281  * Page flags stored in oflags:
282  *
283  * Access to these page flags is synchronized by the lock on the object
284  * containing the page (O).
285  *
286  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
287  * 	 indicates that the page is not under PV management but
288  * 	 otherwise should be treated as a normal page.  Pages not
289  * 	 under PV management cannot be paged out via the
290  * 	 object/vm_page_t because there is no knowledge of their pte
291  * 	 mappings, and such pages are also not on any PQ queue.
292  *
293  */
294 #define	VPO_KMEM_EXEC	0x01		/* kmem mapping allows execution */
295 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
296 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
297 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
298 
299 /*
300  * Busy page implementation details.
301  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
302  * even if the support for owner identity is removed because of size
303  * constraints.  Checks on lock recursion are then not possible, while the
304  * lock assertions effectiveness is someway reduced.
305  */
306 #define	VPB_BIT_SHARED		0x01
307 #define	VPB_BIT_EXCLUSIVE	0x02
308 #define	VPB_BIT_WAITERS		0x04
309 #define	VPB_BIT_FLAGMASK						\
310 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
311 
312 #define	VPB_SHARERS_SHIFT	3
313 #define	VPB_SHARERS(x)							\
314 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
315 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
316 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
317 
318 #define	VPB_SINGLE_EXCLUSIVE	VPB_BIT_EXCLUSIVE
319 #ifdef INVARIANTS
320 #define	VPB_CURTHREAD_EXCLUSIVE						\
321 	(VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK))
322 #else
323 #define	VPB_CURTHREAD_EXCLUSIVE	VPB_SINGLE_EXCLUSIVE
324 #endif
325 
326 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
327 
328 /* Freed lock blocks both shared and exclusive. */
329 #define	VPB_FREED		(0xffffffff - VPB_BIT_SHARED)
330 
331 #define	PQ_NONE		255
332 #define	PQ_INACTIVE	0
333 #define	PQ_ACTIVE	1
334 #define	PQ_LAUNDRY	2
335 #define	PQ_UNSWAPPABLE	3
336 #define	PQ_COUNT	4
337 
338 #ifndef VM_PAGE_HAVE_PGLIST
339 TAILQ_HEAD(pglist, vm_page);
340 #define VM_PAGE_HAVE_PGLIST
341 #endif
342 SLIST_HEAD(spglist, vm_page);
343 
344 #ifdef _KERNEL
345 extern vm_page_t bogus_page;
346 #endif	/* _KERNEL */
347 
348 extern struct mtx_padalign pa_lock[];
349 
350 #if defined(__arm__)
351 #define	PDRSHIFT	PDR_SHIFT
352 #elif !defined(PDRSHIFT)
353 #define PDRSHIFT	21
354 #endif
355 
356 #define	pa_index(pa)	((pa) >> PDRSHIFT)
357 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
358 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
359 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
360 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
361 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
362 #define	PA_UNLOCK_COND(pa) 			\
363 	do {		   			\
364 		if ((pa) != 0) {		\
365 			PA_UNLOCK((pa));	\
366 			(pa) = 0;		\
367 		}				\
368 	} while (0)
369 
370 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
371 
372 #if defined(KLD_MODULE) && !defined(KLD_TIED)
373 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
374 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
375 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
376 #else	/* !KLD_MODULE */
377 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
378 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
379 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
380 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
381 #endif
382 #if defined(INVARIANTS)
383 #define	vm_page_assert_locked(m)		\
384     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
385 #define	vm_page_lock_assert(m, a)		\
386     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
387 #else
388 #define	vm_page_assert_locked(m)
389 #define	vm_page_lock_assert(m, a)
390 #endif
391 
392 /*
393  * The vm_page's aflags are updated using atomic operations.  To set or clear
394  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
395  * must be used.  Neither these flags nor these functions are part of the KBI.
396  *
397  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
398  * both the MI and MD VM layers.  However, kernel loadable modules should not
399  * directly set this flag.  They should call vm_page_reference() instead.
400  *
401  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
402  * When it does so, the object must be locked, or the page must be
403  * exclusive busied.  The MI VM layer must never access this flag
404  * directly.  Instead, it should call pmap_page_is_write_mapped().
405  *
406  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
407  * at least one executable mapping.  It is not consumed by the MI VM layer.
408  *
409  * PGA_NOSYNC must be set and cleared with the page busy lock held.
410  *
411  * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
412  * from a page queue, respectively.  It determines whether the plinks.q field
413  * of the page is valid.  To set or clear this flag, the queue lock for the
414  * page must be held: the page queue lock corresponding to the page's "queue"
415  * field if its value is not PQ_NONE, and the page lock otherwise.
416  *
417  * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
418  * queue, and cleared when the dequeue request is processed.  A page may
419  * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
420  * is requested after the page is scheduled to be enqueued but before it is
421  * actually inserted into the page queue.  For allocated pages, the page lock
422  * must be held to set this flag, but it may be set by vm_page_free_prep()
423  * without the page lock held.  The page queue lock must be held to clear the
424  * PGA_DEQUEUE flag.
425  *
426  * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
427  * in its page queue.  The page lock must be held to set this flag, and the
428  * queue lock for the page must be held to clear it.
429  *
430  * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
431  * the inactive queue, thus bypassing LRU.  The page lock must be held to
432  * set this flag, and the queue lock for the page must be held to clear it.
433  *
434  * PGA_SWAP_FREE is used to defer freeing swap space to the pageout daemon
435  * when the context that dirties the page does not have the object write lock
436  * held.
437  */
438 #define	PGA_WRITEABLE	0x0001		/* page may be mapped writeable */
439 #define	PGA_REFERENCED	0x0002		/* page has been referenced */
440 #define	PGA_EXECUTABLE	0x0004		/* page may be mapped executable */
441 #define	PGA_ENQUEUED	0x0008		/* page is enqueued in a page queue */
442 #define	PGA_DEQUEUE	0x0010		/* page is due to be dequeued */
443 #define	PGA_REQUEUE	0x0020		/* page is due to be requeued */
444 #define	PGA_REQUEUE_HEAD 0x0040		/* page requeue should bypass LRU */
445 #define	PGA_NOSYNC	0x0080		/* do not collect for syncer */
446 #define	PGA_SWAP_FREE	0x0100		/* page with swap space was dirtied */
447 #define	PGA_SWAP_SPACE	0x0200		/* page has allocated swap space */
448 
449 #define	PGA_QUEUE_OP_MASK	(PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD)
450 #define	PGA_QUEUE_STATE_MASK	(PGA_ENQUEUED | PGA_QUEUE_OP_MASK)
451 
452 /*
453  * Page flags.  If changed at any other time than page allocation or
454  * freeing, the modification must be protected by the vm_page lock.
455  *
456  * The PG_PCPU_CACHE flag is set at allocation time if the page was
457  * allocated from a per-CPU cache.  It is cleared the next time that the
458  * page is allocated from the physical memory allocator.
459  */
460 #define	PG_PCPU_CACHE	0x01		/* was allocated from per-CPU caches */
461 #define	PG_FICTITIOUS	0x02		/* physical page doesn't exist */
462 #define	PG_ZERO		0x04		/* page is zeroed */
463 #define	PG_MARKER	0x08		/* special queue marker page */
464 #define	PG_NODUMP	0x10		/* don't include this page in a dump */
465 
466 /*
467  * Misc constants.
468  */
469 #define ACT_DECLINE		1
470 #define ACT_ADVANCE		3
471 #define ACT_INIT		5
472 #define ACT_MAX			64
473 
474 #ifdef _KERNEL
475 
476 #include <sys/systm.h>
477 
478 #include <machine/atomic.h>
479 
480 /*
481  * Each pageable resident page falls into one of five lists:
482  *
483  *	free
484  *		Available for allocation now.
485  *
486  *	inactive
487  *		Low activity, candidates for reclamation.
488  *		This list is approximately LRU ordered.
489  *
490  *	laundry
491  *		This is the list of pages that should be
492  *		paged out next.
493  *
494  *	unswappable
495  *		Dirty anonymous pages that cannot be paged
496  *		out because no swap device is configured.
497  *
498  *	active
499  *		Pages that are "active", i.e., they have been
500  *		recently referenced.
501  *
502  */
503 
504 extern vm_page_t vm_page_array;		/* First resident page in table */
505 extern long vm_page_array_size;		/* number of vm_page_t's */
506 extern long first_page;			/* first physical page number */
507 
508 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
509 
510 /*
511  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
512  * page to which the given physical address belongs. The correct vm_page_t
513  * object is returned for addresses that are not page-aligned.
514  */
515 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
516 
517 /*
518  * Page allocation parameters for vm_page for the functions
519  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
520  * vm_page_alloc_freelist().  Some functions support only a subset
521  * of the flags, and ignore others, see the flags legend.
522  *
523  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
524  * and the vm_page_grab*() functions.  See these functions for details.
525  *
526  * Bits 0 - 1 define class.
527  * Bits 2 - 15 dedicated for flags.
528  * Legend:
529  * (a) - vm_page_alloc() supports the flag.
530  * (c) - vm_page_alloc_contig() supports the flag.
531  * (f) - vm_page_alloc_freelist() supports the flag.
532  * (g) - vm_page_grab() supports the flag.
533  * (p) - vm_page_grab_pages() supports the flag.
534  * Bits above 15 define the count of additional pages that the caller
535  * intends to allocate.
536  */
537 #define VM_ALLOC_NORMAL		0
538 #define VM_ALLOC_INTERRUPT	1
539 #define VM_ALLOC_SYSTEM		2
540 #define	VM_ALLOC_CLASS_MASK	3
541 #define	VM_ALLOC_WAITOK		0x0008	/* (acf) Sleep and retry */
542 #define	VM_ALLOC_WAITFAIL	0x0010	/* (acf) Sleep and return error */
543 #define	VM_ALLOC_WIRED		0x0020	/* (acfgp) Allocate a wired page */
544 #define	VM_ALLOC_ZERO		0x0040	/* (acfgp) Allocate a prezeroed page */
545 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
546 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
547 #define	VM_ALLOC_NOCREAT	0x0400	/* (gp) Don't create a page */
548 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
549 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
550 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
551 #define	VM_ALLOC_NOWAIT		0x8000	/* (acfgp) Do not sleep */
552 #define	VM_ALLOC_COUNT_SHIFT	16
553 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
554 
555 #ifdef M_NOWAIT
556 static inline int
557 malloc2vm_flags(int malloc_flags)
558 {
559 	int pflags;
560 
561 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
562 	    (malloc_flags & M_NOWAIT) != 0,
563 	    ("M_USE_RESERVE requires M_NOWAIT"));
564 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
565 	    VM_ALLOC_SYSTEM;
566 	if ((malloc_flags & M_ZERO) != 0)
567 		pflags |= VM_ALLOC_ZERO;
568 	if ((malloc_flags & M_NODUMP) != 0)
569 		pflags |= VM_ALLOC_NODUMP;
570 	if ((malloc_flags & M_NOWAIT))
571 		pflags |= VM_ALLOC_NOWAIT;
572 	if ((malloc_flags & M_WAITOK))
573 		pflags |= VM_ALLOC_WAITOK;
574 	return (pflags);
575 }
576 #endif
577 
578 /*
579  * Predicates supported by vm_page_ps_test():
580  *
581  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
582  *	However, it can be spuriously false when the (super)page has become
583  *	dirty in the pmap but that information has not been propagated to the
584  *	machine-independent layer.
585  */
586 #define	PS_ALL_DIRTY	0x1
587 #define	PS_ALL_VALID	0x2
588 #define	PS_NONE_BUSY	0x4
589 
590 bool vm_page_busy_acquire(vm_page_t m, int allocflags);
591 void vm_page_busy_downgrade(vm_page_t m);
592 int vm_page_busy_tryupgrade(vm_page_t m);
593 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
594 void vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m,
595     vm_pindex_t pindex, const char *wmesg, bool nonshared);
596 void vm_page_free(vm_page_t m);
597 void vm_page_free_zero(vm_page_t m);
598 
599 void vm_page_activate (vm_page_t);
600 void vm_page_advise(vm_page_t m, int advice);
601 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
602 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int);
603 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
604 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
605     vm_page_t);
606 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
607     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
608     vm_paddr_t boundary, vm_memattr_t memattr);
609 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
610     vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
611     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
612     vm_memattr_t memattr);
613 vm_page_t vm_page_alloc_freelist(int, int);
614 vm_page_t vm_page_alloc_freelist_domain(int, int, int);
615 void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set);
616 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
617 vm_page_t vm_page_grab(vm_object_t, vm_pindex_t, int);
618 vm_page_t vm_page_grab_unlocked(vm_object_t, vm_pindex_t, int);
619 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
620     vm_page_t *ma, int count);
621 int vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex,
622     int allocflags, vm_page_t *ma, int count);
623 int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex,
624     int allocflags);
625 int vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object,
626     vm_pindex_t pindex, int allocflags);
627 void vm_page_deactivate(vm_page_t);
628 void vm_page_deactivate_noreuse(vm_page_t);
629 void vm_page_dequeue(vm_page_t m);
630 void vm_page_dequeue_deferred(vm_page_t m);
631 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
632 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
633 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
634 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
635 void vm_page_invalid(vm_page_t m);
636 void vm_page_launder(vm_page_t m);
637 vm_page_t vm_page_lookup(vm_object_t, vm_pindex_t);
638 vm_page_t vm_page_next(vm_page_t m);
639 void vm_page_pqbatch_drain(void);
640 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
641 bool vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old,
642     vm_page_astate_t new);
643 vm_page_t vm_page_prev(vm_page_t m);
644 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
645 void vm_page_putfake(vm_page_t m);
646 void vm_page_readahead_finish(vm_page_t m);
647 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
648     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
649 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
650     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
651 void vm_page_reference(vm_page_t m);
652 #define	VPR_TRYFREE	0x01
653 #define	VPR_NOREUSE	0x02
654 void vm_page_release(vm_page_t m, int flags);
655 void vm_page_release_locked(vm_page_t m, int flags);
656 vm_page_t vm_page_relookup(vm_object_t, vm_pindex_t);
657 bool vm_page_remove(vm_page_t);
658 bool vm_page_remove_xbusy(vm_page_t);
659 int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t);
660 void vm_page_replace(vm_page_t mnew, vm_object_t object,
661     vm_pindex_t pindex, vm_page_t mold);
662 int vm_page_sbusied(vm_page_t m);
663 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
664     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
665 vm_page_bits_t vm_page_set_dirty(vm_page_t m);
666 void vm_page_set_valid_range(vm_page_t m, int base, int size);
667 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
668 int vm_page_sleep_if_xbusy(vm_page_t m, const char *msg);
669 vm_offset_t vm_page_startup(vm_offset_t vaddr);
670 void vm_page_sunbusy(vm_page_t m);
671 bool vm_page_try_remove_all(vm_page_t m);
672 bool vm_page_try_remove_write(vm_page_t m);
673 int vm_page_trysbusy(vm_page_t m);
674 int vm_page_tryxbusy(vm_page_t m);
675 void vm_page_unhold_pages(vm_page_t *ma, int count);
676 void vm_page_unswappable(vm_page_t m);
677 void vm_page_unwire(vm_page_t m, uint8_t queue);
678 bool vm_page_unwire_noq(vm_page_t m);
679 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
680 void vm_page_wire(vm_page_t);
681 bool vm_page_wire_mapped(vm_page_t m);
682 void vm_page_xunbusy_hard(vm_page_t m);
683 void vm_page_xunbusy_hard_unchecked(vm_page_t m);
684 void vm_page_set_validclean (vm_page_t, int, int);
685 void vm_page_clear_dirty(vm_page_t, int, int);
686 void vm_page_set_invalid(vm_page_t, int, int);
687 void vm_page_valid(vm_page_t m);
688 int vm_page_is_valid(vm_page_t, int, int);
689 void vm_page_test_dirty(vm_page_t);
690 vm_page_bits_t vm_page_bits(int base, int size);
691 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
692 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
693 
694 void vm_page_dirty_KBI(vm_page_t m);
695 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
696 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
697 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
698 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
699 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
700 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
701 #endif
702 
703 #define	vm_page_assert_busied(m)					\
704 	KASSERT(vm_page_busied(m),					\
705 	    ("vm_page_assert_busied: page %p not busy @ %s:%d", \
706 	    (m), __FILE__, __LINE__))
707 
708 #define	vm_page_assert_sbusied(m)					\
709 	KASSERT(vm_page_sbusied(m),					\
710 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
711 	    (m), __FILE__, __LINE__))
712 
713 #define	vm_page_assert_unbusied(m)					\
714 	KASSERT((m->busy_lock & ~VPB_BIT_WAITERS) != 			\
715 	    VPB_CURTHREAD_EXCLUSIVE,					\
716 	    ("vm_page_assert_xbusied: page %p busy_lock %#x owned"	\
717             " by me @ %s:%d",						\
718 	    (m), (m)->busy_lock, __FILE__, __LINE__));			\
719 
720 #define	vm_page_assert_xbusied_unchecked(m) do {			\
721 	KASSERT(vm_page_xbusied(m),					\
722 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
723 	    (m), __FILE__, __LINE__));					\
724 } while (0)
725 #define	vm_page_assert_xbusied(m) do {					\
726 	vm_page_assert_xbusied_unchecked(m);				\
727 	KASSERT((m->busy_lock & ~VPB_BIT_WAITERS) == 			\
728 	    VPB_CURTHREAD_EXCLUSIVE,					\
729 	    ("vm_page_assert_xbusied: page %p busy_lock %#x not owned"	\
730             " by me @ %s:%d",						\
731 	    (m), (m)->busy_lock, __FILE__, __LINE__));			\
732 } while (0)
733 
734 #define	vm_page_busied(m)						\
735 	((m)->busy_lock != VPB_UNBUSIED)
736 
737 #define	vm_page_sbusy(m) do {						\
738 	if (!vm_page_trysbusy(m))					\
739 		panic("%s: page %p failed shared busying", __func__,	\
740 		    (m));						\
741 } while (0)
742 
743 #define	vm_page_xbusied(m)						\
744 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVE) != 0)
745 
746 #define	vm_page_busy_freed(m)						\
747 	((m)->busy_lock == VPB_FREED)
748 
749 #define	vm_page_xbusy(m) do {						\
750 	if (!vm_page_tryxbusy(m))					\
751 		panic("%s: page %p failed exclusive busying", __func__,	\
752 		    (m));						\
753 } while (0)
754 
755 /* Note: page m's lock must not be owned by the caller. */
756 #define	vm_page_xunbusy(m) do {						\
757 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
758 	    VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED))			\
759 		vm_page_xunbusy_hard(m);				\
760 } while (0)
761 #define	vm_page_xunbusy_unchecked(m) do {				\
762 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
763 	    VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED))			\
764 		vm_page_xunbusy_hard_unchecked(m);			\
765 } while (0)
766 
767 #ifdef INVARIANTS
768 void vm_page_object_busy_assert(vm_page_t m);
769 #define	VM_PAGE_OBJECT_BUSY_ASSERT(m)	vm_page_object_busy_assert(m)
770 void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits);
771 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
772 	vm_page_assert_pga_writeable(m, bits)
773 #define	vm_page_xbusy_claim(m) do {					\
774 	vm_page_assert_xbusied_unchecked((m));				\
775 	(m)->busy_lock = VPB_CURTHREAD_EXCLUSIVE;			\
776 } while (0)
777 #else
778 #define	VM_PAGE_OBJECT_BUSY_ASSERT(m)	(void)0
779 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
780 #define	vm_page_xbusy_claim(m)
781 #endif
782 
783 #if BYTE_ORDER == BIG_ENDIAN
784 #define	VM_PAGE_AFLAG_SHIFT	16
785 #else
786 #define	VM_PAGE_AFLAG_SHIFT	0
787 #endif
788 
789 /*
790  *	Load a snapshot of a page's 32-bit atomic state.
791  */
792 static inline vm_page_astate_t
793 vm_page_astate_load(vm_page_t m)
794 {
795 	vm_page_astate_t a;
796 
797 	a._bits = atomic_load_32(&m->a._bits);
798 	return (a);
799 }
800 
801 /*
802  *	Atomically compare and set a page's atomic state.
803  */
804 static inline bool
805 vm_page_astate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new)
806 {
807 
808 	KASSERT(new.queue == PQ_INACTIVE || (new.flags & PGA_REQUEUE_HEAD) == 0,
809 	    ("%s: invalid head requeue request for page %p", __func__, m));
810 	KASSERT((new.flags & PGA_ENQUEUED) == 0 || new.queue != PQ_NONE,
811 	    ("%s: setting PGA_ENQUEUED with PQ_NONE in page %p", __func__, m));
812 	KASSERT(new._bits != old->_bits,
813 	    ("%s: bits are unchanged", __func__));
814 
815 	return (atomic_fcmpset_32(&m->a._bits, &old->_bits, new._bits) != 0);
816 }
817 
818 /*
819  *	Clear the given bits in the specified page.
820  */
821 static inline void
822 vm_page_aflag_clear(vm_page_t m, uint16_t bits)
823 {
824 	uint32_t *addr, val;
825 
826 	/*
827 	 * Access the whole 32-bit word containing the aflags field with an
828 	 * atomic update.  Parallel non-atomic updates to the other fields
829 	 * within this word are handled properly by the atomic update.
830 	 */
831 	addr = (void *)&m->a;
832 	val = bits << VM_PAGE_AFLAG_SHIFT;
833 	atomic_clear_32(addr, val);
834 }
835 
836 /*
837  *	Set the given bits in the specified page.
838  */
839 static inline void
840 vm_page_aflag_set(vm_page_t m, uint16_t bits)
841 {
842 	uint32_t *addr, val;
843 
844 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
845 
846 	/*
847 	 * Access the whole 32-bit word containing the aflags field with an
848 	 * atomic update.  Parallel non-atomic updates to the other fields
849 	 * within this word are handled properly by the atomic update.
850 	 */
851 	addr = (void *)&m->a;
852 	val = bits << VM_PAGE_AFLAG_SHIFT;
853 	atomic_set_32(addr, val);
854 }
855 
856 /*
857  *	vm_page_dirty:
858  *
859  *	Set all bits in the page's dirty field.
860  *
861  *	The object containing the specified page must be locked if the
862  *	call is made from the machine-independent layer.
863  *
864  *	See vm_page_clear_dirty_mask().
865  */
866 static __inline void
867 vm_page_dirty(vm_page_t m)
868 {
869 
870 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
871 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
872 	vm_page_dirty_KBI(m);
873 #else
874 	m->dirty = VM_PAGE_BITS_ALL;
875 #endif
876 }
877 
878 /*
879  *	vm_page_undirty:
880  *
881  *	Set page to not be dirty.  Note: does not clear pmap modify bits
882  */
883 static __inline void
884 vm_page_undirty(vm_page_t m)
885 {
886 
887 	VM_PAGE_OBJECT_BUSY_ASSERT(m);
888 	m->dirty = 0;
889 }
890 
891 static inline uint8_t
892 _vm_page_queue(vm_page_astate_t as)
893 {
894 
895 	if ((as.flags & PGA_DEQUEUE) != 0)
896 		return (PQ_NONE);
897 	return (as.queue);
898 }
899 
900 /*
901  *	vm_page_queue:
902  *
903  *	Return the index of the queue containing m.
904  */
905 static inline uint8_t
906 vm_page_queue(vm_page_t m)
907 {
908 
909 	return (_vm_page_queue(vm_page_astate_load(m)));
910 }
911 
912 static inline bool
913 vm_page_active(vm_page_t m)
914 {
915 
916 	return (vm_page_queue(m) == PQ_ACTIVE);
917 }
918 
919 static inline bool
920 vm_page_inactive(vm_page_t m)
921 {
922 
923 	return (vm_page_queue(m) == PQ_INACTIVE);
924 }
925 
926 static inline bool
927 vm_page_in_laundry(vm_page_t m)
928 {
929 	uint8_t queue;
930 
931 	queue = vm_page_queue(m);
932 	return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
933 }
934 
935 /*
936  *	vm_page_drop:
937  *
938  *	Release a reference to a page and return the old reference count.
939  */
940 static inline u_int
941 vm_page_drop(vm_page_t m, u_int val)
942 {
943 	u_int old;
944 
945 	/*
946 	 * Synchronize with vm_page_free_prep(): ensure that all updates to the
947 	 * page structure are visible before it is freed.
948 	 */
949 	atomic_thread_fence_rel();
950 	old = atomic_fetchadd_int(&m->ref_count, -val);
951 	KASSERT(old != VPRC_BLOCKED,
952 	    ("vm_page_drop: page %p has an invalid refcount value", m));
953 	return (old);
954 }
955 
956 /*
957  *	vm_page_wired:
958  *
959  *	Perform a racy check to determine whether a reference prevents the page
960  *	from being reclaimable.  If the page's object is locked, and the page is
961  *	unmapped and unbusied or exclusively busied by the current thread, no
962  *	new wirings may be created.
963  */
964 static inline bool
965 vm_page_wired(vm_page_t m)
966 {
967 
968 	return (VPRC_WIRE_COUNT(m->ref_count) > 0);
969 }
970 
971 static inline bool
972 vm_page_all_valid(vm_page_t m)
973 {
974 
975 	return (m->valid == VM_PAGE_BITS_ALL);
976 }
977 
978 static inline bool
979 vm_page_none_valid(vm_page_t m)
980 {
981 
982 	return (m->valid == 0);
983 }
984 
985 #endif				/* _KERNEL */
986 #endif				/* !_VM_PAGE_ */
987