xref: /freebsd/sys/vm/vm_page.h (revision 746af69907b9215c63245eceb7c5e86f863f3e37)
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 field is the index of the page queue containing the page,
175  *	or PQ_NONE if the page is not enqueued.  The queue lock of a page is
176  *	the page queue lock corresponding to the page queue index, or the
177  *	page lock (P) for the page if it is not enqueued.  To modify the
178  *	queue field, the queue lock for the old value of the field must be
179  *	held.  There is one exception to this rule: the page daemon may
180  *	transition the queue field from PQ_INACTIVE to PQ_NONE immediately
181  *	prior to freeing a page during an inactive queue scan.  At that
182  *	point the page has already been physically dequeued and no other
183  *	references to that vm_page structure exist.
184  *
185  *	To avoid contention on page queue locks, page queue operations
186  *	(enqueue, dequeue, requeue) are batched using per-CPU queues.  A
187  *	deferred operation is requested by inserting an entry into a batch
188  *	queue; the entry is simply a pointer to the page, and the request
189  *	type is encoded in the page's aflags field using the values in
190  *	PGA_QUEUE_STATE_MASK.  The type-stability of struct vm_pages is
191  *	crucial to this scheme since the processing of entries in a given
192  *	batch queue may be deferred indefinitely.  In particular, a page may
193  *	be freed before its pending batch queue entries have been processed.
194  *	The page lock (P) must be held to schedule a batched queue
195  *	operation, and the page queue lock must be held in order to process
196  *	batch queue entries for the page queue.  There is one exception to
197  *	this rule: the thread freeing a page may schedule a dequeue without
198  *	holding the page lock.  In this scenario the only other thread which
199  *	may hold a reference to the page is the page daemon, which is
200  *	careful to avoid modifying the page's queue state once the dequeue
201  *	has been requested by setting PGA_DEQUEUE.
202  */
203 
204 #if PAGE_SIZE == 4096
205 #define VM_PAGE_BITS_ALL 0xffu
206 typedef uint8_t vm_page_bits_t;
207 #elif PAGE_SIZE == 8192
208 #define VM_PAGE_BITS_ALL 0xffffu
209 typedef uint16_t vm_page_bits_t;
210 #elif PAGE_SIZE == 16384
211 #define VM_PAGE_BITS_ALL 0xffffffffu
212 typedef uint32_t vm_page_bits_t;
213 #elif PAGE_SIZE == 32768
214 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
215 typedef uint64_t vm_page_bits_t;
216 #endif
217 
218 typedef union vm_page_astate {
219 	struct {
220 		uint16_t flags;
221 		uint8_t	queue;
222 		uint8_t act_count;
223 	};
224 	uint32_t _bits;
225 } vm_page_astate_t;
226 
227 struct vm_page {
228 	union {
229 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
230 		struct {
231 			SLIST_ENTRY(vm_page) ss; /* private slists */
232 		} s;
233 		struct {
234 			u_long p;
235 			u_long v;
236 		} memguard;
237 		struct {
238 			void *slab;
239 			void *zone;
240 		} uma;
241 	} plinks;
242 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
243 	vm_object_t object;		/* which object am I in (O) */
244 	vm_pindex_t pindex;		/* offset into object (O,P) */
245 	vm_paddr_t phys_addr;		/* physical address of page (C) */
246 	struct md_page md;		/* machine dependent stuff */
247 	u_int ref_count;		/* page references (A) */
248 	volatile u_int busy_lock;	/* busy owners lock */
249 	union vm_page_astate a;		/* state accessed atomically */
250 	uint8_t order;			/* index of the buddy queue (F) */
251 	uint8_t pool;			/* vm_phys freepool index (F) */
252 	uint8_t flags;			/* page PG_* flags (P) */
253 	uint8_t oflags;			/* page VPO_* flags (O) */
254 	int8_t psind;			/* pagesizes[] index (O) */
255 	int8_t segind;			/* vm_phys segment index (C) */
256 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
257 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
258 	vm_page_bits_t valid;		/* valid DEV_BSIZE chunk map (O,B) */
259 	vm_page_bits_t dirty;		/* dirty DEV_BSIZE chunk map (M,B) */
260 };
261 
262 /*
263  * Special bits used in the ref_count field.
264  *
265  * ref_count is normally used to count wirings that prevent the page from being
266  * reclaimed, but also supports several special types of references that do not
267  * prevent reclamation.  Accesses to the ref_count field must be atomic unless
268  * the page is unallocated.
269  *
270  * VPRC_OBJREF is the reference held by the containing object.  It can set or
271  * cleared only when the corresponding object's write lock is held.
272  *
273  * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while
274  * attempting to tear down all mappings of a given page.  The page lock and
275  * object write lock must both be held in order to set or clear this bit.
276  */
277 #define	VPRC_BLOCKED	0x40000000u	/* mappings are being removed */
278 #define	VPRC_OBJREF	0x80000000u	/* object reference, cleared with (O) */
279 #define	VPRC_WIRE_COUNT(c)	((c) & ~(VPRC_BLOCKED | VPRC_OBJREF))
280 #define	VPRC_WIRE_COUNT_MAX	(~(VPRC_BLOCKED | VPRC_OBJREF))
281 
282 /*
283  * Page flags stored in oflags:
284  *
285  * Access to these page flags is synchronized by the lock on the object
286  * containing the page (O).
287  *
288  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
289  * 	 indicates that the page is not under PV management but
290  * 	 otherwise should be treated as a normal page.  Pages not
291  * 	 under PV management cannot be paged out via the
292  * 	 object/vm_page_t because there is no knowledge of their pte
293  * 	 mappings, and such pages are also not on any PQ queue.
294  *
295  */
296 #define	VPO_KMEM_EXEC	0x01		/* kmem mapping allows execution */
297 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
298 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
299 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
300 
301 /*
302  * Busy page implementation details.
303  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
304  * even if the support for owner identity is removed because of size
305  * constraints.  Checks on lock recursion are then not possible, while the
306  * lock assertions effectiveness is someway reduced.
307  */
308 #define	VPB_BIT_SHARED		0x01
309 #define	VPB_BIT_EXCLUSIVE	0x02
310 #define	VPB_BIT_WAITERS		0x04
311 #define	VPB_BIT_FLAGMASK						\
312 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
313 
314 #define	VPB_SHARERS_SHIFT	3
315 #define	VPB_SHARERS(x)							\
316 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
317 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
318 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
319 
320 #define	VPB_SINGLE_EXCLUSIVE	VPB_BIT_EXCLUSIVE
321 #ifdef INVARIANTS
322 #define	VPB_CURTHREAD_EXCLUSIVE						\
323 	(VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK))
324 #else
325 #define	VPB_CURTHREAD_EXCLUSIVE	VPB_SINGLE_EXCLUSIVE
326 #endif
327 
328 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
329 
330 #define	PQ_NONE		255
331 #define	PQ_INACTIVE	0
332 #define	PQ_ACTIVE	1
333 #define	PQ_LAUNDRY	2
334 #define	PQ_UNSWAPPABLE	3
335 #define	PQ_COUNT	4
336 
337 #ifndef VM_PAGE_HAVE_PGLIST
338 TAILQ_HEAD(pglist, vm_page);
339 #define VM_PAGE_HAVE_PGLIST
340 #endif
341 SLIST_HEAD(spglist, vm_page);
342 
343 #ifdef _KERNEL
344 extern vm_page_t bogus_page;
345 #endif	/* _KERNEL */
346 
347 extern struct mtx_padalign pa_lock[];
348 
349 #if defined(__arm__)
350 #define	PDRSHIFT	PDR_SHIFT
351 #elif !defined(PDRSHIFT)
352 #define PDRSHIFT	21
353 #endif
354 
355 #define	pa_index(pa)	((pa) >> PDRSHIFT)
356 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
357 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
358 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
359 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
360 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
361 #define	PA_UNLOCK_COND(pa) 			\
362 	do {		   			\
363 		if ((pa) != 0) {		\
364 			PA_UNLOCK((pa));	\
365 			(pa) = 0;		\
366 		}				\
367 	} while (0)
368 
369 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
370 
371 #if defined(KLD_MODULE) && !defined(KLD_TIED)
372 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
373 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
374 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
375 #else	/* !KLD_MODULE */
376 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
377 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
378 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
379 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
380 #endif
381 #if defined(INVARIANTS)
382 #define	vm_page_assert_locked(m)		\
383     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
384 #define	vm_page_lock_assert(m, a)		\
385     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
386 #else
387 #define	vm_page_assert_locked(m)
388 #define	vm_page_lock_assert(m, a)
389 #endif
390 
391 /*
392  * The vm_page's aflags are updated using atomic operations.  To set or clear
393  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
394  * must be used.  Neither these flags nor these functions are part of the KBI.
395  *
396  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
397  * both the MI and MD VM layers.  However, kernel loadable modules should not
398  * directly set this flag.  They should call vm_page_reference() instead.
399  *
400  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
401  * When it does so, the object must be locked, or the page must be
402  * exclusive busied.  The MI VM layer must never access this flag
403  * directly.  Instead, it should call pmap_page_is_write_mapped().
404  *
405  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
406  * at least one executable mapping.  It is not consumed by the MI VM layer.
407  *
408  * PGA_NOSYNC must be set and cleared with the page busy lock held.
409  *
410  * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
411  * from a page queue, respectively.  It determines whether the plinks.q field
412  * of the page is valid.  To set or clear this flag, the queue lock for the
413  * page must be held: the page queue lock corresponding to the page's "queue"
414  * field if its value is not PQ_NONE, and the page lock otherwise.
415  *
416  * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
417  * queue, and cleared when the dequeue request is processed.  A page may
418  * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
419  * is requested after the page is scheduled to be enqueued but before it is
420  * actually inserted into the page queue.  For allocated pages, the page lock
421  * must be held to set this flag, but it may be set by vm_page_free_prep()
422  * without the page lock held.  The page queue lock must be held to clear the
423  * PGA_DEQUEUE flag.
424  *
425  * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
426  * in its page queue.  The page lock must be held to set this flag, and the
427  * queue lock for the page must be held to clear it.
428  *
429  * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
430  * the inactive queue, thus bypassing LRU.  The page lock must be held to
431  * set this flag, and the queue lock for the page must be held to clear it.
432  */
433 #define	PGA_WRITEABLE	0x0001		/* page may be mapped writeable */
434 #define	PGA_REFERENCED	0x0002		/* page has been referenced */
435 #define	PGA_EXECUTABLE	0x0004		/* page may be mapped executable */
436 #define	PGA_ENQUEUED	0x0008		/* page is enqueued in a page queue */
437 #define	PGA_DEQUEUE	0x0010		/* page is due to be dequeued */
438 #define	PGA_REQUEUE	0x0020		/* page is due to be requeued */
439 #define	PGA_REQUEUE_HEAD 0x0040		/* page requeue should bypass LRU */
440 #define	PGA_NOSYNC	0x0080		/* do not collect for syncer */
441 
442 #define	PGA_QUEUE_STATE_MASK	(PGA_ENQUEUED | PGA_DEQUEUE | PGA_REQUEUE | \
443 				PGA_REQUEUE_HEAD)
444 
445 /*
446  * Page flags.  If changed at any other time than page allocation or
447  * freeing, the modification must be protected by the vm_page lock.
448  *
449  * The PG_PCPU_CACHE flag is set at allocation time if the page was
450  * allocated from a per-CPU cache.  It is cleared the next time that the
451  * page is allocated from the physical memory allocator.
452  */
453 #define	PG_PCPU_CACHE	0x01		/* was allocated from per-CPU caches */
454 #define	PG_FICTITIOUS	0x02		/* physical page doesn't exist */
455 #define	PG_ZERO		0x04		/* page is zeroed */
456 #define	PG_MARKER	0x08		/* special queue marker page */
457 #define	PG_NODUMP	0x10		/* don't include this page in a dump */
458 
459 /*
460  * Misc constants.
461  */
462 #define ACT_DECLINE		1
463 #define ACT_ADVANCE		3
464 #define ACT_INIT		5
465 #define ACT_MAX			64
466 
467 #ifdef _KERNEL
468 
469 #include <sys/systm.h>
470 
471 #include <machine/atomic.h>
472 
473 /*
474  * Each pageable resident page falls into one of five lists:
475  *
476  *	free
477  *		Available for allocation now.
478  *
479  *	inactive
480  *		Low activity, candidates for reclamation.
481  *		This list is approximately LRU ordered.
482  *
483  *	laundry
484  *		This is the list of pages that should be
485  *		paged out next.
486  *
487  *	unswappable
488  *		Dirty anonymous pages that cannot be paged
489  *		out because no swap device is configured.
490  *
491  *	active
492  *		Pages that are "active", i.e., they have been
493  *		recently referenced.
494  *
495  */
496 
497 extern vm_page_t vm_page_array;		/* First resident page in table */
498 extern long vm_page_array_size;		/* number of vm_page_t's */
499 extern long first_page;			/* first physical page number */
500 
501 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
502 
503 /*
504  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
505  * page to which the given physical address belongs. The correct vm_page_t
506  * object is returned for addresses that are not page-aligned.
507  */
508 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
509 
510 /*
511  * Page allocation parameters for vm_page for the functions
512  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
513  * vm_page_alloc_freelist().  Some functions support only a subset
514  * of the flags, and ignore others, see the flags legend.
515  *
516  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
517  * and the vm_page_grab*() functions.  See these functions for details.
518  *
519  * Bits 0 - 1 define class.
520  * Bits 2 - 15 dedicated for flags.
521  * Legend:
522  * (a) - vm_page_alloc() supports the flag.
523  * (c) - vm_page_alloc_contig() supports the flag.
524  * (f) - vm_page_alloc_freelist() supports the flag.
525  * (g) - vm_page_grab() supports the flag.
526  * (p) - vm_page_grab_pages() supports the flag.
527  * Bits above 15 define the count of additional pages that the caller
528  * intends to allocate.
529  */
530 #define VM_ALLOC_NORMAL		0
531 #define VM_ALLOC_INTERRUPT	1
532 #define VM_ALLOC_SYSTEM		2
533 #define	VM_ALLOC_CLASS_MASK	3
534 #define	VM_ALLOC_WAITOK		0x0008	/* (acf) Sleep and retry */
535 #define	VM_ALLOC_WAITFAIL	0x0010	/* (acf) Sleep and return error */
536 #define	VM_ALLOC_WIRED		0x0020	/* (acfgp) Allocate a wired page */
537 #define	VM_ALLOC_ZERO		0x0040	/* (acfgp) Allocate a prezeroed page */
538 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
539 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
540 #define	VM_ALLOC_NOCREAT	0x0400	/* (gp) Don't create a page */
541 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
542 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
543 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
544 #define	VM_ALLOC_NOWAIT		0x8000	/* (acfgp) Do not sleep */
545 #define	VM_ALLOC_COUNT_SHIFT	16
546 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
547 
548 #ifdef M_NOWAIT
549 static inline int
550 malloc2vm_flags(int malloc_flags)
551 {
552 	int pflags;
553 
554 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
555 	    (malloc_flags & M_NOWAIT) != 0,
556 	    ("M_USE_RESERVE requires M_NOWAIT"));
557 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
558 	    VM_ALLOC_SYSTEM;
559 	if ((malloc_flags & M_ZERO) != 0)
560 		pflags |= VM_ALLOC_ZERO;
561 	if ((malloc_flags & M_NODUMP) != 0)
562 		pflags |= VM_ALLOC_NODUMP;
563 	if ((malloc_flags & M_NOWAIT))
564 		pflags |= VM_ALLOC_NOWAIT;
565 	if ((malloc_flags & M_WAITOK))
566 		pflags |= VM_ALLOC_WAITOK;
567 	return (pflags);
568 }
569 #endif
570 
571 /*
572  * Predicates supported by vm_page_ps_test():
573  *
574  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
575  *	However, it can be spuriously false when the (super)page has become
576  *	dirty in the pmap but that information has not been propagated to the
577  *	machine-independent layer.
578  */
579 #define	PS_ALL_DIRTY	0x1
580 #define	PS_ALL_VALID	0x2
581 #define	PS_NONE_BUSY	0x4
582 
583 bool vm_page_busy_acquire(vm_page_t m, int allocflags);
584 void vm_page_busy_downgrade(vm_page_t m);
585 int vm_page_busy_tryupgrade(vm_page_t m);
586 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
587 void vm_page_free(vm_page_t m);
588 void vm_page_free_zero(vm_page_t m);
589 
590 void vm_page_activate (vm_page_t);
591 void vm_page_advise(vm_page_t m, int advice);
592 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
593 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int);
594 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
595 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
596     vm_page_t);
597 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
598     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
599     vm_paddr_t boundary, vm_memattr_t memattr);
600 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
601     vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
602     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
603     vm_memattr_t memattr);
604 vm_page_t vm_page_alloc_freelist(int, int);
605 vm_page_t vm_page_alloc_freelist_domain(int, int, int);
606 void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set);
607 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
608 void vm_page_change_lock(vm_page_t m, struct mtx **mtx);
609 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
610 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
611     vm_page_t *ma, int count);
612 int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex,
613     int allocflags);
614 void vm_page_deactivate(vm_page_t);
615 void vm_page_deactivate_noreuse(vm_page_t);
616 void vm_page_dequeue(vm_page_t m);
617 void vm_page_dequeue_deferred(vm_page_t m);
618 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
619 bool vm_page_free_prep(vm_page_t m);
620 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
621 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
622 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
623 void vm_page_invalid(vm_page_t m);
624 void vm_page_launder(vm_page_t m);
625 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
626 vm_page_t vm_page_next(vm_page_t m);
627 void vm_page_pqbatch_drain(void);
628 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
629 vm_page_t vm_page_prev(vm_page_t m);
630 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
631 void vm_page_putfake(vm_page_t m);
632 void vm_page_readahead_finish(vm_page_t m);
633 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
634     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
635 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
636     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
637 void vm_page_reference(vm_page_t m);
638 #define	VPR_TRYFREE	0x01
639 #define	VPR_NOREUSE	0x02
640 void vm_page_release(vm_page_t m, int flags);
641 void vm_page_release_locked(vm_page_t m, int flags);
642 bool vm_page_remove(vm_page_t);
643 int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t);
644 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
645     vm_pindex_t pindex);
646 void vm_page_requeue(vm_page_t m);
647 int vm_page_sbusied(vm_page_t m);
648 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
649     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
650 void vm_page_set_valid_range(vm_page_t m, int base, int size);
651 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
652 int vm_page_sleep_if_xbusy(vm_page_t m, const char *msg);
653 vm_offset_t vm_page_startup(vm_offset_t vaddr);
654 void vm_page_sunbusy(vm_page_t m);
655 void vm_page_swapqueue(vm_page_t m, uint8_t oldq, uint8_t newq);
656 bool vm_page_try_remove_all(vm_page_t m);
657 bool vm_page_try_remove_write(vm_page_t m);
658 int vm_page_trysbusy(vm_page_t m);
659 int vm_page_tryxbusy(vm_page_t m);
660 void vm_page_unhold_pages(vm_page_t *ma, int count);
661 void vm_page_unswappable(vm_page_t m);
662 void vm_page_unwire(vm_page_t m, uint8_t queue);
663 bool vm_page_unwire_noq(vm_page_t m);
664 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
665 void vm_page_wire(vm_page_t);
666 bool vm_page_wire_mapped(vm_page_t m);
667 void vm_page_xunbusy_hard(vm_page_t m);
668 void vm_page_xunbusy_hard_unchecked(vm_page_t m);
669 void vm_page_set_validclean (vm_page_t, int, int);
670 void vm_page_clear_dirty(vm_page_t, int, int);
671 void vm_page_set_invalid(vm_page_t, int, int);
672 void vm_page_valid(vm_page_t m);
673 int vm_page_is_valid(vm_page_t, int, int);
674 void vm_page_test_dirty(vm_page_t);
675 vm_page_bits_t vm_page_bits(int base, int size);
676 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
677 void vm_page_free_toq(vm_page_t m);
678 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
679 
680 void vm_page_dirty_KBI(vm_page_t m);
681 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
682 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
683 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
684 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
685 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
686 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
687 #endif
688 
689 #define	vm_page_assert_busied(m)					\
690 	KASSERT(vm_page_busied(m),					\
691 	    ("vm_page_assert_busied: page %p not busy @ %s:%d", \
692 	    (m), __FILE__, __LINE__))
693 
694 #define	vm_page_assert_sbusied(m)					\
695 	KASSERT(vm_page_sbusied(m),					\
696 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
697 	    (m), __FILE__, __LINE__))
698 
699 #define	vm_page_assert_unbusied(m)					\
700 	KASSERT(!vm_page_busied(m),					\
701 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
702 	    (m), __FILE__, __LINE__))
703 
704 #define	vm_page_assert_xbusied_unchecked(m) do {			\
705 	KASSERT(vm_page_xbusied(m),					\
706 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
707 	    (m), __FILE__, __LINE__));					\
708 } while (0)
709 #define	vm_page_assert_xbusied(m) do {					\
710 	vm_page_assert_xbusied_unchecked(m);				\
711 	KASSERT((m->busy_lock & ~VPB_BIT_WAITERS) == 			\
712 	    VPB_CURTHREAD_EXCLUSIVE,					\
713 	    ("vm_page_assert_xbusied: page %p busy_lock %#x not owned"	\
714             " by me @ %s:%d",						\
715 	    (m), (m)->busy_lock, __FILE__, __LINE__));			\
716 } while (0)
717 
718 #define	vm_page_busied(m)						\
719 	((m)->busy_lock != VPB_UNBUSIED)
720 
721 #define	vm_page_sbusy(m) do {						\
722 	if (!vm_page_trysbusy(m))					\
723 		panic("%s: page %p failed shared busying", __func__,	\
724 		    (m));						\
725 } while (0)
726 
727 #define	vm_page_xbusied(m)						\
728 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVE) != 0)
729 
730 #define	vm_page_xbusy(m) do {						\
731 	if (!vm_page_tryxbusy(m))					\
732 		panic("%s: page %p failed exclusive busying", __func__,	\
733 		    (m));						\
734 } while (0)
735 
736 /* Note: page m's lock must not be owned by the caller. */
737 #define	vm_page_xunbusy(m) do {						\
738 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
739 	    VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED))			\
740 		vm_page_xunbusy_hard(m);				\
741 } while (0)
742 #define	vm_page_xunbusy_unchecked(m) do {				\
743 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
744 	    VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED))			\
745 		vm_page_xunbusy_hard_unchecked(m);			\
746 } while (0)
747 
748 #ifdef INVARIANTS
749 void vm_page_object_busy_assert(vm_page_t m);
750 #define	VM_PAGE_OBJECT_BUSY_ASSERT(m)	vm_page_object_busy_assert(m)
751 void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits);
752 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
753 	vm_page_assert_pga_writeable(m, bits)
754 #else
755 #define	VM_PAGE_OBJECT_BUSY_ASSERT(m)	(void)0
756 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
757 #endif
758 
759 /*
760  * We want to use atomic updates for the aflags field, which is 8 bits wide.
761  * However, not all architectures support atomic operations on 8-bit
762  * destinations.  In order that we can easily use a 32-bit operation, we
763  * require that the aflags field be 32-bit aligned.
764  */
765 _Static_assert(offsetof(struct vm_page, a.flags) % sizeof(uint32_t) == 0,
766     "aflags field is not 32-bit aligned");
767 
768 /*
769  * We want to be able to update the aflags and queue fields atomically in
770  * the same operation.
771  */
772 _Static_assert(offsetof(struct vm_page, a.flags) / sizeof(uint32_t) ==
773     offsetof(struct vm_page, a.queue) / sizeof(uint32_t),
774     "aflags and queue fields do not belong to the same 32-bit word");
775 _Static_assert(offsetof(struct vm_page, a.queue) % sizeof(uint32_t) == 2,
776     "queue field is at an unexpected offset");
777 _Static_assert(sizeof(((struct vm_page *)NULL)->a.queue) == 1,
778     "queue field has an unexpected size");
779 
780 #if BYTE_ORDER == LITTLE_ENDIAN
781 #define	VM_PAGE_AFLAG_SHIFT	0
782 #define	VM_PAGE_QUEUE_SHIFT	16
783 #else
784 #define	VM_PAGE_AFLAG_SHIFT	16
785 #define	VM_PAGE_QUEUE_SHIFT	8
786 #endif
787 #define	VM_PAGE_QUEUE_MASK	(0xff << VM_PAGE_QUEUE_SHIFT)
788 
789 /*
790  *	Clear the given bits in the specified page.
791  */
792 static inline void
793 vm_page_aflag_clear(vm_page_t m, uint16_t bits)
794 {
795 	uint32_t *addr, val;
796 
797 	/*
798 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
799 	 */
800 	if ((bits & PGA_REFERENCED) != 0)
801 		vm_page_assert_locked(m);
802 
803 	/*
804 	 * Access the whole 32-bit word containing the aflags field with an
805 	 * atomic update.  Parallel non-atomic updates to the other fields
806 	 * within this word are handled properly by the atomic update.
807 	 */
808 	addr = (void *)&m->a.flags;
809 	val = bits << VM_PAGE_AFLAG_SHIFT;
810 	atomic_clear_32(addr, val);
811 }
812 
813 /*
814  *	Set the given bits in the specified page.
815  */
816 static inline void
817 vm_page_aflag_set(vm_page_t m, uint16_t bits)
818 {
819 	uint32_t *addr, val;
820 
821 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
822 
823 	/*
824 	 * Access the whole 32-bit word containing the aflags field with an
825 	 * atomic update.  Parallel non-atomic updates to the other fields
826 	 * within this word are handled properly by the atomic update.
827 	 */
828 	addr = (void *)&m->a.flags;
829 	val = bits << VM_PAGE_AFLAG_SHIFT;
830 	atomic_set_32(addr, val);
831 }
832 
833 /*
834  *	Atomically update the queue state of the page.  The operation fails if
835  *	any of the queue flags in "fflags" are set or if the "queue" field of
836  *	the page does not match the expected value; if the operation is
837  *	successful, the flags in "nflags" are set and all other queue state
838  *	flags are cleared.
839  */
840 static inline bool
841 vm_page_pqstate_cmpset(vm_page_t m, uint32_t oldq, uint32_t newq,
842     uint32_t fflags, uint32_t nflags)
843 {
844 	uint32_t *addr, nval, oval, qsmask;
845 
846 	fflags <<= VM_PAGE_AFLAG_SHIFT;
847 	nflags <<= VM_PAGE_AFLAG_SHIFT;
848 	newq <<= VM_PAGE_QUEUE_SHIFT;
849 	oldq <<= VM_PAGE_QUEUE_SHIFT;
850 	qsmask = ((PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD) <<
851 	    VM_PAGE_AFLAG_SHIFT) | VM_PAGE_QUEUE_MASK;
852 
853 	addr = (void *)&m->a.flags;
854 	oval = atomic_load_32(addr);
855 	do {
856 		if ((oval & fflags) != 0)
857 			return (false);
858 		if ((oval & VM_PAGE_QUEUE_MASK) != oldq)
859 			return (false);
860 		nval = (oval & ~qsmask) | nflags | newq;
861 	} while (!atomic_fcmpset_32(addr, &oval, nval));
862 
863 	return (true);
864 }
865 
866 /*
867  *	vm_page_dirty:
868  *
869  *	Set all bits in the page's dirty field.
870  *
871  *	The object containing the specified page must be locked if the
872  *	call is made from the machine-independent layer.
873  *
874  *	See vm_page_clear_dirty_mask().
875  */
876 static __inline void
877 vm_page_dirty(vm_page_t m)
878 {
879 
880 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
881 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
882 	vm_page_dirty_KBI(m);
883 #else
884 	m->dirty = VM_PAGE_BITS_ALL;
885 #endif
886 }
887 
888 /*
889  *	vm_page_undirty:
890  *
891  *	Set page to not be dirty.  Note: does not clear pmap modify bits
892  */
893 static __inline void
894 vm_page_undirty(vm_page_t m)
895 {
896 
897 	VM_PAGE_OBJECT_BUSY_ASSERT(m);
898 	m->dirty = 0;
899 }
900 
901 static inline void
902 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
903     vm_page_t mold)
904 {
905 	vm_page_t mret;
906 
907 	mret = vm_page_replace(mnew, object, pindex);
908 	KASSERT(mret == mold,
909 	    ("invalid page replacement, mold=%p, mret=%p", mold, mret));
910 
911 	/* Unused if !INVARIANTS. */
912 	(void)mold;
913 	(void)mret;
914 }
915 
916 /*
917  *	vm_page_queue:
918  *
919  *	Return the index of the queue containing m.  This index is guaranteed
920  *	not to change while the page lock is held.
921  */
922 static inline uint8_t
923 vm_page_queue(vm_page_t m)
924 {
925 
926 	vm_page_assert_locked(m);
927 
928 	if ((m->a.flags & PGA_DEQUEUE) != 0)
929 		return (PQ_NONE);
930 	atomic_thread_fence_acq();
931 	return (m->a.queue);
932 }
933 
934 static inline bool
935 vm_page_active(vm_page_t m)
936 {
937 
938 	return (vm_page_queue(m) == PQ_ACTIVE);
939 }
940 
941 static inline bool
942 vm_page_inactive(vm_page_t m)
943 {
944 
945 	return (vm_page_queue(m) == PQ_INACTIVE);
946 }
947 
948 static inline bool
949 vm_page_in_laundry(vm_page_t m)
950 {
951 	uint8_t queue;
952 
953 	queue = vm_page_queue(m);
954 	return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
955 }
956 
957 /*
958  *	vm_page_drop:
959  *
960  *	Release a reference to a page and return the old reference count.
961  */
962 static inline u_int
963 vm_page_drop(vm_page_t m, u_int val)
964 {
965 	u_int old;
966 
967 	/*
968 	 * Synchronize with vm_page_free_prep(): ensure that all updates to the
969 	 * page structure are visible before it is freed.
970 	 */
971 	atomic_thread_fence_rel();
972 	old = atomic_fetchadd_int(&m->ref_count, -val);
973 	KASSERT(old != VPRC_BLOCKED,
974 	    ("vm_page_drop: page %p has an invalid refcount value", m));
975 	return (old);
976 }
977 
978 /*
979  *	vm_page_wired:
980  *
981  *	Perform a racy check to determine whether a reference prevents the page
982  *	from being reclaimable.  If the page's object is locked, and the page is
983  *	unmapped and unbusied or exclusively busied by the current thread, no
984  *	new wirings may be created.
985  */
986 static inline bool
987 vm_page_wired(vm_page_t m)
988 {
989 
990 	return (VPRC_WIRE_COUNT(m->ref_count) > 0);
991 }
992 
993 static inline bool
994 vm_page_all_valid(vm_page_t m)
995 {
996 
997 	return (m->valid == VM_PAGE_BITS_ALL);
998 }
999 
1000 static inline bool
1001 vm_page_none_valid(vm_page_t m)
1002 {
1003 
1004 	return (m->valid == 0);
1005 }
1006 
1007 #endif				/* _KERNEL */
1008 #endif				/* !_VM_PAGE_ */
1009