xref: /freebsd/sys/vm/vm_page.h (revision 2a2234c0f41da33b8cfc938e46b54a8234b64135)
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 pool lock for the page (P),
97  *	or the lock for either the free or paging queue (Q).  If a field is
98  *	annotated below with two of these locks, then holding either lock is
99  *	sufficient for read access, but both locks are required for write
100  *	access.  An annotation of (C) indicates that the field is immutable.
101  *
102  *	In contrast, the synchronization of accesses to the page's
103  *	dirty field is machine dependent (M).  In the
104  *	machine-independent layer, the lock on the object that the
105  *	page belongs to must be held in order to operate on the field.
106  *	However, the pmap layer is permitted to set all bits within
107  *	the field without holding that lock.  If the underlying
108  *	architecture does not support atomic read-modify-write
109  *	operations on the field's type, then the machine-independent
110  *	layer uses a 32-bit atomic on the aligned 32-bit word that
111  *	contains the dirty field.  In the machine-independent layer,
112  *	the implementation of read-modify-write operations on the
113  *	field is encapsulated in vm_page_clear_dirty_mask().
114  *
115  *	The page structure contains two counters which prevent page reuse.
116  *	Both counters are protected by the page lock (P).  The hold
117  *	counter counts transient references obtained via a pmap lookup, and
118  *	is also used to prevent page reclamation in situations where it is
119  *	undesirable to block other accesses to the page.  The wire counter
120  *	is used to implement mlock(2) and is non-zero for pages containing
121  *	kernel memory.  Pages that are wired or held will not be reclaimed
122  *	or laundered by the page daemon, but are treated differently during
123  *	a page queue scan: held pages remain at their position in the queue,
124  *	while wired pages are removed from the queue and must later be
125  *	re-enqueued appropriately by the unwiring thread.  It is legal to
126  *	call vm_page_free() on a held page; doing so causes it to be removed
127  *	from its object and page queue, and the page is released to the
128  *	allocator once the last hold reference is dropped.  In contrast,
129  *	wired pages may not be freed.
130  *
131  *	In some pmap implementations, the wire count of a page table page is
132  *	used to track the number of populated entries.
133  *
134  *	The busy lock is an embedded reader-writer lock which protects the
135  *	page's contents and identity (i.e., its <object, pindex> tuple) and
136  *	interlocks with the object lock (O).  In particular, a page may be
137  *	busied or unbusied only with the object write lock held.  To avoid
138  *	bloating the page structure, the busy lock lacks some of the
139  *	features available to the kernel's general-purpose synchronization
140  *	primitives.  As a result, busy lock ordering rules are not verified,
141  *	lock recursion is not detected, and an attempt to xbusy a busy page
142  *	or sbusy an xbusy page results will trigger a panic rather than
143  *	causing the thread to block.  vm_page_sleep_if_busy() can be used to
144  *	sleep until the page's busy state changes, after which the caller
145  *	must re-lookup the page and re-evaluate its state.
146  */
147 
148 #if PAGE_SIZE == 4096
149 #define VM_PAGE_BITS_ALL 0xffu
150 typedef uint8_t vm_page_bits_t;
151 #elif PAGE_SIZE == 8192
152 #define VM_PAGE_BITS_ALL 0xffffu
153 typedef uint16_t vm_page_bits_t;
154 #elif PAGE_SIZE == 16384
155 #define VM_PAGE_BITS_ALL 0xffffffffu
156 typedef uint32_t vm_page_bits_t;
157 #elif PAGE_SIZE == 32768
158 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
159 typedef uint64_t vm_page_bits_t;
160 #endif
161 
162 struct vm_page {
163 	union {
164 		TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
165 		struct {
166 			SLIST_ENTRY(vm_page) ss; /* private slists */
167 			void *pv;
168 		} s;
169 		struct {
170 			u_long p;
171 			u_long v;
172 		} memguard;
173 	} plinks;
174 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) */
175 	vm_object_t object;		/* which object am I in (O,P) */
176 	vm_pindex_t pindex;		/* offset into object (O,P) */
177 	vm_paddr_t phys_addr;		/* physical address of page */
178 	struct md_page md;		/* machine dependent stuff */
179 	u_int wire_count;		/* wired down maps refs (P) */
180 	volatile u_int busy_lock;	/* busy owners lock */
181 	uint16_t hold_count;		/* page hold count (P) */
182 	uint16_t flags;			/* page PG_* flags (P) */
183 	uint8_t aflags;			/* access is atomic */
184 	uint8_t oflags;			/* page VPO_* flags (O) */
185 	uint8_t	queue;			/* page queue index (P,Q) */
186 	int8_t psind;			/* pagesizes[] index (O) */
187 	int8_t segind;			/* vm_phys segment index (C) */
188 	uint8_t	order;			/* index of the buddy queue */
189 	uint8_t pool;			/* vm_phys freepool index (Q) */
190 	u_char	act_count;		/* page usage count (P) */
191 	/* NOTE that these must support one bit per DEV_BSIZE in a page */
192 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
193 	vm_page_bits_t valid;		/* map of valid DEV_BSIZE chunks (O) */
194 	vm_page_bits_t dirty;		/* map of dirty DEV_BSIZE chunks (M) */
195 };
196 
197 /*
198  * Page flags stored in oflags:
199  *
200  * Access to these page flags is synchronized by the lock on the object
201  * containing the page (O).
202  *
203  * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
204  * 	 indicates that the page is not under PV management but
205  * 	 otherwise should be treated as a normal page.  Pages not
206  * 	 under PV management cannot be paged out via the
207  * 	 object/vm_page_t because there is no knowledge of their pte
208  * 	 mappings, and such pages are also not on any PQ queue.
209  *
210  */
211 #define	VPO_UNUSED01	0x01		/* --available-- */
212 #define	VPO_SWAPSLEEP	0x02		/* waiting for swap to finish */
213 #define	VPO_UNMANAGED	0x04		/* no PV management for page */
214 #define	VPO_SWAPINPROG	0x08		/* swap I/O in progress on page */
215 #define	VPO_NOSYNC	0x10		/* do not collect for syncer */
216 
217 /*
218  * Busy page implementation details.
219  * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
220  * even if the support for owner identity is removed because of size
221  * constraints.  Checks on lock recursion are then not possible, while the
222  * lock assertions effectiveness is someway reduced.
223  */
224 #define	VPB_BIT_SHARED		0x01
225 #define	VPB_BIT_EXCLUSIVE	0x02
226 #define	VPB_BIT_WAITERS		0x04
227 #define	VPB_BIT_FLAGMASK						\
228 	(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
229 
230 #define	VPB_SHARERS_SHIFT	3
231 #define	VPB_SHARERS(x)							\
232 	(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
233 #define	VPB_SHARERS_WORD(x)	((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
234 #define	VPB_ONE_SHARER		(1 << VPB_SHARERS_SHIFT)
235 
236 #define	VPB_SINGLE_EXCLUSIVER	VPB_BIT_EXCLUSIVE
237 
238 #define	VPB_UNBUSIED		VPB_SHARERS_WORD(0)
239 
240 #define	PQ_NONE		255
241 #define	PQ_INACTIVE	0
242 #define	PQ_ACTIVE	1
243 #define	PQ_LAUNDRY	2
244 #define	PQ_UNSWAPPABLE	3
245 #define	PQ_COUNT	4
246 
247 #ifndef VM_PAGE_HAVE_PGLIST
248 TAILQ_HEAD(pglist, vm_page);
249 #define VM_PAGE_HAVE_PGLIST
250 #endif
251 SLIST_HEAD(spglist, vm_page);
252 
253 #ifdef _KERNEL
254 extern vm_page_t bogus_page;
255 #endif	/* _KERNEL */
256 
257 extern struct mtx_padalign pa_lock[];
258 
259 #if defined(__arm__)
260 #define	PDRSHIFT	PDR_SHIFT
261 #elif !defined(PDRSHIFT)
262 #define PDRSHIFT	21
263 #endif
264 
265 #define	pa_index(pa)	((pa) >> PDRSHIFT)
266 #define	PA_LOCKPTR(pa)	((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
267 #define	PA_LOCKOBJPTR(pa)	((struct lock_object *)PA_LOCKPTR((pa)))
268 #define	PA_LOCK(pa)	mtx_lock(PA_LOCKPTR(pa))
269 #define	PA_TRYLOCK(pa)	mtx_trylock(PA_LOCKPTR(pa))
270 #define	PA_UNLOCK(pa)	mtx_unlock(PA_LOCKPTR(pa))
271 #define	PA_UNLOCK_COND(pa) 			\
272 	do {		   			\
273 		if ((pa) != 0) {		\
274 			PA_UNLOCK((pa));	\
275 			(pa) = 0;		\
276 		}				\
277 	} while (0)
278 
279 #define	PA_LOCK_ASSERT(pa, a)	mtx_assert(PA_LOCKPTR(pa), (a))
280 
281 #ifdef KLD_MODULE
282 #define	vm_page_lock(m)		vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
283 #define	vm_page_unlock(m)	vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
284 #define	vm_page_trylock(m)	vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
285 #else	/* !KLD_MODULE */
286 #define	vm_page_lockptr(m)	(PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
287 #define	vm_page_lock(m)		mtx_lock(vm_page_lockptr((m)))
288 #define	vm_page_unlock(m)	mtx_unlock(vm_page_lockptr((m)))
289 #define	vm_page_trylock(m)	mtx_trylock(vm_page_lockptr((m)))
290 #endif
291 #if defined(INVARIANTS)
292 #define	vm_page_assert_locked(m)		\
293     vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
294 #define	vm_page_lock_assert(m, a)		\
295     vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
296 #else
297 #define	vm_page_assert_locked(m)
298 #define	vm_page_lock_assert(m, a)
299 #endif
300 
301 /*
302  * The vm_page's aflags are updated using atomic operations.  To set or clear
303  * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
304  * must be used.  Neither these flags nor these functions are part of the KBI.
305  *
306  * PGA_REFERENCED may be cleared only if the page is locked.  It is set by
307  * both the MI and MD VM layers.  However, kernel loadable modules should not
308  * directly set this flag.  They should call vm_page_reference() instead.
309  *
310  * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
311  * When it does so, the object must be locked, or the page must be
312  * exclusive busied.  The MI VM layer must never access this flag
313  * directly.  Instead, it should call pmap_page_is_write_mapped().
314  *
315  * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
316  * at least one executable mapping.  It is not consumed by the MI VM layer.
317  */
318 #define	PGA_WRITEABLE	0x01		/* page may be mapped writeable */
319 #define	PGA_REFERENCED	0x02		/* page has been referenced */
320 #define	PGA_EXECUTABLE	0x04		/* page may be mapped executable */
321 
322 /*
323  * Page flags.  If changed at any other time than page allocation or
324  * freeing, the modification must be protected by the vm_page lock.
325  */
326 #define	PG_FICTITIOUS	0x0004		/* physical page doesn't exist */
327 #define	PG_ZERO		0x0008		/* page is zeroed */
328 #define	PG_MARKER	0x0010		/* special queue marker page */
329 #define	PG_NODUMP	0x0080		/* don't include this page in a dump */
330 #define	PG_UNHOLDFREE	0x0100		/* delayed free of a held page */
331 
332 /*
333  * Misc constants.
334  */
335 #define ACT_DECLINE		1
336 #define ACT_ADVANCE		3
337 #define ACT_INIT		5
338 #define ACT_MAX			64
339 
340 #ifdef _KERNEL
341 
342 #include <sys/systm.h>
343 
344 #include <machine/atomic.h>
345 
346 /*
347  * Each pageable resident page falls into one of five lists:
348  *
349  *	free
350  *		Available for allocation now.
351  *
352  *	inactive
353  *		Low activity, candidates for reclamation.
354  *		This list is approximately LRU ordered.
355  *
356  *	laundry
357  *		This is the list of pages that should be
358  *		paged out next.
359  *
360  *	unswappable
361  *		Dirty anonymous pages that cannot be paged
362  *		out because no swap device is configured.
363  *
364  *	active
365  *		Pages that are "active", i.e., they have been
366  *		recently referenced.
367  *
368  */
369 
370 extern vm_page_t vm_page_array;		/* First resident page in table */
371 extern long vm_page_array_size;		/* number of vm_page_t's */
372 extern long first_page;			/* first physical page number */
373 
374 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
375 
376 /*
377  * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
378  * page to which the given physical address belongs. The correct vm_page_t
379  * object is returned for addresses that are not page-aligned.
380  */
381 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
382 
383 /*
384  * Page allocation parameters for vm_page for the functions
385  * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
386  * vm_page_alloc_freelist().  Some functions support only a subset
387  * of the flags, and ignore others, see the flags legend.
388  *
389  * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
390  * and the vm_page_grab*() functions.  See these functions for details.
391  *
392  * Bits 0 - 1 define class.
393  * Bits 2 - 15 dedicated for flags.
394  * Legend:
395  * (a) - vm_page_alloc() supports the flag.
396  * (c) - vm_page_alloc_contig() supports the flag.
397  * (f) - vm_page_alloc_freelist() supports the flag.
398  * (g) - vm_page_grab() supports the flag.
399  * (p) - vm_page_grab_pages() supports the flag.
400  * Bits above 15 define the count of additional pages that the caller
401  * intends to allocate.
402  */
403 #define VM_ALLOC_NORMAL		0
404 #define VM_ALLOC_INTERRUPT	1
405 #define VM_ALLOC_SYSTEM		2
406 #define	VM_ALLOC_CLASS_MASK	3
407 #define	VM_ALLOC_WAITOK		0x0008	/* (acf) Sleep and retry */
408 #define	VM_ALLOC_WAITFAIL	0x0010	/* (acf) Sleep and return error */
409 #define	VM_ALLOC_WIRED		0x0020	/* (acfgp) Allocate a wired page */
410 #define	VM_ALLOC_ZERO		0x0040	/* (acfgp) Allocate a prezeroed page */
411 #define	VM_ALLOC_NOOBJ		0x0100	/* (acg) No associated object */
412 #define	VM_ALLOC_NOBUSY		0x0200	/* (acgp) Do not excl busy the page */
413 #define	VM_ALLOC_IGN_SBUSY	0x1000	/* (gp) Ignore shared busy flag */
414 #define	VM_ALLOC_NODUMP		0x2000	/* (ag) don't include in dump */
415 #define	VM_ALLOC_SBUSY		0x4000	/* (acgp) Shared busy the page */
416 #define	VM_ALLOC_NOWAIT		0x8000	/* (acfgp) Do not sleep */
417 #define	VM_ALLOC_COUNT_SHIFT	16
418 #define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
419 
420 #ifdef M_NOWAIT
421 static inline int
422 malloc2vm_flags(int malloc_flags)
423 {
424 	int pflags;
425 
426 	KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
427 	    (malloc_flags & M_NOWAIT) != 0,
428 	    ("M_USE_RESERVE requires M_NOWAIT"));
429 	pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
430 	    VM_ALLOC_SYSTEM;
431 	if ((malloc_flags & M_ZERO) != 0)
432 		pflags |= VM_ALLOC_ZERO;
433 	if ((malloc_flags & M_NODUMP) != 0)
434 		pflags |= VM_ALLOC_NODUMP;
435 	if ((malloc_flags & M_NOWAIT))
436 		pflags |= VM_ALLOC_NOWAIT;
437 	if ((malloc_flags & M_WAITOK))
438 		pflags |= VM_ALLOC_WAITOK;
439 	return (pflags);
440 }
441 #endif
442 
443 /*
444  * Predicates supported by vm_page_ps_test():
445  *
446  *	PS_ALL_DIRTY is true only if the entire (super)page is dirty.
447  *	However, it can be spuriously false when the (super)page has become
448  *	dirty in the pmap but that information has not been propagated to the
449  *	machine-independent layer.
450  */
451 #define	PS_ALL_DIRTY	0x1
452 #define	PS_ALL_VALID	0x2
453 #define	PS_NONE_BUSY	0x4
454 
455 void vm_page_busy_downgrade(vm_page_t m);
456 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared);
457 void vm_page_flash(vm_page_t m);
458 void vm_page_hold(vm_page_t mem);
459 void vm_page_unhold(vm_page_t mem);
460 void vm_page_free(vm_page_t m);
461 void vm_page_free_zero(vm_page_t m);
462 
463 void vm_page_activate (vm_page_t);
464 void vm_page_advise(vm_page_t m, int advice);
465 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
466 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int);
467 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
468 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
469     vm_page_t);
470 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
471     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
472     vm_paddr_t boundary, vm_memattr_t memattr);
473 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
474     vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
475     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
476     vm_memattr_t memattr);
477 vm_page_t vm_page_alloc_freelist(int, int);
478 vm_page_t vm_page_alloc_freelist_domain(int, int, int);
479 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
480 void vm_page_change_lock(vm_page_t m, struct mtx **mtx);
481 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
482 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
483     vm_page_t *ma, int count);
484 void vm_page_deactivate(vm_page_t);
485 void vm_page_deactivate_noreuse(vm_page_t);
486 void vm_page_dequeue(vm_page_t m);
487 void vm_page_dequeue_locked(vm_page_t m);
488 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
489 void vm_page_free_phys_pglist(struct pglist *tq);
490 bool vm_page_free_prep(vm_page_t m, bool pagequeue_locked);
491 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
492 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
493 void vm_page_init_marker(vm_page_t m, int queue);
494 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
495 void vm_page_launder(vm_page_t m);
496 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
497 vm_page_t vm_page_next(vm_page_t m);
498 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *);
499 struct vm_pagequeue *vm_page_pagequeue(vm_page_t m);
500 vm_page_t vm_page_prev(vm_page_t m);
501 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
502 void vm_page_putfake(vm_page_t m);
503 void vm_page_readahead_finish(vm_page_t m);
504 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
505     vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
506 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
507     vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
508 void vm_page_reference(vm_page_t m);
509 void vm_page_remove (vm_page_t);
510 int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
511 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object,
512     vm_pindex_t pindex);
513 void vm_page_requeue(vm_page_t m);
514 void vm_page_requeue_locked(vm_page_t m);
515 int vm_page_sbusied(vm_page_t m);
516 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
517     vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
518 void vm_page_set_valid_range(vm_page_t m, int base, int size);
519 int vm_page_sleep_if_busy(vm_page_t m, const char *msg);
520 vm_offset_t vm_page_startup(vm_offset_t vaddr);
521 void vm_page_sunbusy(vm_page_t m);
522 bool vm_page_try_to_free(vm_page_t m);
523 int vm_page_trysbusy(vm_page_t m);
524 void vm_page_unhold_pages(vm_page_t *ma, int count);
525 void vm_page_unswappable(vm_page_t m);
526 bool vm_page_unwire(vm_page_t m, uint8_t queue);
527 bool vm_page_unwire_noq(vm_page_t m);
528 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
529 void vm_page_wire (vm_page_t);
530 void vm_page_xunbusy_hard(vm_page_t m);
531 void vm_page_xunbusy_maybelocked(vm_page_t m);
532 void vm_page_set_validclean (vm_page_t, int, int);
533 void vm_page_clear_dirty (vm_page_t, int, int);
534 void vm_page_set_invalid (vm_page_t, int, int);
535 int vm_page_is_valid (vm_page_t, int, int);
536 void vm_page_test_dirty (vm_page_t);
537 vm_page_bits_t vm_page_bits(int base, int size);
538 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
539 void vm_page_free_toq(vm_page_t m);
540 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
541 
542 void vm_page_dirty_KBI(vm_page_t m);
543 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
544 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
545 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
546 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
547 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
548 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
549 #endif
550 
551 #define	vm_page_assert_sbusied(m)					\
552 	KASSERT(vm_page_sbusied(m),					\
553 	    ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
554 	    (m), __FILE__, __LINE__))
555 
556 #define	vm_page_assert_unbusied(m)					\
557 	KASSERT(!vm_page_busied(m),					\
558 	    ("vm_page_assert_unbusied: page %p busy @ %s:%d",		\
559 	    (m), __FILE__, __LINE__))
560 
561 #define	vm_page_assert_xbusied(m)					\
562 	KASSERT(vm_page_xbusied(m),					\
563 	    ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
564 	    (m), __FILE__, __LINE__))
565 
566 #define	vm_page_busied(m)						\
567 	((m)->busy_lock != VPB_UNBUSIED)
568 
569 #define	vm_page_sbusy(m) do {						\
570 	if (!vm_page_trysbusy(m))					\
571 		panic("%s: page %p failed shared busying", __func__,	\
572 		    (m));						\
573 } while (0)
574 
575 #define	vm_page_tryxbusy(m)						\
576 	(atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED,		\
577 	    VPB_SINGLE_EXCLUSIVER))
578 
579 #define	vm_page_xbusied(m)						\
580 	(((m)->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0)
581 
582 #define	vm_page_xbusy(m) do {						\
583 	if (!vm_page_tryxbusy(m))					\
584 		panic("%s: page %p failed exclusive busying", __func__,	\
585 		    (m));						\
586 } while (0)
587 
588 /* Note: page m's lock must not be owned by the caller. */
589 #define	vm_page_xunbusy(m) do {						\
590 	if (!atomic_cmpset_rel_int(&(m)->busy_lock,			\
591 	    VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED))			\
592 		vm_page_xunbusy_hard(m);				\
593 } while (0)
594 
595 #ifdef INVARIANTS
596 void vm_page_object_lock_assert(vm_page_t m);
597 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	vm_page_object_lock_assert(m)
598 void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits);
599 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)				\
600 	vm_page_assert_pga_writeable(m, bits)
601 #else
602 #define	VM_PAGE_OBJECT_LOCK_ASSERT(m)	(void)0
603 #define	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits)	(void)0
604 #endif
605 
606 /*
607  * We want to use atomic updates for the aflags field, which is 8 bits wide.
608  * However, not all architectures support atomic operations on 8-bit
609  * destinations.  In order that we can easily use a 32-bit operation, we
610  * require that the aflags field be 32-bit aligned.
611  */
612 CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0);
613 
614 /*
615  *	Clear the given bits in the specified page.
616  */
617 static inline void
618 vm_page_aflag_clear(vm_page_t m, uint8_t bits)
619 {
620 	uint32_t *addr, val;
621 
622 	/*
623 	 * The PGA_REFERENCED flag can only be cleared if the page is locked.
624 	 */
625 	if ((bits & PGA_REFERENCED) != 0)
626 		vm_page_assert_locked(m);
627 
628 	/*
629 	 * Access the whole 32-bit word containing the aflags field with an
630 	 * atomic update.  Parallel non-atomic updates to the other fields
631 	 * within this word are handled properly by the atomic update.
632 	 */
633 	addr = (void *)&m->aflags;
634 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
635 	    ("vm_page_aflag_clear: aflags is misaligned"));
636 	val = bits;
637 #if BYTE_ORDER == BIG_ENDIAN
638 	val <<= 24;
639 #endif
640 	atomic_clear_32(addr, val);
641 }
642 
643 /*
644  *	Set the given bits in the specified page.
645  */
646 static inline void
647 vm_page_aflag_set(vm_page_t m, uint8_t bits)
648 {
649 	uint32_t *addr, val;
650 
651 	VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
652 
653 	/*
654 	 * Access the whole 32-bit word containing the aflags field with an
655 	 * atomic update.  Parallel non-atomic updates to the other fields
656 	 * within this word are handled properly by the atomic update.
657 	 */
658 	addr = (void *)&m->aflags;
659 	KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0,
660 	    ("vm_page_aflag_set: aflags is misaligned"));
661 	val = bits;
662 #if BYTE_ORDER == BIG_ENDIAN
663 	val <<= 24;
664 #endif
665 	atomic_set_32(addr, val);
666 }
667 
668 /*
669  *	vm_page_dirty:
670  *
671  *	Set all bits in the page's dirty field.
672  *
673  *	The object containing the specified page must be locked if the
674  *	call is made from the machine-independent layer.
675  *
676  *	See vm_page_clear_dirty_mask().
677  */
678 static __inline void
679 vm_page_dirty(vm_page_t m)
680 {
681 
682 	/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
683 #if defined(KLD_MODULE) || defined(INVARIANTS)
684 	vm_page_dirty_KBI(m);
685 #else
686 	m->dirty = VM_PAGE_BITS_ALL;
687 #endif
688 }
689 
690 /*
691  *	vm_page_remque:
692  *
693  *	If the given page is in a page queue, then remove it from that page
694  *	queue.
695  *
696  *	The page must be locked.
697  */
698 static inline void
699 vm_page_remque(vm_page_t m)
700 {
701 
702 	if (m->queue != PQ_NONE)
703 		vm_page_dequeue(m);
704 }
705 
706 /*
707  *	vm_page_undirty:
708  *
709  *	Set page to not be dirty.  Note: does not clear pmap modify bits
710  */
711 static __inline void
712 vm_page_undirty(vm_page_t m)
713 {
714 
715 	VM_PAGE_OBJECT_LOCK_ASSERT(m);
716 	m->dirty = 0;
717 }
718 
719 static inline void
720 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
721     vm_page_t mold)
722 {
723 	vm_page_t mret;
724 
725 	mret = vm_page_replace(mnew, object, pindex);
726 	KASSERT(mret == mold,
727 	    ("invalid page replacement, mold=%p, mret=%p", mold, mret));
728 
729 	/* Unused if !INVARIANTS. */
730 	(void)mold;
731 	(void)mret;
732 }
733 
734 static inline bool
735 vm_page_active(vm_page_t m)
736 {
737 
738 	return (m->queue == PQ_ACTIVE);
739 }
740 
741 static inline bool
742 vm_page_inactive(vm_page_t m)
743 {
744 
745 	return (m->queue == PQ_INACTIVE);
746 }
747 
748 static inline bool
749 vm_page_in_laundry(vm_page_t m)
750 {
751 
752 	return (m->queue == PQ_LAUNDRY || m->queue == PQ_UNSWAPPABLE);
753 }
754 
755 /*
756  *	vm_page_held:
757  *
758  *	Return true if a reference prevents the page from being reclaimable.
759  */
760 static inline bool
761 vm_page_held(vm_page_t m)
762 {
763 
764 	return (m->hold_count > 0 || m->wire_count > 0);
765 }
766 
767 #endif				/* _KERNEL */
768 #endif				/* !_VM_PAGE_ */
769