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 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
45 *
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61/*
62 * Resident memory system definitions.
63 */
64
65#ifndef _VM_PAGE_
66#define _VM_PAGE_
67
68#include <vm/pmap.h>
69#include <vm/_vm_phys.h>
70
71/*
72 * Management of resident (logical) pages.
73 *
74 * A small structure is kept for each resident
75 * page, indexed by page number. Each structure
76 * is an element of several collections:
77 *
78 * A radix tree used to quickly
79 * perform object/offset lookups
80 *
81 * An ordered list of pages due for pageout.
82 *
83 * In addition, the structure contains the object
84 * and offset to which this page belongs (for pageout),
85 * and sundry status bits.
86 *
87 * In general, operations on this structure's mutable fields are
88 * synchronized using either one of or a combination of locks. If a
89 * field is annotated with two of these locks then holding either is
90 * sufficient for read access but both are required for write access.
91 * The queue lock for a page depends on the value of its queue field and is
92 * described in detail below.
93 *
94 * The following annotations are possible:
95 * (A) the field must be accessed using atomic(9) and may require
96 * additional synchronization.
97 * (B) the page busy lock.
98 * (C) the field is immutable.
99 * (F) the per-domain lock for the free queues.
100 * (M) Machine dependent, defined by pmap layer.
101 * (O) the object that the page belongs to.
102 * (Q) the page's queue lock.
103 *
104 * The busy lock is an embedded reader-writer lock that protects the
105 * page's contents and identity (i.e., its <object, pindex> tuple) as
106 * well as certain valid/dirty modifications. To avoid bloating the
107 * the page structure, the busy lock lacks some of the features available
108 * the kernel's general-purpose synchronization primitives. As a result,
109 * busy lock ordering rules are not verified, lock recursion is not
110 * detected, and an attempt to xbusy a busy page or sbusy an xbusy page
111 * results will trigger a panic rather than causing the thread to block.
112 * vm_page_sleep_if_busy() can be used to sleep until the page's busy
113 * state changes, after which the caller must re-lookup the page and
114 * re-evaluate its state. vm_page_busy_acquire() will block until
115 * the lock is acquired.
116 *
117 * The valid field is protected by the page busy lock (B) and object
118 * lock (O). Transitions from invalid to valid are generally done
119 * via I/O or zero filling and do not require the object lock.
120 * These must be protected with the busy lock to prevent page-in or
121 * creation races. Page invalidation generally happens as a result
122 * of truncate or msync. When invalidated, pages must not be present
123 * in pmap and must hold the object lock to prevent concurrent
124 * speculative read-only mappings that do not require busy. I/O
125 * routines may check for validity without a lock if they are prepared
126 * to handle invalidation races with higher level locks (vnode) or are
127 * unconcerned with races so long as they hold a reference to prevent
128 * recycling. When a valid bit is set while holding a shared busy
129 * lock (A) atomic operations are used to protect against concurrent
130 * modification.
131 *
132 * In contrast, the synchronization of accesses to the page's
133 * dirty field is a mix of machine dependent (M) and busy (B). In
134 * the machine-independent layer, the page busy must be held to
135 * operate on the field. However, the pmap layer is permitted to
136 * set all bits within the field without holding that lock. If the
137 * underlying architecture does not support atomic read-modify-write
138 * operations on the field's type, then the machine-independent
139 * layer uses a 32-bit atomic on the aligned 32-bit word that
140 * contains the dirty field. In the machine-independent layer,
141 * the implementation of read-modify-write operations on the
142 * field is encapsulated in vm_page_clear_dirty_mask(). An
143 * exclusive busy lock combined with pmap_remove_{write/all}() is the
144 * only way to ensure a page can not become dirty. I/O generally
145 * removes the page from pmap to ensure exclusive access and atomic
146 * writes.
147 *
148 * The ref_count field tracks references to the page. References that
149 * prevent the page from being reclaimable are called wirings and are
150 * counted in the low bits of ref_count. The containing object's
151 * reference, if one exists, is counted using the VPRC_OBJREF bit in the
152 * ref_count field. Additionally, the VPRC_BLOCKED bit is used to
153 * atomically check for wirings and prevent new wirings via
154 * pmap_extract_and_hold(). When a page belongs to an object, it may be
155 * wired only when the object is locked, or the page is busy, or by
156 * pmap_extract_and_hold(). As a result, if the object is locked and the
157 * page is not busy (or is exclusively busied by the current thread), and
158 * the page is unmapped, its wire count will not increase. The ref_count
159 * field is updated using atomic operations in most cases, except when it
160 * is known that no other references to the page exist, such as in the page
161 * allocator. A page may be present in the page queues, or even actively
162 * scanned by the page daemon, without an explicitly counted referenced.
163 * The page daemon must therefore handle the possibility of a concurrent
164 * free of the page.
165 *
166 * The queue state of a page consists of the queue and act_count fields of
167 * its atomically updated state, and the subset of atomic flags specified
168 * by PGA_QUEUE_STATE_MASK. The queue field contains the page's page queue
169 * index, or PQ_NONE if it does not belong to a page queue. To modify the
170 * queue field, the page queue lock corresponding to the old value must be
171 * held, unless that value is PQ_NONE, in which case the queue index must
172 * be updated using an atomic RMW operation. There is one exception to
173 * this rule: the page daemon may transition the queue field from
174 * PQ_INACTIVE to PQ_NONE immediately prior to freeing the page during an
175 * inactive queue scan. At that point the page is already dequeued and no
176 * other references to that vm_page structure can exist. The PGA_ENQUEUED
177 * flag, when set, indicates that the page structure is physically inserted
178 * into the queue corresponding to the page's queue index, and may only be
179 * set or cleared with the corresponding page queue lock held.
180 *
181 * To avoid contention on page queue locks, page queue operations (enqueue,
182 * dequeue, requeue) are batched using fixed-size per-CPU queues. A
183 * deferred operation is requested by setting one of the flags in
184 * PGA_QUEUE_OP_MASK and inserting an entry into a batch queue. When a
185 * queue is full, an attempt to insert a new entry will lock the page
186 * queues and trigger processing of the pending entries. The
187 * type-stability of vm_page structures is crucial to this scheme since the
188 * processing of entries in a given batch queue may be deferred
189 * indefinitely. In particular, a page may be freed with pending batch
190 * queue entries. The page queue operation flags must be set using atomic
191 * RWM operations.
192 */
193
194#if PAGE_SIZE == 4096
195#define VM_PAGE_BITS_ALL 0xffu
196typedef uint8_t vm_page_bits_t;
197#elif PAGE_SIZE == 8192
198#define VM_PAGE_BITS_ALL 0xffffu
199typedef uint16_t vm_page_bits_t;
200#elif PAGE_SIZE == 16384
201#define VM_PAGE_BITS_ALL 0xffffffffu
202typedef uint32_t vm_page_bits_t;
203#elif PAGE_SIZE == 32768
204#define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
205typedef uint64_t vm_page_bits_t;
206#endif
207
208typedef union vm_page_astate {
209 struct {
210 uint16_t flags;
211 uint8_t queue;
212 uint8_t act_count;
213 };
214 uint32_t _bits;
215} vm_page_astate_t;
216
217struct vm_page {
218 union {
219 TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
220 struct {
221 SLIST_ENTRY(vm_page) ss; /* private slists */
222 } s;
223 struct {
224 u_long p;
225 u_long v;
226 } memguard;
227 struct {
228 void *slab;
229 void *zone;
230 } uma;
231 } plinks;
232 vm_object_t object; /* which object am I in (O) */
233 vm_pindex_t pindex; /* offset into object (O,P) */
234 vm_paddr_t phys_addr; /* physical address of page (C) */
235 struct md_page md; /* machine dependent stuff */
236 u_int ref_count; /* page references (A) */
237 u_int busy_lock; /* busy owners lock (A) */
238 union vm_page_astate a; /* state accessed atomically (A) */
239 uint8_t order; /* index of the buddy queue (F) */
240 uint8_t pool; /* vm_phys freepool index (F) */
241 uint8_t flags; /* page PG_* flags (P) */
242 uint8_t oflags; /* page VPO_* flags (O) */
243 int8_t psind; /* pagesizes[] index (O) */
244 int8_t segind; /* vm_phys segment index (C) */
245 /* NOTE that these must support one bit per DEV_BSIZE in a page */
246 /* so, on normal X86 kernels, they must be at least 8 bits wide */
247 vm_page_bits_t valid; /* valid DEV_BSIZE chunk map (O,B) */
248 vm_page_bits_t dirty; /* dirty DEV_BSIZE chunk map (M,B) */
249};
250
251/*
252 * Special bits used in the ref_count field.
253 *
254 * ref_count is normally used to count wirings that prevent the page from being
255 * reclaimed, but also supports several special types of references that do not
256 * prevent reclamation. Accesses to the ref_count field must be atomic unless
257 * the page is unallocated.
258 *
259 * VPRC_OBJREF is the reference held by the containing object. It can set or
260 * cleared only when the corresponding object's write lock is held.
261 *
262 * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while
263 * attempting to tear down all mappings of a given page. The page busy lock and
264 * object write lock must both be held in order to set or clear this bit.
265 */
266#define VPRC_BLOCKED 0x40000000u /* mappings are being removed */
267#define VPRC_OBJREF 0x80000000u /* object reference, cleared with (O) */
268#define VPRC_WIRE_COUNT(c) ((c) & ~(VPRC_BLOCKED | VPRC_OBJREF))
269#define VPRC_WIRE_COUNT_MAX (~(VPRC_BLOCKED | VPRC_OBJREF))
270
271/*
272 * Page flags stored in oflags:
273 *
274 * Access to these page flags is synchronized by the lock on the object
275 * containing the page (O).
276 *
277 * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
278 * indicates that the page is not under PV management but
279 * otherwise should be treated as a normal page. Pages not
280 * under PV management cannot be paged out via the
281 * object/vm_page_t because there is no knowledge of their pte
282 * mappings, and such pages are also not on any PQ queue.
283 *
284 */
285#define VPO_KMEM_EXEC 0x01 /* kmem mapping allows execution */
286#define VPO_SWAPSLEEP 0x02 /* waiting for swap to finish */
287#define VPO_UNMANAGED 0x04 /* no PV management for page */
288#define VPO_SWAPINPROG 0x08 /* swap I/O in progress on page */
289
290/*
291 * Busy page implementation details.
292 * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
293 * even if the support for owner identity is removed because of size
294 * constraints. Checks on lock recursion are then not possible, while the
295 * lock assertions effectiveness is someway reduced.
296 */
297#define VPB_BIT_SHARED 0x01
298#define VPB_BIT_EXCLUSIVE 0x02
299#define VPB_BIT_WAITERS 0x04
300#define VPB_BIT_FLAGMASK \
301 (VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
302
303#define VPB_SHARERS_SHIFT 3
304#define VPB_SHARERS(x) \
305 (((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
306#define VPB_SHARERS_WORD(x) ((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
307#define VPB_ONE_SHARER (1 << VPB_SHARERS_SHIFT)
308
309#define VPB_SINGLE_EXCLUSIVE VPB_BIT_EXCLUSIVE
310#ifdef INVARIANTS
311#define VPB_CURTHREAD_EXCLUSIVE \
312 (VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK))
313#else
314#define VPB_CURTHREAD_EXCLUSIVE VPB_SINGLE_EXCLUSIVE
315#endif
316
317#define VPB_UNBUSIED VPB_SHARERS_WORD(0)
318
319/* Freed lock blocks both shared and exclusive. */
320#define VPB_FREED (0xffffffff - VPB_BIT_SHARED)
321
322#define PQ_NONE 255
323#define PQ_INACTIVE 0
324#define PQ_ACTIVE 1
325#define PQ_LAUNDRY 2
326#define PQ_UNSWAPPABLE 3
327#define PQ_COUNT 4
328
329#ifndef VM_PAGE_HAVE_PGLIST
330TAILQ_HEAD(pglist, vm_page);
331#define VM_PAGE_HAVE_PGLIST
332#endif
333SLIST_HEAD(spglist, vm_page);
334
335#ifdef _KERNEL
336extern vm_page_t bogus_page;
337#endif /* _KERNEL */
338
339/*
340 * The vm_page's aflags are updated using atomic operations. To set or clear
341 * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
342 * must be used. Neither these flags nor these functions are part of the KBI.
343 *
344 * PGA_REFERENCED may be cleared only if the page is locked. It is set by
345 * both the MI and MD VM layers. However, kernel loadable modules should not
346 * directly set this flag. They should call vm_page_reference() instead.
347 *
348 * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
349 * When it does so, the object must be locked, or the page must be
350 * exclusive busied. The MI VM layer must never access this flag
351 * directly. Instead, it should call pmap_page_is_write_mapped().
352 *
353 * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
354 * at least one executable mapping. It is not consumed by the MI VM layer.
355 *
356 * PGA_NOSYNC must be set and cleared with the page busy lock held.
357 *
358 * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
359 * from a page queue, respectively. It determines whether the plinks.q field
360 * of the page is valid. To set or clear this flag, page's "queue" field must
361 * be a valid queue index, and the corresponding page queue lock must be held.
362 *
363 * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
364 * queue, and cleared when the dequeue request is processed. A page may
365 * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
366 * is requested after the page is scheduled to be enqueued but before it is
367 * actually inserted into the page queue.
368 *
369 * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
370 * in its page queue.
371 *
372 * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
373 * the inactive queue, thus bypassing LRU.
374 *
375 * The PGA_DEQUEUE, PGA_REQUEUE and PGA_REQUEUE_HEAD flags must be set using an
376 * atomic RMW operation to ensure that the "queue" field is a valid queue index,
377 * and the corresponding page queue lock must be held when clearing any of the
378 * flags.
379 *
380 * PGA_SWAP_FREE is used to defer freeing swap space to the pageout daemon
381 * when the context that dirties the page does not have the object write lock
382 * held.
383 */
384#define PGA_WRITEABLE 0x0001 /* page may be mapped writeable */
385#define PGA_REFERENCED 0x0002 /* page has been referenced */
386#define PGA_EXECUTABLE 0x0004 /* page may be mapped executable */
387#define PGA_ENQUEUED 0x0008 /* page is enqueued in a page queue */
388#define PGA_DEQUEUE 0x0010 /* page is due to be dequeued */
389#define PGA_REQUEUE 0x0020 /* page is due to be requeued */
390#define PGA_REQUEUE_HEAD 0x0040 /* page requeue should bypass LRU */
391#define PGA_NOSYNC 0x0080 /* do not collect for syncer */
392#define PGA_SWAP_FREE 0x0100 /* page with swap space was dirtied */
393#define PGA_SWAP_SPACE 0x0200 /* page has allocated swap space */
394
395#define PGA_QUEUE_OP_MASK (PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD)
396#define PGA_QUEUE_STATE_MASK (PGA_ENQUEUED | PGA_QUEUE_OP_MASK)
397
398/*
399 * Page flags. Updates to these flags are not synchronized, and thus they must
400 * be set during page allocation or free to avoid races.
401 *
402 * The PG_PCPU_CACHE flag is set at allocation time if the page was
403 * allocated from a per-CPU cache. It is cleared the next time that the
404 * page is allocated from the physical memory allocator.
405 */
406#define PG_PCPU_CACHE 0x01 /* was allocated from per-CPU caches */
407#define PG_FICTITIOUS 0x02 /* physical page doesn't exist */
408#define PG_ZERO 0x04 /* page is zeroed */
409#define PG_MARKER 0x08 /* special queue marker page */
410#define PG_NODUMP 0x10 /* don't include this page in a dump */
411#define PG_NOFREE 0x20 /* page should never be freed. */
412
413/*
414 * Misc constants.
415 */
416#define ACT_DECLINE 1
417#define ACT_ADVANCE 3
418#define ACT_INIT 5
419#define ACT_MAX 64
420
421#ifdef _KERNEL
422
423#include <sys/kassert.h>
424#include <machine/atomic.h>
425struct pctrie_iter;
426
427/*
428 * Each pageable resident page falls into one of five lists:
429 *
430 * free
431 * Available for allocation now.
432 *
433 * inactive
434 * Low activity, candidates for reclamation.
435 * This list is approximately LRU ordered.
436 *
437 * laundry
438 * This is the list of pages that should be
439 * paged out next.
440 *
441 * unswappable
442 * Dirty anonymous pages that cannot be paged
443 * out because no swap device is configured.
444 *
445 * active
446 * Pages that are "active", i.e., they have been
447 * recently referenced.
448 *
449 */
450
451extern vm_page_t vm_page_array; /* First resident page in table */
452extern long vm_page_array_size; /* number of vm_page_t's */
453extern long first_page; /* first physical page number */
454
455#define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr)
456
457/*
458 * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
459 * page to which the given physical address belongs. The correct vm_page_t
460 * object is returned for addresses that are not page-aligned.
461 */
462vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
463
464/*
465 * vm_page allocation arguments for the functions vm_page_alloc(),
466 * vm_page_alloc_contig(), vm_page_alloc_noobj(), vm_page_grab(), and
467 * vm_page_grab_pages(). Each function supports only a subset of the flags.
468 * See the flags legend.
469 *
470 * The meaning of VM_ALLOC_ZERO varies: vm_page_alloc_noobj(), vm_page_grab(),
471 * and vm_page_grab_pages() guarantee that the returned pages are zeroed; in
472 * contrast vm_page_alloc() and vm_page_alloc_contig() do not, leaving it to
473 * the caller to test the page's flags for PG_ZERO.
474 *
475 * Bits 0 - 1 define class.
476 * Bits 2 - 15 dedicated for flags.
477 * Legend:
478 * (a) - vm_page_alloc() supports the flag.
479 * (c) - vm_page_alloc_contig() supports the flag.
480 * (g) - vm_page_grab() supports the flag.
481 * (n) - vm_page_alloc_noobj() supports the flag.
482 * (p) - vm_page_grab_pages() supports the flag.
483 * Bits above 15 define the count of additional pages that the caller
484 * intends to allocate.
485 */
486#define VM_ALLOC_NORMAL 0
487#define VM_ALLOC_INTERRUPT 1
488#define VM_ALLOC_SYSTEM 2
489#define VM_ALLOC_CLASS_MASK 3
490#define VM_ALLOC_WAITOK 0x0008 /* (gnp) Sleep and retry */
491#define VM_ALLOC_WAITFAIL 0x0010 /* (acgnp) Sleep and return error */
492#define VM_ALLOC_WIRED 0x0020 /* (acgnp) Allocate a wired page */
493#define VM_ALLOC_ZERO 0x0040 /* (acgnp) Allocate a zeroed page */
494#define VM_ALLOC_NORECLAIM 0x0080 /* (c) Do not reclaim after failure */
495#define VM_ALLOC_NOFREE 0x0100 /* (agnp) Page will never be freed */
496#define VM_ALLOC_NOBUSY 0x0200 /* (acgp) Do not excl busy the page */
497#define VM_ALLOC_NOCREAT 0x0400 /* (gp) Do not allocate a page */
498#define VM_ALLOC_AVAIL1 0x0800
499#define VM_ALLOC_IGN_SBUSY 0x1000 /* (gp) Ignore shared busy state */
500#define VM_ALLOC_NODUMP 0x2000 /* (acgnp) Do not include in dump */
501#define VM_ALLOC_SBUSY 0x4000 /* (acgp) Shared busy the page */
502#define VM_ALLOC_NOWAIT 0x8000 /* (acgnp) Do not sleep */
503#define VM_ALLOC_COUNT_MAX 0xffff
504#define VM_ALLOC_COUNT_SHIFT 16
505#define VM_ALLOC_COUNT_MASK (VM_ALLOC_COUNT(VM_ALLOC_COUNT_MAX))
506#define VM_ALLOC_COUNT(count) ({ /* (acgn) Additional pages */ \
507 KASSERT((count) <= VM_ALLOC_COUNT_MAX, \
508 ("%s: invalid VM_ALLOC_COUNT value", __func__)); \
509 (count) << VM_ALLOC_COUNT_SHIFT; \
510})
511
512#ifdef M_NOWAIT
513static inline int
514malloc2vm_flags(int malloc_flags)
515{
516 int pflags;
517
518 KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
519 (malloc_flags & M_NOWAIT) != 0,
520 ("M_USE_RESERVE requires M_NOWAIT"));
521 pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
522 VM_ALLOC_SYSTEM;
523 if ((malloc_flags & M_ZERO) != 0)
524 pflags |= VM_ALLOC_ZERO;
525 if ((malloc_flags & M_NODUMP) != 0)
526 pflags |= VM_ALLOC_NODUMP;
527 if ((malloc_flags & M_NOWAIT))
528 pflags |= VM_ALLOC_NOWAIT;
529 if ((malloc_flags & M_WAITOK))
530 pflags |= VM_ALLOC_WAITOK;
531 if ((malloc_flags & M_NORECLAIM))
532 pflags |= VM_ALLOC_NORECLAIM;
533 if ((malloc_flags & M_NEVERFREED))
534 pflags |= VM_ALLOC_NOFREE;
535 return (pflags);
536}
537#endif
538
539/*
540 * Predicates supported by vm_page_ps_test():
541 *
542 * PS_ALL_DIRTY is true only if the entire (super)page is dirty.
543 * However, it can be spuriously false when the (super)page has become
544 * dirty in the pmap but that information has not been propagated to the
545 * machine-independent layer.
546 */
547#define PS_ALL_DIRTY 0x1
548#define PS_ALL_VALID 0x2
549#define PS_NONE_BUSY 0x4
550
551void vm_page_activate (vm_page_t);
552void vm_page_advise(vm_page_t m, int advice);
553vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
554vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
555 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
556 vm_paddr_t boundary, vm_memattr_t memattr);
557vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
558 vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
559 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
560 vm_memattr_t memattr);
561vm_page_t vm_page_alloc_domain_iter(vm_object_t object, vm_pindex_t pindex,
562 int domain, int req, struct pctrie_iter *pages);
563vm_page_t vm_page_alloc_iter(vm_object_t object, vm_pindex_t pindex, int req,
564 struct pctrie_iter *pages);
565vm_page_t vm_page_alloc_noobj(int);
566vm_page_t vm_page_alloc_noobj_domain(int, int);
567vm_page_t vm_page_alloc_noobj_contig(int req, u_long npages, vm_paddr_t low,
568 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
569 vm_memattr_t memattr);
570vm_page_t vm_page_alloc_noobj_contig_domain(int domain, int req, u_long npages,
571 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
572 vm_memattr_t memattr);
573void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set);
574bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
575bool vm_page_busy_acquire(vm_page_t m, int allocflags);
576void vm_page_busy_downgrade(vm_page_t m);
577int vm_page_busy_tryupgrade(vm_page_t m);
578bool vm_page_busy_sleep(vm_page_t m, const char *msg, int allocflags);
579void vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m,
580 vm_pindex_t pindex, const char *wmesg, int allocflags);
581void vm_page_deactivate(vm_page_t m);
582void vm_page_deactivate_noreuse(vm_page_t m);
583void vm_page_dequeue(vm_page_t m);
584void vm_page_dequeue_deferred(vm_page_t m);
585void vm_page_free(vm_page_t m);
586void vm_page_free_invalid(vm_page_t m);
587int vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
588void vm_page_free_zero(vm_page_t m);
589vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
590int vm_page_grab_zero_partial(vm_object_t object, vm_pindex_t pindex, int base,
591 int end);
592vm_page_t vm_page_grab(vm_object_t, vm_pindex_t, int);
593vm_page_t vm_page_grab_iter(vm_object_t object, vm_pindex_t pindex,
594 int allocflags, struct pctrie_iter *pages);
595vm_page_t vm_page_grab_unlocked(vm_object_t, vm_pindex_t, int);
596int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
597 vm_page_t *ma, int count);
598int vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex,
599 int allocflags, vm_page_t *ma, int count);
600int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex,
601 int allocflags);
602int vm_page_grab_valid_iter(vm_page_t *mp, vm_object_t object,
603 vm_pindex_t pindex, int allocflags, struct pctrie_iter *pages);
604int vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object,
605 vm_pindex_t pindex, int allocflags);
606void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
607void vm_page_init_marker(vm_page_t marker, int queue, uint16_t aflags);
608void vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind, int pool);
609int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
610void vm_page_invalid(vm_page_t m);
611void vm_page_iter_free(struct pctrie_iter *pages, vm_page_t m);
612void vm_page_iter_init(struct pctrie_iter *, vm_object_t);
613int vm_page_iter_insert(vm_page_t m, vm_object_t, vm_pindex_t,
614 struct pctrie_iter *);
615void vm_page_iter_limit_init(struct pctrie_iter *, vm_object_t, vm_pindex_t);
616bool vm_page_iter_remove(struct pctrie_iter *pages, vm_page_t m);
617bool vm_page_iter_rename(struct pctrie_iter *old_pages, vm_page_t m,
618 vm_object_t new_object, vm_pindex_t new_pindex);
619void vm_page_launder(vm_page_t m);
620vm_page_t vm_page_lookup(vm_object_t, vm_pindex_t);
621vm_page_t vm_page_lookup_unlocked(vm_object_t, vm_pindex_t);
622void vm_page_pqbatch_drain(void);
623void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
624bool vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old,
625 vm_page_astate_t new);
626bool vm_page_ps_test(vm_page_t m, int psind, int flags, vm_page_t skip_m);
627void vm_page_putfake(vm_page_t m);
628void vm_page_readahead_finish(vm_page_t m);
629int vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
630 vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
631int vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
632 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
633int vm_page_reclaim_contig_domain_ext(int domain, int req, u_long npages,
634 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
635 int desired_runs);
636void vm_page_reference(vm_page_t m);
637#define VPR_TRYFREE 0x01
638#define VPR_NOREUSE 0x02
639void vm_page_release(vm_page_t m, int flags);
640void vm_page_release_locked(vm_page_t m, int flags);
641vm_page_t vm_page_relookup(vm_object_t, vm_pindex_t);
642bool vm_page_remove(vm_page_t);
643bool vm_page_remove_xbusy(vm_page_t);
644void vm_page_replace(vm_page_t mnew, vm_object_t object,
645 vm_pindex_t pindex, vm_page_t mold);
646int vm_page_sbusied(vm_page_t m);
647vm_page_bits_t vm_page_set_dirty(vm_page_t m);
648void vm_page_set_valid_range(vm_page_t m, int base, int size);
649vm_offset_t vm_page_startup(vm_offset_t vaddr);
650void vm_page_sunbusy(vm_page_t m);
651bool vm_page_try_remove_all(vm_page_t m);
652bool vm_page_try_remove_write(vm_page_t m);
653int vm_page_trysbusy(vm_page_t m);
654int vm_page_tryxbusy(vm_page_t m);
655void vm_page_unhold_pages(vm_page_t *ma, int count);
656void vm_page_unswappable(vm_page_t m);
657void vm_page_unwire(vm_page_t m, uint8_t queue);
658bool vm_page_unwire_noq(vm_page_t m);
659void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
660void vm_page_wire(vm_page_t);
661bool vm_page_wire_mapped(vm_page_t m);
662void vm_page_xunbusy_hard(vm_page_t m);
663void vm_page_xunbusy_hard_unchecked(vm_page_t m);
664void vm_page_set_validclean (vm_page_t, int, int);
665void vm_page_clear_dirty(vm_page_t, int, int);
666void vm_page_set_invalid(vm_page_t, int, int);
667void vm_page_valid(vm_page_t m);
668int vm_page_is_valid(vm_page_t, int, int);
669void vm_page_test_dirty(vm_page_t);
670vm_page_bits_t vm_page_bits(int base, int size);
671void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
672
673void vm_page_dirty_KBI(vm_page_t m);
674
675#define vm_page_busy_fetch(m) atomic_load_int(&(m)->busy_lock)
676
677#define vm_page_assert_busied(m) \
678 KASSERT(vm_page_busied(m), \
679 ("vm_page_assert_busied: page %p not busy @ %s:%d", \
680 (m), __FILE__, __LINE__))
681
682#define vm_page_assert_sbusied(m) \
683 KASSERT(vm_page_sbusied(m), \
684 ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
685 (m), __FILE__, __LINE__))
686
687#define vm_page_assert_unbusied(m) \
688 KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) != \
689 VPB_CURTHREAD_EXCLUSIVE, \
690 ("vm_page_assert_unbusied: page %p busy_lock %#x owned" \
691 " by me (%p) @ %s:%d", \
692 (m), (m)->busy_lock, curthread, __FILE__, __LINE__)); \
693
694#define vm_page_assert_xbusied_unchecked(m) do { \
695 KASSERT(vm_page_xbusied(m), \
696 ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
697 (m), __FILE__, __LINE__)); \
698} while (0)
699#define vm_page_assert_xbusied(m) do { \
700 vm_page_assert_xbusied_unchecked(m); \
701 KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) == \
702 VPB_CURTHREAD_EXCLUSIVE, \
703 ("vm_page_assert_xbusied: page %p busy_lock %#x not owned" \
704 " by me (%p) @ %s:%d", \
705 (m), (m)->busy_lock, curthread, __FILE__, __LINE__)); \
706} while (0)
707
708#define vm_page_busied(m) \
709 (vm_page_busy_fetch(m) != VPB_UNBUSIED)
710
711#define vm_page_xbusied(m) \
712 ((vm_page_busy_fetch(m) & VPB_SINGLE_EXCLUSIVE) != 0)
713
714#define vm_page_busy_freed(m) \
715 (vm_page_busy_fetch(m) == VPB_FREED)
716
717/* Note: page m's lock must not be owned by the caller. */
718#define vm_page_xunbusy(m) do { \
719 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \
720 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \
721 vm_page_xunbusy_hard(m); \
722} while (0)
723#define vm_page_xunbusy_unchecked(m) do { \
724 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \
725 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \
726 vm_page_xunbusy_hard_unchecked(m); \
727} while (0)
728
729#ifdef INVARIANTS
730void vm_page_object_busy_assert(vm_page_t m);
731#define VM_PAGE_OBJECT_BUSY_ASSERT(m) vm_page_object_busy_assert(m)
732void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits);
733#define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) \
734 vm_page_assert_pga_writeable(m, bits)
735/*
736 * Claim ownership of a page's xbusy state. In non-INVARIANTS kernels this
737 * operation is a no-op since ownership is not tracked. In particular
738 * this macro does not provide any synchronization with the previous owner.
739 */
740#define vm_page_xbusy_claim(m) do { \
741 u_int _busy_lock; \
742 \
743 vm_page_assert_xbusied_unchecked((m)); \
744 do { \
745 _busy_lock = vm_page_busy_fetch(m); \
746 } while (!atomic_cmpset_int(&(m)->busy_lock, _busy_lock, \
747 (_busy_lock & VPB_BIT_FLAGMASK) | VPB_CURTHREAD_EXCLUSIVE)); \
748} while (0)
749#else
750#define VM_PAGE_OBJECT_BUSY_ASSERT(m) (void)0
751#define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) (void)0
752#define vm_page_xbusy_claim(m)
753#endif
754
755#if BYTE_ORDER == BIG_ENDIAN
756#define VM_PAGE_AFLAG_SHIFT 16
757#else
758#define VM_PAGE_AFLAG_SHIFT 0
759#endif
760
761/*
762 * Load a snapshot of a page's 32-bit atomic state.
763 */
764static inline vm_page_astate_t
765vm_page_astate_load(vm_page_t m)
766{
767 vm_page_astate_t a;
768
769 a._bits = atomic_load_32(&m->a._bits);
770 return (a);
771}
772
773/*
774 * Atomically compare and set a page's atomic state.
775 */
776static inline bool
777vm_page_astate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new)
778{
779
780 KASSERT(new.queue == PQ_INACTIVE || (new.flags & PGA_REQUEUE_HEAD) == 0,
781 ("%s: invalid head requeue request for page %p", __func__, m));
782 KASSERT((new.flags & PGA_ENQUEUED) == 0 || new.queue != PQ_NONE,
783 ("%s: setting PGA_ENQUEUED with PQ_NONE in page %p", __func__, m));
784 KASSERT(new._bits != old->_bits,
785 ("%s: bits are unchanged", __func__));
786
787 return (atomic_fcmpset_32(&m->a._bits, &old->_bits, new._bits) != 0);
788}
789
790/*
791 * Clear the given bits in the specified page.
792 */
793static inline void
794vm_page_aflag_clear(vm_page_t m, uint16_t bits)
795{
796 uint32_t *addr, val;
797
798 /*
799 * Access the whole 32-bit word containing the aflags field with an
800 * atomic update. Parallel non-atomic updates to the other fields
801 * within this word are handled properly by the atomic update.
802 */
803 addr = (void *)&m->a;
804 val = bits << VM_PAGE_AFLAG_SHIFT;
805 atomic_clear_32(addr, val);
806}
807
808/*
809 * Set the given bits in the specified page.
810 */
811static inline void
812vm_page_aflag_set(vm_page_t m, uint16_t bits)
813{
814 uint32_t *addr, val;
815
816 VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
817
818 /*
819 * Access the whole 32-bit word containing the aflags field with an
820 * atomic update. Parallel non-atomic updates to the other fields
821 * within this word are handled properly by the atomic update.
822 */
823 addr = (void *)&m->a;
824 val = bits << VM_PAGE_AFLAG_SHIFT;
825 atomic_set_32(addr, val);
826}
827
828/*
829 * vm_page_dirty:
830 *
831 * Set all bits in the page's dirty field.
832 *
833 * The object containing the specified page must be locked if the
834 * call is made from the machine-independent layer.
835 *
836 * See vm_page_clear_dirty_mask().
837 */
838static __inline void
839vm_page_dirty(vm_page_t m)
840{
841
842 /* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
843#if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
844 vm_page_dirty_KBI(m);
845#else
846 m->dirty = VM_PAGE_BITS_ALL;
847#endif
848}
849
850/*
851 * vm_page_undirty:
852 *
853 * Set page to not be dirty. Note: does not clear pmap modify bits
854 */
855static __inline void
856vm_page_undirty(vm_page_t m)
857{
858
859 VM_PAGE_OBJECT_BUSY_ASSERT(m);
860 m->dirty = 0;
861}
862
863static inline uint8_t
864_vm_page_queue(vm_page_astate_t as)
865{
866
867 if ((as.flags & PGA_DEQUEUE) != 0)
868 return (PQ_NONE);
869 return (as.queue);
870}
871
872/*
873 * vm_page_queue:
874 *
875 * Return the index of the queue containing m.
876 */
877static inline uint8_t
878vm_page_queue(vm_page_t m)
879{
880
881 return (_vm_page_queue(vm_page_astate_load(m)));
882}
883
884static inline bool
885vm_page_active(vm_page_t m)
886{
887
888 return (vm_page_queue(m) == PQ_ACTIVE);
889}
890
891static inline bool
892vm_page_inactive(vm_page_t m)
893{
894
895 return (vm_page_queue(m) == PQ_INACTIVE);
896}
897
898static inline bool
899vm_page_in_laundry(vm_page_t m)
900{
901 uint8_t queue;
902
903 queue = vm_page_queue(m);
904 return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
905}
906
907static inline void
908vm_page_clearref(vm_page_t m)
909{
910 u_int r;
911
912 r = m->ref_count;
913 while (atomic_fcmpset_int(&m->ref_count, &r, r & (VPRC_BLOCKED |
914 VPRC_OBJREF)) == 0)
915 ;
916}
917
918/*
919 * vm_page_drop:
920 *
921 * Release a reference to a page and return the old reference count.
922 */
923static inline u_int
924vm_page_drop(vm_page_t m, u_int val)
925{
926 u_int old;
927
928 /*
929 * Synchronize with vm_page_free_prep(): ensure that all updates to the
930 * page structure are visible before it is freed.
931 */
932 atomic_thread_fence_rel();
933 old = atomic_fetchadd_int(&m->ref_count, -val);
934 KASSERT(old != VPRC_BLOCKED,
935 ("vm_page_drop: page %p has an invalid refcount value", m));
936 return (old);
937}
938
939/*
940 * vm_page_wired:
941 *
942 * Perform a racy check to determine whether a reference prevents the page
943 * from being reclaimable. If the page's object is locked, and the page is
944 * unmapped and exclusively busied by the current thread, no new wirings
945 * may be created.
946 */
947static inline bool
948vm_page_wired(vm_page_t m)
949{
950
951 return (VPRC_WIRE_COUNT(m->ref_count) > 0);
952}
953
954static inline bool
955vm_page_all_valid(vm_page_t m)
956{
957
958 return (m->valid == VM_PAGE_BITS_ALL);
959}
960
961static inline bool
962vm_page_any_valid(vm_page_t m)
963{
964
965 return (m->valid != 0);
966}
967
968static inline bool
969vm_page_none_valid(vm_page_t m)
970{
971
972 return (m->valid == 0);
973}
974
975static inline int
976vm_page_domain(vm_page_t m __numa_used)
977{
978#ifdef NUMA
979 int domn, segind;
980
981 segind = m->segind;
982 KASSERT(segind < vm_phys_nsegs, ("segind %d m %p", segind, m));
983 domn = vm_phys_segs[segind].domain;
984 KASSERT(domn >= 0 && domn < vm_ndomains, ("domain %d m %p", domn, m));
985 return (domn);
986#else
987 return (0);
988#endif
989}
990
991#endif /* _KERNEL */
992#endif /* !_VM_PAGE_ */