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
63 /*
64 * Resident memory system definitions.
65 */
66
67 #ifndef _VM_PAGE_
68 #define _VM_PAGE_
69
70 #include <vm/pmap.h>
71 #include <vm/_vm_phys.h>
72
73 /*
74 * Management of resident (logical) pages.
75 *
76 * A small structure is kept for each resident
77 * page, indexed by page number. Each structure
78 * is an element of several collections:
79 *
80 * A radix tree used to quickly
81 * perform object/offset lookups
82 *
83 * A list of all pages for a given object,
84 * so they can be quickly deactivated at
85 * time of deallocation.
86 *
87 * An ordered list of pages due for pageout.
88 *
89 * In addition, the structure contains the object
90 * and offset to which this page belongs (for pageout),
91 * and sundry status bits.
92 *
93 * In general, operations on this structure's mutable fields are
94 * synchronized using either one of or a combination of locks. If a
95 * field is annotated with two of these locks then holding either is
96 * sufficient for read access but both are required for write access.
97 * The queue lock for a page depends on the value of its queue field and is
98 * described in detail below.
99 *
100 * The following annotations are possible:
101 * (A) the field must be accessed using atomic(9) and may require
102 * additional synchronization.
103 * (B) the page busy lock.
104 * (C) the field is immutable.
105 * (F) the per-domain lock for the free queues.
106 * (M) Machine dependent, defined by pmap layer.
107 * (O) the object that the page belongs to.
108 * (Q) the page's queue lock.
109 *
110 * The busy lock is an embedded reader-writer lock that protects the
111 * page's contents and identity (i.e., its <object, pindex> tuple) as
112 * well as certain valid/dirty modifications. To avoid bloating the
113 * the page structure, the busy lock lacks some of the features available
114 * the kernel's general-purpose synchronization primitives. As a result,
115 * busy lock ordering rules are not verified, lock recursion is not
116 * detected, and an attempt to xbusy a busy page or sbusy an xbusy page
117 * results will trigger a panic rather than causing the thread to block.
118 * vm_page_sleep_if_busy() can be used to sleep until the page's busy
119 * state changes, after which the caller must re-lookup the page and
120 * re-evaluate its state. vm_page_busy_acquire() will block until
121 * the lock is acquired.
122 *
123 * The valid field is protected by the page busy lock (B) and object
124 * lock (O). Transitions from invalid to valid are generally done
125 * via I/O or zero filling and do not require the object lock.
126 * These must be protected with the busy lock to prevent page-in or
127 * creation races. Page invalidation generally happens as a result
128 * of truncate or msync. When invalidated, pages must not be present
129 * in pmap and must hold the object lock to prevent concurrent
130 * speculative read-only mappings that do not require busy. I/O
131 * routines may check for validity without a lock if they are prepared
132 * to handle invalidation races with higher level locks (vnode) or are
133 * unconcerned with races so long as they hold a reference to prevent
134 * recycling. When a valid bit is set while holding a shared busy
135 * lock (A) atomic operations are used to protect against concurrent
136 * modification.
137 *
138 * In contrast, the synchronization of accesses to the page's
139 * dirty field is a mix of machine dependent (M) and busy (B). In
140 * the machine-independent layer, the page busy must be held to
141 * operate on the field. However, the pmap layer is permitted to
142 * set all bits within the field without holding that lock. If the
143 * underlying architecture does not support atomic read-modify-write
144 * operations on the field's type, then the machine-independent
145 * layer uses a 32-bit atomic on the aligned 32-bit word that
146 * contains the dirty field. In the machine-independent layer,
147 * the implementation of read-modify-write operations on the
148 * field is encapsulated in vm_page_clear_dirty_mask(). An
149 * exclusive busy lock combined with pmap_remove_{write/all}() is the
150 * only way to ensure a page can not become dirty. I/O generally
151 * removes the page from pmap to ensure exclusive access and atomic
152 * writes.
153 *
154 * The ref_count field tracks references to the page. References that
155 * prevent the page from being reclaimable are called wirings and are
156 * counted in the low bits of ref_count. The containing object's
157 * reference, if one exists, is counted using the VPRC_OBJREF bit in the
158 * ref_count field. Additionally, the VPRC_BLOCKED bit is used to
159 * atomically check for wirings and prevent new wirings via
160 * pmap_extract_and_hold(). When a page belongs to an object, it may be
161 * wired only when the object is locked, or the page is busy, or by
162 * pmap_extract_and_hold(). As a result, if the object is locked and the
163 * page is not busy (or is exclusively busied by the current thread), and
164 * the page is unmapped, its wire count will not increase. The ref_count
165 * field is updated using atomic operations in most cases, except when it
166 * is known that no other references to the page exist, such as in the page
167 * allocator. A page may be present in the page queues, or even actively
168 * scanned by the page daemon, without an explicitly counted referenced.
169 * The page daemon must therefore handle the possibility of a concurrent
170 * free of the page.
171 *
172 * The queue state of a page consists of the queue and act_count fields of
173 * its atomically updated state, and the subset of atomic flags specified
174 * by PGA_QUEUE_STATE_MASK. The queue field contains the page's page queue
175 * index, or PQ_NONE if it does not belong to a page queue. To modify the
176 * queue field, the page queue lock corresponding to the old value must be
177 * held, unless that value is PQ_NONE, in which case the queue index must
178 * be updated using an atomic RMW operation. There is one exception to
179 * this rule: the page daemon may transition the queue field from
180 * PQ_INACTIVE to PQ_NONE immediately prior to freeing the page during an
181 * inactive queue scan. At that point the page is already dequeued and no
182 * other references to that vm_page structure can exist. The PGA_ENQUEUED
183 * flag, when set, indicates that the page structure is physically inserted
184 * into the queue corresponding to the page's queue index, and may only be
185 * set or cleared with the corresponding page queue lock held.
186 *
187 * To avoid contention on page queue locks, page queue operations (enqueue,
188 * dequeue, requeue) are batched using fixed-size per-CPU queues. A
189 * deferred operation is requested by setting one of the flags in
190 * PGA_QUEUE_OP_MASK and inserting an entry into a batch queue. When a
191 * queue is full, an attempt to insert a new entry will lock the page
192 * queues and trigger processing of the pending entries. The
193 * type-stability of vm_page structures is crucial to this scheme since the
194 * processing of entries in a given batch queue may be deferred
195 * indefinitely. In particular, a page may be freed with pending batch
196 * queue entries. The page queue operation flags must be set using atomic
197 * RWM operations.
198 */
199
200 #if PAGE_SIZE == 4096
201 #define VM_PAGE_BITS_ALL 0xffu
202 typedef uint8_t vm_page_bits_t;
203 #elif PAGE_SIZE == 8192
204 #define VM_PAGE_BITS_ALL 0xffffu
205 typedef uint16_t vm_page_bits_t;
206 #elif PAGE_SIZE == 16384
207 #define VM_PAGE_BITS_ALL 0xffffffffu
208 typedef uint32_t vm_page_bits_t;
209 #elif PAGE_SIZE == 32768
210 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
211 typedef uint64_t vm_page_bits_t;
212 #endif
213
214 typedef union vm_page_astate {
215 struct {
216 uint16_t flags;
217 uint8_t queue;
218 uint8_t act_count;
219 };
220 uint32_t _bits;
221 } vm_page_astate_t;
222
223 struct vm_page {
224 union {
225 TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
226 struct {
227 SLIST_ENTRY(vm_page) ss; /* private slists */
228 } s;
229 struct {
230 u_long p;
231 u_long v;
232 } memguard;
233 struct {
234 void *slab;
235 void *zone;
236 } uma;
237 } plinks;
238 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */
239 vm_object_t object; /* which object am I in (O) */
240 vm_pindex_t pindex; /* offset into object (O,P) */
241 vm_paddr_t phys_addr; /* physical address of page (C) */
242 struct md_page md; /* machine dependent stuff */
243 u_int ref_count; /* page references (A) */
244 u_int busy_lock; /* busy owners lock (A) */
245 union vm_page_astate a; /* state accessed atomically (A) */
246 uint8_t order; /* index of the buddy queue (F) */
247 uint8_t pool; /* vm_phys freepool index (F) */
248 uint8_t flags; /* page PG_* flags (P) */
249 uint8_t oflags; /* page VPO_* flags (O) */
250 int8_t psind; /* pagesizes[] index (O) */
251 int8_t segind; /* vm_phys segment index (C) */
252 /* NOTE that these must support one bit per DEV_BSIZE in a page */
253 /* so, on normal X86 kernels, they must be at least 8 bits wide */
254 vm_page_bits_t valid; /* valid DEV_BSIZE chunk map (O,B) */
255 vm_page_bits_t dirty; /* dirty DEV_BSIZE chunk map (M,B) */
256 };
257
258 /*
259 * Special bits used in the ref_count field.
260 *
261 * ref_count is normally used to count wirings that prevent the page from being
262 * reclaimed, but also supports several special types of references that do not
263 * prevent reclamation. Accesses to the ref_count field must be atomic unless
264 * the page is unallocated.
265 *
266 * VPRC_OBJREF is the reference held by the containing object. It can set or
267 * cleared only when the corresponding object's write lock is held.
268 *
269 * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while
270 * attempting to tear down all mappings of a given page. The page busy lock and
271 * object write lock must both be held in order to set or clear this bit.
272 */
273 #define VPRC_BLOCKED 0x40000000u /* mappings are being removed */
274 #define VPRC_OBJREF 0x80000000u /* object reference, cleared with (O) */
275 #define VPRC_WIRE_COUNT(c) ((c) & ~(VPRC_BLOCKED | VPRC_OBJREF))
276 #define VPRC_WIRE_COUNT_MAX (~(VPRC_BLOCKED | VPRC_OBJREF))
277
278 /*
279 * Page flags stored in oflags:
280 *
281 * Access to these page flags is synchronized by the lock on the object
282 * containing the page (O).
283 *
284 * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
285 * indicates that the page is not under PV management but
286 * otherwise should be treated as a normal page. Pages not
287 * under PV management cannot be paged out via the
288 * object/vm_page_t because there is no knowledge of their pte
289 * mappings, and such pages are also not on any PQ queue.
290 *
291 */
292 #define VPO_KMEM_EXEC 0x01 /* kmem mapping allows execution */
293 #define VPO_SWAPSLEEP 0x02 /* waiting for swap to finish */
294 #define VPO_UNMANAGED 0x04 /* no PV management for page */
295 #define VPO_SWAPINPROG 0x08 /* swap I/O in progress on page */
296
297 /*
298 * Busy page implementation details.
299 * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
300 * even if the support for owner identity is removed because of size
301 * constraints. Checks on lock recursion are then not possible, while the
302 * lock assertions effectiveness is someway reduced.
303 */
304 #define VPB_BIT_SHARED 0x01
305 #define VPB_BIT_EXCLUSIVE 0x02
306 #define VPB_BIT_WAITERS 0x04
307 #define VPB_BIT_FLAGMASK \
308 (VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
309
310 #define VPB_SHARERS_SHIFT 3
311 #define VPB_SHARERS(x) \
312 (((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
313 #define VPB_SHARERS_WORD(x) ((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
314 #define VPB_ONE_SHARER (1 << VPB_SHARERS_SHIFT)
315
316 #define VPB_SINGLE_EXCLUSIVE VPB_BIT_EXCLUSIVE
317 #ifdef INVARIANTS
318 #define VPB_CURTHREAD_EXCLUSIVE \
319 (VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK))
320 #else
321 #define VPB_CURTHREAD_EXCLUSIVE VPB_SINGLE_EXCLUSIVE
322 #endif
323
324 #define VPB_UNBUSIED VPB_SHARERS_WORD(0)
325
326 /* Freed lock blocks both shared and exclusive. */
327 #define VPB_FREED (0xffffffff - VPB_BIT_SHARED)
328
329 #define PQ_NONE 255
330 #define PQ_INACTIVE 0
331 #define PQ_ACTIVE 1
332 #define PQ_LAUNDRY 2
333 #define PQ_UNSWAPPABLE 3
334 #define PQ_COUNT 4
335
336 #ifndef VM_PAGE_HAVE_PGLIST
337 TAILQ_HEAD(pglist, vm_page);
338 #define VM_PAGE_HAVE_PGLIST
339 #endif
340 SLIST_HEAD(spglist, vm_page);
341
342 #ifdef _KERNEL
343 extern vm_page_t bogus_page;
344 #endif /* _KERNEL */
345
346 extern struct mtx_padalign pa_lock[];
347
348 #if defined(__arm__)
349 #define PDRSHIFT PDR_SHIFT
350 #elif !defined(PDRSHIFT)
351 #define PDRSHIFT 21
352 #endif
353
354 #define pa_index(pa) ((pa) >> PDRSHIFT)
355 #define PA_LOCKPTR(pa) ((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
356 #define PA_LOCKOBJPTR(pa) ((struct lock_object *)PA_LOCKPTR((pa)))
357 #define PA_LOCK(pa) mtx_lock(PA_LOCKPTR(pa))
358 #define PA_TRYLOCK(pa) mtx_trylock(PA_LOCKPTR(pa))
359 #define PA_UNLOCK(pa) mtx_unlock(PA_LOCKPTR(pa))
360 #define PA_UNLOCK_COND(pa) \
361 do { \
362 if ((pa) != 0) { \
363 PA_UNLOCK((pa)); \
364 (pa) = 0; \
365 } \
366 } while (0)
367
368 #define PA_LOCK_ASSERT(pa, a) mtx_assert(PA_LOCKPTR(pa), (a))
369
370 #if defined(KLD_MODULE) && !defined(KLD_TIED)
371 #define vm_page_lock(m) vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
372 #define vm_page_unlock(m) vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
373 #define vm_page_trylock(m) vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
374 #else /* !KLD_MODULE */
375 #define vm_page_lockptr(m) (PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
376 #define vm_page_lock(m) mtx_lock(vm_page_lockptr((m)))
377 #define vm_page_unlock(m) mtx_unlock(vm_page_lockptr((m)))
378 #define vm_page_trylock(m) mtx_trylock(vm_page_lockptr((m)))
379 #endif
380 #if defined(INVARIANTS)
381 #define vm_page_assert_locked(m) \
382 vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
383 #define vm_page_lock_assert(m, a) \
384 vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
385 #else
386 #define vm_page_assert_locked(m)
387 #define vm_page_lock_assert(m, a)
388 #endif
389
390 /*
391 * The vm_page's aflags are updated using atomic operations. To set or clear
392 * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
393 * must be used. Neither these flags nor these functions are part of the KBI.
394 *
395 * PGA_REFERENCED may be cleared only if the page is locked. It is set by
396 * both the MI and MD VM layers. However, kernel loadable modules should not
397 * directly set this flag. They should call vm_page_reference() instead.
398 *
399 * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
400 * When it does so, the object must be locked, or the page must be
401 * exclusive busied. The MI VM layer must never access this flag
402 * directly. Instead, it should call pmap_page_is_write_mapped().
403 *
404 * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
405 * at least one executable mapping. It is not consumed by the MI VM layer.
406 *
407 * PGA_NOSYNC must be set and cleared with the page busy lock held.
408 *
409 * PGA_ENQUEUED is set and cleared when a page is inserted into or removed
410 * from a page queue, respectively. It determines whether the plinks.q field
411 * of the page is valid. To set or clear this flag, page's "queue" field must
412 * be a valid queue index, and the corresponding page queue lock must be held.
413 *
414 * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
415 * queue, and cleared when the dequeue request is processed. A page may
416 * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
417 * is requested after the page is scheduled to be enqueued but before it is
418 * actually inserted into the page queue.
419 *
420 * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
421 * in its page queue.
422 *
423 * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
424 * the inactive queue, thus bypassing LRU.
425 *
426 * The PGA_DEQUEUE, PGA_REQUEUE and PGA_REQUEUE_HEAD flags must be set using an
427 * atomic RMW operation to ensure that the "queue" field is a valid queue index,
428 * and the corresponding page queue lock must be held when clearing any of the
429 * flags.
430 *
431 * PGA_SWAP_FREE is used to defer freeing swap space to the pageout daemon
432 * when the context that dirties the page does not have the object write lock
433 * held.
434 */
435 #define PGA_WRITEABLE 0x0001 /* page may be mapped writeable */
436 #define PGA_REFERENCED 0x0002 /* page has been referenced */
437 #define PGA_EXECUTABLE 0x0004 /* page may be mapped executable */
438 #define PGA_ENQUEUED 0x0008 /* page is enqueued in a page queue */
439 #define PGA_DEQUEUE 0x0010 /* page is due to be dequeued */
440 #define PGA_REQUEUE 0x0020 /* page is due to be requeued */
441 #define PGA_REQUEUE_HEAD 0x0040 /* page requeue should bypass LRU */
442 #define PGA_NOSYNC 0x0080 /* do not collect for syncer */
443 #define PGA_SWAP_FREE 0x0100 /* page with swap space was dirtied */
444 #define PGA_SWAP_SPACE 0x0200 /* page has allocated swap space */
445
446 #define PGA_QUEUE_OP_MASK (PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD)
447 #define PGA_QUEUE_STATE_MASK (PGA_ENQUEUED | PGA_QUEUE_OP_MASK)
448
449 /*
450 * Page flags. Updates to these flags are not synchronized, and thus they must
451 * be set during page allocation or free to avoid races.
452 *
453 * The PG_PCPU_CACHE flag is set at allocation time if the page was
454 * allocated from a per-CPU cache. It is cleared the next time that the
455 * page is allocated from the physical memory allocator.
456 */
457 #define PG_PCPU_CACHE 0x01 /* was allocated from per-CPU caches */
458 #define PG_FICTITIOUS 0x02 /* physical page doesn't exist */
459 #define PG_ZERO 0x04 /* page is zeroed */
460 #define PG_MARKER 0x08 /* special queue marker page */
461 #define PG_NODUMP 0x10 /* don't include this page in a dump */
462
463 /*
464 * Misc constants.
465 */
466 #define ACT_DECLINE 1
467 #define ACT_ADVANCE 3
468 #define ACT_INIT 5
469 #define ACT_MAX 64
470
471 #ifdef _KERNEL
472
473 #include <sys/kassert.h>
474 #include <machine/atomic.h>
475
476 /*
477 * Each pageable resident page falls into one of five lists:
478 *
479 * free
480 * Available for allocation now.
481 *
482 * inactive
483 * Low activity, candidates for reclamation.
484 * This list is approximately LRU ordered.
485 *
486 * laundry
487 * This is the list of pages that should be
488 * paged out next.
489 *
490 * unswappable
491 * Dirty anonymous pages that cannot be paged
492 * out because no swap device is configured.
493 *
494 * active
495 * Pages that are "active", i.e., they have been
496 * recently referenced.
497 *
498 */
499
500 extern vm_page_t vm_page_array; /* First resident page in table */
501 extern long vm_page_array_size; /* number of vm_page_t's */
502 extern long first_page; /* first physical page number */
503
504 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr)
505
506 /*
507 * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
508 * page to which the given physical address belongs. The correct vm_page_t
509 * object is returned for addresses that are not page-aligned.
510 */
511 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
512
513 /*
514 * Page allocation parameters for vm_page for the functions
515 * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
516 * vm_page_alloc_freelist(). Some functions support only a subset
517 * of the flags, and ignore others, see the flags legend.
518 *
519 * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
520 * and the vm_page_grab*() functions. See these functions for details.
521 *
522 * Bits 0 - 1 define class.
523 * Bits 2 - 15 dedicated for flags.
524 * Legend:
525 * (a) - vm_page_alloc() supports the flag.
526 * (c) - vm_page_alloc_contig() supports the flag.
527 * (g) - vm_page_grab() supports the flag.
528 * (n) - vm_page_alloc_noobj() and vm_page_alloc_freelist() support the flag.
529 * (p) - vm_page_grab_pages() supports the flag.
530 * Bits above 15 define the count of additional pages that the caller
531 * intends to allocate.
532 */
533 #define VM_ALLOC_NORMAL 0
534 #define VM_ALLOC_INTERRUPT 1
535 #define VM_ALLOC_SYSTEM 2
536 #define VM_ALLOC_CLASS_MASK 3
537 #define VM_ALLOC_WAITOK 0x0008 /* (acn) Sleep and retry */
538 #define VM_ALLOC_WAITFAIL 0x0010 /* (acn) Sleep and return error */
539 #define VM_ALLOC_WIRED 0x0020 /* (acgnp) Allocate a wired page */
540 #define VM_ALLOC_ZERO 0x0040 /* (acgnp) Allocate a zeroed page */
541 #define VM_ALLOC_NORECLAIM 0x0080 /* (c) Do not reclaim after failure */
542 #define VM_ALLOC_NOOBJ 0x0100 /* (acg) No associated object */
543 #define VM_ALLOC_NOBUSY 0x0200 /* (acgp) Do not excl busy the page */
544 #define VM_ALLOC_NOCREAT 0x0400 /* (gp) Don't create a page */
545 #define VM_ALLOC_IGN_SBUSY 0x1000 /* (gp) Ignore shared busy flag */
546 #define VM_ALLOC_NODUMP 0x2000 /* (ag) don't include in dump */
547 #define VM_ALLOC_SBUSY 0x4000 /* (acgp) Shared busy the page */
548 #define VM_ALLOC_NOWAIT 0x8000 /* (acgnp) Do not sleep */
549 #define VM_ALLOC_COUNT_SHIFT 16
550 #define VM_ALLOC_COUNT(count) ((count) << VM_ALLOC_COUNT_SHIFT)
551
552 #ifdef M_NOWAIT
553 static inline int
malloc2vm_flags(int malloc_flags)554 malloc2vm_flags(int malloc_flags)
555 {
556 int pflags;
557
558 KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
559 (malloc_flags & M_NOWAIT) != 0,
560 ("M_USE_RESERVE requires M_NOWAIT"));
561 pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
562 VM_ALLOC_SYSTEM;
563 if ((malloc_flags & M_ZERO) != 0)
564 pflags |= VM_ALLOC_ZERO;
565 if ((malloc_flags & M_NODUMP) != 0)
566 pflags |= VM_ALLOC_NODUMP;
567 if ((malloc_flags & M_NOWAIT))
568 pflags |= VM_ALLOC_NOWAIT;
569 if ((malloc_flags & M_WAITOK))
570 pflags |= VM_ALLOC_WAITOK;
571 if ((malloc_flags & M_NORECLAIM))
572 pflags |= VM_ALLOC_NORECLAIM;
573 return (pflags);
574 }
575 #endif
576
577 /*
578 * Predicates supported by vm_page_ps_test():
579 *
580 * PS_ALL_DIRTY is true only if the entire (super)page is dirty.
581 * However, it can be spuriously false when the (super)page has become
582 * dirty in the pmap but that information has not been propagated to the
583 * machine-independent layer.
584 */
585 #define PS_ALL_DIRTY 0x1
586 #define PS_ALL_VALID 0x2
587 #define PS_NONE_BUSY 0x4
588
589 bool vm_page_busy_acquire(vm_page_t m, int allocflags);
590 void vm_page_busy_downgrade(vm_page_t m);
591 int vm_page_busy_tryupgrade(vm_page_t m);
592 bool vm_page_busy_sleep(vm_page_t m, const char *msg, int allocflags);
593 void vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m,
594 vm_pindex_t pindex, const char *wmesg, int allocflags);
595 void vm_page_free(vm_page_t m);
596 void vm_page_free_zero(vm_page_t m);
597
598 void vm_page_activate (vm_page_t);
599 void vm_page_advise(vm_page_t m, int advice);
600 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
601 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int);
602 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t);
603 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
604 vm_page_t);
605 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
606 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
607 vm_paddr_t boundary, vm_memattr_t memattr);
608 vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
609 vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
610 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
611 vm_memattr_t memattr);
612 vm_page_t vm_page_alloc_freelist(int, int);
613 vm_page_t vm_page_alloc_freelist_domain(int, int, int);
614 vm_page_t vm_page_alloc_noobj(int);
615 vm_page_t vm_page_alloc_noobj_domain(int, int);
616 vm_page_t vm_page_alloc_noobj_contig(int req, u_long npages, vm_paddr_t low,
617 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
618 vm_memattr_t memattr);
619 vm_page_t vm_page_alloc_noobj_contig_domain(int domain, int req, u_long npages,
620 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
621 vm_memattr_t memattr);
622 void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set);
623 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
624 vm_page_t vm_page_grab(vm_object_t, vm_pindex_t, int);
625 vm_page_t vm_page_grab_unlocked(vm_object_t, vm_pindex_t, int);
626 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
627 vm_page_t *ma, int count);
628 int vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex,
629 int allocflags, vm_page_t *ma, int count);
630 int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex,
631 int allocflags);
632 int vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object,
633 vm_pindex_t pindex, int allocflags);
634 void vm_page_deactivate(vm_page_t);
635 void vm_page_deactivate_noreuse(vm_page_t);
636 void vm_page_dequeue(vm_page_t m);
637 void vm_page_dequeue_deferred(vm_page_t m);
638 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
639 void vm_page_free_invalid(vm_page_t);
640 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
641 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
642 void vm_page_init_marker(vm_page_t marker, int queue, uint16_t aflags);
643 void vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind);
644 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
645 void vm_page_invalid(vm_page_t m);
646 void vm_page_launder(vm_page_t m);
647 vm_page_t vm_page_lookup(vm_object_t, vm_pindex_t);
648 vm_page_t vm_page_lookup_unlocked(vm_object_t, vm_pindex_t);
649 vm_page_t vm_page_next(vm_page_t m);
650 void vm_page_pqbatch_drain(void);
651 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
652 bool vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old,
653 vm_page_astate_t new);
654 vm_page_t vm_page_prev(vm_page_t m);
655 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m);
656 void vm_page_putfake(vm_page_t m);
657 void vm_page_readahead_finish(vm_page_t m);
658 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
659 vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
660 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
661 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
662 void vm_page_reference(vm_page_t m);
663 #define VPR_TRYFREE 0x01
664 #define VPR_NOREUSE 0x02
665 void vm_page_release(vm_page_t m, int flags);
666 void vm_page_release_locked(vm_page_t m, int flags);
667 vm_page_t vm_page_relookup(vm_object_t, vm_pindex_t);
668 bool vm_page_remove(vm_page_t);
669 bool vm_page_remove_xbusy(vm_page_t);
670 int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t);
671 void vm_page_replace(vm_page_t mnew, vm_object_t object,
672 vm_pindex_t pindex, vm_page_t mold);
673 int vm_page_sbusied(vm_page_t m);
674 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start,
675 vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options);
676 vm_page_bits_t vm_page_set_dirty(vm_page_t m);
677 void vm_page_set_valid_range(vm_page_t m, int base, int size);
678 vm_offset_t vm_page_startup(vm_offset_t vaddr);
679 void vm_page_sunbusy(vm_page_t m);
680 bool vm_page_try_remove_all(vm_page_t m);
681 bool vm_page_try_remove_write(vm_page_t m);
682 int vm_page_trysbusy(vm_page_t m);
683 int vm_page_tryxbusy(vm_page_t m);
684 void vm_page_unhold_pages(vm_page_t *ma, int count);
685 void vm_page_unswappable(vm_page_t m);
686 void vm_page_unwire(vm_page_t m, uint8_t queue);
687 bool vm_page_unwire_noq(vm_page_t m);
688 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
689 void vm_page_wire(vm_page_t);
690 bool vm_page_wire_mapped(vm_page_t m);
691 void vm_page_xunbusy_hard(vm_page_t m);
692 void vm_page_xunbusy_hard_unchecked(vm_page_t m);
693 void vm_page_set_validclean (vm_page_t, int, int);
694 void vm_page_clear_dirty(vm_page_t, int, int);
695 void vm_page_set_invalid(vm_page_t, int, int);
696 void vm_page_valid(vm_page_t m);
697 int vm_page_is_valid(vm_page_t, int, int);
698 void vm_page_test_dirty(vm_page_t);
699 vm_page_bits_t vm_page_bits(int base, int size);
700 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
701 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
702
703 void vm_page_dirty_KBI(vm_page_t m);
704 void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
705 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
706 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
707 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
708 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
709 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
710 #endif
711
712 #define vm_page_busy_fetch(m) atomic_load_int(&(m)->busy_lock)
713
714 #define vm_page_assert_busied(m) \
715 KASSERT(vm_page_busied(m), \
716 ("vm_page_assert_busied: page %p not busy @ %s:%d", \
717 (m), __FILE__, __LINE__))
718
719 #define vm_page_assert_sbusied(m) \
720 KASSERT(vm_page_sbusied(m), \
721 ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
722 (m), __FILE__, __LINE__))
723
724 #define vm_page_assert_unbusied(m) \
725 KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) != \
726 VPB_CURTHREAD_EXCLUSIVE, \
727 ("vm_page_assert_xbusied: page %p busy_lock %#x owned" \
728 " by me @ %s:%d", \
729 (m), (m)->busy_lock, __FILE__, __LINE__)); \
730
731 #define vm_page_assert_xbusied_unchecked(m) do { \
732 KASSERT(vm_page_xbusied(m), \
733 ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
734 (m), __FILE__, __LINE__)); \
735 } while (0)
736 #define vm_page_assert_xbusied(m) do { \
737 vm_page_assert_xbusied_unchecked(m); \
738 KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) == \
739 VPB_CURTHREAD_EXCLUSIVE, \
740 ("vm_page_assert_xbusied: page %p busy_lock %#x not owned" \
741 " by me @ %s:%d", \
742 (m), (m)->busy_lock, __FILE__, __LINE__)); \
743 } while (0)
744
745 #define vm_page_busied(m) \
746 (vm_page_busy_fetch(m) != VPB_UNBUSIED)
747
748 #define vm_page_xbusied(m) \
749 ((vm_page_busy_fetch(m) & VPB_SINGLE_EXCLUSIVE) != 0)
750
751 #define vm_page_busy_freed(m) \
752 (vm_page_busy_fetch(m) == VPB_FREED)
753
754 /* Note: page m's lock must not be owned by the caller. */
755 #define vm_page_xunbusy(m) do { \
756 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \
757 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \
758 vm_page_xunbusy_hard(m); \
759 } while (0)
760 #define vm_page_xunbusy_unchecked(m) do { \
761 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \
762 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \
763 vm_page_xunbusy_hard_unchecked(m); \
764 } while (0)
765
766 #ifdef INVARIANTS
767 void vm_page_object_busy_assert(vm_page_t m);
768 #define VM_PAGE_OBJECT_BUSY_ASSERT(m) vm_page_object_busy_assert(m)
769 void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits);
770 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) \
771 vm_page_assert_pga_writeable(m, bits)
772 /*
773 * Claim ownership of a page's xbusy state. In non-INVARIANTS kernels this
774 * operation is a no-op since ownership is not tracked. In particular
775 * this macro does not provide any synchronization with the previous owner.
776 */
777 #define vm_page_xbusy_claim(m) do { \
778 u_int _busy_lock; \
779 \
780 vm_page_assert_xbusied_unchecked((m)); \
781 do { \
782 _busy_lock = vm_page_busy_fetch(m); \
783 } while (!atomic_cmpset_int(&(m)->busy_lock, _busy_lock, \
784 (_busy_lock & VPB_BIT_FLAGMASK) | VPB_CURTHREAD_EXCLUSIVE)); \
785 } while (0)
786 #else
787 #define VM_PAGE_OBJECT_BUSY_ASSERT(m) (void)0
788 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) (void)0
789 #define vm_page_xbusy_claim(m)
790 #endif
791
792 #if BYTE_ORDER == BIG_ENDIAN
793 #define VM_PAGE_AFLAG_SHIFT 16
794 #else
795 #define VM_PAGE_AFLAG_SHIFT 0
796 #endif
797
798 /*
799 * Load a snapshot of a page's 32-bit atomic state.
800 */
801 static inline vm_page_astate_t
vm_page_astate_load(vm_page_t m)802 vm_page_astate_load(vm_page_t m)
803 {
804 vm_page_astate_t a;
805
806 a._bits = atomic_load_32(&m->a._bits);
807 return (a);
808 }
809
810 /*
811 * Atomically compare and set a page's atomic state.
812 */
813 static inline bool
vm_page_astate_fcmpset(vm_page_t m,vm_page_astate_t * old,vm_page_astate_t new)814 vm_page_astate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new)
815 {
816
817 KASSERT(new.queue == PQ_INACTIVE || (new.flags & PGA_REQUEUE_HEAD) == 0,
818 ("%s: invalid head requeue request for page %p", __func__, m));
819 KASSERT((new.flags & PGA_ENQUEUED) == 0 || new.queue != PQ_NONE,
820 ("%s: setting PGA_ENQUEUED with PQ_NONE in page %p", __func__, m));
821 KASSERT(new._bits != old->_bits,
822 ("%s: bits are unchanged", __func__));
823
824 return (atomic_fcmpset_32(&m->a._bits, &old->_bits, new._bits) != 0);
825 }
826
827 /*
828 * Clear the given bits in the specified page.
829 */
830 static inline void
vm_page_aflag_clear(vm_page_t m,uint16_t bits)831 vm_page_aflag_clear(vm_page_t m, uint16_t bits)
832 {
833 uint32_t *addr, val;
834
835 /*
836 * Access the whole 32-bit word containing the aflags field with an
837 * atomic update. Parallel non-atomic updates to the other fields
838 * within this word are handled properly by the atomic update.
839 */
840 addr = (void *)&m->a;
841 val = bits << VM_PAGE_AFLAG_SHIFT;
842 atomic_clear_32(addr, val);
843 }
844
845 /*
846 * Set the given bits in the specified page.
847 */
848 static inline void
vm_page_aflag_set(vm_page_t m,uint16_t bits)849 vm_page_aflag_set(vm_page_t m, uint16_t bits)
850 {
851 uint32_t *addr, val;
852
853 VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
854
855 /*
856 * Access the whole 32-bit word containing the aflags field with an
857 * atomic update. Parallel non-atomic updates to the other fields
858 * within this word are handled properly by the atomic update.
859 */
860 addr = (void *)&m->a;
861 val = bits << VM_PAGE_AFLAG_SHIFT;
862 atomic_set_32(addr, val);
863 }
864
865 /*
866 * vm_page_dirty:
867 *
868 * Set all bits in the page's dirty field.
869 *
870 * The object containing the specified page must be locked if the
871 * call is made from the machine-independent layer.
872 *
873 * See vm_page_clear_dirty_mask().
874 */
875 static __inline void
vm_page_dirty(vm_page_t m)876 vm_page_dirty(vm_page_t m)
877 {
878
879 /* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
880 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
881 vm_page_dirty_KBI(m);
882 #else
883 m->dirty = VM_PAGE_BITS_ALL;
884 #endif
885 }
886
887 /*
888 * vm_page_undirty:
889 *
890 * Set page to not be dirty. Note: does not clear pmap modify bits
891 */
892 static __inline void
vm_page_undirty(vm_page_t m)893 vm_page_undirty(vm_page_t m)
894 {
895
896 VM_PAGE_OBJECT_BUSY_ASSERT(m);
897 m->dirty = 0;
898 }
899
900 static inline uint8_t
_vm_page_queue(vm_page_astate_t as)901 _vm_page_queue(vm_page_astate_t as)
902 {
903
904 if ((as.flags & PGA_DEQUEUE) != 0)
905 return (PQ_NONE);
906 return (as.queue);
907 }
908
909 /*
910 * vm_page_queue:
911 *
912 * Return the index of the queue containing m.
913 */
914 static inline uint8_t
vm_page_queue(vm_page_t m)915 vm_page_queue(vm_page_t m)
916 {
917
918 return (_vm_page_queue(vm_page_astate_load(m)));
919 }
920
921 static inline bool
vm_page_active(vm_page_t m)922 vm_page_active(vm_page_t m)
923 {
924
925 return (vm_page_queue(m) == PQ_ACTIVE);
926 }
927
928 static inline bool
vm_page_inactive(vm_page_t m)929 vm_page_inactive(vm_page_t m)
930 {
931
932 return (vm_page_queue(m) == PQ_INACTIVE);
933 }
934
935 static inline bool
vm_page_in_laundry(vm_page_t m)936 vm_page_in_laundry(vm_page_t m)
937 {
938 uint8_t queue;
939
940 queue = vm_page_queue(m);
941 return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
942 }
943
944 /*
945 * vm_page_drop:
946 *
947 * Release a reference to a page and return the old reference count.
948 */
949 static inline u_int
vm_page_drop(vm_page_t m,u_int val)950 vm_page_drop(vm_page_t m, u_int val)
951 {
952 u_int old;
953
954 /*
955 * Synchronize with vm_page_free_prep(): ensure that all updates to the
956 * page structure are visible before it is freed.
957 */
958 atomic_thread_fence_rel();
959 old = atomic_fetchadd_int(&m->ref_count, -val);
960 KASSERT(old != VPRC_BLOCKED,
961 ("vm_page_drop: page %p has an invalid refcount value", m));
962 return (old);
963 }
964
965 /*
966 * vm_page_wired:
967 *
968 * Perform a racy check to determine whether a reference prevents the page
969 * from being reclaimable. If the page's object is locked, and the page is
970 * unmapped and exclusively busied by the current thread, no new wirings
971 * may be created.
972 */
973 static inline bool
vm_page_wired(vm_page_t m)974 vm_page_wired(vm_page_t m)
975 {
976
977 return (VPRC_WIRE_COUNT(m->ref_count) > 0);
978 }
979
980 static inline bool
vm_page_all_valid(vm_page_t m)981 vm_page_all_valid(vm_page_t m)
982 {
983
984 return (m->valid == VM_PAGE_BITS_ALL);
985 }
986
987 static inline bool
vm_page_any_valid(vm_page_t m)988 vm_page_any_valid(vm_page_t m)
989 {
990
991 return (m->valid != 0);
992 }
993
994 static inline bool
vm_page_none_valid(vm_page_t m)995 vm_page_none_valid(vm_page_t m)
996 {
997
998 return (m->valid == 0);
999 }
1000
1001 static inline int
vm_page_domain(vm_page_t m)1002 vm_page_domain(vm_page_t m)
1003 {
1004 #ifdef NUMA
1005 int domn, segind;
1006
1007 segind = m->segind;
1008 KASSERT(segind < vm_phys_nsegs, ("segind %d m %p", segind, m));
1009 domn = vm_phys_segs[segind].domain;
1010 KASSERT(domn >= 0 && domn < vm_ndomains, ("domain %d m %p", domn, m));
1011 return (domn);
1012 #else
1013 return (0);
1014 #endif
1015 }
1016
1017 #endif /* _KERNEL */
1018 #endif /* !_VM_PAGE_ */
1019