1 /*- 2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_page.h 8.2 (Berkeley) 12/13/93 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 * 62 * $FreeBSD$ 63 */ 64 65 /* 66 * Resident memory system definitions. 67 */ 68 69 #ifndef _VM_PAGE_ 70 #define _VM_PAGE_ 71 72 #include <vm/pmap.h> 73 74 /* 75 * Management of resident (logical) pages. 76 * 77 * A small structure is kept for each resident 78 * page, indexed by page number. Each structure 79 * is an element of several collections: 80 * 81 * A radix tree used to quickly 82 * perform object/offset lookups 83 * 84 * A list of all pages for a given object, 85 * so they can be quickly deactivated at 86 * time of deallocation. 87 * 88 * An ordered list of pages due for pageout. 89 * 90 * In addition, the structure contains the object 91 * and offset to which this page belongs (for pageout), 92 * and sundry status bits. 93 * 94 * In general, operations on this structure's mutable fields are 95 * synchronized using either one of or a combination of locks. If a 96 * field is annotated with two of these locks then holding either is 97 * sufficient for read access but both are required for write access. 98 * The physical address of a page is used to select its page lock from 99 * a pool. The queue lock for a page depends on the value of its queue 100 * field and is described in detail below. 101 * 102 * The following annotations are possible: 103 * (A) the field is atomic and may require additional synchronization. 104 * (B) the page busy lock. 105 * (C) the field is immutable. 106 * (F) the per-domain lock for the free queues 107 * (M) Machine dependent, defined by pmap layer. 108 * (O) the object that the page belongs to. 109 * (P) the page lock. 110 * (Q) the page's queue lock. 111 * 112 * The busy lock is an embedded reader-writer lock that protects the 113 * page's contents and identity (i.e., its <object, pindex> tuple) as 114 * well as certain valid/dirty modifications. To avoid bloating the 115 * the page structure, the busy lock lacks some of the features available 116 * the kernel's general-purpose synchronization primitives. As a result, 117 * busy lock ordering rules are not verified, lock recursion is not 118 * detected, and an attempt to xbusy a busy page or sbusy an xbusy page 119 * results will trigger a panic rather than causing the thread to block. 120 * vm_page_sleep_if_busy() can be used to sleep until the page's busy 121 * state changes, after which the caller must re-lookup the page and 122 * re-evaluate its state. vm_page_busy_acquire() will block until 123 * the lock is acquired. 124 * 125 * The valid field is protected by the page busy lock (B) and object 126 * lock (O). Transitions from invalid to valid are generally done 127 * via I/O or zero filling and do not require the object lock. 128 * These must be protected with the busy lock to prevent page-in or 129 * creation races. Page invalidation generally happens as a result 130 * of truncate or msync. When invalidated, pages must not be present 131 * in pmap and must hold the object lock to prevent concurrent 132 * speculative read-only mappings that do not require busy. I/O 133 * routines may check for validity without a lock if they are prepared 134 * to handle invalidation races with higher level locks (vnode) or are 135 * unconcerned with races so long as they hold a reference to prevent 136 * recycling. When a valid bit is set while holding a shared busy 137 * lock (A) atomic operations are used to protect against concurrent 138 * modification. 139 * 140 * In contrast, the synchronization of accesses to the page's 141 * dirty field is a mix of machine dependent (M) and busy (B). In 142 * the machine-independent layer, the page busy must be held to 143 * operate on the field. However, the pmap layer is permitted to 144 * set all bits within the field without holding that lock. If the 145 * underlying architecture does not support atomic read-modify-write 146 * operations on the field's type, then the machine-independent 147 * layer uses a 32-bit atomic on the aligned 32-bit word that 148 * contains the dirty field. In the machine-independent layer, 149 * the implementation of read-modify-write operations on the 150 * field is encapsulated in vm_page_clear_dirty_mask(). An 151 * exclusive busy lock combined with pmap_remove_{write/all}() is the 152 * only way to ensure a page can not become dirty. I/O generally 153 * removes the page from pmap to ensure exclusive access and atomic 154 * writes. 155 * 156 * The ref_count field tracks references to the page. References that 157 * prevent the page from being reclaimable are called wirings and are 158 * counted in the low bits of ref_count. The containing object's 159 * reference, if one exists, is counted using the VPRC_OBJREF bit in the 160 * ref_count field. Additionally, the VPRC_BLOCKED bit is used to 161 * atomically check for wirings and prevent new wirings via 162 * pmap_extract_and_hold(). When a page belongs to an object, it may be 163 * wired only when the object is locked, or the page is busy, or by 164 * pmap_extract_and_hold(). As a result, if the object is locked and the 165 * page is not busy (or is exclusively busied by the current thread), and 166 * the page is unmapped, its wire count will not increase. The ref_count 167 * field is updated using atomic operations in most cases, except when it 168 * is known that no other references to the page exist, such as in the page 169 * allocator. A page may be present in the page queues, or even actively 170 * scanned by the page daemon, without an explicitly counted referenced. 171 * The page daemon must therefore handle the possibility of a concurrent 172 * free of the page. 173 * 174 * The queue field is the index of the page queue containing the page, 175 * or PQ_NONE if the page is not enqueued. The queue lock of a page is 176 * the page queue lock corresponding to the page queue index, or the 177 * page lock (P) for the page if it is not enqueued. To modify the 178 * queue field, the queue lock for the old value of the field must be 179 * held. There is one exception to this rule: the page daemon may 180 * transition the queue field from PQ_INACTIVE to PQ_NONE immediately 181 * prior to freeing a page during an inactive queue scan. At that 182 * point the page has already been physically dequeued and no other 183 * references to that vm_page structure exist. 184 * 185 * To avoid contention on page queue locks, page queue operations 186 * (enqueue, dequeue, requeue) are batched using per-CPU queues. A 187 * deferred operation is requested by inserting an entry into a batch 188 * queue; the entry is simply a pointer to the page, and the request 189 * type is encoded in the page's aflags field using the values in 190 * PGA_QUEUE_STATE_MASK. The type-stability of struct vm_pages is 191 * crucial to this scheme since the processing of entries in a given 192 * batch queue may be deferred indefinitely. In particular, a page may 193 * be freed before its pending batch queue entries have been processed. 194 * The page lock (P) must be held to schedule a batched queue 195 * operation, and the page queue lock must be held in order to process 196 * batch queue entries for the page queue. There is one exception to 197 * this rule: the thread freeing a page may schedule a dequeue without 198 * holding the page lock. In this scenario the only other thread which 199 * may hold a reference to the page is the page daemon, which is 200 * careful to avoid modifying the page's queue state once the dequeue 201 * has been requested by setting PGA_DEQUEUE. 202 */ 203 204 #if PAGE_SIZE == 4096 205 #define VM_PAGE_BITS_ALL 0xffu 206 typedef uint8_t vm_page_bits_t; 207 #elif PAGE_SIZE == 8192 208 #define VM_PAGE_BITS_ALL 0xffffu 209 typedef uint16_t vm_page_bits_t; 210 #elif PAGE_SIZE == 16384 211 #define VM_PAGE_BITS_ALL 0xffffffffu 212 typedef uint32_t vm_page_bits_t; 213 #elif PAGE_SIZE == 32768 214 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu 215 typedef uint64_t vm_page_bits_t; 216 #endif 217 218 struct vm_page { 219 union { 220 TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */ 221 struct { 222 SLIST_ENTRY(vm_page) ss; /* private slists */ 223 void *pv; 224 } s; 225 struct { 226 u_long p; 227 u_long v; 228 } memguard; 229 } plinks; 230 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */ 231 vm_object_t object; /* which object am I in (O) */ 232 vm_pindex_t pindex; /* offset into object (O,P) */ 233 vm_paddr_t phys_addr; /* physical address of page (C) */ 234 struct md_page md; /* machine dependent stuff */ 235 u_int ref_count; /* page references (A) */ 236 volatile u_int busy_lock; /* busy owners lock */ 237 uint16_t aflags; /* atomic flags (A) */ 238 uint8_t queue; /* page queue index (Q) */ 239 uint8_t act_count; /* page usage count (P) */ 240 uint8_t order; /* index of the buddy queue (F) */ 241 uint8_t pool; /* vm_phys freepool index (F) */ 242 uint8_t flags; /* page PG_* flags (P) */ 243 uint8_t oflags; /* page VPO_* flags (O) */ 244 int8_t psind; /* pagesizes[] index (O) */ 245 int8_t segind; /* vm_phys segment index (C) */ 246 /* NOTE that these must support one bit per DEV_BSIZE in a page */ 247 /* so, on normal X86 kernels, they must be at least 8 bits wide */ 248 vm_page_bits_t valid; /* valid DEV_BSIZE chunk map (O,B) */ 249 vm_page_bits_t dirty; /* dirty DEV_BSIZE chunk map (M,B) */ 250 }; 251 252 /* 253 * Special bits used in the ref_count field. 254 * 255 * ref_count is normally used to count wirings that prevent the page from being 256 * reclaimed, but also supports several special types of references that do not 257 * prevent reclamation. Accesses to the ref_count field must be atomic unless 258 * the page is unallocated. 259 * 260 * VPRC_OBJREF is the reference held by the containing object. It can set or 261 * cleared only when the corresponding object's write lock is held. 262 * 263 * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while 264 * attempting to tear down all mappings of a given page. The page lock and 265 * object write lock must both be held in order to set or clear this bit. 266 */ 267 #define VPRC_BLOCKED 0x40000000u /* mappings are being removed */ 268 #define VPRC_OBJREF 0x80000000u /* object reference, cleared with (O) */ 269 #define VPRC_WIRE_COUNT(c) ((c) & ~(VPRC_BLOCKED | VPRC_OBJREF)) 270 #define VPRC_WIRE_COUNT_MAX (~(VPRC_BLOCKED | VPRC_OBJREF)) 271 272 /* 273 * Page flags stored in oflags: 274 * 275 * Access to these page flags is synchronized by the lock on the object 276 * containing the page (O). 277 * 278 * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG) 279 * indicates that the page is not under PV management but 280 * otherwise should be treated as a normal page. Pages not 281 * under PV management cannot be paged out via the 282 * object/vm_page_t because there is no knowledge of their pte 283 * mappings, and such pages are also not on any PQ queue. 284 * 285 */ 286 #define VPO_KMEM_EXEC 0x01 /* kmem mapping allows execution */ 287 #define VPO_SWAPSLEEP 0x02 /* waiting for swap to finish */ 288 #define VPO_UNMANAGED 0x04 /* no PV management for page */ 289 #define VPO_SWAPINPROG 0x08 /* swap I/O in progress on page */ 290 291 /* 292 * Busy page implementation details. 293 * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation, 294 * even if the support for owner identity is removed because of size 295 * constraints. Checks on lock recursion are then not possible, while the 296 * lock assertions effectiveness is someway reduced. 297 */ 298 #define VPB_BIT_SHARED 0x01 299 #define VPB_BIT_EXCLUSIVE 0x02 300 #define VPB_BIT_WAITERS 0x04 301 #define VPB_BIT_FLAGMASK \ 302 (VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS) 303 304 #define VPB_SHARERS_SHIFT 3 305 #define VPB_SHARERS(x) \ 306 (((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT) 307 #define VPB_SHARERS_WORD(x) ((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED) 308 #define VPB_ONE_SHARER (1 << VPB_SHARERS_SHIFT) 309 310 #define VPB_SINGLE_EXCLUSIVE VPB_BIT_EXCLUSIVE 311 #ifdef INVARIANTS 312 #define VPB_CURTHREAD_EXCLUSIVE \ 313 (VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK)) 314 #else 315 #define VPB_CURTHREAD_EXCLUSIVE VPB_SINGLE_EXCLUSIVE 316 #endif 317 318 #define VPB_UNBUSIED VPB_SHARERS_WORD(0) 319 320 #define PQ_NONE 255 321 #define PQ_INACTIVE 0 322 #define PQ_ACTIVE 1 323 #define PQ_LAUNDRY 2 324 #define PQ_UNSWAPPABLE 3 325 #define PQ_COUNT 4 326 327 #ifndef VM_PAGE_HAVE_PGLIST 328 TAILQ_HEAD(pglist, vm_page); 329 #define VM_PAGE_HAVE_PGLIST 330 #endif 331 SLIST_HEAD(spglist, vm_page); 332 333 #ifdef _KERNEL 334 extern vm_page_t bogus_page; 335 #endif /* _KERNEL */ 336 337 extern struct mtx_padalign pa_lock[]; 338 339 #if defined(__arm__) 340 #define PDRSHIFT PDR_SHIFT 341 #elif !defined(PDRSHIFT) 342 #define PDRSHIFT 21 343 #endif 344 345 #define pa_index(pa) ((pa) >> PDRSHIFT) 346 #define PA_LOCKPTR(pa) ((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT])) 347 #define PA_LOCKOBJPTR(pa) ((struct lock_object *)PA_LOCKPTR((pa))) 348 #define PA_LOCK(pa) mtx_lock(PA_LOCKPTR(pa)) 349 #define PA_TRYLOCK(pa) mtx_trylock(PA_LOCKPTR(pa)) 350 #define PA_UNLOCK(pa) mtx_unlock(PA_LOCKPTR(pa)) 351 #define PA_UNLOCK_COND(pa) \ 352 do { \ 353 if ((pa) != 0) { \ 354 PA_UNLOCK((pa)); \ 355 (pa) = 0; \ 356 } \ 357 } while (0) 358 359 #define PA_LOCK_ASSERT(pa, a) mtx_assert(PA_LOCKPTR(pa), (a)) 360 361 #if defined(KLD_MODULE) && !defined(KLD_TIED) 362 #define vm_page_lock(m) vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE) 363 #define vm_page_unlock(m) vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE) 364 #define vm_page_trylock(m) vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE) 365 #else /* !KLD_MODULE */ 366 #define vm_page_lockptr(m) (PA_LOCKPTR(VM_PAGE_TO_PHYS((m)))) 367 #define vm_page_lock(m) mtx_lock(vm_page_lockptr((m))) 368 #define vm_page_unlock(m) mtx_unlock(vm_page_lockptr((m))) 369 #define vm_page_trylock(m) mtx_trylock(vm_page_lockptr((m))) 370 #endif 371 #if defined(INVARIANTS) 372 #define vm_page_assert_locked(m) \ 373 vm_page_assert_locked_KBI((m), __FILE__, __LINE__) 374 #define vm_page_lock_assert(m, a) \ 375 vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__) 376 #else 377 #define vm_page_assert_locked(m) 378 #define vm_page_lock_assert(m, a) 379 #endif 380 381 /* 382 * The vm_page's aflags are updated using atomic operations. To set or clear 383 * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear() 384 * must be used. Neither these flags nor these functions are part of the KBI. 385 * 386 * PGA_REFERENCED may be cleared only if the page is locked. It is set by 387 * both the MI and MD VM layers. However, kernel loadable modules should not 388 * directly set this flag. They should call vm_page_reference() instead. 389 * 390 * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter(). 391 * When it does so, the object must be locked, or the page must be 392 * exclusive busied. The MI VM layer must never access this flag 393 * directly. Instead, it should call pmap_page_is_write_mapped(). 394 * 395 * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has 396 * at least one executable mapping. It is not consumed by the MI VM layer. 397 * 398 * PGA_NOSYNC must be set and cleared with the page busy lock held. 399 * 400 * PGA_ENQUEUED is set and cleared when a page is inserted into or removed 401 * from a page queue, respectively. It determines whether the plinks.q field 402 * of the page is valid. To set or clear this flag, the queue lock for the 403 * page must be held: the page queue lock corresponding to the page's "queue" 404 * field if its value is not PQ_NONE, and the page lock otherwise. 405 * 406 * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page 407 * queue, and cleared when the dequeue request is processed. A page may 408 * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue 409 * is requested after the page is scheduled to be enqueued but before it is 410 * actually inserted into the page queue. For allocated pages, the page lock 411 * must be held to set this flag, but it may be set by vm_page_free_prep() 412 * without the page lock held. The page queue lock must be held to clear the 413 * PGA_DEQUEUE flag. 414 * 415 * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued 416 * in its page queue. The page lock must be held to set this flag, and the 417 * queue lock for the page must be held to clear it. 418 * 419 * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of 420 * the inactive queue, thus bypassing LRU. The page lock must be held to 421 * set this flag, and the queue lock for the page must be held to clear it. 422 */ 423 #define PGA_WRITEABLE 0x0001 /* page may be mapped writeable */ 424 #define PGA_REFERENCED 0x0002 /* page has been referenced */ 425 #define PGA_EXECUTABLE 0x0004 /* page may be mapped executable */ 426 #define PGA_ENQUEUED 0x0008 /* page is enqueued in a page queue */ 427 #define PGA_DEQUEUE 0x0010 /* page is due to be dequeued */ 428 #define PGA_REQUEUE 0x0020 /* page is due to be requeued */ 429 #define PGA_REQUEUE_HEAD 0x0040 /* page requeue should bypass LRU */ 430 #define PGA_NOSYNC 0x0080 /* do not collect for syncer */ 431 432 #define PGA_QUEUE_STATE_MASK (PGA_ENQUEUED | PGA_DEQUEUE | PGA_REQUEUE | \ 433 PGA_REQUEUE_HEAD) 434 435 /* 436 * Page flags. If changed at any other time than page allocation or 437 * freeing, the modification must be protected by the vm_page lock. 438 * 439 * The PG_PCPU_CACHE flag is set at allocation time if the page was 440 * allocated from a per-CPU cache. It is cleared the next time that the 441 * page is allocated from the physical memory allocator. 442 */ 443 #define PG_PCPU_CACHE 0x01 /* was allocated from per-CPU caches */ 444 #define PG_FICTITIOUS 0x02 /* physical page doesn't exist */ 445 #define PG_ZERO 0x04 /* page is zeroed */ 446 #define PG_MARKER 0x08 /* special queue marker page */ 447 #define PG_NODUMP 0x10 /* don't include this page in a dump */ 448 449 /* 450 * Misc constants. 451 */ 452 #define ACT_DECLINE 1 453 #define ACT_ADVANCE 3 454 #define ACT_INIT 5 455 #define ACT_MAX 64 456 457 #ifdef _KERNEL 458 459 #include <sys/systm.h> 460 461 #include <machine/atomic.h> 462 463 /* 464 * Each pageable resident page falls into one of five lists: 465 * 466 * free 467 * Available for allocation now. 468 * 469 * inactive 470 * Low activity, candidates for reclamation. 471 * This list is approximately LRU ordered. 472 * 473 * laundry 474 * This is the list of pages that should be 475 * paged out next. 476 * 477 * unswappable 478 * Dirty anonymous pages that cannot be paged 479 * out because no swap device is configured. 480 * 481 * active 482 * Pages that are "active", i.e., they have been 483 * recently referenced. 484 * 485 */ 486 487 extern vm_page_t vm_page_array; /* First resident page in table */ 488 extern long vm_page_array_size; /* number of vm_page_t's */ 489 extern long first_page; /* first physical page number */ 490 491 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 492 493 /* 494 * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory 495 * page to which the given physical address belongs. The correct vm_page_t 496 * object is returned for addresses that are not page-aligned. 497 */ 498 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa); 499 500 /* 501 * Page allocation parameters for vm_page for the functions 502 * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and 503 * vm_page_alloc_freelist(). Some functions support only a subset 504 * of the flags, and ignore others, see the flags legend. 505 * 506 * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*() 507 * and the vm_page_grab*() functions. See these functions for details. 508 * 509 * Bits 0 - 1 define class. 510 * Bits 2 - 15 dedicated for flags. 511 * Legend: 512 * (a) - vm_page_alloc() supports the flag. 513 * (c) - vm_page_alloc_contig() supports the flag. 514 * (f) - vm_page_alloc_freelist() supports the flag. 515 * (g) - vm_page_grab() supports the flag. 516 * (p) - vm_page_grab_pages() supports the flag. 517 * Bits above 15 define the count of additional pages that the caller 518 * intends to allocate. 519 */ 520 #define VM_ALLOC_NORMAL 0 521 #define VM_ALLOC_INTERRUPT 1 522 #define VM_ALLOC_SYSTEM 2 523 #define VM_ALLOC_CLASS_MASK 3 524 #define VM_ALLOC_WAITOK 0x0008 /* (acf) Sleep and retry */ 525 #define VM_ALLOC_WAITFAIL 0x0010 /* (acf) Sleep and return error */ 526 #define VM_ALLOC_WIRED 0x0020 /* (acfgp) Allocate a wired page */ 527 #define VM_ALLOC_ZERO 0x0040 /* (acfgp) Allocate a prezeroed page */ 528 #define VM_ALLOC_NOOBJ 0x0100 /* (acg) No associated object */ 529 #define VM_ALLOC_NOBUSY 0x0200 /* (acgp) Do not excl busy the page */ 530 #define VM_ALLOC_NOCREAT 0x0400 /* (gp) Don't create a page */ 531 #define VM_ALLOC_IGN_SBUSY 0x1000 /* (gp) Ignore shared busy flag */ 532 #define VM_ALLOC_NODUMP 0x2000 /* (ag) don't include in dump */ 533 #define VM_ALLOC_SBUSY 0x4000 /* (acgp) Shared busy the page */ 534 #define VM_ALLOC_NOWAIT 0x8000 /* (acfgp) Do not sleep */ 535 #define VM_ALLOC_COUNT_SHIFT 16 536 #define VM_ALLOC_COUNT(count) ((count) << VM_ALLOC_COUNT_SHIFT) 537 538 #ifdef M_NOWAIT 539 static inline int 540 malloc2vm_flags(int malloc_flags) 541 { 542 int pflags; 543 544 KASSERT((malloc_flags & M_USE_RESERVE) == 0 || 545 (malloc_flags & M_NOWAIT) != 0, 546 ("M_USE_RESERVE requires M_NOWAIT")); 547 pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT : 548 VM_ALLOC_SYSTEM; 549 if ((malloc_flags & M_ZERO) != 0) 550 pflags |= VM_ALLOC_ZERO; 551 if ((malloc_flags & M_NODUMP) != 0) 552 pflags |= VM_ALLOC_NODUMP; 553 if ((malloc_flags & M_NOWAIT)) 554 pflags |= VM_ALLOC_NOWAIT; 555 if ((malloc_flags & M_WAITOK)) 556 pflags |= VM_ALLOC_WAITOK; 557 return (pflags); 558 } 559 #endif 560 561 /* 562 * Predicates supported by vm_page_ps_test(): 563 * 564 * PS_ALL_DIRTY is true only if the entire (super)page is dirty. 565 * However, it can be spuriously false when the (super)page has become 566 * dirty in the pmap but that information has not been propagated to the 567 * machine-independent layer. 568 */ 569 #define PS_ALL_DIRTY 0x1 570 #define PS_ALL_VALID 0x2 571 #define PS_NONE_BUSY 0x4 572 573 int vm_page_busy_acquire(vm_page_t m, int allocflags); 574 void vm_page_busy_downgrade(vm_page_t m); 575 int vm_page_busy_tryupgrade(vm_page_t m); 576 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared); 577 void vm_page_free(vm_page_t m); 578 void vm_page_free_zero(vm_page_t m); 579 580 void vm_page_activate (vm_page_t); 581 void vm_page_advise(vm_page_t m, int advice); 582 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int); 583 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int); 584 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t); 585 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int, 586 vm_page_t); 587 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, 588 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, 589 vm_paddr_t boundary, vm_memattr_t memattr); 590 vm_page_t vm_page_alloc_contig_domain(vm_object_t object, 591 vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low, 592 vm_paddr_t high, u_long alignment, vm_paddr_t boundary, 593 vm_memattr_t memattr); 594 vm_page_t vm_page_alloc_freelist(int, int); 595 vm_page_t vm_page_alloc_freelist_domain(int, int, int); 596 void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set); 597 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose); 598 void vm_page_change_lock(vm_page_t m, struct mtx **mtx); 599 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int); 600 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags, 601 vm_page_t *ma, int count); 602 int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex, 603 int allocflags); 604 void vm_page_deactivate(vm_page_t); 605 void vm_page_deactivate_noreuse(vm_page_t); 606 void vm_page_dequeue(vm_page_t m); 607 void vm_page_dequeue_deferred(vm_page_t m); 608 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t); 609 bool vm_page_free_prep(vm_page_t m); 610 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr); 611 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); 612 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t); 613 void vm_page_invalid(vm_page_t m); 614 void vm_page_launder(vm_page_t m); 615 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t); 616 vm_page_t vm_page_next(vm_page_t m); 617 void vm_page_pqbatch_drain(void); 618 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue); 619 vm_page_t vm_page_prev(vm_page_t m); 620 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m); 621 void vm_page_putfake(vm_page_t m); 622 void vm_page_readahead_finish(vm_page_t m); 623 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, 624 vm_paddr_t high, u_long alignment, vm_paddr_t boundary); 625 bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages, 626 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary); 627 void vm_page_reference(vm_page_t m); 628 #define VPR_TRYFREE 0x01 629 #define VPR_NOREUSE 0x02 630 void vm_page_release(vm_page_t m, int flags); 631 void vm_page_release_locked(vm_page_t m, int flags); 632 bool vm_page_remove(vm_page_t); 633 int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t); 634 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object, 635 vm_pindex_t pindex); 636 void vm_page_requeue(vm_page_t m); 637 int vm_page_sbusied(vm_page_t m); 638 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start, 639 vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options); 640 void vm_page_set_valid_range(vm_page_t m, int base, int size); 641 int vm_page_sleep_if_busy(vm_page_t m, const char *msg); 642 int vm_page_sleep_if_xbusy(vm_page_t m, const char *msg); 643 vm_offset_t vm_page_startup(vm_offset_t vaddr); 644 void vm_page_sunbusy(vm_page_t m); 645 void vm_page_swapqueue(vm_page_t m, uint8_t oldq, uint8_t newq); 646 bool vm_page_try_remove_all(vm_page_t m); 647 bool vm_page_try_remove_write(vm_page_t m); 648 int vm_page_trysbusy(vm_page_t m); 649 int vm_page_tryxbusy(vm_page_t m); 650 void vm_page_unhold_pages(vm_page_t *ma, int count); 651 void vm_page_unswappable(vm_page_t m); 652 void vm_page_unwire(vm_page_t m, uint8_t queue); 653 bool vm_page_unwire_noq(vm_page_t m); 654 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); 655 void vm_page_wire(vm_page_t); 656 bool vm_page_wire_mapped(vm_page_t m); 657 void vm_page_xunbusy_hard(vm_page_t m); 658 void vm_page_xunbusy_hard_unchecked(vm_page_t m); 659 void vm_page_set_validclean (vm_page_t, int, int); 660 void vm_page_clear_dirty(vm_page_t, int, int); 661 void vm_page_set_invalid(vm_page_t, int, int); 662 void vm_page_valid(vm_page_t m); 663 int vm_page_is_valid(vm_page_t, int, int); 664 void vm_page_test_dirty(vm_page_t); 665 vm_page_bits_t vm_page_bits(int base, int size); 666 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 667 void vm_page_free_toq(vm_page_t m); 668 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count); 669 670 void vm_page_dirty_KBI(vm_page_t m); 671 void vm_page_lock_KBI(vm_page_t m, const char *file, int line); 672 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line); 673 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line); 674 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 675 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line); 676 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line); 677 #endif 678 679 #define vm_page_assert_busied(m) \ 680 KASSERT(vm_page_busied(m), \ 681 ("vm_page_assert_busied: page %p not busy @ %s:%d", \ 682 (m), __FILE__, __LINE__)) 683 684 #define vm_page_assert_sbusied(m) \ 685 KASSERT(vm_page_sbusied(m), \ 686 ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \ 687 (m), __FILE__, __LINE__)) 688 689 #define vm_page_assert_unbusied(m) \ 690 KASSERT(!vm_page_busied(m), \ 691 ("vm_page_assert_unbusied: page %p busy @ %s:%d", \ 692 (m), __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((m->busy_lock & ~VPB_BIT_WAITERS) == \ 702 VPB_CURTHREAD_EXCLUSIVE, \ 703 ("vm_page_assert_xbusied: page %p busy_lock %#x not owned" \ 704 " by me @ %s:%d", \ 705 (m), (m)->busy_lock, __FILE__, __LINE__)); \ 706 } while (0) 707 708 #define vm_page_busied(m) \ 709 ((m)->busy_lock != VPB_UNBUSIED) 710 711 #define vm_page_sbusy(m) do { \ 712 if (!vm_page_trysbusy(m)) \ 713 panic("%s: page %p failed shared busying", __func__, \ 714 (m)); \ 715 } while (0) 716 717 #define vm_page_xbusied(m) \ 718 (((m)->busy_lock & VPB_SINGLE_EXCLUSIVE) != 0) 719 720 #define vm_page_xbusy(m) do { \ 721 if (!vm_page_tryxbusy(m)) \ 722 panic("%s: page %p failed exclusive busying", __func__, \ 723 (m)); \ 724 } while (0) 725 726 /* Note: page m's lock must not be owned by the caller. */ 727 #define vm_page_xunbusy(m) do { \ 728 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \ 729 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \ 730 vm_page_xunbusy_hard(m); \ 731 } while (0) 732 #define vm_page_xunbusy_unchecked(m) do { \ 733 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \ 734 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \ 735 vm_page_xunbusy_hard_unchecked(m); \ 736 } while (0) 737 738 #ifdef INVARIANTS 739 void vm_page_object_busy_assert(vm_page_t m); 740 #define VM_PAGE_OBJECT_BUSY_ASSERT(m) vm_page_object_busy_assert(m) 741 void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits); 742 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) \ 743 vm_page_assert_pga_writeable(m, bits) 744 #else 745 #define VM_PAGE_OBJECT_BUSY_ASSERT(m) (void)0 746 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) (void)0 747 #endif 748 749 /* 750 * We want to use atomic updates for the aflags field, which is 8 bits wide. 751 * However, not all architectures support atomic operations on 8-bit 752 * destinations. In order that we can easily use a 32-bit operation, we 753 * require that the aflags field be 32-bit aligned. 754 */ 755 _Static_assert(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0, 756 "aflags field is not 32-bit aligned"); 757 758 /* 759 * We want to be able to update the aflags and queue fields atomically in 760 * the same operation. 761 */ 762 _Static_assert(offsetof(struct vm_page, aflags) / sizeof(uint32_t) == 763 offsetof(struct vm_page, queue) / sizeof(uint32_t), 764 "aflags and queue fields do not belong to the same 32-bit word"); 765 _Static_assert(offsetof(struct vm_page, queue) % sizeof(uint32_t) == 2, 766 "queue field is at an unexpected offset"); 767 _Static_assert(sizeof(((struct vm_page *)NULL)->queue) == 1, 768 "queue field has an unexpected size"); 769 770 #if BYTE_ORDER == LITTLE_ENDIAN 771 #define VM_PAGE_AFLAG_SHIFT 0 772 #define VM_PAGE_QUEUE_SHIFT 16 773 #else 774 #define VM_PAGE_AFLAG_SHIFT 16 775 #define VM_PAGE_QUEUE_SHIFT 8 776 #endif 777 #define VM_PAGE_QUEUE_MASK (0xff << VM_PAGE_QUEUE_SHIFT) 778 779 /* 780 * Clear the given bits in the specified page. 781 */ 782 static inline void 783 vm_page_aflag_clear(vm_page_t m, uint16_t bits) 784 { 785 uint32_t *addr, val; 786 787 /* 788 * The PGA_REFERENCED flag can only be cleared if the page is locked. 789 */ 790 if ((bits & PGA_REFERENCED) != 0) 791 vm_page_assert_locked(m); 792 793 /* 794 * Access the whole 32-bit word containing the aflags field with an 795 * atomic update. Parallel non-atomic updates to the other fields 796 * within this word are handled properly by the atomic update. 797 */ 798 addr = (void *)&m->aflags; 799 val = bits << VM_PAGE_AFLAG_SHIFT; 800 atomic_clear_32(addr, val); 801 } 802 803 /* 804 * Set the given bits in the specified page. 805 */ 806 static inline void 807 vm_page_aflag_set(vm_page_t m, uint16_t bits) 808 { 809 uint32_t *addr, val; 810 811 VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits); 812 813 /* 814 * Access the whole 32-bit word containing the aflags field with an 815 * atomic update. Parallel non-atomic updates to the other fields 816 * within this word are handled properly by the atomic update. 817 */ 818 addr = (void *)&m->aflags; 819 val = bits << VM_PAGE_AFLAG_SHIFT; 820 atomic_set_32(addr, val); 821 } 822 823 /* 824 * Atomically update the queue state of the page. The operation fails if 825 * any of the queue flags in "fflags" are set or if the "queue" field of 826 * the page does not match the expected value; if the operation is 827 * successful, the flags in "nflags" are set and all other queue state 828 * flags are cleared. 829 */ 830 static inline bool 831 vm_page_pqstate_cmpset(vm_page_t m, uint32_t oldq, uint32_t newq, 832 uint32_t fflags, uint32_t nflags) 833 { 834 uint32_t *addr, nval, oval, qsmask; 835 836 fflags <<= VM_PAGE_AFLAG_SHIFT; 837 nflags <<= VM_PAGE_AFLAG_SHIFT; 838 newq <<= VM_PAGE_QUEUE_SHIFT; 839 oldq <<= VM_PAGE_QUEUE_SHIFT; 840 qsmask = ((PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD) << 841 VM_PAGE_AFLAG_SHIFT) | VM_PAGE_QUEUE_MASK; 842 843 addr = (void *)&m->aflags; 844 oval = atomic_load_32(addr); 845 do { 846 if ((oval & fflags) != 0) 847 return (false); 848 if ((oval & VM_PAGE_QUEUE_MASK) != oldq) 849 return (false); 850 nval = (oval & ~qsmask) | nflags | newq; 851 } while (!atomic_fcmpset_32(addr, &oval, nval)); 852 853 return (true); 854 } 855 856 /* 857 * vm_page_dirty: 858 * 859 * Set all bits in the page's dirty field. 860 * 861 * The object containing the specified page must be locked if the 862 * call is made from the machine-independent layer. 863 * 864 * See vm_page_clear_dirty_mask(). 865 */ 866 static __inline void 867 vm_page_dirty(vm_page_t m) 868 { 869 870 /* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */ 871 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS) 872 vm_page_dirty_KBI(m); 873 #else 874 m->dirty = VM_PAGE_BITS_ALL; 875 #endif 876 } 877 878 /* 879 * vm_page_undirty: 880 * 881 * Set page to not be dirty. Note: does not clear pmap modify bits 882 */ 883 static __inline void 884 vm_page_undirty(vm_page_t m) 885 { 886 887 VM_PAGE_OBJECT_BUSY_ASSERT(m); 888 m->dirty = 0; 889 } 890 891 static inline void 892 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex, 893 vm_page_t mold) 894 { 895 vm_page_t mret; 896 897 mret = vm_page_replace(mnew, object, pindex); 898 KASSERT(mret == mold, 899 ("invalid page replacement, mold=%p, mret=%p", mold, mret)); 900 901 /* Unused if !INVARIANTS. */ 902 (void)mold; 903 (void)mret; 904 } 905 906 /* 907 * vm_page_queue: 908 * 909 * Return the index of the queue containing m. This index is guaranteed 910 * not to change while the page lock is held. 911 */ 912 static inline uint8_t 913 vm_page_queue(vm_page_t m) 914 { 915 916 vm_page_assert_locked(m); 917 918 if ((m->aflags & PGA_DEQUEUE) != 0) 919 return (PQ_NONE); 920 atomic_thread_fence_acq(); 921 return (m->queue); 922 } 923 924 static inline bool 925 vm_page_active(vm_page_t m) 926 { 927 928 return (vm_page_queue(m) == PQ_ACTIVE); 929 } 930 931 static inline bool 932 vm_page_inactive(vm_page_t m) 933 { 934 935 return (vm_page_queue(m) == PQ_INACTIVE); 936 } 937 938 static inline bool 939 vm_page_in_laundry(vm_page_t m) 940 { 941 uint8_t queue; 942 943 queue = vm_page_queue(m); 944 return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE); 945 } 946 947 /* 948 * vm_page_drop: 949 * 950 * Release a reference to a page and return the old reference count. 951 */ 952 static inline u_int 953 vm_page_drop(vm_page_t m, u_int val) 954 { 955 u_int old; 956 957 /* 958 * Synchronize with vm_page_free_prep(): ensure that all updates to the 959 * page structure are visible before it is freed. 960 */ 961 atomic_thread_fence_rel(); 962 old = atomic_fetchadd_int(&m->ref_count, -val); 963 KASSERT(old != VPRC_BLOCKED, 964 ("vm_page_drop: page %p has an invalid refcount value", m)); 965 return (old); 966 } 967 968 /* 969 * vm_page_wired: 970 * 971 * Perform a racy check to determine whether a reference prevents the page 972 * from being reclaimable. If the page's object is locked, and the page is 973 * unmapped and unbusied or exclusively busied by the current thread, no 974 * new wirings may be created. 975 */ 976 static inline bool 977 vm_page_wired(vm_page_t m) 978 { 979 980 return (VPRC_WIRE_COUNT(m->ref_count) > 0); 981 } 982 983 static inline bool 984 vm_page_all_valid(vm_page_t m) 985 { 986 987 return (m->valid == VM_PAGE_BITS_ALL); 988 } 989 990 static inline bool 991 vm_page_none_valid(vm_page_t m) 992 { 993 994 return (m->valid == 0); 995 } 996 997 #endif /* _KERNEL */ 998 #endif /* !_VM_PAGE_ */ 999