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 * A list of all pages for a given object, 82 * so they can be quickly deactivated at 83 * time of deallocation. 84 * 85 * An ordered list of pages due for pageout. 86 * 87 * In addition, the structure contains the object 88 * and offset to which this page belongs (for pageout), 89 * and sundry status bits. 90 * 91 * In general, operations on this structure's mutable fields are 92 * synchronized using either one of or a combination of locks. If a 93 * field is annotated with two of these locks then holding either is 94 * sufficient for read access but both are required for write access. 95 * The queue lock for a page depends on the value of its queue field and is 96 * described in detail below. 97 * 98 * The following annotations are possible: 99 * (A) the field must be accessed using atomic(9) and may require 100 * additional synchronization. 101 * (B) the page busy lock. 102 * (C) the field is immutable. 103 * (F) the per-domain lock for the free queues. 104 * (M) Machine dependent, defined by pmap layer. 105 * (O) the object that the page belongs to. 106 * (Q) the page's queue lock. 107 * 108 * The busy lock is an embedded reader-writer lock that protects the 109 * page's contents and identity (i.e., its <object, pindex> tuple) as 110 * well as certain valid/dirty modifications. To avoid bloating the 111 * the page structure, the busy lock lacks some of the features available 112 * the kernel's general-purpose synchronization primitives. As a result, 113 * busy lock ordering rules are not verified, lock recursion is not 114 * detected, and an attempt to xbusy a busy page or sbusy an xbusy page 115 * results will trigger a panic rather than causing the thread to block. 116 * vm_page_sleep_if_busy() can be used to sleep until the page's busy 117 * state changes, after which the caller must re-lookup the page and 118 * re-evaluate its state. vm_page_busy_acquire() will block until 119 * the lock is acquired. 120 * 121 * The valid field is protected by the page busy lock (B) and object 122 * lock (O). Transitions from invalid to valid are generally done 123 * via I/O or zero filling and do not require the object lock. 124 * These must be protected with the busy lock to prevent page-in or 125 * creation races. Page invalidation generally happens as a result 126 * of truncate or msync. When invalidated, pages must not be present 127 * in pmap and must hold the object lock to prevent concurrent 128 * speculative read-only mappings that do not require busy. I/O 129 * routines may check for validity without a lock if they are prepared 130 * to handle invalidation races with higher level locks (vnode) or are 131 * unconcerned with races so long as they hold a reference to prevent 132 * recycling. When a valid bit is set while holding a shared busy 133 * lock (A) atomic operations are used to protect against concurrent 134 * modification. 135 * 136 * In contrast, the synchronization of accesses to the page's 137 * dirty field is a mix of machine dependent (M) and busy (B). In 138 * the machine-independent layer, the page busy must be held to 139 * operate on the field. However, the pmap layer is permitted to 140 * set all bits within the field without holding that lock. If the 141 * underlying architecture does not support atomic read-modify-write 142 * operations on the field's type, then the machine-independent 143 * layer uses a 32-bit atomic on the aligned 32-bit word that 144 * contains the dirty field. In the machine-independent layer, 145 * the implementation of read-modify-write operations on the 146 * field is encapsulated in vm_page_clear_dirty_mask(). An 147 * exclusive busy lock combined with pmap_remove_{write/all}() is the 148 * only way to ensure a page can not become dirty. I/O generally 149 * removes the page from pmap to ensure exclusive access and atomic 150 * writes. 151 * 152 * The ref_count field tracks references to the page. References that 153 * prevent the page from being reclaimable are called wirings and are 154 * counted in the low bits of ref_count. The containing object's 155 * reference, if one exists, is counted using the VPRC_OBJREF bit in the 156 * ref_count field. Additionally, the VPRC_BLOCKED bit is used to 157 * atomically check for wirings and prevent new wirings via 158 * pmap_extract_and_hold(). When a page belongs to an object, it may be 159 * wired only when the object is locked, or the page is busy, or by 160 * pmap_extract_and_hold(). As a result, if the object is locked and the 161 * page is not busy (or is exclusively busied by the current thread), and 162 * the page is unmapped, its wire count will not increase. The ref_count 163 * field is updated using atomic operations in most cases, except when it 164 * is known that no other references to the page exist, such as in the page 165 * allocator. A page may be present in the page queues, or even actively 166 * scanned by the page daemon, without an explicitly counted referenced. 167 * The page daemon must therefore handle the possibility of a concurrent 168 * free of the page. 169 * 170 * The queue state of a page consists of the queue and act_count fields of 171 * its atomically updated state, and the subset of atomic flags specified 172 * by PGA_QUEUE_STATE_MASK. The queue field contains the page's page queue 173 * index, or PQ_NONE if it does not belong to a page queue. To modify the 174 * queue field, the page queue lock corresponding to the old value must be 175 * held, unless that value is PQ_NONE, in which case the queue index must 176 * be updated using an atomic RMW operation. There is one exception to 177 * this rule: the page daemon may transition the queue field from 178 * PQ_INACTIVE to PQ_NONE immediately prior to freeing the page during an 179 * inactive queue scan. At that point the page is already dequeued and no 180 * other references to that vm_page structure can exist. The PGA_ENQUEUED 181 * flag, when set, indicates that the page structure is physically inserted 182 * into the queue corresponding to the page's queue index, and may only be 183 * set or cleared with the corresponding page queue lock held. 184 * 185 * To avoid contention on page queue locks, page queue operations (enqueue, 186 * dequeue, requeue) are batched using fixed-size per-CPU queues. A 187 * deferred operation is requested by setting one of the flags in 188 * PGA_QUEUE_OP_MASK and inserting an entry into a batch queue. When a 189 * queue is full, an attempt to insert a new entry will lock the page 190 * queues and trigger processing of the pending entries. The 191 * type-stability of vm_page structures is crucial to this scheme since the 192 * processing of entries in a given batch queue may be deferred 193 * indefinitely. In particular, a page may be freed with pending batch 194 * queue entries. The page queue operation flags must be set using atomic 195 * RWM operations. 196 */ 197 198 #if PAGE_SIZE == 4096 199 #define VM_PAGE_BITS_ALL 0xffu 200 typedef uint8_t vm_page_bits_t; 201 #elif PAGE_SIZE == 8192 202 #define VM_PAGE_BITS_ALL 0xffffu 203 typedef uint16_t vm_page_bits_t; 204 #elif PAGE_SIZE == 16384 205 #define VM_PAGE_BITS_ALL 0xffffffffu 206 typedef uint32_t vm_page_bits_t; 207 #elif PAGE_SIZE == 32768 208 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu 209 typedef uint64_t vm_page_bits_t; 210 #endif 211 212 typedef union vm_page_astate { 213 struct { 214 uint16_t flags; 215 uint8_t queue; 216 uint8_t act_count; 217 }; 218 uint32_t _bits; 219 } vm_page_astate_t; 220 221 struct vm_page { 222 union { 223 TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */ 224 struct { 225 SLIST_ENTRY(vm_page) ss; /* private slists */ 226 } s; 227 struct { 228 u_long p; 229 u_long v; 230 } memguard; 231 struct { 232 void *slab; 233 void *zone; 234 } uma; 235 } plinks; 236 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */ 237 vm_object_t object; /* which object am I in (O) */ 238 vm_pindex_t pindex; /* offset into object (O,P) */ 239 vm_paddr_t phys_addr; /* physical address of page (C) */ 240 struct md_page md; /* machine dependent stuff */ 241 u_int ref_count; /* page references (A) */ 242 u_int busy_lock; /* busy owners lock (A) */ 243 union vm_page_astate a; /* state accessed atomically (A) */ 244 uint8_t order; /* index of the buddy queue (F) */ 245 uint8_t pool; /* vm_phys freepool index (F) */ 246 uint8_t flags; /* page PG_* flags (P) */ 247 uint8_t oflags; /* page VPO_* flags (O) */ 248 int8_t psind; /* pagesizes[] index (O) */ 249 int8_t segind; /* vm_phys segment index (C) */ 250 /* NOTE that these must support one bit per DEV_BSIZE in a page */ 251 /* so, on normal X86 kernels, they must be at least 8 bits wide */ 252 vm_page_bits_t valid; /* valid DEV_BSIZE chunk map (O,B) */ 253 vm_page_bits_t dirty; /* dirty DEV_BSIZE chunk map (M,B) */ 254 }; 255 256 /* 257 * Special bits used in the ref_count field. 258 * 259 * ref_count is normally used to count wirings that prevent the page from being 260 * reclaimed, but also supports several special types of references that do not 261 * prevent reclamation. Accesses to the ref_count field must be atomic unless 262 * the page is unallocated. 263 * 264 * VPRC_OBJREF is the reference held by the containing object. It can set or 265 * cleared only when the corresponding object's write lock is held. 266 * 267 * VPRC_BLOCKED is used to atomically block wirings via pmap lookups while 268 * attempting to tear down all mappings of a given page. The page busy lock and 269 * object write lock must both be held in order to set or clear this bit. 270 */ 271 #define VPRC_BLOCKED 0x40000000u /* mappings are being removed */ 272 #define VPRC_OBJREF 0x80000000u /* object reference, cleared with (O) */ 273 #define VPRC_WIRE_COUNT(c) ((c) & ~(VPRC_BLOCKED | VPRC_OBJREF)) 274 #define VPRC_WIRE_COUNT_MAX (~(VPRC_BLOCKED | VPRC_OBJREF)) 275 276 /* 277 * Page flags stored in oflags: 278 * 279 * Access to these page flags is synchronized by the lock on the object 280 * containing the page (O). 281 * 282 * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG) 283 * indicates that the page is not under PV management but 284 * otherwise should be treated as a normal page. Pages not 285 * under PV management cannot be paged out via the 286 * object/vm_page_t because there is no knowledge of their pte 287 * mappings, and such pages are also not on any PQ queue. 288 * 289 */ 290 #define VPO_KMEM_EXEC 0x01 /* kmem mapping allows execution */ 291 #define VPO_SWAPSLEEP 0x02 /* waiting for swap to finish */ 292 #define VPO_UNMANAGED 0x04 /* no PV management for page */ 293 #define VPO_SWAPINPROG 0x08 /* swap I/O in progress on page */ 294 295 /* 296 * Busy page implementation details. 297 * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation, 298 * even if the support for owner identity is removed because of size 299 * constraints. Checks on lock recursion are then not possible, while the 300 * lock assertions effectiveness is someway reduced. 301 */ 302 #define VPB_BIT_SHARED 0x01 303 #define VPB_BIT_EXCLUSIVE 0x02 304 #define VPB_BIT_WAITERS 0x04 305 #define VPB_BIT_FLAGMASK \ 306 (VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS) 307 308 #define VPB_SHARERS_SHIFT 3 309 #define VPB_SHARERS(x) \ 310 (((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT) 311 #define VPB_SHARERS_WORD(x) ((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED) 312 #define VPB_ONE_SHARER (1 << VPB_SHARERS_SHIFT) 313 314 #define VPB_SINGLE_EXCLUSIVE VPB_BIT_EXCLUSIVE 315 #ifdef INVARIANTS 316 #define VPB_CURTHREAD_EXCLUSIVE \ 317 (VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK)) 318 #else 319 #define VPB_CURTHREAD_EXCLUSIVE VPB_SINGLE_EXCLUSIVE 320 #endif 321 322 #define VPB_UNBUSIED VPB_SHARERS_WORD(0) 323 324 /* Freed lock blocks both shared and exclusive. */ 325 #define VPB_FREED (0xffffffff - VPB_BIT_SHARED) 326 327 #define PQ_NONE 255 328 #define PQ_INACTIVE 0 329 #define PQ_ACTIVE 1 330 #define PQ_LAUNDRY 2 331 #define PQ_UNSWAPPABLE 3 332 #define PQ_COUNT 4 333 334 #ifndef VM_PAGE_HAVE_PGLIST 335 TAILQ_HEAD(pglist, vm_page); 336 #define VM_PAGE_HAVE_PGLIST 337 #endif 338 SLIST_HEAD(spglist, vm_page); 339 340 #ifdef _KERNEL 341 extern vm_page_t bogus_page; 342 #endif /* _KERNEL */ 343 344 extern struct mtx_padalign pa_lock[]; 345 346 #if defined(__arm__) 347 #define PDRSHIFT PDR_SHIFT 348 #elif !defined(PDRSHIFT) 349 #define PDRSHIFT 21 350 #endif 351 352 #define pa_index(pa) ((pa) >> PDRSHIFT) 353 #define PA_LOCKPTR(pa) ((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT])) 354 #define PA_LOCKOBJPTR(pa) ((struct lock_object *)PA_LOCKPTR((pa))) 355 #define PA_LOCK(pa) mtx_lock(PA_LOCKPTR(pa)) 356 #define PA_TRYLOCK(pa) mtx_trylock(PA_LOCKPTR(pa)) 357 #define PA_UNLOCK(pa) mtx_unlock(PA_LOCKPTR(pa)) 358 #define PA_UNLOCK_COND(pa) \ 359 do { \ 360 if ((pa) != 0) { \ 361 PA_UNLOCK((pa)); \ 362 (pa) = 0; \ 363 } \ 364 } while (0) 365 366 #define PA_LOCK_ASSERT(pa, a) mtx_assert(PA_LOCKPTR(pa), (a)) 367 368 #if defined(KLD_MODULE) && !defined(KLD_TIED) 369 #define vm_page_lock(m) vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE) 370 #define vm_page_unlock(m) vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE) 371 #define vm_page_trylock(m) vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE) 372 #else /* !KLD_MODULE */ 373 #define vm_page_lockptr(m) (PA_LOCKPTR(VM_PAGE_TO_PHYS((m)))) 374 #define vm_page_lock(m) mtx_lock(vm_page_lockptr((m))) 375 #define vm_page_unlock(m) mtx_unlock(vm_page_lockptr((m))) 376 #define vm_page_trylock(m) mtx_trylock(vm_page_lockptr((m))) 377 #endif 378 #if defined(INVARIANTS) 379 #define vm_page_assert_locked(m) \ 380 vm_page_assert_locked_KBI((m), __FILE__, __LINE__) 381 #define vm_page_lock_assert(m, a) \ 382 vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__) 383 #else 384 #define vm_page_assert_locked(m) 385 #define vm_page_lock_assert(m, a) 386 #endif 387 388 /* 389 * The vm_page's aflags are updated using atomic operations. To set or clear 390 * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear() 391 * must be used. Neither these flags nor these functions are part of the KBI. 392 * 393 * PGA_REFERENCED may be cleared only if the page is locked. It is set by 394 * both the MI and MD VM layers. However, kernel loadable modules should not 395 * directly set this flag. They should call vm_page_reference() instead. 396 * 397 * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter(). 398 * When it does so, the object must be locked, or the page must be 399 * exclusive busied. The MI VM layer must never access this flag 400 * directly. Instead, it should call pmap_page_is_write_mapped(). 401 * 402 * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has 403 * at least one executable mapping. It is not consumed by the MI VM layer. 404 * 405 * PGA_NOSYNC must be set and cleared with the page busy lock held. 406 * 407 * PGA_ENQUEUED is set and cleared when a page is inserted into or removed 408 * from a page queue, respectively. It determines whether the plinks.q field 409 * of the page is valid. To set or clear this flag, page's "queue" field must 410 * be a valid queue index, and the corresponding page queue lock must be held. 411 * 412 * PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page 413 * queue, and cleared when the dequeue request is processed. A page may 414 * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue 415 * is requested after the page is scheduled to be enqueued but before it is 416 * actually inserted into the page queue. 417 * 418 * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued 419 * in its page queue. 420 * 421 * PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of 422 * the inactive queue, thus bypassing LRU. 423 * 424 * The PGA_DEQUEUE, PGA_REQUEUE and PGA_REQUEUE_HEAD flags must be set using an 425 * atomic RMW operation to ensure that the "queue" field is a valid queue index, 426 * and the corresponding page queue lock must be held when clearing any of the 427 * flags. 428 * 429 * PGA_SWAP_FREE is used to defer freeing swap space to the pageout daemon 430 * when the context that dirties the page does not have the object write lock 431 * held. 432 */ 433 #define PGA_WRITEABLE 0x0001 /* page may be mapped writeable */ 434 #define PGA_REFERENCED 0x0002 /* page has been referenced */ 435 #define PGA_EXECUTABLE 0x0004 /* page may be mapped executable */ 436 #define PGA_ENQUEUED 0x0008 /* page is enqueued in a page queue */ 437 #define PGA_DEQUEUE 0x0010 /* page is due to be dequeued */ 438 #define PGA_REQUEUE 0x0020 /* page is due to be requeued */ 439 #define PGA_REQUEUE_HEAD 0x0040 /* page requeue should bypass LRU */ 440 #define PGA_NOSYNC 0x0080 /* do not collect for syncer */ 441 #define PGA_SWAP_FREE 0x0100 /* page with swap space was dirtied */ 442 #define PGA_SWAP_SPACE 0x0200 /* page has allocated swap space */ 443 444 #define PGA_QUEUE_OP_MASK (PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD) 445 #define PGA_QUEUE_STATE_MASK (PGA_ENQUEUED | PGA_QUEUE_OP_MASK) 446 447 /* 448 * Page flags. Updates to these flags are not synchronized, and thus they must 449 * be set during page allocation or free to avoid races. 450 * 451 * The PG_PCPU_CACHE flag is set at allocation time if the page was 452 * allocated from a per-CPU cache. It is cleared the next time that the 453 * page is allocated from the physical memory allocator. 454 */ 455 #define PG_PCPU_CACHE 0x01 /* was allocated from per-CPU caches */ 456 #define PG_FICTITIOUS 0x02 /* physical page doesn't exist */ 457 #define PG_ZERO 0x04 /* page is zeroed */ 458 #define PG_MARKER 0x08 /* special queue marker page */ 459 #define PG_NODUMP 0x10 /* don't include this page in a dump */ 460 #define PG_NOFREE 0x20 /* page should never be freed. */ 461 462 /* 463 * Misc constants. 464 */ 465 #define ACT_DECLINE 1 466 #define ACT_ADVANCE 3 467 #define ACT_INIT 5 468 #define ACT_MAX 64 469 470 #ifdef _KERNEL 471 472 #include <sys/kassert.h> 473 #include <machine/atomic.h> 474 475 /* 476 * Each pageable resident page falls into one of five lists: 477 * 478 * free 479 * Available for allocation now. 480 * 481 * inactive 482 * Low activity, candidates for reclamation. 483 * This list is approximately LRU ordered. 484 * 485 * laundry 486 * This is the list of pages that should be 487 * paged out next. 488 * 489 * unswappable 490 * Dirty anonymous pages that cannot be paged 491 * out because no swap device is configured. 492 * 493 * active 494 * Pages that are "active", i.e., they have been 495 * recently referenced. 496 * 497 */ 498 499 extern vm_page_t vm_page_array; /* First resident page in table */ 500 extern long vm_page_array_size; /* number of vm_page_t's */ 501 extern long first_page; /* first physical page number */ 502 503 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 504 505 /* 506 * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory 507 * page to which the given physical address belongs. The correct vm_page_t 508 * object is returned for addresses that are not page-aligned. 509 */ 510 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa); 511 512 /* 513 * Page allocation parameters for vm_page for the functions 514 * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and 515 * vm_page_alloc_freelist(). Some functions support only a subset 516 * of the flags, and ignore others, see the flags legend. 517 * 518 * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*() 519 * and the vm_page_grab*() functions. See these functions for details. 520 * 521 * Bits 0 - 1 define class. 522 * Bits 2 - 15 dedicated for flags. 523 * Legend: 524 * (a) - vm_page_alloc() supports the flag. 525 * (c) - vm_page_alloc_contig() supports the flag. 526 * (g) - vm_page_grab() supports the flag. 527 * (n) - vm_page_alloc_noobj() and vm_page_alloc_freelist() support the flag. 528 * (p) - vm_page_grab_pages() supports the flag. 529 * Bits above 15 define the count of additional pages that the caller 530 * intends to allocate. 531 */ 532 #define VM_ALLOC_NORMAL 0 533 #define VM_ALLOC_INTERRUPT 1 534 #define VM_ALLOC_SYSTEM 2 535 #define VM_ALLOC_CLASS_MASK 3 536 #define VM_ALLOC_WAITOK 0x0008 /* (acn) Sleep and retry */ 537 #define VM_ALLOC_WAITFAIL 0x0010 /* (acn) Sleep and return error */ 538 #define VM_ALLOC_WIRED 0x0020 /* (acgnp) Allocate a wired page */ 539 #define VM_ALLOC_ZERO 0x0040 /* (acgnp) Allocate a zeroed page */ 540 #define VM_ALLOC_NORECLAIM 0x0080 /* (c) Do not reclaim after failure */ 541 #define VM_ALLOC_NOFREE 0x0100 /* (an) Page will never be released */ 542 #define VM_ALLOC_NOBUSY 0x0200 /* (acgp) Do not excl busy the page */ 543 #define VM_ALLOC_NOCREAT 0x0400 /* (gp) Don't create a page */ 544 #define VM_ALLOC_AVAIL1 0x0800 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_MAX 0xffff 550 #define VM_ALLOC_COUNT_SHIFT 16 551 #define VM_ALLOC_COUNT_MASK (VM_ALLOC_COUNT(VM_ALLOC_COUNT_MAX)) 552 #define VM_ALLOC_COUNT(count) ({ \ 553 KASSERT((count) <= VM_ALLOC_COUNT_MAX, \ 554 ("%s: invalid VM_ALLOC_COUNT value", __func__)); \ 555 (count) << VM_ALLOC_COUNT_SHIFT; \ 556 }) 557 558 #ifdef M_NOWAIT 559 static inline int 560 malloc2vm_flags(int malloc_flags) 561 { 562 int pflags; 563 564 KASSERT((malloc_flags & M_USE_RESERVE) == 0 || 565 (malloc_flags & M_NOWAIT) != 0, 566 ("M_USE_RESERVE requires M_NOWAIT")); 567 pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT : 568 VM_ALLOC_SYSTEM; 569 if ((malloc_flags & M_ZERO) != 0) 570 pflags |= VM_ALLOC_ZERO; 571 if ((malloc_flags & M_NODUMP) != 0) 572 pflags |= VM_ALLOC_NODUMP; 573 if ((malloc_flags & M_NOWAIT)) 574 pflags |= VM_ALLOC_NOWAIT; 575 if ((malloc_flags & M_WAITOK)) 576 pflags |= VM_ALLOC_WAITOK; 577 if ((malloc_flags & M_NORECLAIM)) 578 pflags |= VM_ALLOC_NORECLAIM; 579 if ((malloc_flags & M_NEVERFREED)) 580 pflags |= VM_ALLOC_NOFREE; 581 return (pflags); 582 } 583 #endif 584 585 /* 586 * Predicates supported by vm_page_ps_test(): 587 * 588 * PS_ALL_DIRTY is true only if the entire (super)page is dirty. 589 * However, it can be spuriously false when the (super)page has become 590 * dirty in the pmap but that information has not been propagated to the 591 * machine-independent layer. 592 */ 593 #define PS_ALL_DIRTY 0x1 594 #define PS_ALL_VALID 0x2 595 #define PS_NONE_BUSY 0x4 596 597 bool vm_page_busy_acquire(vm_page_t m, int allocflags); 598 void vm_page_busy_downgrade(vm_page_t m); 599 int vm_page_busy_tryupgrade(vm_page_t m); 600 bool vm_page_busy_sleep(vm_page_t m, const char *msg, int allocflags); 601 void vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m, 602 vm_pindex_t pindex, const char *wmesg, int allocflags); 603 void vm_page_free(vm_page_t m); 604 void vm_page_free_zero(vm_page_t m); 605 606 void vm_page_activate (vm_page_t); 607 void vm_page_advise(vm_page_t m, int advice); 608 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int); 609 vm_page_t vm_page_alloc_domain(vm_object_t, vm_pindex_t, int, int); 610 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t); 611 vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int, 612 vm_page_t); 613 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, 614 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, 615 vm_paddr_t boundary, vm_memattr_t memattr); 616 vm_page_t vm_page_alloc_contig_domain(vm_object_t object, 617 vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low, 618 vm_paddr_t high, u_long alignment, vm_paddr_t boundary, 619 vm_memattr_t memattr); 620 vm_page_t vm_page_alloc_noobj(int); 621 vm_page_t vm_page_alloc_noobj_domain(int, int); 622 vm_page_t vm_page_alloc_noobj_contig(int req, u_long npages, vm_paddr_t low, 623 vm_paddr_t high, u_long alignment, vm_paddr_t boundary, 624 vm_memattr_t memattr); 625 vm_page_t vm_page_alloc_noobj_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 vm_memattr_t memattr); 628 void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set); 629 bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose); 630 vm_page_t vm_page_grab(vm_object_t, vm_pindex_t, int); 631 vm_page_t vm_page_grab_unlocked(vm_object_t, vm_pindex_t, int); 632 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags, 633 vm_page_t *ma, int count); 634 int vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex, 635 int allocflags, vm_page_t *ma, int count); 636 int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex, 637 int allocflags); 638 int vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object, 639 vm_pindex_t pindex, int allocflags); 640 void vm_page_deactivate(vm_page_t); 641 void vm_page_deactivate_noreuse(vm_page_t); 642 void vm_page_dequeue(vm_page_t m); 643 void vm_page_dequeue_deferred(vm_page_t m); 644 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t); 645 void vm_page_free_invalid(vm_page_t); 646 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr); 647 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); 648 void vm_page_init_marker(vm_page_t marker, int queue, uint16_t aflags); 649 void vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind, int pool); 650 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t); 651 void vm_page_invalid(vm_page_t m); 652 void vm_page_launder(vm_page_t m); 653 vm_page_t vm_page_lookup(vm_object_t, vm_pindex_t); 654 vm_page_t vm_page_lookup_unlocked(vm_object_t, vm_pindex_t); 655 vm_page_t vm_page_next(vm_page_t m); 656 void vm_page_pqbatch_drain(void); 657 void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue); 658 bool vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old, 659 vm_page_astate_t new); 660 vm_page_t vm_page_prev(vm_page_t m); 661 bool vm_page_ps_test(vm_page_t m, int psind, int flags, vm_page_t skip_m); 662 void vm_page_putfake(vm_page_t m); 663 void vm_page_readahead_finish(vm_page_t m); 664 int vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, 665 vm_paddr_t high, u_long alignment, vm_paddr_t boundary); 666 int vm_page_reclaim_contig_domain(int domain, int req, u_long npages, 667 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary); 668 int vm_page_reclaim_contig_domain_ext(int domain, int req, u_long npages, 669 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, 670 int desired_runs); 671 void vm_page_reference(vm_page_t m); 672 #define VPR_TRYFREE 0x01 673 #define VPR_NOREUSE 0x02 674 void vm_page_release(vm_page_t m, int flags); 675 void vm_page_release_locked(vm_page_t m, int flags); 676 vm_page_t vm_page_relookup(vm_object_t, vm_pindex_t); 677 bool vm_page_remove(vm_page_t); 678 bool vm_page_remove_xbusy(vm_page_t); 679 int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t); 680 void vm_page_replace(vm_page_t mnew, vm_object_t object, 681 vm_pindex_t pindex, vm_page_t mold); 682 int vm_page_sbusied(vm_page_t m); 683 vm_page_bits_t vm_page_set_dirty(vm_page_t m); 684 void vm_page_set_valid_range(vm_page_t m, int base, int size); 685 vm_offset_t vm_page_startup(vm_offset_t vaddr); 686 void vm_page_sunbusy(vm_page_t m); 687 bool vm_page_try_remove_all(vm_page_t m); 688 bool vm_page_try_remove_write(vm_page_t m); 689 int vm_page_trysbusy(vm_page_t m); 690 int vm_page_tryxbusy(vm_page_t m); 691 void vm_page_unhold_pages(vm_page_t *ma, int count); 692 void vm_page_unswappable(vm_page_t m); 693 void vm_page_unwire(vm_page_t m, uint8_t queue); 694 bool vm_page_unwire_noq(vm_page_t m); 695 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); 696 void vm_page_wire(vm_page_t); 697 bool vm_page_wire_mapped(vm_page_t m); 698 void vm_page_xunbusy_hard(vm_page_t m); 699 void vm_page_xunbusy_hard_unchecked(vm_page_t m); 700 void vm_page_set_validclean (vm_page_t, int, int); 701 void vm_page_clear_dirty(vm_page_t, int, int); 702 void vm_page_set_invalid(vm_page_t, int, int); 703 void vm_page_valid(vm_page_t m); 704 int vm_page_is_valid(vm_page_t, int, int); 705 void vm_page_test_dirty(vm_page_t); 706 vm_page_bits_t vm_page_bits(int base, int size); 707 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 708 void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count); 709 710 void vm_page_dirty_KBI(vm_page_t m); 711 void vm_page_lock_KBI(vm_page_t m, const char *file, int line); 712 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line); 713 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line); 714 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 715 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line); 716 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line); 717 #endif 718 719 #define vm_page_busy_fetch(m) atomic_load_int(&(m)->busy_lock) 720 721 #define vm_page_assert_busied(m) \ 722 KASSERT(vm_page_busied(m), \ 723 ("vm_page_assert_busied: page %p not busy @ %s:%d", \ 724 (m), __FILE__, __LINE__)) 725 726 #define vm_page_assert_sbusied(m) \ 727 KASSERT(vm_page_sbusied(m), \ 728 ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \ 729 (m), __FILE__, __LINE__)) 730 731 #define vm_page_assert_unbusied(m) \ 732 KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) != \ 733 VPB_CURTHREAD_EXCLUSIVE, \ 734 ("vm_page_assert_unbusied: page %p busy_lock %#x owned" \ 735 " by me (%p) @ %s:%d", \ 736 (m), (m)->busy_lock, curthread, __FILE__, __LINE__)); \ 737 738 #define vm_page_assert_xbusied_unchecked(m) do { \ 739 KASSERT(vm_page_xbusied(m), \ 740 ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \ 741 (m), __FILE__, __LINE__)); \ 742 } while (0) 743 #define vm_page_assert_xbusied(m) do { \ 744 vm_page_assert_xbusied_unchecked(m); \ 745 KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) == \ 746 VPB_CURTHREAD_EXCLUSIVE, \ 747 ("vm_page_assert_xbusied: page %p busy_lock %#x not owned" \ 748 " by me (%p) @ %s:%d", \ 749 (m), (m)->busy_lock, curthread, __FILE__, __LINE__)); \ 750 } while (0) 751 752 #define vm_page_busied(m) \ 753 (vm_page_busy_fetch(m) != VPB_UNBUSIED) 754 755 #define vm_page_xbusied(m) \ 756 ((vm_page_busy_fetch(m) & VPB_SINGLE_EXCLUSIVE) != 0) 757 758 #define vm_page_busy_freed(m) \ 759 (vm_page_busy_fetch(m) == VPB_FREED) 760 761 /* Note: page m's lock must not be owned by the caller. */ 762 #define vm_page_xunbusy(m) do { \ 763 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \ 764 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \ 765 vm_page_xunbusy_hard(m); \ 766 } while (0) 767 #define vm_page_xunbusy_unchecked(m) do { \ 768 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \ 769 VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \ 770 vm_page_xunbusy_hard_unchecked(m); \ 771 } while (0) 772 773 #ifdef INVARIANTS 774 void vm_page_object_busy_assert(vm_page_t m); 775 #define VM_PAGE_OBJECT_BUSY_ASSERT(m) vm_page_object_busy_assert(m) 776 void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits); 777 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) \ 778 vm_page_assert_pga_writeable(m, bits) 779 /* 780 * Claim ownership of a page's xbusy state. In non-INVARIANTS kernels this 781 * operation is a no-op since ownership is not tracked. In particular 782 * this macro does not provide any synchronization with the previous owner. 783 */ 784 #define vm_page_xbusy_claim(m) do { \ 785 u_int _busy_lock; \ 786 \ 787 vm_page_assert_xbusied_unchecked((m)); \ 788 do { \ 789 _busy_lock = vm_page_busy_fetch(m); \ 790 } while (!atomic_cmpset_int(&(m)->busy_lock, _busy_lock, \ 791 (_busy_lock & VPB_BIT_FLAGMASK) | VPB_CURTHREAD_EXCLUSIVE)); \ 792 } while (0) 793 #else 794 #define VM_PAGE_OBJECT_BUSY_ASSERT(m) (void)0 795 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) (void)0 796 #define vm_page_xbusy_claim(m) 797 #endif 798 799 #if BYTE_ORDER == BIG_ENDIAN 800 #define VM_PAGE_AFLAG_SHIFT 16 801 #else 802 #define VM_PAGE_AFLAG_SHIFT 0 803 #endif 804 805 /* 806 * Load a snapshot of a page's 32-bit atomic state. 807 */ 808 static inline vm_page_astate_t 809 vm_page_astate_load(vm_page_t m) 810 { 811 vm_page_astate_t a; 812 813 a._bits = atomic_load_32(&m->a._bits); 814 return (a); 815 } 816 817 /* 818 * Atomically compare and set a page's atomic state. 819 */ 820 static inline bool 821 vm_page_astate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new) 822 { 823 824 KASSERT(new.queue == PQ_INACTIVE || (new.flags & PGA_REQUEUE_HEAD) == 0, 825 ("%s: invalid head requeue request for page %p", __func__, m)); 826 KASSERT((new.flags & PGA_ENQUEUED) == 0 || new.queue != PQ_NONE, 827 ("%s: setting PGA_ENQUEUED with PQ_NONE in page %p", __func__, m)); 828 KASSERT(new._bits != old->_bits, 829 ("%s: bits are unchanged", __func__)); 830 831 return (atomic_fcmpset_32(&m->a._bits, &old->_bits, new._bits) != 0); 832 } 833 834 /* 835 * Clear the given bits in the specified page. 836 */ 837 static inline void 838 vm_page_aflag_clear(vm_page_t m, uint16_t bits) 839 { 840 uint32_t *addr, val; 841 842 /* 843 * Access the whole 32-bit word containing the aflags field with an 844 * atomic update. Parallel non-atomic updates to the other fields 845 * within this word are handled properly by the atomic update. 846 */ 847 addr = (void *)&m->a; 848 val = bits << VM_PAGE_AFLAG_SHIFT; 849 atomic_clear_32(addr, val); 850 } 851 852 /* 853 * Set the given bits in the specified page. 854 */ 855 static inline void 856 vm_page_aflag_set(vm_page_t m, uint16_t bits) 857 { 858 uint32_t *addr, val; 859 860 VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits); 861 862 /* 863 * Access the whole 32-bit word containing the aflags field with an 864 * atomic update. Parallel non-atomic updates to the other fields 865 * within this word are handled properly by the atomic update. 866 */ 867 addr = (void *)&m->a; 868 val = bits << VM_PAGE_AFLAG_SHIFT; 869 atomic_set_32(addr, val); 870 } 871 872 /* 873 * vm_page_dirty: 874 * 875 * Set all bits in the page's dirty field. 876 * 877 * The object containing the specified page must be locked if the 878 * call is made from the machine-independent layer. 879 * 880 * See vm_page_clear_dirty_mask(). 881 */ 882 static __inline void 883 vm_page_dirty(vm_page_t m) 884 { 885 886 /* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */ 887 #if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS) 888 vm_page_dirty_KBI(m); 889 #else 890 m->dirty = VM_PAGE_BITS_ALL; 891 #endif 892 } 893 894 /* 895 * vm_page_undirty: 896 * 897 * Set page to not be dirty. Note: does not clear pmap modify bits 898 */ 899 static __inline void 900 vm_page_undirty(vm_page_t m) 901 { 902 903 VM_PAGE_OBJECT_BUSY_ASSERT(m); 904 m->dirty = 0; 905 } 906 907 static inline uint8_t 908 _vm_page_queue(vm_page_astate_t as) 909 { 910 911 if ((as.flags & PGA_DEQUEUE) != 0) 912 return (PQ_NONE); 913 return (as.queue); 914 } 915 916 /* 917 * vm_page_queue: 918 * 919 * Return the index of the queue containing m. 920 */ 921 static inline uint8_t 922 vm_page_queue(vm_page_t m) 923 { 924 925 return (_vm_page_queue(vm_page_astate_load(m))); 926 } 927 928 static inline bool 929 vm_page_active(vm_page_t m) 930 { 931 932 return (vm_page_queue(m) == PQ_ACTIVE); 933 } 934 935 static inline bool 936 vm_page_inactive(vm_page_t m) 937 { 938 939 return (vm_page_queue(m) == PQ_INACTIVE); 940 } 941 942 static inline bool 943 vm_page_in_laundry(vm_page_t m) 944 { 945 uint8_t queue; 946 947 queue = vm_page_queue(m); 948 return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE); 949 } 950 951 static inline void 952 vm_page_clearref(vm_page_t m) 953 { 954 u_int r; 955 956 r = m->ref_count; 957 while (atomic_fcmpset_int(&m->ref_count, &r, r & (VPRC_BLOCKED | 958 VPRC_OBJREF)) == 0) 959 ; 960 } 961 962 /* 963 * vm_page_drop: 964 * 965 * Release a reference to a page and return the old reference count. 966 */ 967 static inline u_int 968 vm_page_drop(vm_page_t m, u_int val) 969 { 970 u_int old; 971 972 /* 973 * Synchronize with vm_page_free_prep(): ensure that all updates to the 974 * page structure are visible before it is freed. 975 */ 976 atomic_thread_fence_rel(); 977 old = atomic_fetchadd_int(&m->ref_count, -val); 978 KASSERT(old != VPRC_BLOCKED, 979 ("vm_page_drop: page %p has an invalid refcount value", m)); 980 return (old); 981 } 982 983 /* 984 * vm_page_wired: 985 * 986 * Perform a racy check to determine whether a reference prevents the page 987 * from being reclaimable. If the page's object is locked, and the page is 988 * unmapped and exclusively busied by the current thread, no new wirings 989 * may be created. 990 */ 991 static inline bool 992 vm_page_wired(vm_page_t m) 993 { 994 995 return (VPRC_WIRE_COUNT(m->ref_count) > 0); 996 } 997 998 static inline bool 999 vm_page_all_valid(vm_page_t m) 1000 { 1001 1002 return (m->valid == VM_PAGE_BITS_ALL); 1003 } 1004 1005 static inline bool 1006 vm_page_any_valid(vm_page_t m) 1007 { 1008 1009 return (m->valid != 0); 1010 } 1011 1012 static inline bool 1013 vm_page_none_valid(vm_page_t m) 1014 { 1015 1016 return (m->valid == 0); 1017 } 1018 1019 static inline int 1020 vm_page_domain(vm_page_t m __numa_used) 1021 { 1022 #ifdef NUMA 1023 int domn, segind; 1024 1025 segind = m->segind; 1026 KASSERT(segind < vm_phys_nsegs, ("segind %d m %p", segind, m)); 1027 domn = vm_phys_segs[segind].domain; 1028 KASSERT(domn >= 0 && domn < vm_ndomains, ("domain %d m %p", domn, m)); 1029 return (domn); 1030 #else 1031 return (0); 1032 #endif 1033 } 1034 1035 #endif /* _KERNEL */ 1036 #endif /* !_VM_PAGE_ */ 1037