1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_page.h 8.2 (Berkeley) 12/13/93 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 * 62 * $FreeBSD$ 63 */ 64 65 /* 66 * Resident memory system definitions. 67 */ 68 69 #ifndef _VM_PAGE_ 70 #define _VM_PAGE_ 71 72 #include <vm/pmap.h> 73 74 /* 75 * Management of resident (logical) pages. 76 * 77 * A small structure is kept for each resident 78 * page, indexed by page number. Each structure 79 * is an element of several collections: 80 * 81 * A radix tree used to quickly 82 * perform object/offset lookups 83 * 84 * A list of all pages for a given object, 85 * so they can be quickly deactivated at 86 * time of deallocation. 87 * 88 * An ordered list of pages due for pageout. 89 * 90 * In addition, the structure contains the object 91 * and offset to which this page belongs (for pageout), 92 * and sundry status bits. 93 * 94 * In general, operations on this structure's mutable fields are 95 * synchronized using either one of or a combination of the lock on the 96 * object that the page belongs to (O), the pool lock for the page (P), 97 * or the lock for either the free or paging queue (Q). If a field is 98 * annotated below with two of these locks, then holding either lock is 99 * sufficient for read access, but both locks are required for write 100 * access. 101 * 102 * In contrast, the synchronization of accesses to the page's 103 * dirty field is machine dependent (M). In the 104 * machine-independent layer, the lock on the object that the 105 * page belongs to must be held in order to operate on the field. 106 * However, the pmap layer is permitted to set all bits within 107 * the field without holding that lock. If the underlying 108 * architecture does not support atomic read-modify-write 109 * operations on the field's type, then the machine-independent 110 * layer uses a 32-bit atomic on the aligned 32-bit word that 111 * contains the dirty field. In the machine-independent layer, 112 * the implementation of read-modify-write operations on the 113 * field is encapsulated in vm_page_clear_dirty_mask(). 114 */ 115 116 #if PAGE_SIZE == 4096 117 #define VM_PAGE_BITS_ALL 0xffu 118 typedef uint8_t vm_page_bits_t; 119 #elif PAGE_SIZE == 8192 120 #define VM_PAGE_BITS_ALL 0xffffu 121 typedef uint16_t vm_page_bits_t; 122 #elif PAGE_SIZE == 16384 123 #define VM_PAGE_BITS_ALL 0xffffffffu 124 typedef uint32_t vm_page_bits_t; 125 #elif PAGE_SIZE == 32768 126 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu 127 typedef uint64_t vm_page_bits_t; 128 #endif 129 130 struct vm_page { 131 union { 132 TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */ 133 struct { 134 SLIST_ENTRY(vm_page) ss; /* private slists */ 135 void *pv; 136 } s; 137 struct { 138 u_long p; 139 u_long v; 140 } memguard; 141 } plinks; 142 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */ 143 vm_object_t object; /* which object am I in (O,P) */ 144 vm_pindex_t pindex; /* offset into object (O,P) */ 145 vm_paddr_t phys_addr; /* physical address of page */ 146 struct md_page md; /* machine dependent stuff */ 147 u_int wire_count; /* wired down maps refs (P) */ 148 volatile u_int busy_lock; /* busy owners lock */ 149 uint16_t hold_count; /* page hold count (P) */ 150 uint16_t flags; /* page PG_* flags (P) */ 151 uint8_t aflags; /* access is atomic */ 152 uint8_t oflags; /* page VPO_* flags (O) */ 153 uint8_t queue; /* page queue index (P,Q) */ 154 int8_t psind; /* pagesizes[] index (O) */ 155 int8_t segind; 156 uint8_t order; /* index of the buddy queue */ 157 uint8_t pool; 158 u_char act_count; /* page usage count (P) */ 159 /* NOTE that these must support one bit per DEV_BSIZE in a page */ 160 /* so, on normal X86 kernels, they must be at least 8 bits wide */ 161 vm_page_bits_t valid; /* map of valid DEV_BSIZE chunks (O) */ 162 vm_page_bits_t dirty; /* map of dirty DEV_BSIZE chunks (M) */ 163 }; 164 165 /* 166 * Page flags stored in oflags: 167 * 168 * Access to these page flags is synchronized by the lock on the object 169 * containing the page (O). 170 * 171 * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG) 172 * indicates that the page is not under PV management but 173 * otherwise should be treated as a normal page. Pages not 174 * under PV management cannot be paged out via the 175 * object/vm_page_t because there is no knowledge of their pte 176 * mappings, and such pages are also not on any PQ queue. 177 * 178 */ 179 #define VPO_UNUSED01 0x01 /* --available-- */ 180 #define VPO_SWAPSLEEP 0x02 /* waiting for swap to finish */ 181 #define VPO_UNMANAGED 0x04 /* no PV management for page */ 182 #define VPO_SWAPINPROG 0x08 /* swap I/O in progress on page */ 183 #define VPO_NOSYNC 0x10 /* do not collect for syncer */ 184 185 /* 186 * Busy page implementation details. 187 * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation, 188 * even if the support for owner identity is removed because of size 189 * constraints. Checks on lock recursion are then not possible, while the 190 * lock assertions effectiveness is someway reduced. 191 */ 192 #define VPB_BIT_SHARED 0x01 193 #define VPB_BIT_EXCLUSIVE 0x02 194 #define VPB_BIT_WAITERS 0x04 195 #define VPB_BIT_FLAGMASK \ 196 (VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS) 197 198 #define VPB_SHARERS_SHIFT 3 199 #define VPB_SHARERS(x) \ 200 (((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT) 201 #define VPB_SHARERS_WORD(x) ((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED) 202 #define VPB_ONE_SHARER (1 << VPB_SHARERS_SHIFT) 203 204 #define VPB_SINGLE_EXCLUSIVER VPB_BIT_EXCLUSIVE 205 206 #define VPB_UNBUSIED VPB_SHARERS_WORD(0) 207 208 #define PQ_NONE 255 209 #define PQ_INACTIVE 0 210 #define PQ_ACTIVE 1 211 #define PQ_LAUNDRY 2 212 #define PQ_UNSWAPPABLE 3 213 #define PQ_COUNT 4 214 215 #ifndef VM_PAGE_HAVE_PGLIST 216 TAILQ_HEAD(pglist, vm_page); 217 #define VM_PAGE_HAVE_PGLIST 218 #endif 219 SLIST_HEAD(spglist, vm_page); 220 221 struct vm_pagequeue { 222 struct mtx pq_mutex; 223 struct pglist pq_pl; 224 int pq_cnt; 225 u_int * const pq_vcnt; 226 const char * const pq_name; 227 } __aligned(CACHE_LINE_SIZE); 228 229 230 struct vm_domain { 231 struct vm_pagequeue vmd_pagequeues[PQ_COUNT]; 232 u_int vmd_page_count; 233 u_int vmd_free_count; 234 long vmd_segs; /* bitmask of the segments */ 235 boolean_t vmd_oom; 236 int vmd_oom_seq; 237 int vmd_last_active_scan; 238 struct vm_page vmd_laundry_marker; 239 struct vm_page vmd_marker; /* marker for pagedaemon private use */ 240 struct vm_page vmd_inacthead; /* marker for LRU-defeating insertions */ 241 }; 242 243 extern struct vm_domain vm_dom[MAXMEMDOM]; 244 245 #define vm_pagequeue_assert_locked(pq) mtx_assert(&(pq)->pq_mutex, MA_OWNED) 246 #define vm_pagequeue_lock(pq) mtx_lock(&(pq)->pq_mutex) 247 #define vm_pagequeue_lockptr(pq) (&(pq)->pq_mutex) 248 #define vm_pagequeue_unlock(pq) mtx_unlock(&(pq)->pq_mutex) 249 250 #ifdef _KERNEL 251 extern vm_page_t bogus_page; 252 253 static __inline void 254 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend) 255 { 256 257 #ifdef notyet 258 vm_pagequeue_assert_locked(pq); 259 #endif 260 pq->pq_cnt += addend; 261 atomic_add_int(pq->pq_vcnt, addend); 262 } 263 #define vm_pagequeue_cnt_inc(pq) vm_pagequeue_cnt_add((pq), 1) 264 #define vm_pagequeue_cnt_dec(pq) vm_pagequeue_cnt_add((pq), -1) 265 #endif /* _KERNEL */ 266 267 extern struct mtx_padalign vm_page_queue_free_mtx; 268 extern struct mtx_padalign pa_lock[]; 269 270 #if defined(__arm__) 271 #define PDRSHIFT PDR_SHIFT 272 #elif !defined(PDRSHIFT) 273 #define PDRSHIFT 21 274 #endif 275 276 #define pa_index(pa) ((pa) >> PDRSHIFT) 277 #define PA_LOCKPTR(pa) ((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT])) 278 #define PA_LOCKOBJPTR(pa) ((struct lock_object *)PA_LOCKPTR((pa))) 279 #define PA_LOCK(pa) mtx_lock(PA_LOCKPTR(pa)) 280 #define PA_TRYLOCK(pa) mtx_trylock(PA_LOCKPTR(pa)) 281 #define PA_UNLOCK(pa) mtx_unlock(PA_LOCKPTR(pa)) 282 #define PA_UNLOCK_COND(pa) \ 283 do { \ 284 if ((pa) != 0) { \ 285 PA_UNLOCK((pa)); \ 286 (pa) = 0; \ 287 } \ 288 } while (0) 289 290 #define PA_LOCK_ASSERT(pa, a) mtx_assert(PA_LOCKPTR(pa), (a)) 291 292 #ifdef KLD_MODULE 293 #define vm_page_lock(m) vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE) 294 #define vm_page_unlock(m) vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE) 295 #define vm_page_trylock(m) vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE) 296 #else /* !KLD_MODULE */ 297 #define vm_page_lockptr(m) (PA_LOCKPTR(VM_PAGE_TO_PHYS((m)))) 298 #define vm_page_lock(m) mtx_lock(vm_page_lockptr((m))) 299 #define vm_page_unlock(m) mtx_unlock(vm_page_lockptr((m))) 300 #define vm_page_trylock(m) mtx_trylock(vm_page_lockptr((m))) 301 #endif 302 #if defined(INVARIANTS) 303 #define vm_page_assert_locked(m) \ 304 vm_page_assert_locked_KBI((m), __FILE__, __LINE__) 305 #define vm_page_lock_assert(m, a) \ 306 vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__) 307 #else 308 #define vm_page_assert_locked(m) 309 #define vm_page_lock_assert(m, a) 310 #endif 311 312 /* 313 * The vm_page's aflags are updated using atomic operations. To set or clear 314 * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear() 315 * must be used. Neither these flags nor these functions are part of the KBI. 316 * 317 * PGA_REFERENCED may be cleared only if the page is locked. It is set by 318 * both the MI and MD VM layers. However, kernel loadable modules should not 319 * directly set this flag. They should call vm_page_reference() instead. 320 * 321 * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter(). 322 * When it does so, the object must be locked, or the page must be 323 * exclusive busied. The MI VM layer must never access this flag 324 * directly. Instead, it should call pmap_page_is_write_mapped(). 325 * 326 * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has 327 * at least one executable mapping. It is not consumed by the MI VM layer. 328 */ 329 #define PGA_WRITEABLE 0x01 /* page may be mapped writeable */ 330 #define PGA_REFERENCED 0x02 /* page has been referenced */ 331 #define PGA_EXECUTABLE 0x04 /* page may be mapped executable */ 332 333 /* 334 * Page flags. If changed at any other time than page allocation or 335 * freeing, the modification must be protected by the vm_page lock. 336 */ 337 #define PG_FICTITIOUS 0x0004 /* physical page doesn't exist */ 338 #define PG_ZERO 0x0008 /* page is zeroed */ 339 #define PG_MARKER 0x0010 /* special queue marker page */ 340 #define PG_NODUMP 0x0080 /* don't include this page in a dump */ 341 #define PG_UNHOLDFREE 0x0100 /* delayed free of a held page */ 342 343 /* 344 * Misc constants. 345 */ 346 #define ACT_DECLINE 1 347 #define ACT_ADVANCE 3 348 #define ACT_INIT 5 349 #define ACT_MAX 64 350 351 #ifdef _KERNEL 352 353 #include <sys/systm.h> 354 355 #include <machine/atomic.h> 356 357 /* 358 * Each pageable resident page falls into one of five lists: 359 * 360 * free 361 * Available for allocation now. 362 * 363 * inactive 364 * Low activity, candidates for reclamation. 365 * This list is approximately LRU ordered. 366 * 367 * laundry 368 * This is the list of pages that should be 369 * paged out next. 370 * 371 * unswappable 372 * Dirty anonymous pages that cannot be paged 373 * out because no swap device is configured. 374 * 375 * active 376 * Pages that are "active", i.e., they have been 377 * recently referenced. 378 * 379 */ 380 381 extern int vm_page_zero_count; 382 383 extern vm_page_t vm_page_array; /* First resident page in table */ 384 extern long vm_page_array_size; /* number of vm_page_t's */ 385 extern long first_page; /* first physical page number */ 386 387 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 388 389 /* 390 * PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory 391 * page to which the given physical address belongs. The correct vm_page_t 392 * object is returned for addresses that are not page-aligned. 393 */ 394 vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa); 395 396 /* 397 * Page allocation parameters for vm_page for the functions 398 * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and 399 * vm_page_alloc_freelist(). Some functions support only a subset 400 * of the flags, and ignore others, see the flags legend. 401 * 402 * The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*() 403 * and the vm_page_grab*() functions. See these functions for details. 404 * 405 * Bits 0 - 1 define class. 406 * Bits 2 - 15 dedicated for flags. 407 * Legend: 408 * (a) - vm_page_alloc() supports the flag. 409 * (c) - vm_page_alloc_contig() supports the flag. 410 * (f) - vm_page_alloc_freelist() supports the flag. 411 * (g) - vm_page_grab() supports the flag. 412 * (p) - vm_page_grab_pages() supports the flag. 413 * Bits above 15 define the count of additional pages that the caller 414 * intends to allocate. 415 */ 416 #define VM_ALLOC_NORMAL 0 417 #define VM_ALLOC_INTERRUPT 1 418 #define VM_ALLOC_SYSTEM 2 419 #define VM_ALLOC_CLASS_MASK 3 420 #define VM_ALLOC_WAITOK 0x0008 /* (acf) Sleep and retry */ 421 #define VM_ALLOC_WAITFAIL 0x0010 /* (acf) Sleep and return error */ 422 #define VM_ALLOC_WIRED 0x0020 /* (acfgp) Allocate a wired page */ 423 #define VM_ALLOC_ZERO 0x0040 /* (acfgp) Allocate a prezeroed page */ 424 #define VM_ALLOC_NOOBJ 0x0100 /* (acg) No associated object */ 425 #define VM_ALLOC_NOBUSY 0x0200 /* (acgp) Do not excl busy the page */ 426 #define VM_ALLOC_IGN_SBUSY 0x1000 /* (gp) Ignore shared busy flag */ 427 #define VM_ALLOC_NODUMP 0x2000 /* (ag) don't include in dump */ 428 #define VM_ALLOC_SBUSY 0x4000 /* (acgp) Shared busy the page */ 429 #define VM_ALLOC_NOWAIT 0x8000 /* (acfgp) Do not sleep */ 430 #define VM_ALLOC_COUNT_SHIFT 16 431 #define VM_ALLOC_COUNT(count) ((count) << VM_ALLOC_COUNT_SHIFT) 432 433 #ifdef M_NOWAIT 434 static inline int 435 malloc2vm_flags(int malloc_flags) 436 { 437 int pflags; 438 439 KASSERT((malloc_flags & M_USE_RESERVE) == 0 || 440 (malloc_flags & M_NOWAIT) != 0, 441 ("M_USE_RESERVE requires M_NOWAIT")); 442 pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT : 443 VM_ALLOC_SYSTEM; 444 if ((malloc_flags & M_ZERO) != 0) 445 pflags |= VM_ALLOC_ZERO; 446 if ((malloc_flags & M_NODUMP) != 0) 447 pflags |= VM_ALLOC_NODUMP; 448 if ((malloc_flags & M_NOWAIT)) 449 pflags |= VM_ALLOC_NOWAIT; 450 if ((malloc_flags & M_WAITOK)) 451 pflags |= VM_ALLOC_WAITOK; 452 return (pflags); 453 } 454 #endif 455 456 /* 457 * Predicates supported by vm_page_ps_test(): 458 * 459 * PS_ALL_DIRTY is true only if the entire (super)page is dirty. 460 * However, it can be spuriously false when the (super)page has become 461 * dirty in the pmap but that information has not been propagated to the 462 * machine-independent layer. 463 */ 464 #define PS_ALL_DIRTY 0x1 465 #define PS_ALL_VALID 0x2 466 #define PS_NONE_BUSY 0x4 467 468 void vm_page_busy_downgrade(vm_page_t m); 469 void vm_page_busy_sleep(vm_page_t m, const char *msg, bool nonshared); 470 void vm_page_flash(vm_page_t m); 471 void vm_page_hold(vm_page_t mem); 472 void vm_page_unhold(vm_page_t mem); 473 void vm_page_free(vm_page_t m); 474 void vm_page_free_zero(vm_page_t m); 475 476 void vm_page_activate (vm_page_t); 477 void vm_page_advise(vm_page_t m, int advice); 478 vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int); 479 vm_page_t vm_page_alloc_after(vm_object_t, vm_pindex_t, int, vm_page_t); 480 vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, 481 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, 482 vm_paddr_t boundary, vm_memattr_t memattr); 483 vm_page_t vm_page_alloc_freelist(int, int); 484 void vm_page_change_lock(vm_page_t m, struct mtx **mtx); 485 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int); 486 int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags, 487 vm_page_t *ma, int count); 488 void vm_page_deactivate (vm_page_t); 489 void vm_page_deactivate_noreuse(vm_page_t); 490 void vm_page_dequeue(vm_page_t m); 491 void vm_page_dequeue_locked(vm_page_t m); 492 vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t); 493 void vm_page_free_phys_pglist(struct pglist *tq); 494 bool vm_page_free_prep(vm_page_t m, bool pagequeue_locked); 495 vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr); 496 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); 497 int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t); 498 void vm_page_launder(vm_page_t m); 499 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t); 500 vm_page_t vm_page_next(vm_page_t m); 501 int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *); 502 struct vm_pagequeue *vm_page_pagequeue(vm_page_t m); 503 vm_page_t vm_page_prev(vm_page_t m); 504 bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m); 505 void vm_page_putfake(vm_page_t m); 506 void vm_page_readahead_finish(vm_page_t m); 507 bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, 508 vm_paddr_t high, u_long alignment, vm_paddr_t boundary); 509 void vm_page_reference(vm_page_t m); 510 void vm_page_remove (vm_page_t); 511 int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t); 512 vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object, 513 vm_pindex_t pindex); 514 void vm_page_requeue(vm_page_t m); 515 void vm_page_requeue_locked(vm_page_t m); 516 int vm_page_sbusied(vm_page_t m); 517 vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start, 518 vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options); 519 void vm_page_set_valid_range(vm_page_t m, int base, int size); 520 int vm_page_sleep_if_busy(vm_page_t m, const char *msg); 521 vm_offset_t vm_page_startup(vm_offset_t vaddr); 522 void vm_page_sunbusy(vm_page_t m); 523 bool vm_page_try_to_free(vm_page_t m); 524 int vm_page_trysbusy(vm_page_t m); 525 void vm_page_unhold_pages(vm_page_t *ma, int count); 526 void vm_page_unswappable(vm_page_t m); 527 boolean_t vm_page_unwire(vm_page_t m, uint8_t queue); 528 void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); 529 void vm_page_wire (vm_page_t); 530 void vm_page_xunbusy_hard(vm_page_t m); 531 void vm_page_xunbusy_maybelocked(vm_page_t m); 532 void vm_page_set_validclean (vm_page_t, int, int); 533 void vm_page_clear_dirty (vm_page_t, int, int); 534 void vm_page_set_invalid (vm_page_t, int, int); 535 int vm_page_is_valid (vm_page_t, int, int); 536 void vm_page_test_dirty (vm_page_t); 537 vm_page_bits_t vm_page_bits(int base, int size); 538 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 539 void vm_page_free_toq(vm_page_t m); 540 541 void vm_page_dirty_KBI(vm_page_t m); 542 void vm_page_lock_KBI(vm_page_t m, const char *file, int line); 543 void vm_page_unlock_KBI(vm_page_t m, const char *file, int line); 544 int vm_page_trylock_KBI(vm_page_t m, const char *file, int line); 545 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 546 void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line); 547 void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line); 548 #endif 549 550 #define vm_page_assert_sbusied(m) \ 551 KASSERT(vm_page_sbusied(m), \ 552 ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \ 553 (m), __FILE__, __LINE__)) 554 555 #define vm_page_assert_unbusied(m) \ 556 KASSERT(!vm_page_busied(m), \ 557 ("vm_page_assert_unbusied: page %p busy @ %s:%d", \ 558 (m), __FILE__, __LINE__)) 559 560 #define vm_page_assert_xbusied(m) \ 561 KASSERT(vm_page_xbusied(m), \ 562 ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \ 563 (m), __FILE__, __LINE__)) 564 565 #define vm_page_busied(m) \ 566 ((m)->busy_lock != VPB_UNBUSIED) 567 568 #define vm_page_sbusy(m) do { \ 569 if (!vm_page_trysbusy(m)) \ 570 panic("%s: page %p failed shared busying", __func__, \ 571 (m)); \ 572 } while (0) 573 574 #define vm_page_tryxbusy(m) \ 575 (atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED, \ 576 VPB_SINGLE_EXCLUSIVER)) 577 578 #define vm_page_xbusied(m) \ 579 (((m)->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0) 580 581 #define vm_page_xbusy(m) do { \ 582 if (!vm_page_tryxbusy(m)) \ 583 panic("%s: page %p failed exclusive busying", __func__, \ 584 (m)); \ 585 } while (0) 586 587 /* Note: page m's lock must not be owned by the caller. */ 588 #define vm_page_xunbusy(m) do { \ 589 if (!atomic_cmpset_rel_int(&(m)->busy_lock, \ 590 VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED)) \ 591 vm_page_xunbusy_hard(m); \ 592 } while (0) 593 594 #ifdef INVARIANTS 595 void vm_page_object_lock_assert(vm_page_t m); 596 #define VM_PAGE_OBJECT_LOCK_ASSERT(m) vm_page_object_lock_assert(m) 597 void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits); 598 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) \ 599 vm_page_assert_pga_writeable(m, bits) 600 #else 601 #define VM_PAGE_OBJECT_LOCK_ASSERT(m) (void)0 602 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) (void)0 603 #endif 604 605 /* 606 * We want to use atomic updates for the aflags field, which is 8 bits wide. 607 * However, not all architectures support atomic operations on 8-bit 608 * destinations. In order that we can easily use a 32-bit operation, we 609 * require that the aflags field be 32-bit aligned. 610 */ 611 CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0); 612 613 /* 614 * Clear the given bits in the specified page. 615 */ 616 static inline void 617 vm_page_aflag_clear(vm_page_t m, uint8_t bits) 618 { 619 uint32_t *addr, val; 620 621 /* 622 * The PGA_REFERENCED flag can only be cleared if the page is locked. 623 */ 624 if ((bits & PGA_REFERENCED) != 0) 625 vm_page_assert_locked(m); 626 627 /* 628 * Access the whole 32-bit word containing the aflags field with an 629 * atomic update. Parallel non-atomic updates to the other fields 630 * within this word are handled properly by the atomic update. 631 */ 632 addr = (void *)&m->aflags; 633 KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0, 634 ("vm_page_aflag_clear: aflags is misaligned")); 635 val = bits; 636 #if BYTE_ORDER == BIG_ENDIAN 637 val <<= 24; 638 #endif 639 atomic_clear_32(addr, val); 640 } 641 642 /* 643 * Set the given bits in the specified page. 644 */ 645 static inline void 646 vm_page_aflag_set(vm_page_t m, uint8_t bits) 647 { 648 uint32_t *addr, val; 649 650 VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits); 651 652 /* 653 * Access the whole 32-bit word containing the aflags field with an 654 * atomic update. Parallel non-atomic updates to the other fields 655 * within this word are handled properly by the atomic update. 656 */ 657 addr = (void *)&m->aflags; 658 KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0, 659 ("vm_page_aflag_set: aflags is misaligned")); 660 val = bits; 661 #if BYTE_ORDER == BIG_ENDIAN 662 val <<= 24; 663 #endif 664 atomic_set_32(addr, val); 665 } 666 667 /* 668 * vm_page_dirty: 669 * 670 * Set all bits in the page's dirty field. 671 * 672 * The object containing the specified page must be locked if the 673 * call is made from the machine-independent layer. 674 * 675 * See vm_page_clear_dirty_mask(). 676 */ 677 static __inline void 678 vm_page_dirty(vm_page_t m) 679 { 680 681 /* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */ 682 #if defined(KLD_MODULE) || defined(INVARIANTS) 683 vm_page_dirty_KBI(m); 684 #else 685 m->dirty = VM_PAGE_BITS_ALL; 686 #endif 687 } 688 689 /* 690 * vm_page_remque: 691 * 692 * If the given page is in a page queue, then remove it from that page 693 * queue. 694 * 695 * The page must be locked. 696 */ 697 static inline void 698 vm_page_remque(vm_page_t m) 699 { 700 701 if (m->queue != PQ_NONE) 702 vm_page_dequeue(m); 703 } 704 705 /* 706 * vm_page_undirty: 707 * 708 * Set page to not be dirty. Note: does not clear pmap modify bits 709 */ 710 static __inline void 711 vm_page_undirty(vm_page_t m) 712 { 713 714 VM_PAGE_OBJECT_LOCK_ASSERT(m); 715 m->dirty = 0; 716 } 717 718 static inline void 719 vm_page_replace_checked(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex, 720 vm_page_t mold) 721 { 722 vm_page_t mret; 723 724 mret = vm_page_replace(mnew, object, pindex); 725 KASSERT(mret == mold, 726 ("invalid page replacement, mold=%p, mret=%p", mold, mret)); 727 728 /* Unused if !INVARIANTS. */ 729 (void)mold; 730 (void)mret; 731 } 732 733 static inline bool 734 vm_page_active(vm_page_t m) 735 { 736 737 return (m->queue == PQ_ACTIVE); 738 } 739 740 static inline bool 741 vm_page_inactive(vm_page_t m) 742 { 743 744 return (m->queue == PQ_INACTIVE); 745 } 746 747 static inline bool 748 vm_page_in_laundry(vm_page_t m) 749 { 750 751 return (m->queue == PQ_LAUNDRY || m->queue == PQ_UNSWAPPABLE); 752 } 753 754 #endif /* _KERNEL */ 755 #endif /* !_VM_PAGE_ */ 756