vm_page.h revision 43752
1/* 2 * Copyright (c) 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * The Mach Operating System project at Carnegie-Mellon University. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: @(#)vm_page.h 8.2 (Berkeley) 12/13/93 37 * 38 * 39 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40 * All rights reserved. 41 * 42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 43 * 44 * Permission to use, copy, modify and distribute this software and 45 * its documentation is hereby granted, provided that both the copyright 46 * notice and this permission notice appear in all copies of the 47 * software, derivative works or modified versions, and any portions 48 * thereof, and that both notices appear in supporting documentation. 49 * 50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 53 * 54 * Carnegie Mellon requests users of this software to return to 55 * 56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57 * School of Computer Science 58 * Carnegie Mellon University 59 * Pittsburgh PA 15213-3890 60 * 61 * any improvements or extensions that they make and grant Carnegie the 62 * rights to redistribute these changes. 63 * 64 * $Id: vm_page.h,v 1.54 1999/02/07 20:45:15 dillon Exp $ 65 */ 66 67/* 68 * Resident memory system definitions. 69 */ 70 71#ifndef _VM_PAGE_ 72#define _VM_PAGE_ 73 74#include "opt_vmpage.h" 75 76#include <vm/pmap.h> 77#include <machine/atomic.h> 78 79/* 80 * Management of resident (logical) pages. 81 * 82 * A small structure is kept for each resident 83 * page, indexed by page number. Each structure 84 * is an element of several lists: 85 * 86 * A hash table bucket used to quickly 87 * perform object/offset lookups 88 * 89 * A list of all pages for a given object, 90 * so they can be quickly deactivated at 91 * time of deallocation. 92 * 93 * An ordered list of pages due for pageout. 94 * 95 * In addition, the structure contains the object 96 * and offset to which this page belongs (for pageout), 97 * and sundry status bits. 98 * 99 * Fields in this structure are locked either by the lock on the 100 * object that the page belongs to (O) or by the lock on the page 101 * queues (P). 102 */ 103 104TAILQ_HEAD(pglist, vm_page); 105 106struct vm_page { 107 TAILQ_ENTRY(vm_page) pageq; /* queue info for FIFO queue or free list (P) */ 108 struct vm_page *hnext; /* hash table link (O,P) */ 109 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */ 110 111 vm_object_t object; /* which object am I in (O,P)*/ 112 vm_pindex_t pindex; /* offset into object (O,P) */ 113 vm_offset_t phys_addr; /* physical address of page */ 114 u_short queue; /* page queue index */ 115 u_short flags, /* see below */ 116 pc; /* page color */ 117 u_short wire_count; /* wired down maps refs (P) */ 118 short hold_count; /* page hold count */ 119 u_char act_count; /* page usage count */ 120 u_char busy; /* page busy count */ 121 /* NOTE that these must support one bit per DEV_BSIZE in a page!!! */ 122 /* so, on normal X86 kernels, they must be at least 8 bits wide */ 123#if PAGE_SIZE == 4096 124 u_char valid; /* map of valid DEV_BSIZE chunks */ 125 u_char dirty; /* map of dirty DEV_BSIZE chunks */ 126#elif PAGE_SIZE == 8192 127 u_short valid; /* map of valid DEV_BSIZE chunks */ 128 u_short dirty; /* map of dirty DEV_BSIZE chunks */ 129#endif 130}; 131 132/* 133 * note SWAPBLK_NONE is a flag, basically the high bit. 134 */ 135 136#define SWAPBLK_MASK ((daddr_t)((u_daddr_t)-1 >> 1)) /* mask */ 137#define SWAPBLK_NONE ((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */ 138 139/* 140 * Page coloring parameters 141 */ 142/* Each of PQ_FREE, and PQ_CACHE have PQ_HASH_SIZE entries */ 143 144/* Define one of the following */ 145#if defined(PQ_HUGECACHE) 146#define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */ 147#define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */ 148#define PQ_PRIME3 17 /* Prime number somewhat less than PQ_HASH_SIZE */ 149#define PQ_L2_SIZE 256 /* A number of colors opt for 1M cache */ 150#endif 151 152/* Define one of the following */ 153#if defined(PQ_LARGECACHE) 154#define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */ 155#define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */ 156#define PQ_PRIME3 17 /* Prime number somewhat less than PQ_HASH_SIZE */ 157#define PQ_L2_SIZE 128 /* A number of colors opt for 512K cache */ 158#endif 159 160 161/* 162 * Use 'options PQ_NOOPT' to disable page coloring 163 */ 164#if defined(PQ_NOOPT) 165#define PQ_PRIME1 1 166#define PQ_PRIME2 1 167#define PQ_PRIME3 1 168#define PQ_L2_SIZE 1 169#endif 170 171#if defined(PQ_NORMALCACHE) 172#define PQ_PRIME1 5 /* Prime number somewhat less than PQ_HASH_SIZE */ 173#define PQ_PRIME2 3 /* Prime number somewhat less than PQ_HASH_SIZE */ 174#define PQ_PRIME3 11 /* Prime number somewhat less than PQ_HASH_SIZE */ 175#define PQ_L2_SIZE 16 /* A reasonable number of colors (opt for 64K cache) */ 176#endif 177 178#if defined(PQ_MEDIUMCACHE) || !defined(PQ_L2_SIZE) 179#define PQ_PRIME1 13 /* Prime number somewhat less than PQ_HASH_SIZE */ 180#define PQ_PRIME2 7 /* Prime number somewhat less than PQ_HASH_SIZE */ 181#define PQ_PRIME3 5 /* Prime number somewhat less than PQ_HASH_SIZE */ 182#define PQ_L2_SIZE 64 /* A number of colors opt for 256K cache */ 183#endif 184 185#define PQ_L2_MASK (PQ_L2_SIZE - 1) 186 187#define PQ_NONE 0 188#define PQ_FREE 1 189/* #define PQ_ZERO (1 + PQ_L2_SIZE) */ 190#define PQ_INACTIVE (1 + 1*PQ_L2_SIZE) 191#define PQ_ACTIVE (2 + 1*PQ_L2_SIZE) 192#define PQ_CACHE (3 + 1*PQ_L2_SIZE) 193#define PQ_COUNT (3 + 2*PQ_L2_SIZE) 194 195extern struct vpgqueues { 196 struct pglist *pl; 197 int *cnt; 198 int *lcnt; 199} vm_page_queues[PQ_COUNT]; 200 201/* 202 * These are the flags defined for vm_page. 203 * 204 * Note: PG_FILLED and PG_DIRTY are added for the filesystems. 205 */ 206#define PG_BUSY 0x0001 /* page is in transit (O) */ 207#define PG_WANTED 0x0002 /* someone is waiting for page (O) */ 208#define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */ 209#define PG_WRITEABLE 0x0010 /* page is mapped writeable */ 210#define PG_MAPPED 0x0020 /* page is mapped */ 211#define PG_ZERO 0x0040 /* page is zeroed */ 212#define PG_REFERENCED 0x0080 /* page has been referenced */ 213#define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */ 214#define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */ 215 216/* 217 * Misc constants. 218 */ 219 220#define ACT_DECLINE 1 221#define ACT_ADVANCE 3 222#define ACT_INIT 5 223#define ACT_MAX 64 224#define PFCLUSTER_BEHIND 3 225#define PFCLUSTER_AHEAD 3 226 227#ifdef KERNEL 228/* 229 * Each pageable resident page falls into one of four lists: 230 * 231 * free 232 * Available for allocation now. 233 * 234 * The following are all LRU sorted: 235 * 236 * cache 237 * Almost available for allocation. Still in an 238 * object, but clean and immediately freeable at 239 * non-interrupt times. 240 * 241 * inactive 242 * Low activity, candidates for reclamation. 243 * This is the list of pages that should be 244 * paged out next. 245 * 246 * active 247 * Pages that are "active" i.e. they have been 248 * recently referenced. 249 * 250 * zero 251 * Pages that are really free and have been pre-zeroed 252 * 253 */ 254 255extern struct pglist vm_page_queue_free[PQ_L2_SIZE];/* memory free queue */ 256extern struct pglist vm_page_queue_active; /* active memory queue */ 257extern struct pglist vm_page_queue_inactive; /* inactive memory queue */ 258extern struct pglist vm_page_queue_cache[PQ_L2_SIZE];/* cache memory queue */ 259 260extern int vm_page_zero_count; 261 262extern vm_page_t vm_page_array; /* First resident page in table */ 263extern long first_page; /* first physical page number */ 264 265 /* ... represented in vm_page_array */ 266extern long last_page; /* last physical page number */ 267 268 /* ... represented in vm_page_array */ 269 /* [INCLUSIVE] */ 270extern vm_offset_t first_phys_addr; /* physical address for first_page */ 271extern vm_offset_t last_phys_addr; /* physical address for last_page */ 272 273#define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 274 275#define IS_VM_PHYSADDR(pa) \ 276 ((pa) >= first_phys_addr && (pa) <= last_phys_addr) 277 278#define PHYS_TO_VM_PAGE(pa) \ 279 (&vm_page_array[atop(pa) - first_page ]) 280 281/* 282 * Functions implemented as macros 283 */ 284 285static __inline void 286vm_page_flag_set(vm_page_t m, unsigned int bits) 287{ 288 atomic_set_short(&(m)->flags, bits); 289} 290 291static __inline void 292vm_page_flag_clear(vm_page_t m, unsigned int bits) 293{ 294 atomic_clear_short(&(m)->flags, bits); 295} 296 297#if 0 298static __inline void 299vm_page_assert_wait(vm_page_t m, int interruptible) 300{ 301 vm_page_flag_set(m, PG_WANTED); 302 assert_wait((int) m, interruptible); 303} 304#endif 305 306static __inline void 307vm_page_busy(vm_page_t m) 308{ 309 KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!")); 310 vm_page_flag_set(m, PG_BUSY); 311} 312 313/* 314 * vm_page_flash: 315 * 316 * wakeup anyone waiting for the page. 317 */ 318 319static __inline void 320vm_page_flash(vm_page_t m) 321{ 322 if (m->flags & PG_WANTED) { 323 vm_page_flag_clear(m, PG_WANTED); 324 wakeup(m); 325 } 326} 327 328/* 329 * vm_page_wakeup: 330 * 331 * clear the PG_BUSY flag and wakeup anyone waiting for the 332 * page. 333 * 334 */ 335 336static __inline void 337vm_page_wakeup(vm_page_t m) 338{ 339 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!")); 340 vm_page_flag_clear(m, PG_BUSY); 341 vm_page_flash(m); 342} 343 344/* 345 * 346 * 347 */ 348 349static __inline void 350vm_page_io_start(vm_page_t m) 351{ 352 atomic_add_char(&(m)->busy, 1); 353} 354 355static __inline void 356vm_page_io_finish(vm_page_t m) 357{ 358 atomic_subtract_char(&m->busy, 1); 359 if (m->busy == 0) 360 vm_page_flash(m); 361} 362 363 364#if PAGE_SIZE == 4096 365#define VM_PAGE_BITS_ALL 0xff 366#endif 367 368#if PAGE_SIZE == 8192 369#define VM_PAGE_BITS_ALL 0xffff 370#endif 371 372#define VM_ALLOC_NORMAL 0 373#define VM_ALLOC_INTERRUPT 1 374#define VM_ALLOC_SYSTEM 2 375#define VM_ALLOC_ZERO 3 376#define VM_ALLOC_RETRY 0x80 377 378void vm_page_activate __P((vm_page_t)); 379vm_page_t vm_page_alloc __P((vm_object_t, vm_pindex_t, int)); 380vm_page_t vm_page_grab __P((vm_object_t, vm_pindex_t, int)); 381void vm_page_cache __P((register vm_page_t)); 382static __inline void vm_page_copy __P((vm_page_t, vm_page_t)); 383static __inline void vm_page_free __P((vm_page_t)); 384static __inline void vm_page_free_zero __P((vm_page_t)); 385void vm_page_destroy __P((vm_page_t)); 386void vm_page_deactivate __P((vm_page_t)); 387void vm_page_insert __P((vm_page_t, vm_object_t, vm_pindex_t)); 388vm_page_t vm_page_lookup __P((vm_object_t, vm_pindex_t)); 389vm_object_t vm_page_remove __P((vm_page_t)); 390void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t)); 391vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t)); 392void vm_page_unwire __P((vm_page_t, int)); 393void vm_page_wire __P((vm_page_t)); 394void vm_page_unqueue __P((vm_page_t)); 395void vm_page_unqueue_nowakeup __P((vm_page_t)); 396void vm_page_set_validclean __P((vm_page_t, int, int)); 397void vm_page_set_invalid __P((vm_page_t, int, int)); 398static __inline boolean_t vm_page_zero_fill __P((vm_page_t)); 399int vm_page_is_valid __P((vm_page_t, int, int)); 400void vm_page_test_dirty __P((vm_page_t)); 401int vm_page_bits __P((int, int)); 402vm_page_t _vm_page_list_find __P((int, int)); 403int vm_page_queue_index __P((vm_offset_t, int)); 404#if 0 405int vm_page_sleep(vm_page_t m, char *msg, char *busy); 406int vm_page_asleep(vm_page_t m, char *msg, char *busy); 407#endif 408void vm_page_free_toq(vm_page_t m); 409 410/* 411 * Keep page from being freed by the page daemon 412 * much of the same effect as wiring, except much lower 413 * overhead and should be used only for *very* temporary 414 * holding ("wiring"). 415 */ 416static __inline void 417vm_page_hold(vm_page_t mem) 418{ 419 mem->hold_count++; 420} 421 422static __inline void 423vm_page_unhold(vm_page_t mem) 424{ 425 --mem->hold_count; 426 KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!")); 427} 428 429/* 430 * vm_page_protect: 431 * 432 * Reduce the protection of a page. This routine never 433 * raises the protection and therefore can be safely 434 * called if the page is already at VM_PROT_NONE ( it 435 * will be a NOP effectively ). 436 */ 437 438static __inline void 439vm_page_protect(vm_page_t mem, int prot) 440{ 441 if (prot == VM_PROT_NONE) { 442 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) { 443 pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_NONE); 444 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED); 445 } 446 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) { 447 pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_READ); 448 vm_page_flag_clear(mem, PG_WRITEABLE); 449 } 450} 451 452/* 453 * vm_page_zero_fill: 454 * 455 * Zero-fill the specified page. 456 * Written as a standard pagein routine, to 457 * be used by the zero-fill object. 458 */ 459static __inline boolean_t 460vm_page_zero_fill(m) 461 vm_page_t m; 462{ 463 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 464 return (TRUE); 465} 466 467/* 468 * vm_page_copy: 469 * 470 * Copy one page to another 471 */ 472static __inline void 473vm_page_copy(src_m, dest_m) 474 vm_page_t src_m; 475 vm_page_t dest_m; 476{ 477 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 478 dest_m->valid = VM_PAGE_BITS_ALL; 479} 480 481/* 482 * vm_page_free: 483 * 484 * Free a page 485 * 486 * The clearing of PG_ZERO is a temporary safety until the code can be 487 * reviewed to determine that PG_ZERO is being properly cleared on 488 * write faults or maps. PG_ZERO was previously cleared in 489 * vm_page_alloc(). 490 */ 491static __inline void 492vm_page_free(m) 493 vm_page_t m; 494{ 495 vm_page_flag_clear(m, PG_ZERO); 496 vm_page_free_toq(m); 497} 498 499/* 500 * vm_page_free_zero: 501 * 502 * Free a page to the zerod-pages queue 503 */ 504static __inline void 505vm_page_free_zero(m) 506 vm_page_t m; 507{ 508 vm_page_flag_set(m, PG_ZERO); 509 vm_page_free_toq(m); 510} 511 512/* 513 * vm_page_sleep_busy: 514 * 515 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE) 516 * m->busy is zero. Returns TRUE if it had to sleep ( including if 517 * it almost had to sleep and made temporary spl*() mods), FALSE 518 * otherwise. 519 * 520 * This routine assumes that interrupts can only remove the busy 521 * status from a page, not set the busy status or change it from 522 * PG_BUSY to m->busy or vise versa (which would create a timing 523 * window). 524 * 525 * Note that being an inline, this code will be well optimized. 526 */ 527 528static __inline int 529vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg) 530{ 531 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 532 int s = splvm(); 533 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 534 /* 535 * Page is busy. Wait and retry. 536 */ 537 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED); 538 tsleep(m, PVM, msg, 0); 539 } 540 splx(s); 541 return(TRUE); 542 /* not reached */ 543 } 544 return(FALSE); 545} 546 547/* 548 * vm_page_dirty: 549 * 550 * make page all dirty 551 */ 552 553static __inline void 554vm_page_dirty(vm_page_t m) 555{ 556 KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!")); 557 m->dirty = VM_PAGE_BITS_ALL; 558} 559 560static __inline vm_page_t 561vm_page_list_find(int basequeue, int index, boolean_t prefer_zero) 562{ 563 vm_page_t m; 564 565#if PQ_L2_SIZE > 1 566 if (prefer_zero) { 567 m = TAILQ_LAST(vm_page_queues[basequeue+index].pl, pglist); 568 } else { 569 m = TAILQ_FIRST(vm_page_queues[basequeue+index].pl); 570 } 571 if (m == NULL) 572 m = _vm_page_list_find(basequeue, index); 573#else 574 if (prefer_zero) { 575 m = TAILQ_LAST(vm_page_queues[basequeue].pl, pglist); 576 } else { 577 m = TAILQ_FIRST(vm_page_queues[basequeue].pl); 578 } 579#endif 580 return(m); 581} 582 583#endif /* KERNEL */ 584#endif /* !_VM_PAGE_ */ 585