vm_page.h revision 49720
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 1.65 1999/08/12 21:16:53 alc Exp $ 65 */ 66 67/* 68 * Resident memory system definitions. 69 */ 70 71#ifndef _VM_PAGE_ 72#define _VM_PAGE_ 73 74#if !defined(KLD_MODULE) 75#include "opt_vmpage.h" 76#endif 77 78#include <vm/pmap.h> 79#include <machine/atomic.h> 80 81/* 82 * Management of resident (logical) pages. 83 * 84 * A small structure is kept for each resident 85 * page, indexed by page number. Each structure 86 * is an element of several lists: 87 * 88 * A hash table bucket used to quickly 89 * perform object/offset lookups 90 * 91 * A list of all pages for a given object, 92 * so they can be quickly deactivated at 93 * time of deallocation. 94 * 95 * An ordered list of pages due for pageout. 96 * 97 * In addition, the structure contains the object 98 * and offset to which this page belongs (for pageout), 99 * and sundry status bits. 100 * 101 * Fields in this structure are locked either by the lock on the 102 * object that the page belongs to (O) or by the lock on the page 103 * queues (P). 104 * 105 * The 'valid' and 'dirty' fields are distinct. A page may have dirty 106 * bits set without having associated valid bits set. This is used by 107 * NFS to implement piecemeal writes. 108 */ 109 110TAILQ_HEAD(pglist, vm_page); 111 112struct vm_page { 113 TAILQ_ENTRY(vm_page) pageq; /* queue info for FIFO queue or free list (P) */ 114 struct vm_page *hnext; /* hash table link (O,P) */ 115 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */ 116 117 vm_object_t object; /* which object am I in (O,P)*/ 118 vm_pindex_t pindex; /* offset into object (O,P) */ 119 vm_offset_t phys_addr; /* physical address of page */ 120 u_short queue; /* page queue index */ 121 u_short flags, /* see below */ 122 pc; /* page color */ 123 u_short wire_count; /* wired down maps refs (P) */ 124 short hold_count; /* page hold count */ 125 u_char act_count; /* page usage count */ 126 u_char busy; /* page busy count */ 127 /* NOTE that these must support one bit per DEV_BSIZE in a page!!! */ 128 /* so, on normal X86 kernels, they must be at least 8 bits wide */ 129#if PAGE_SIZE == 4096 130 u_char valid; /* map of valid DEV_BSIZE chunks */ 131 u_char dirty; /* map of dirty DEV_BSIZE chunks */ 132#elif PAGE_SIZE == 8192 133 u_short valid; /* map of valid DEV_BSIZE chunks */ 134 u_short dirty; /* map of dirty DEV_BSIZE chunks */ 135#endif 136}; 137 138/* 139 * note SWAPBLK_NONE is a flag, basically the high bit. 140 */ 141 142#define SWAPBLK_MASK ((daddr_t)((u_daddr_t)-1 >> 1)) /* mask */ 143#define SWAPBLK_NONE ((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */ 144 145#if !defined(KLD_MODULE) 146 147/* 148 * Page coloring parameters 149 */ 150/* Each of PQ_FREE, and PQ_CACHE have PQ_HASH_SIZE entries */ 151 152/* Define one of the following */ 153#if defined(PQ_HUGECACHE) 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_L2_SIZE 256 /* A number of colors opt for 1M cache */ 157#endif 158 159/* Define one of the following */ 160#if defined(PQ_LARGECACHE) 161#define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */ 162#define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */ 163#define PQ_L2_SIZE 128 /* A number of colors opt for 512K cache */ 164#endif 165 166 167/* 168 * Use 'options PQ_NOOPT' to disable page coloring 169 */ 170#if defined(PQ_NOOPT) 171#define PQ_PRIME1 1 172#define PQ_PRIME2 1 173#define PQ_L2_SIZE 1 174#endif 175 176#if defined(PQ_NORMALCACHE) 177#define PQ_PRIME1 5 /* Prime number somewhat less than PQ_HASH_SIZE */ 178#define PQ_PRIME2 3 /* Prime number somewhat less than PQ_HASH_SIZE */ 179#define PQ_L2_SIZE 16 /* A reasonable number of colors (opt for 64K cache) */ 180#endif 181 182#if defined(PQ_MEDIUMCACHE) 183#define PQ_PRIME1 13 /* Prime number somewhat less than PQ_HASH_SIZE */ 184#define PQ_PRIME2 7 /* Prime number somewhat less than PQ_HASH_SIZE */ 185#define PQ_L2_SIZE 64 /* A number of colors opt for 256K cache */ 186#endif 187 188#if !defined(PQ_L2_SIZE) 189#define PQ_PRIME1 9 /* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */ 190#define PQ_PRIME2 5 /* Prime number somewhat less than PQ_HASH_SIZE */ 191#define PQ_L2_SIZE 32 /* 512KB or smaller, 4-way set-associative cache */ 192#endif 193 194#define PQ_L2_MASK (PQ_L2_SIZE - 1) 195 196#define PQ_NONE PQ_COUNT 197#define PQ_FREE 0 198#define PQ_INACTIVE PQ_L2_SIZE 199#define PQ_ACTIVE (1 + PQ_L2_SIZE) 200#define PQ_CACHE (2 + PQ_L2_SIZE) 201#define PQ_COUNT (2 + 2*PQ_L2_SIZE) 202 203extern struct vpgqueues { 204 struct pglist *pl; 205 int *cnt; 206 int lcnt; 207} vm_page_queues[PQ_COUNT]; 208 209#endif 210 211/* 212 * These are the flags defined for vm_page. 213 * 214 * Note: PG_FILLED and PG_DIRTY are added for the filesystems. 215 */ 216#define PG_BUSY 0x0001 /* page is in transit (O) */ 217#define PG_WANTED 0x0002 /* someone is waiting for page (O) */ 218#define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */ 219#define PG_WRITEABLE 0x0010 /* page is mapped writeable */ 220#define PG_MAPPED 0x0020 /* page is mapped */ 221#define PG_ZERO 0x0040 /* page is zeroed */ 222#define PG_REFERENCED 0x0080 /* page has been referenced */ 223#define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */ 224#define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */ 225 226/* 227 * Misc constants. 228 */ 229 230#define ACT_DECLINE 1 231#define ACT_ADVANCE 3 232#define ACT_INIT 5 233#define ACT_MAX 64 234#define PFCLUSTER_BEHIND 3 235#define PFCLUSTER_AHEAD 3 236 237#ifdef KERNEL 238/* 239 * Each pageable resident page falls into one of four lists: 240 * 241 * free 242 * Available for allocation now. 243 * 244 * The following are all LRU sorted: 245 * 246 * cache 247 * Almost available for allocation. Still in an 248 * object, but clean and immediately freeable at 249 * non-interrupt times. 250 * 251 * inactive 252 * Low activity, candidates for reclamation. 253 * This is the list of pages that should be 254 * paged out next. 255 * 256 * active 257 * Pages that are "active" i.e. they have been 258 * recently referenced. 259 * 260 * zero 261 * Pages that are really free and have been pre-zeroed 262 * 263 */ 264 265#if !defined(KLD_MODULE) 266 267extern struct pglist vm_page_queue_free[PQ_L2_SIZE];/* memory free queue */ 268extern struct pglist vm_page_queue_active; /* active memory queue */ 269extern struct pglist vm_page_queue_inactive; /* inactive memory queue */ 270extern struct pglist vm_page_queue_cache[PQ_L2_SIZE];/* cache memory queue */ 271 272#endif 273 274extern int vm_page_zero_count; 275 276extern vm_page_t vm_page_array; /* First resident page in table */ 277extern long first_page; /* first physical page number */ 278 279#define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 280 281#define PHYS_TO_VM_PAGE(pa) \ 282 (&vm_page_array[atop(pa) - first_page ]) 283 284/* 285 * Functions implemented as macros 286 */ 287 288static __inline void 289vm_page_flag_set(vm_page_t m, unsigned int bits) 290{ 291 atomic_set_short(&(m)->flags, bits); 292} 293 294static __inline void 295vm_page_flag_clear(vm_page_t m, unsigned int bits) 296{ 297 atomic_clear_short(&(m)->flags, bits); 298} 299 300#if 0 301static __inline void 302vm_page_assert_wait(vm_page_t m, int interruptible) 303{ 304 vm_page_flag_set(m, PG_WANTED); 305 assert_wait((int) m, interruptible); 306} 307#endif 308 309static __inline void 310vm_page_busy(vm_page_t m) 311{ 312 KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!")); 313 vm_page_flag_set(m, PG_BUSY); 314} 315 316/* 317 * vm_page_flash: 318 * 319 * wakeup anyone waiting for the page. 320 */ 321 322static __inline void 323vm_page_flash(vm_page_t m) 324{ 325 if (m->flags & PG_WANTED) { 326 vm_page_flag_clear(m, PG_WANTED); 327 wakeup(m); 328 } 329} 330 331/* 332 * vm_page_wakeup: 333 * 334 * clear the PG_BUSY flag and wakeup anyone waiting for the 335 * page. 336 * 337 */ 338 339static __inline void 340vm_page_wakeup(vm_page_t m) 341{ 342 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!")); 343 vm_page_flag_clear(m, PG_BUSY); 344 vm_page_flash(m); 345} 346 347/* 348 * 349 * 350 */ 351 352static __inline void 353vm_page_io_start(vm_page_t m) 354{ 355 atomic_add_char(&(m)->busy, 1); 356} 357 358static __inline void 359vm_page_io_finish(vm_page_t m) 360{ 361 atomic_subtract_char(&m->busy, 1); 362 if (m->busy == 0) 363 vm_page_flash(m); 364} 365 366 367#if PAGE_SIZE == 4096 368#define VM_PAGE_BITS_ALL 0xff 369#endif 370 371#if PAGE_SIZE == 8192 372#define VM_PAGE_BITS_ALL 0xffff 373#endif 374 375#define VM_ALLOC_NORMAL 0 376#define VM_ALLOC_INTERRUPT 1 377#define VM_ALLOC_SYSTEM 2 378#define VM_ALLOC_ZERO 3 379#define VM_ALLOC_RETRY 0x80 380 381void vm_page_activate __P((vm_page_t)); 382vm_page_t vm_page_alloc __P((vm_object_t, vm_pindex_t, int)); 383vm_page_t vm_page_grab __P((vm_object_t, vm_pindex_t, int)); 384void vm_page_cache __P((register vm_page_t)); 385static __inline void vm_page_copy __P((vm_page_t, vm_page_t)); 386static __inline void vm_page_free __P((vm_page_t)); 387static __inline void vm_page_free_zero __P((vm_page_t)); 388void vm_page_deactivate __P((vm_page_t)); 389void vm_page_insert __P((vm_page_t, vm_object_t, vm_pindex_t)); 390vm_page_t vm_page_lookup __P((vm_object_t, vm_pindex_t)); 391void vm_page_remove __P((vm_page_t)); 392void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t)); 393vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t)); 394void vm_page_unwire __P((vm_page_t, int)); 395void vm_page_wire __P((vm_page_t)); 396void vm_page_unqueue __P((vm_page_t)); 397void vm_page_unqueue_nowakeup __P((vm_page_t)); 398void vm_page_set_validclean __P((vm_page_t, int, int)); 399void vm_page_set_dirty __P((vm_page_t, int, int)); 400void vm_page_clear_dirty __P((vm_page_t, int, int)); 401void vm_page_set_invalid __P((vm_page_t, int, int)); 402static __inline boolean_t vm_page_zero_fill __P((vm_page_t)); 403int vm_page_is_valid __P((vm_page_t, int, int)); 404void vm_page_test_dirty __P((vm_page_t)); 405int vm_page_bits __P((int, int)); 406vm_page_t _vm_page_list_find __P((int, int)); 407#if 0 408int vm_page_sleep(vm_page_t m, char *msg, char *busy); 409int vm_page_asleep(vm_page_t m, char *msg, char *busy); 410#endif 411void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 412void vm_page_free_toq(vm_page_t m); 413 414/* 415 * Keep page from being freed by the page daemon 416 * much of the same effect as wiring, except much lower 417 * overhead and should be used only for *very* temporary 418 * holding ("wiring"). 419 */ 420static __inline void 421vm_page_hold(vm_page_t mem) 422{ 423 mem->hold_count++; 424} 425 426static __inline void 427vm_page_unhold(vm_page_t mem) 428{ 429 --mem->hold_count; 430 KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!")); 431} 432 433/* 434 * vm_page_protect: 435 * 436 * Reduce the protection of a page. This routine never 437 * raises the protection and therefore can be safely 438 * called if the page is already at VM_PROT_NONE ( it 439 * will be a NOP effectively ). 440 */ 441 442static __inline void 443vm_page_protect(vm_page_t mem, int prot) 444{ 445 if (prot == VM_PROT_NONE) { 446 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) { 447 pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_NONE); 448 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED); 449 } 450 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) { 451 pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_READ); 452 vm_page_flag_clear(mem, PG_WRITEABLE); 453 } 454} 455 456/* 457 * vm_page_zero_fill: 458 * 459 * Zero-fill the specified page. 460 * Written as a standard pagein routine, to 461 * be used by the zero-fill object. 462 */ 463static __inline boolean_t 464vm_page_zero_fill(m) 465 vm_page_t m; 466{ 467 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 468 return (TRUE); 469} 470 471/* 472 * vm_page_copy: 473 * 474 * Copy one page to another 475 */ 476static __inline void 477vm_page_copy(src_m, dest_m) 478 vm_page_t src_m; 479 vm_page_t dest_m; 480{ 481 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 482 dest_m->valid = VM_PAGE_BITS_ALL; 483} 484 485/* 486 * vm_page_free: 487 * 488 * Free a page 489 * 490 * The clearing of PG_ZERO is a temporary safety until the code can be 491 * reviewed to determine that PG_ZERO is being properly cleared on 492 * write faults or maps. PG_ZERO was previously cleared in 493 * vm_page_alloc(). 494 */ 495static __inline void 496vm_page_free(m) 497 vm_page_t m; 498{ 499 vm_page_flag_clear(m, PG_ZERO); 500 vm_page_free_toq(m); 501} 502 503/* 504 * vm_page_free_zero: 505 * 506 * Free a page to the zerod-pages queue 507 */ 508static __inline void 509vm_page_free_zero(m) 510 vm_page_t m; 511{ 512 vm_page_flag_set(m, PG_ZERO); 513 vm_page_free_toq(m); 514} 515 516/* 517 * vm_page_sleep_busy: 518 * 519 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE) 520 * m->busy is zero. Returns TRUE if it had to sleep ( including if 521 * it almost had to sleep and made temporary spl*() mods), FALSE 522 * otherwise. 523 * 524 * This routine assumes that interrupts can only remove the busy 525 * status from a page, not set the busy status or change it from 526 * PG_BUSY to m->busy or vise versa (which would create a timing 527 * window). 528 * 529 * Note that being an inline, this code will be well optimized. 530 */ 531 532static __inline int 533vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg) 534{ 535 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 536 int s = splvm(); 537 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 538 /* 539 * Page is busy. Wait and retry. 540 */ 541 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED); 542 tsleep(m, PVM, msg, 0); 543 } 544 splx(s); 545 return(TRUE); 546 /* not reached */ 547 } 548 return(FALSE); 549} 550 551#if !defined(KLD_MODULE) 552 553/* 554 * vm_page_dirty: 555 * 556 * make page all dirty 557 */ 558 559static __inline void 560vm_page_dirty(vm_page_t m) 561{ 562 KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!")); 563 m->dirty = VM_PAGE_BITS_ALL; 564} 565 566static __inline vm_page_t 567vm_page_list_find(int basequeue, int index, boolean_t prefer_zero) 568{ 569 vm_page_t m; 570 571#if PQ_L2_SIZE > 1 572 if (prefer_zero) { 573 m = TAILQ_LAST(vm_page_queues[basequeue+index].pl, pglist); 574 } else { 575 m = TAILQ_FIRST(vm_page_queues[basequeue+index].pl); 576 } 577 if (m == NULL) 578 m = _vm_page_list_find(basequeue, index); 579#else 580 if (prefer_zero) { 581 m = TAILQ_LAST(vm_page_queues[basequeue].pl, pglist); 582 } else { 583 m = TAILQ_FIRST(vm_page_queues[basequeue].pl); 584 } 585#endif 586 return(m); 587} 588 589#endif 590 591#endif /* KERNEL */ 592#endif /* !_VM_PAGE_ */ 593