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 * $FreeBSD: src/sys/vm/vm_page.h,v 1.75.2.8 2002/03/06 01:07:09 dillon 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 */ 109 110 TAILQ_HEAD(pglist, vm_page); 111 112 struct 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 struct md_page md; /* machine dependant stuff */ 121 u_short queue; /* page queue index */ 122 u_short flags, /* see below */ 123 pc; /* page color */ 124 u_short wire_count; /* wired down maps refs (P) */ 125 short hold_count; /* page hold count */ 126 u_char act_count; /* page usage count */ 127 u_char busy; /* page busy count */ 128 /* NOTE that these must support one bit per DEV_BSIZE in a page!!! */ 129 /* so, on normal X86 kernels, they must be at least 8 bits wide */ 130 #if PAGE_SIZE == 4096 131 u_char valid; /* map of valid DEV_BSIZE chunks */ 132 u_char dirty; /* map of dirty DEV_BSIZE chunks */ 133 #elif PAGE_SIZE == 8192 134 u_short valid; /* map of valid DEV_BSIZE chunks */ 135 u_short dirty; /* map of dirty DEV_BSIZE chunks */ 136 #endif 137 }; 138 139 /* 140 * note: currently use SWAPBLK_NONE as an absolute value rather then 141 * a flag bit. 142 */ 143 144 #define SWAPBLK_MASK ((daddr_t)((u_daddr_t)-1 >> 1)) /* mask */ 145 #define SWAPBLK_NONE ((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */ 146 147 #if !defined(KLD_MODULE) 148 149 /* 150 * Page coloring parameters 151 */ 152 /* Each of PQ_FREE, and PQ_CACHE have PQ_HASH_SIZE entries */ 153 154 /* Backward compatibility for existing PQ_*CACHE config options. */ 155 #if !defined(PQ_CACHESIZE) 156 #if defined(PQ_HUGECACHE) 157 #define PQ_CACHESIZE 1024 158 #elif defined(PQ_LARGECACHE) 159 #define PQ_CACHESIZE 512 160 #elif defined(PQ_MEDIUMCACHE) 161 #define PQ_CACHESIZE 256 162 #elif defined(PQ_NORMALCACHE) 163 #define PQ_CACHESIZE 64 164 #elif defined(PQ_NOOPT) 165 #define PQ_CACHESIZE 0 166 #else 167 #define PQ_CACHESIZE 128 168 #endif 169 #endif 170 171 #if PQ_CACHESIZE >= 1024 172 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */ 173 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */ 174 #define PQ_L2_SIZE 256 /* A number of colors opt for 1M cache */ 175 176 #elif PQ_CACHESIZE >= 512 177 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */ 178 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */ 179 #define PQ_L2_SIZE 128 /* A number of colors opt for 512K cache */ 180 181 #elif PQ_CACHESIZE >= 256 182 #define PQ_PRIME1 13 /* Prime number somewhat less than PQ_HASH_SIZE */ 183 #define PQ_PRIME2 7 /* Prime number somewhat less than PQ_HASH_SIZE */ 184 #define PQ_L2_SIZE 64 /* A number of colors opt for 256K cache */ 185 186 #elif PQ_CACHESIZE >= 128 187 #define PQ_PRIME1 9 /* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */ 188 #define PQ_PRIME2 5 /* Prime number somewhat less than PQ_HASH_SIZE */ 189 #define PQ_L2_SIZE 32 /* A number of colors opt for 128k cache */ 190 191 #elif PQ_CACHESIZE >= 64 192 #define PQ_PRIME1 5 /* Prime number somewhat less than PQ_HASH_SIZE */ 193 #define PQ_PRIME2 3 /* Prime number somewhat less than PQ_HASH_SIZE */ 194 #define PQ_L2_SIZE 16 /* A reasonable number of colors (opt for 64K cache) */ 195 196 #else 197 #define PQ_PRIME1 1 /* Disable page coloring. */ 198 #define PQ_PRIME2 1 199 #define PQ_L2_SIZE 1 200 201 #endif 202 203 #define PQ_L2_MASK (PQ_L2_SIZE - 1) 204 205 #define PQ_NONE 0 206 #define PQ_FREE 1 207 #define PQ_INACTIVE (1 + 1*PQ_L2_SIZE) 208 #define PQ_ACTIVE (2 + 1*PQ_L2_SIZE) 209 #define PQ_CACHE (3 + 1*PQ_L2_SIZE) 210 #define PQ_HOLD (3 + 2*PQ_L2_SIZE) 211 #define PQ_COUNT (4 + 2*PQ_L2_SIZE) 212 213 struct vpgqueues { 214 struct pglist pl; 215 int *cnt; 216 int lcnt; 217 }; 218 219 extern struct vpgqueues vm_page_queues[PQ_COUNT]; 220 221 #endif 222 223 /* 224 * These are the flags defined for vm_page. 225 * 226 * Note: PG_FILLED and PG_DIRTY are added for the filesystems. 227 * 228 * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is 229 * not under PV management but otherwise should be treated as a 230 * normal page. Pages not under PV management cannot be paged out 231 * via the object/vm_page_t because there is no knowledge of their 232 * pte mappings, nor can they be removed from their objects via 233 * the object, and such pages are also not on any PQ queue. 234 */ 235 #define PG_BUSY 0x0001 /* page is in transit (O) */ 236 #define PG_WANTED 0x0002 /* someone is waiting for page (O) */ 237 #define PG_WINATCFLS 0x0004 /* flush dirty page on inactive q */ 238 #define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */ 239 #define PG_WRITEABLE 0x0010 /* page is mapped writeable */ 240 #define PG_MAPPED 0x0020 /* page is mapped */ 241 #define PG_ZERO 0x0040 /* page is zeroed */ 242 #define PG_REFERENCED 0x0080 /* page has been referenced */ 243 #define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */ 244 #define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */ 245 #define PG_NOSYNC 0x0400 /* do not collect for syncer */ 246 #define PG_UNMANAGED 0x0800 /* No PV management for page */ 247 #define PG_MARKER 0x1000 /* special queue marker page */ 248 249 /* 250 * Misc constants. 251 */ 252 253 #define ACT_DECLINE 1 254 #define ACT_ADVANCE 3 255 #define ACT_INIT 5 256 #define ACT_MAX 64 257 #define PFCLUSTER_BEHIND 3 258 #define PFCLUSTER_AHEAD 3 259 260 #ifdef _KERNEL 261 /* 262 * Each pageable resident page falls into one of four lists: 263 * 264 * free 265 * Available for allocation now. 266 * 267 * The following are all LRU sorted: 268 * 269 * cache 270 * Almost available for allocation. Still in an 271 * object, but clean and immediately freeable at 272 * non-interrupt times. 273 * 274 * inactive 275 * Low activity, candidates for reclamation. 276 * This is the list of pages that should be 277 * paged out next. 278 * 279 * active 280 * Pages that are "active" i.e. they have been 281 * recently referenced. 282 * 283 * zero 284 * Pages that are really free and have been pre-zeroed 285 * 286 */ 287 288 extern int vm_page_zero_count; 289 290 extern vm_page_t vm_page_array; /* First resident page in table */ 291 extern int vm_page_array_size; /* number of vm_page_t's */ 292 extern long first_page; /* first physical page number */ 293 294 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 295 296 #define PHYS_TO_VM_PAGE(pa) \ 297 (&vm_page_array[atop(pa) - first_page ]) 298 299 /* 300 * Functions implemented as macros 301 */ 302 303 static __inline void 304 vm_page_flag_set(vm_page_t m, unsigned int bits) 305 { 306 atomic_set_short(&(m)->flags, bits); 307 } 308 309 static __inline void 310 vm_page_flag_clear(vm_page_t m, unsigned int bits) 311 { 312 atomic_clear_short(&(m)->flags, bits); 313 } 314 315 #if 0 316 static __inline void 317 vm_page_assert_wait(vm_page_t m, int interruptible) 318 { 319 vm_page_flag_set(m, PG_WANTED); 320 assert_wait((int) m, interruptible); 321 } 322 #endif 323 324 static __inline void 325 vm_page_busy(vm_page_t m) 326 { 327 KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!")); 328 vm_page_flag_set(m, PG_BUSY); 329 } 330 331 /* 332 * vm_page_flash: 333 * 334 * wakeup anyone waiting for the page. 335 */ 336 337 static __inline void 338 vm_page_flash(vm_page_t m) 339 { 340 if (m->flags & PG_WANTED) { 341 vm_page_flag_clear(m, PG_WANTED); 342 wakeup(m); 343 } 344 } 345 346 /* 347 * vm_page_wakeup: 348 * 349 * clear the PG_BUSY flag and wakeup anyone waiting for the 350 * page. 351 * 352 */ 353 354 static __inline void 355 vm_page_wakeup(vm_page_t m) 356 { 357 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!")); 358 vm_page_flag_clear(m, PG_BUSY); 359 vm_page_flash(m); 360 } 361 362 /* 363 * 364 * 365 */ 366 367 static __inline void 368 vm_page_io_start(vm_page_t m) 369 { 370 atomic_add_char(&(m)->busy, 1); 371 } 372 373 static __inline void 374 vm_page_io_finish(vm_page_t m) 375 { 376 atomic_subtract_char(&m->busy, 1); 377 if (m->busy == 0) 378 vm_page_flash(m); 379 } 380 381 382 #if PAGE_SIZE == 4096 383 #define VM_PAGE_BITS_ALL 0xff 384 #endif 385 386 #if PAGE_SIZE == 8192 387 #define VM_PAGE_BITS_ALL 0xffff 388 #endif 389 390 #define VM_ALLOC_NORMAL 0 391 #define VM_ALLOC_INTERRUPT 1 392 #define VM_ALLOC_SYSTEM 2 393 #define VM_ALLOC_ZERO 3 394 #define VM_ALLOC_RETRY 0x80 395 396 void vm_page_unhold(vm_page_t mem); 397 398 void vm_page_activate (vm_page_t); 399 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int); 400 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int); 401 void vm_page_cache (register vm_page_t); 402 int vm_page_try_to_cache (vm_page_t); 403 int vm_page_try_to_free (vm_page_t); 404 void vm_page_dontneed (register vm_page_t); 405 static __inline void vm_page_copy (vm_page_t, vm_page_t); 406 static __inline void vm_page_free (vm_page_t); 407 static __inline void vm_page_free_zero (vm_page_t); 408 void vm_page_deactivate (vm_page_t); 409 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t); 410 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t); 411 void vm_page_remove (vm_page_t); 412 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t); 413 vm_offset_t vm_page_startup (vm_offset_t, vm_offset_t, vm_offset_t); 414 vm_page_t vm_add_new_page (vm_offset_t pa); 415 void vm_page_unmanage (vm_page_t); 416 void vm_page_unwire (vm_page_t, int); 417 void vm_page_wire (vm_page_t); 418 void vm_page_unqueue (vm_page_t); 419 void vm_page_unqueue_nowakeup (vm_page_t); 420 void vm_page_set_validclean (vm_page_t, int, int); 421 void vm_page_set_dirty (vm_page_t, int, int); 422 void vm_page_clear_dirty (vm_page_t, int, int); 423 void vm_page_set_invalid (vm_page_t, int, int); 424 static __inline boolean_t vm_page_zero_fill (vm_page_t); 425 int vm_page_is_valid (vm_page_t, int, int); 426 void vm_page_test_dirty (vm_page_t); 427 int vm_page_bits (int, int); 428 vm_page_t _vm_page_list_find (int, int); 429 #if 0 430 int vm_page_sleep(vm_page_t m, char *msg, char *busy); 431 int vm_page_asleep(vm_page_t m, char *msg, char *busy); 432 #endif 433 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 434 void vm_page_free_toq(vm_page_t m); 435 436 /* 437 * Keep page from being freed by the page daemon 438 * much of the same effect as wiring, except much lower 439 * overhead and should be used only for *very* temporary 440 * holding ("wiring"). 441 */ 442 static __inline void 443 vm_page_hold(vm_page_t mem) 444 { 445 mem->hold_count++; 446 } 447 448 /* 449 * vm_page_protect: 450 * 451 * Reduce the protection of a page. This routine never raises the 452 * protection and therefore can be safely called if the page is already 453 * at VM_PROT_NONE (it will be a NOP effectively ). 454 */ 455 456 static __inline void 457 vm_page_protect(vm_page_t mem, int prot) 458 { 459 if (prot == VM_PROT_NONE) { 460 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) { 461 pmap_page_protect(mem, VM_PROT_NONE); 462 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED); 463 } 464 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) { 465 pmap_page_protect(mem, VM_PROT_READ); 466 vm_page_flag_clear(mem, PG_WRITEABLE); 467 } 468 } 469 470 /* 471 * vm_page_zero_fill: 472 * 473 * Zero-fill the specified page. 474 * Written as a standard pagein routine, to 475 * be used by the zero-fill object. 476 */ 477 static __inline boolean_t 478 vm_page_zero_fill(m) 479 vm_page_t m; 480 { 481 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 482 return (TRUE); 483 } 484 485 /* 486 * vm_page_copy: 487 * 488 * Copy one page to another 489 */ 490 static __inline void 491 vm_page_copy(src_m, dest_m) 492 vm_page_t src_m; 493 vm_page_t dest_m; 494 { 495 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 496 dest_m->valid = VM_PAGE_BITS_ALL; 497 } 498 499 /* 500 * vm_page_free: 501 * 502 * Free a page 503 * 504 * The clearing of PG_ZERO is a temporary safety until the code can be 505 * reviewed to determine that PG_ZERO is being properly cleared on 506 * write faults or maps. PG_ZERO was previously cleared in 507 * vm_page_alloc(). 508 */ 509 static __inline void 510 vm_page_free(m) 511 vm_page_t m; 512 { 513 vm_page_flag_clear(m, PG_ZERO); 514 vm_page_free_toq(m); 515 } 516 517 /* 518 * vm_page_free_zero: 519 * 520 * Free a page to the zerod-pages queue 521 */ 522 static __inline void 523 vm_page_free_zero(m) 524 vm_page_t m; 525 { 526 vm_page_flag_set(m, PG_ZERO); 527 vm_page_free_toq(m); 528 } 529 530 /* 531 * vm_page_sleep_busy: 532 * 533 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE) 534 * m->busy is zero. Returns TRUE if it had to sleep ( including if 535 * it almost had to sleep and made temporary spl*() mods), FALSE 536 * otherwise. 537 * 538 * This routine assumes that interrupts can only remove the busy 539 * status from a page, not set the busy status or change it from 540 * PG_BUSY to m->busy or vise versa (which would create a timing 541 * window). 542 * 543 * Note that being an inline, this code will be well optimized. 544 */ 545 546 static __inline int 547 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg) 548 { 549 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 550 int s = splvm(); 551 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 552 /* 553 * Page is busy. Wait and retry. 554 */ 555 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED); 556 tsleep(m, PVM, msg, 0); 557 } 558 splx(s); 559 return(TRUE); 560 /* not reached */ 561 } 562 return(FALSE); 563 } 564 565 /* 566 * vm_page_dirty: 567 * 568 * make page all dirty 569 */ 570 571 static __inline void 572 vm_page_dirty(vm_page_t m) 573 { 574 #if !defined(KLD_MODULE) 575 KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!")); 576 #endif 577 m->dirty = VM_PAGE_BITS_ALL; 578 } 579 580 /* 581 * vm_page_undirty: 582 * 583 * Set page to not be dirty. Note: does not clear pmap modify bits 584 */ 585 586 static __inline void 587 vm_page_undirty(vm_page_t m) 588 { 589 m->dirty = 0; 590 } 591 592 #if !defined(KLD_MODULE) 593 594 static __inline vm_page_t 595 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero) 596 { 597 vm_page_t m; 598 599 #if PQ_L2_SIZE > 1 600 if (prefer_zero) { 601 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist); 602 } else { 603 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl); 604 } 605 if (m == NULL) 606 m = _vm_page_list_find(basequeue, index); 607 #else 608 if (prefer_zero) { 609 m = TAILQ_LAST(&vm_page_queues[basequeue].pl, pglist); 610 } else { 611 m = TAILQ_FIRST(&vm_page_queues[basequeue].pl); 612 } 613 #endif 614 return(m); 615 } 616 617 #endif 618 619 #endif /* _KERNEL */ 620 #endif /* !_VM_PAGE_ */ 621