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 * $FreeBSD$ 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 #if 1 206 #define PQ_NONE 0 207 #define PQ_FREE 1 208 #define PQ_INACTIVE (1 + 1*PQ_L2_SIZE) 209 #define PQ_ACTIVE (2 + 1*PQ_L2_SIZE) 210 #define PQ_CACHE (3 + 1*PQ_L2_SIZE) 211 #define PQ_COUNT (3 + 2*PQ_L2_SIZE) 212 #else 213 #define PQ_NONE PQ_COUNT 214 #define PQ_FREE 0 215 #define PQ_INACTIVE PQ_L2_SIZE 216 #define PQ_ACTIVE (1 + PQ_L2_SIZE) 217 #define PQ_CACHE (2 + PQ_L2_SIZE) 218 #define PQ_COUNT (2 + 2*PQ_L2_SIZE) 219 #endif 220 221 struct vpgqueues { 222 struct pglist pl; 223 int *cnt; 224 int lcnt; 225 }; 226 227 extern struct vpgqueues vm_page_queues[PQ_COUNT]; 228 229 #endif 230 231 /* 232 * These are the flags defined for vm_page. 233 * 234 * Note: PG_FILLED and PG_DIRTY are added for the filesystems. 235 * 236 * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is 237 * not under PV management but otherwise should be treated as a 238 * normal page. Pages not under PV management cannot be paged out 239 * via the object/vm_page_t because there is no knowledge of their 240 * pte mappings, nor can they be removed from their objects via 241 * the object, and such pages are also not on any PQ queue. 242 */ 243 #define PG_BUSY 0x0001 /* page is in transit (O) */ 244 #define PG_WANTED 0x0002 /* someone is waiting for page (O) */ 245 #define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */ 246 #define PG_WRITEABLE 0x0010 /* page is mapped writeable */ 247 #define PG_MAPPED 0x0020 /* page is mapped */ 248 #define PG_ZERO 0x0040 /* page is zeroed */ 249 #define PG_REFERENCED 0x0080 /* page has been referenced */ 250 #define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */ 251 #define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */ 252 #define PG_NOSYNC 0x0400 /* do not collect for syncer */ 253 #define PG_UNMANAGED 0x0800 /* No PV management for page */ 254 255 /* 256 * Misc constants. 257 */ 258 259 #define ACT_DECLINE 1 260 #define ACT_ADVANCE 3 261 #define ACT_INIT 5 262 #define ACT_MAX 64 263 #define PFCLUSTER_BEHIND 3 264 #define PFCLUSTER_AHEAD 3 265 266 #ifdef _KERNEL 267 /* 268 * Each pageable resident page falls into one of four lists: 269 * 270 * free 271 * Available for allocation now. 272 * 273 * The following are all LRU sorted: 274 * 275 * cache 276 * Almost available for allocation. Still in an 277 * object, but clean and immediately freeable at 278 * non-interrupt times. 279 * 280 * inactive 281 * Low activity, candidates for reclamation. 282 * This is the list of pages that should be 283 * paged out next. 284 * 285 * active 286 * Pages that are "active" i.e. they have been 287 * recently referenced. 288 * 289 * zero 290 * Pages that are really free and have been pre-zeroed 291 * 292 */ 293 294 extern int vm_page_zero_count; 295 296 extern vm_page_t vm_page_array; /* First resident page in table */ 297 extern int vm_page_array_size; /* number of vm_page_t's */ 298 extern long first_page; /* first physical page number */ 299 300 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 301 302 #define PHYS_TO_VM_PAGE(pa) \ 303 (&vm_page_array[atop(pa) - first_page ]) 304 305 /* 306 * Functions implemented as macros 307 */ 308 309 static __inline void 310 vm_page_flag_set(vm_page_t m, unsigned int bits) 311 { 312 atomic_set_short(&(m)->flags, bits); 313 } 314 315 static __inline void 316 vm_page_flag_clear(vm_page_t m, unsigned int bits) 317 { 318 atomic_clear_short(&(m)->flags, bits); 319 } 320 321 #if 0 322 static __inline void 323 vm_page_assert_wait(vm_page_t m, int interruptible) 324 { 325 vm_page_flag_set(m, PG_WANTED); 326 assert_wait((int) m, interruptible); 327 } 328 #endif 329 330 static __inline void 331 vm_page_busy(vm_page_t m) 332 { 333 KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!")); 334 vm_page_flag_set(m, PG_BUSY); 335 } 336 337 /* 338 * vm_page_flash: 339 * 340 * wakeup anyone waiting for the page. 341 */ 342 343 static __inline void 344 vm_page_flash(vm_page_t m) 345 { 346 if (m->flags & PG_WANTED) { 347 vm_page_flag_clear(m, PG_WANTED); 348 wakeup(m); 349 } 350 } 351 352 /* 353 * vm_page_wakeup: 354 * 355 * clear the PG_BUSY flag and wakeup anyone waiting for the 356 * page. 357 * 358 */ 359 360 static __inline void 361 vm_page_wakeup(vm_page_t m) 362 { 363 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!")); 364 vm_page_flag_clear(m, PG_BUSY); 365 vm_page_flash(m); 366 } 367 368 /* 369 * 370 * 371 */ 372 373 static __inline void 374 vm_page_io_start(vm_page_t m) 375 { 376 atomic_add_char(&(m)->busy, 1); 377 } 378 379 static __inline void 380 vm_page_io_finish(vm_page_t m) 381 { 382 atomic_subtract_char(&m->busy, 1); 383 if (m->busy == 0) 384 vm_page_flash(m); 385 } 386 387 388 #if PAGE_SIZE == 4096 389 #define VM_PAGE_BITS_ALL 0xff 390 #endif 391 392 #if PAGE_SIZE == 8192 393 #define VM_PAGE_BITS_ALL 0xffff 394 #endif 395 396 #define VM_ALLOC_NORMAL 0 397 #define VM_ALLOC_INTERRUPT 1 398 #define VM_ALLOC_SYSTEM 2 399 #define VM_ALLOC_ZERO 3 400 #define VM_ALLOC_RETRY 0x80 401 402 void vm_page_activate __P((vm_page_t)); 403 vm_page_t vm_page_alloc __P((vm_object_t, vm_pindex_t, int)); 404 vm_page_t vm_page_grab __P((vm_object_t, vm_pindex_t, int)); 405 void vm_page_cache __P((register vm_page_t)); 406 void vm_page_dontneed __P((register vm_page_t)); 407 static __inline void vm_page_copy __P((vm_page_t, vm_page_t)); 408 static __inline void vm_page_free __P((vm_page_t)); 409 static __inline void vm_page_free_zero __P((vm_page_t)); 410 void vm_page_deactivate __P((vm_page_t)); 411 void vm_page_insert __P((vm_page_t, vm_object_t, vm_pindex_t)); 412 vm_page_t vm_page_lookup __P((vm_object_t, vm_pindex_t)); 413 void vm_page_remove __P((vm_page_t)); 414 void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t)); 415 vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t)); 416 vm_page_t vm_add_new_page __P((vm_offset_t pa)); 417 void vm_page_unmanage __P((vm_page_t)); 418 void vm_page_unwire __P((vm_page_t, int)); 419 void vm_page_wire __P((vm_page_t)); 420 void vm_page_unqueue __P((vm_page_t)); 421 void vm_page_unqueue_nowakeup __P((vm_page_t)); 422 void vm_page_set_validclean __P((vm_page_t, int, int)); 423 void vm_page_set_dirty __P((vm_page_t, int, int)); 424 void vm_page_clear_dirty __P((vm_page_t, int, int)); 425 void vm_page_set_invalid __P((vm_page_t, int, int)); 426 static __inline boolean_t vm_page_zero_fill __P((vm_page_t)); 427 int vm_page_is_valid __P((vm_page_t, int, int)); 428 void vm_page_test_dirty __P((vm_page_t)); 429 int vm_page_bits __P((int, int)); 430 vm_page_t _vm_page_list_find __P((int, int)); 431 #if 0 432 int vm_page_sleep(vm_page_t m, char *msg, char *busy); 433 int vm_page_asleep(vm_page_t m, char *msg, char *busy); 434 #endif 435 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 436 void vm_page_free_toq(vm_page_t m); 437 438 /* 439 * Keep page from being freed by the page daemon 440 * much of the same effect as wiring, except much lower 441 * overhead and should be used only for *very* temporary 442 * holding ("wiring"). 443 */ 444 static __inline void 445 vm_page_hold(vm_page_t mem) 446 { 447 mem->hold_count++; 448 } 449 450 static __inline void 451 vm_page_unhold(vm_page_t mem) 452 { 453 --mem->hold_count; 454 KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!")); 455 } 456 457 /* 458 * vm_page_protect: 459 * 460 * Reduce the protection of a page. This routine never raises the 461 * protection and therefore can be safely called if the page is already 462 * at VM_PROT_NONE (it will be a NOP effectively ). 463 */ 464 465 static __inline void 466 vm_page_protect(vm_page_t mem, int prot) 467 { 468 if (prot == VM_PROT_NONE) { 469 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) { 470 pmap_page_protect(mem, VM_PROT_NONE); 471 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED); 472 } 473 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) { 474 pmap_page_protect(mem, VM_PROT_READ); 475 vm_page_flag_clear(mem, PG_WRITEABLE); 476 } 477 } 478 479 /* 480 * vm_page_zero_fill: 481 * 482 * Zero-fill the specified page. 483 * Written as a standard pagein routine, to 484 * be used by the zero-fill object. 485 */ 486 static __inline boolean_t 487 vm_page_zero_fill(m) 488 vm_page_t m; 489 { 490 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 491 return (TRUE); 492 } 493 494 /* 495 * vm_page_copy: 496 * 497 * Copy one page to another 498 */ 499 static __inline void 500 vm_page_copy(src_m, dest_m) 501 vm_page_t src_m; 502 vm_page_t dest_m; 503 { 504 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 505 dest_m->valid = VM_PAGE_BITS_ALL; 506 } 507 508 /* 509 * vm_page_free: 510 * 511 * Free a page 512 * 513 * The clearing of PG_ZERO is a temporary safety until the code can be 514 * reviewed to determine that PG_ZERO is being properly cleared on 515 * write faults or maps. PG_ZERO was previously cleared in 516 * vm_page_alloc(). 517 */ 518 static __inline void 519 vm_page_free(m) 520 vm_page_t m; 521 { 522 vm_page_flag_clear(m, PG_ZERO); 523 vm_page_free_toq(m); 524 } 525 526 /* 527 * vm_page_free_zero: 528 * 529 * Free a page to the zerod-pages queue 530 */ 531 static __inline void 532 vm_page_free_zero(m) 533 vm_page_t m; 534 { 535 vm_page_flag_set(m, PG_ZERO); 536 vm_page_free_toq(m); 537 } 538 539 /* 540 * vm_page_sleep_busy: 541 * 542 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE) 543 * m->busy is zero. Returns TRUE if it had to sleep ( including if 544 * it almost had to sleep and made temporary spl*() mods), FALSE 545 * otherwise. 546 * 547 * This routine assumes that interrupts can only remove the busy 548 * status from a page, not set the busy status or change it from 549 * PG_BUSY to m->busy or vise versa (which would create a timing 550 * window). 551 * 552 * Note that being an inline, this code will be well optimized. 553 */ 554 555 static __inline int 556 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg) 557 { 558 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 559 int s = splvm(); 560 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 561 /* 562 * Page is busy. Wait and retry. 563 */ 564 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED); 565 tsleep(m, PVM, msg, 0); 566 } 567 splx(s); 568 return(TRUE); 569 /* not reached */ 570 } 571 return(FALSE); 572 } 573 574 /* 575 * vm_page_dirty: 576 * 577 * make page all dirty 578 */ 579 580 static __inline void 581 vm_page_dirty(vm_page_t m) 582 { 583 #if !defined(KLD_MODULE) 584 KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!")); 585 #endif 586 m->dirty = VM_PAGE_BITS_ALL; 587 } 588 589 /* 590 * vm_page_undirty: 591 * 592 * Set page to not be dirty. Note: does not clear pmap modify bits 593 */ 594 595 static __inline void 596 vm_page_undirty(vm_page_t m) 597 { 598 m->dirty = 0; 599 } 600 601 #if !defined(KLD_MODULE) 602 603 static __inline vm_page_t 604 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero) 605 { 606 vm_page_t m; 607 608 #if PQ_L2_SIZE > 1 609 if (prefer_zero) { 610 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist); 611 } else { 612 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl); 613 } 614 if (m == NULL) 615 m = _vm_page_list_find(basequeue, index); 616 #else 617 if (prefer_zero) { 618 m = TAILQ_LAST(&vm_page_queues[basequeue].pl, pglist); 619 } else { 620 m = TAILQ_FIRST(&vm_page_queues[basequeue].pl); 621 } 622 #endif 623 return(m); 624 } 625 626 #endif 627 628 #endif /* _KERNEL */ 629 #endif /* !_VM_PAGE_ */ 630