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_WINATCFLS 0x0004 /* flush dirty page on inactive q */ 246 #define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */ 247 #define PG_WRITEABLE 0x0010 /* page is mapped writeable */ 248 #define PG_MAPPED 0x0020 /* page is mapped */ 249 #define PG_ZERO 0x0040 /* page is zeroed */ 250 #define PG_REFERENCED 0x0080 /* page has been referenced */ 251 #define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */ 252 #define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */ 253 #define PG_NOSYNC 0x0400 /* do not collect for syncer */ 254 #define PG_UNMANAGED 0x0800 /* No PV management for page */ 255 #define PG_MARKER 0x1000 /* special queue marker page */ 256 257 /* 258 * Misc constants. 259 */ 260 261 #define ACT_DECLINE 1 262 #define ACT_ADVANCE 3 263 #define ACT_INIT 5 264 #define ACT_MAX 64 265 #define PFCLUSTER_BEHIND 3 266 #define PFCLUSTER_AHEAD 3 267 268 #ifdef _KERNEL 269 /* 270 * Each pageable resident page falls into one of four lists: 271 * 272 * free 273 * Available for allocation now. 274 * 275 * The following are all LRU sorted: 276 * 277 * cache 278 * Almost available for allocation. Still in an 279 * object, but clean and immediately freeable at 280 * non-interrupt times. 281 * 282 * inactive 283 * Low activity, candidates for reclamation. 284 * This is the list of pages that should be 285 * paged out next. 286 * 287 * active 288 * Pages that are "active" i.e. they have been 289 * recently referenced. 290 * 291 * zero 292 * Pages that are really free and have been pre-zeroed 293 * 294 */ 295 296 extern int vm_page_zero_count; 297 298 extern vm_page_t vm_page_array; /* First resident page in table */ 299 extern int vm_page_array_size; /* number of vm_page_t's */ 300 extern long first_page; /* first physical page number */ 301 302 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 303 304 #define PHYS_TO_VM_PAGE(pa) \ 305 (&vm_page_array[atop(pa) - first_page ]) 306 307 /* 308 * Functions implemented as macros 309 */ 310 311 static __inline void 312 vm_page_flag_set(vm_page_t m, unsigned short bits) 313 { 314 atomic_set_short(&(m)->flags, bits); 315 } 316 317 static __inline void 318 vm_page_flag_clear(vm_page_t m, unsigned short bits) 319 { 320 atomic_clear_short(&(m)->flags, bits); 321 } 322 323 #if 0 324 static __inline void 325 vm_page_assert_wait(vm_page_t m, int interruptible) 326 { 327 vm_page_flag_set(m, PG_WANTED); 328 assert_wait((int) m, interruptible); 329 } 330 #endif 331 332 static __inline void 333 vm_page_busy(vm_page_t m) 334 { 335 KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!")); 336 vm_page_flag_set(m, PG_BUSY); 337 } 338 339 /* 340 * vm_page_flash: 341 * 342 * wakeup anyone waiting for the page. 343 */ 344 345 static __inline void 346 vm_page_flash(vm_page_t m) 347 { 348 if (m->flags & PG_WANTED) { 349 vm_page_flag_clear(m, PG_WANTED); 350 wakeup(m); 351 } 352 } 353 354 /* 355 * vm_page_wakeup: 356 * 357 * clear the PG_BUSY flag and wakeup anyone waiting for the 358 * page. 359 * 360 */ 361 362 static __inline void 363 vm_page_wakeup(vm_page_t m) 364 { 365 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!")); 366 vm_page_flag_clear(m, PG_BUSY); 367 vm_page_flash(m); 368 } 369 370 /* 371 * 372 * 373 */ 374 375 static __inline void 376 vm_page_io_start(vm_page_t m) 377 { 378 atomic_add_char(&(m)->busy, 1); 379 } 380 381 static __inline void 382 vm_page_io_finish(vm_page_t m) 383 { 384 atomic_subtract_char(&m->busy, 1); 385 if (m->busy == 0) 386 vm_page_flash(m); 387 } 388 389 390 #if PAGE_SIZE == 4096 391 #define VM_PAGE_BITS_ALL 0xff 392 #endif 393 394 #if PAGE_SIZE == 8192 395 #define VM_PAGE_BITS_ALL 0xffff 396 #endif 397 398 #define VM_ALLOC_NORMAL 0 399 #define VM_ALLOC_INTERRUPT 1 400 #define VM_ALLOC_SYSTEM 2 401 #define VM_ALLOC_ZERO 3 402 #define VM_ALLOC_RETRY 0x80 403 404 void vm_page_activate __P((vm_page_t)); 405 vm_page_t vm_page_alloc __P((vm_object_t, vm_pindex_t, int)); 406 vm_page_t vm_page_grab __P((vm_object_t, vm_pindex_t, int)); 407 void vm_page_cache __P((register vm_page_t)); 408 int vm_page_try_to_cache __P((vm_page_t)); 409 void vm_page_dontneed __P((register vm_page_t)); 410 static __inline void vm_page_copy __P((vm_page_t, vm_page_t)); 411 static __inline void vm_page_free __P((vm_page_t)); 412 static __inline void vm_page_free_zero __P((vm_page_t)); 413 void vm_page_deactivate __P((vm_page_t)); 414 void vm_page_insert __P((vm_page_t, vm_object_t, vm_pindex_t)); 415 vm_page_t vm_page_lookup __P((vm_object_t, vm_pindex_t)); 416 void vm_page_remove __P((vm_page_t)); 417 void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t)); 418 vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t)); 419 vm_page_t vm_add_new_page __P((vm_offset_t pa)); 420 void vm_page_unmanage __P((vm_page_t)); 421 void vm_page_unwire __P((vm_page_t, int)); 422 void vm_page_wire __P((vm_page_t)); 423 void vm_page_unqueue __P((vm_page_t)); 424 void vm_page_unqueue_nowakeup __P((vm_page_t)); 425 void vm_page_set_validclean __P((vm_page_t, int, int)); 426 void vm_page_set_dirty __P((vm_page_t, int, int)); 427 void vm_page_clear_dirty __P((vm_page_t, int, int)); 428 void vm_page_set_invalid __P((vm_page_t, int, int)); 429 static __inline boolean_t vm_page_zero_fill __P((vm_page_t)); 430 int vm_page_is_valid __P((vm_page_t, int, int)); 431 void vm_page_test_dirty __P((vm_page_t)); 432 int vm_page_bits __P((int, int)); 433 vm_page_t _vm_page_list_find __P((int, int)); 434 #if 0 435 int vm_page_sleep(vm_page_t m, char *msg, char *busy); 436 int vm_page_asleep(vm_page_t m, char *msg, char *busy); 437 #endif 438 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 439 void vm_page_free_toq(vm_page_t m); 440 441 /* 442 * Keep page from being freed by the page daemon 443 * much of the same effect as wiring, except much lower 444 * overhead and should be used only for *very* temporary 445 * holding ("wiring"). 446 */ 447 static __inline void 448 vm_page_hold(vm_page_t mem) 449 { 450 mem->hold_count++; 451 } 452 453 static __inline void 454 vm_page_unhold(vm_page_t mem) 455 { 456 --mem->hold_count; 457 KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!")); 458 } 459 460 /* 461 * vm_page_protect: 462 * 463 * Reduce the protection of a page. This routine never raises the 464 * protection and therefore can be safely called if the page is already 465 * at VM_PROT_NONE (it will be a NOP effectively ). 466 */ 467 468 static __inline void 469 vm_page_protect(vm_page_t mem, int prot) 470 { 471 if (prot == VM_PROT_NONE) { 472 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) { 473 pmap_page_protect(mem, VM_PROT_NONE); 474 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED); 475 } 476 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) { 477 pmap_page_protect(mem, VM_PROT_READ); 478 vm_page_flag_clear(mem, PG_WRITEABLE); 479 } 480 } 481 482 /* 483 * vm_page_zero_fill: 484 * 485 * Zero-fill the specified page. 486 * Written as a standard pagein routine, to 487 * be used by the zero-fill object. 488 */ 489 static __inline boolean_t 490 vm_page_zero_fill(m) 491 vm_page_t m; 492 { 493 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 494 return (TRUE); 495 } 496 497 /* 498 * vm_page_copy: 499 * 500 * Copy one page to another 501 */ 502 static __inline void 503 vm_page_copy(src_m, dest_m) 504 vm_page_t src_m; 505 vm_page_t dest_m; 506 { 507 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 508 dest_m->valid = VM_PAGE_BITS_ALL; 509 } 510 511 /* 512 * vm_page_free: 513 * 514 * Free a page 515 * 516 * The clearing of PG_ZERO is a temporary safety until the code can be 517 * reviewed to determine that PG_ZERO is being properly cleared on 518 * write faults or maps. PG_ZERO was previously cleared in 519 * vm_page_alloc(). 520 */ 521 static __inline void 522 vm_page_free(m) 523 vm_page_t m; 524 { 525 vm_page_flag_clear(m, PG_ZERO); 526 vm_page_free_toq(m); 527 } 528 529 /* 530 * vm_page_free_zero: 531 * 532 * Free a page to the zerod-pages queue 533 */ 534 static __inline void 535 vm_page_free_zero(m) 536 vm_page_t m; 537 { 538 vm_page_flag_set(m, PG_ZERO); 539 vm_page_free_toq(m); 540 } 541 542 /* 543 * vm_page_sleep_busy: 544 * 545 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE) 546 * m->busy is zero. Returns TRUE if it had to sleep ( including if 547 * it almost had to sleep and made temporary spl*() mods), FALSE 548 * otherwise. 549 * 550 * This routine assumes that interrupts can only remove the busy 551 * status from a page, not set the busy status or change it from 552 * PG_BUSY to m->busy or vise versa (which would create a timing 553 * window). 554 * 555 * Note that being an inline, this code will be well optimized. 556 */ 557 558 static __inline int 559 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg) 560 { 561 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 562 int s = splvm(); 563 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 564 /* 565 * Page is busy. Wait and retry. 566 */ 567 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED); 568 tsleep(m, PVM, msg, 0); 569 } 570 splx(s); 571 return(TRUE); 572 /* not reached */ 573 } 574 return(FALSE); 575 } 576 577 /* 578 * vm_page_dirty: 579 * 580 * make page all dirty 581 */ 582 583 static __inline void 584 vm_page_dirty(vm_page_t m) 585 { 586 #if !defined(KLD_MODULE) 587 KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!")); 588 #endif 589 m->dirty = VM_PAGE_BITS_ALL; 590 } 591 592 /* 593 * vm_page_undirty: 594 * 595 * Set page to not be dirty. Note: does not clear pmap modify bits 596 */ 597 598 static __inline void 599 vm_page_undirty(vm_page_t m) 600 { 601 m->dirty = 0; 602 } 603 604 #if !defined(KLD_MODULE) 605 606 static __inline vm_page_t 607 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero) 608 { 609 vm_page_t m; 610 611 #if PQ_L2_SIZE > 1 612 if (prefer_zero) { 613 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist); 614 } else { 615 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl); 616 } 617 if (m == NULL) 618 m = _vm_page_list_find(basequeue, index); 619 #else 620 if (prefer_zero) { 621 m = TAILQ_LAST(&vm_page_queues[basequeue].pl, pglist); 622 } else { 623 m = TAILQ_FIRST(&vm_page_queues[basequeue].pl); 624 } 625 #endif 626 return(m); 627 } 628 629 #endif 630 631 #endif /* _KERNEL */ 632 #endif /* !_VM_PAGE_ */ 633