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 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 #if 1 197 #define PQ_NONE 0 198 #define PQ_FREE 1 199 #define PQ_INACTIVE (1 + 1*PQ_L2_SIZE) 200 #define PQ_ACTIVE (2 + 1*PQ_L2_SIZE) 201 #define PQ_CACHE (3 + 1*PQ_L2_SIZE) 202 #define PQ_COUNT (3 + 2*PQ_L2_SIZE) 203 #else 204 #define PQ_NONE PQ_COUNT 205 #define PQ_FREE 0 206 #define PQ_INACTIVE PQ_L2_SIZE 207 #define PQ_ACTIVE (1 + PQ_L2_SIZE) 208 #define PQ_CACHE (2 + PQ_L2_SIZE) 209 #define PQ_COUNT (2 + 2*PQ_L2_SIZE) 210 #endif 211 212 extern struct vpgqueues { 213 struct pglist *pl; 214 int *cnt; 215 int lcnt; 216 } vm_page_queues[PQ_COUNT]; 217 218 #endif 219 220 /* 221 * These are the flags defined for vm_page. 222 * 223 * Note: PG_FILLED and PG_DIRTY are added for the filesystems. 224 */ 225 #define PG_BUSY 0x0001 /* page is in transit (O) */ 226 #define PG_WANTED 0x0002 /* someone is waiting for page (O) */ 227 #define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */ 228 #define PG_WRITEABLE 0x0010 /* page is mapped writeable */ 229 #define PG_MAPPED 0x0020 /* page is mapped */ 230 #define PG_ZERO 0x0040 /* page is zeroed */ 231 #define PG_REFERENCED 0x0080 /* page has been referenced */ 232 #define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */ 233 #define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */ 234 235 /* 236 * Misc constants. 237 */ 238 239 #define ACT_DECLINE 1 240 #define ACT_ADVANCE 3 241 #define ACT_INIT 5 242 #define ACT_MAX 64 243 #define PFCLUSTER_BEHIND 3 244 #define PFCLUSTER_AHEAD 3 245 246 #ifdef KERNEL 247 /* 248 * Each pageable resident page falls into one of four lists: 249 * 250 * free 251 * Available for allocation now. 252 * 253 * The following are all LRU sorted: 254 * 255 * cache 256 * Almost available for allocation. Still in an 257 * object, but clean and immediately freeable at 258 * non-interrupt times. 259 * 260 * inactive 261 * Low activity, candidates for reclamation. 262 * This is the list of pages that should be 263 * paged out next. 264 * 265 * active 266 * Pages that are "active" i.e. they have been 267 * recently referenced. 268 * 269 * zero 270 * Pages that are really free and have been pre-zeroed 271 * 272 */ 273 274 #if !defined(KLD_MODULE) 275 276 extern struct pglist vm_page_queue_free[PQ_L2_SIZE];/* memory free queue */ 277 extern struct pglist vm_page_queue_active; /* active memory queue */ 278 extern struct pglist vm_page_queue_inactive; /* inactive memory queue */ 279 extern struct pglist vm_page_queue_cache[PQ_L2_SIZE];/* cache memory queue */ 280 281 #endif 282 283 extern int vm_page_zero_count; 284 285 extern vm_page_t vm_page_array; /* First resident page in table */ 286 extern long first_page; /* first physical page number */ 287 288 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 289 290 #define PHYS_TO_VM_PAGE(pa) \ 291 (&vm_page_array[atop(pa) - first_page ]) 292 293 /* 294 * Functions implemented as macros 295 */ 296 297 static __inline void 298 vm_page_flag_set(vm_page_t m, unsigned int bits) 299 { 300 atomic_set_short(&(m)->flags, bits); 301 } 302 303 static __inline void 304 vm_page_flag_clear(vm_page_t m, unsigned int bits) 305 { 306 atomic_clear_short(&(m)->flags, bits); 307 } 308 309 #if 0 310 static __inline void 311 vm_page_assert_wait(vm_page_t m, int interruptible) 312 { 313 vm_page_flag_set(m, PG_WANTED); 314 assert_wait((int) m, interruptible); 315 } 316 #endif 317 318 static __inline void 319 vm_page_busy(vm_page_t m) 320 { 321 KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!")); 322 vm_page_flag_set(m, PG_BUSY); 323 } 324 325 /* 326 * vm_page_flash: 327 * 328 * wakeup anyone waiting for the page. 329 */ 330 331 static __inline void 332 vm_page_flash(vm_page_t m) 333 { 334 if (m->flags & PG_WANTED) { 335 vm_page_flag_clear(m, PG_WANTED); 336 wakeup(m); 337 } 338 } 339 340 /* 341 * vm_page_wakeup: 342 * 343 * clear the PG_BUSY flag and wakeup anyone waiting for the 344 * page. 345 * 346 */ 347 348 static __inline void 349 vm_page_wakeup(vm_page_t m) 350 { 351 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!")); 352 vm_page_flag_clear(m, PG_BUSY); 353 vm_page_flash(m); 354 } 355 356 /* 357 * 358 * 359 */ 360 361 static __inline void 362 vm_page_io_start(vm_page_t m) 363 { 364 atomic_add_char(&(m)->busy, 1); 365 } 366 367 static __inline void 368 vm_page_io_finish(vm_page_t m) 369 { 370 atomic_subtract_char(&m->busy, 1); 371 if (m->busy == 0) 372 vm_page_flash(m); 373 } 374 375 376 #if PAGE_SIZE == 4096 377 #define VM_PAGE_BITS_ALL 0xff 378 #endif 379 380 #if PAGE_SIZE == 8192 381 #define VM_PAGE_BITS_ALL 0xffff 382 #endif 383 384 #define VM_ALLOC_NORMAL 0 385 #define VM_ALLOC_INTERRUPT 1 386 #define VM_ALLOC_SYSTEM 2 387 #define VM_ALLOC_ZERO 3 388 #define VM_ALLOC_RETRY 0x80 389 390 void vm_page_activate __P((vm_page_t)); 391 vm_page_t vm_page_alloc __P((vm_object_t, vm_pindex_t, int)); 392 vm_page_t vm_page_grab __P((vm_object_t, vm_pindex_t, int)); 393 void vm_page_cache __P((register vm_page_t)); 394 static __inline void vm_page_copy __P((vm_page_t, vm_page_t)); 395 static __inline void vm_page_free __P((vm_page_t)); 396 static __inline void vm_page_free_zero __P((vm_page_t)); 397 void vm_page_deactivate __P((vm_page_t)); 398 void vm_page_insert __P((vm_page_t, vm_object_t, vm_pindex_t)); 399 vm_page_t vm_page_lookup __P((vm_object_t, vm_pindex_t)); 400 void vm_page_remove __P((vm_page_t)); 401 void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t)); 402 vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t)); 403 void vm_page_unwire __P((vm_page_t, int)); 404 void vm_page_wire __P((vm_page_t)); 405 void vm_page_unqueue __P((vm_page_t)); 406 void vm_page_unqueue_nowakeup __P((vm_page_t)); 407 void vm_page_set_validclean __P((vm_page_t, int, int)); 408 void vm_page_set_dirty __P((vm_page_t, int, int)); 409 void vm_page_clear_dirty __P((vm_page_t, int, int)); 410 void vm_page_set_invalid __P((vm_page_t, int, int)); 411 static __inline boolean_t vm_page_zero_fill __P((vm_page_t)); 412 int vm_page_is_valid __P((vm_page_t, int, int)); 413 void vm_page_test_dirty __P((vm_page_t)); 414 int vm_page_bits __P((int, int)); 415 vm_page_t _vm_page_list_find __P((int, int)); 416 #if 0 417 int vm_page_sleep(vm_page_t m, char *msg, char *busy); 418 int vm_page_asleep(vm_page_t m, char *msg, char *busy); 419 #endif 420 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); 421 void vm_page_free_toq(vm_page_t m); 422 423 /* 424 * Keep page from being freed by the page daemon 425 * much of the same effect as wiring, except much lower 426 * overhead and should be used only for *very* temporary 427 * holding ("wiring"). 428 */ 429 static __inline void 430 vm_page_hold(vm_page_t mem) 431 { 432 mem->hold_count++; 433 } 434 435 static __inline void 436 vm_page_unhold(vm_page_t mem) 437 { 438 --mem->hold_count; 439 KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!")); 440 } 441 442 /* 443 * vm_page_protect: 444 * 445 * Reduce the protection of a page. This routine never 446 * raises the protection and therefore can be safely 447 * called if the page is already at VM_PROT_NONE ( it 448 * will be a NOP effectively ). 449 */ 450 451 static __inline void 452 vm_page_protect(vm_page_t mem, int prot) 453 { 454 if (prot == VM_PROT_NONE) { 455 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) { 456 pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_NONE); 457 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED); 458 } 459 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) { 460 pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_READ); 461 vm_page_flag_clear(mem, PG_WRITEABLE); 462 } 463 } 464 465 /* 466 * vm_page_zero_fill: 467 * 468 * Zero-fill the specified page. 469 * Written as a standard pagein routine, to 470 * be used by the zero-fill object. 471 */ 472 static __inline boolean_t 473 vm_page_zero_fill(m) 474 vm_page_t m; 475 { 476 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 477 return (TRUE); 478 } 479 480 /* 481 * vm_page_copy: 482 * 483 * Copy one page to another 484 */ 485 static __inline void 486 vm_page_copy(src_m, dest_m) 487 vm_page_t src_m; 488 vm_page_t dest_m; 489 { 490 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 491 dest_m->valid = VM_PAGE_BITS_ALL; 492 } 493 494 /* 495 * vm_page_free: 496 * 497 * Free a page 498 * 499 * The clearing of PG_ZERO is a temporary safety until the code can be 500 * reviewed to determine that PG_ZERO is being properly cleared on 501 * write faults or maps. PG_ZERO was previously cleared in 502 * vm_page_alloc(). 503 */ 504 static __inline void 505 vm_page_free(m) 506 vm_page_t m; 507 { 508 vm_page_flag_clear(m, PG_ZERO); 509 vm_page_free_toq(m); 510 } 511 512 /* 513 * vm_page_free_zero: 514 * 515 * Free a page to the zerod-pages queue 516 */ 517 static __inline void 518 vm_page_free_zero(m) 519 vm_page_t m; 520 { 521 vm_page_flag_set(m, PG_ZERO); 522 vm_page_free_toq(m); 523 } 524 525 /* 526 * vm_page_sleep_busy: 527 * 528 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE) 529 * m->busy is zero. Returns TRUE if it had to sleep ( including if 530 * it almost had to sleep and made temporary spl*() mods), FALSE 531 * otherwise. 532 * 533 * This routine assumes that interrupts can only remove the busy 534 * status from a page, not set the busy status or change it from 535 * PG_BUSY to m->busy or vise versa (which would create a timing 536 * window). 537 * 538 * Note that being an inline, this code will be well optimized. 539 */ 540 541 static __inline int 542 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg) 543 { 544 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 545 int s = splvm(); 546 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) { 547 /* 548 * Page is busy. Wait and retry. 549 */ 550 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED); 551 tsleep(m, PVM, msg, 0); 552 } 553 splx(s); 554 return(TRUE); 555 /* not reached */ 556 } 557 return(FALSE); 558 } 559 560 /* 561 * vm_page_dirty: 562 * 563 * make page all dirty 564 */ 565 566 static __inline void 567 vm_page_dirty(vm_page_t m) 568 { 569 #if !defined(KLD_MODULE) 570 KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!")); 571 #endif 572 m->dirty = VM_PAGE_BITS_ALL; 573 } 574 575 /* 576 * vm_page_undirty: 577 * 578 * Set page to not be dirty. Note: does not clear pmap modify bits 579 */ 580 581 static __inline void 582 vm_page_undirty(vm_page_t m) 583 { 584 m->dirty = 0; 585 } 586 587 #if !defined(KLD_MODULE) 588 589 static __inline vm_page_t 590 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero) 591 { 592 vm_page_t m; 593 594 #if PQ_L2_SIZE > 1 595 if (prefer_zero) { 596 m = TAILQ_LAST(vm_page_queues[basequeue+index].pl, pglist); 597 } else { 598 m = TAILQ_FIRST(vm_page_queues[basequeue+index].pl); 599 } 600 if (m == NULL) 601 m = _vm_page_list_find(basequeue, index); 602 #else 603 if (prefer_zero) { 604 m = TAILQ_LAST(vm_page_queues[basequeue].pl, pglist); 605 } else { 606 m = TAILQ_FIRST(vm_page_queues[basequeue].pl); 607 } 608 #endif 609 return(m); 610 } 611 612 #endif 613 614 #endif /* KERNEL */ 615 #endif /* !_VM_PAGE_ */ 616