vm_page.c revision 15890
1/* 2 * Copyright (c) 1991 Regents of the University of California. 3 * 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.c 7.4 (Berkeley) 5/7/91 37 * $Id: vm_page.c,v 1.51 1996/05/18 03:37:57 dyson Exp $ 38 */ 39 40/* 41 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 42 * All rights reserved. 43 * 44 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 45 * 46 * Permission to use, copy, modify and distribute this software and 47 * its documentation is hereby granted, provided that both the copyright 48 * notice and this permission notice appear in all copies of the 49 * software, derivative works or modified versions, and any portions 50 * thereof, and that both notices appear in supporting documentation. 51 * 52 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 53 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 54 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 55 * 56 * Carnegie Mellon requests users of this software to return to 57 * 58 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 59 * School of Computer Science 60 * Carnegie Mellon University 61 * Pittsburgh PA 15213-3890 62 * 63 * any improvements or extensions that they make and grant Carnegie the 64 * rights to redistribute these changes. 65 */ 66 67/* 68 * Resident memory management module. 69 */ 70#include "opt_ddb.h" 71 72#include <sys/param.h> 73#include <sys/systm.h> 74#include <sys/malloc.h> 75#include <sys/proc.h> 76#include <sys/queue.h> 77#include <sys/vmmeter.h> 78 79#include <vm/vm.h> 80#include <vm/vm_param.h> 81#include <vm/vm_prot.h> 82#include <vm/lock.h> 83#include <vm/vm_kern.h> 84#include <vm/vm_object.h> 85#include <vm/vm_page.h> 86#include <vm/vm_map.h> 87#include <vm/vm_pageout.h> 88#include <vm/vm_extern.h> 89 90#ifdef DDB 91extern void DDB_print_page_info __P((void)); 92#endif 93 94/* 95 * Associated with page of user-allocatable memory is a 96 * page structure. 97 */ 98 99static struct pglist *vm_page_buckets; /* Array of buckets */ 100static int vm_page_bucket_count; /* How big is array? */ 101static int vm_page_hash_mask; /* Mask for hash function */ 102 103struct pglist vm_page_queue_free; 104struct pglist vm_page_queue_zero; 105struct pglist vm_page_queue_active; 106struct pglist vm_page_queue_inactive; 107struct pglist vm_page_queue_cache; 108 109int no_queue; 110 111struct { 112 struct pglist *pl; 113 int *cnt; 114} vm_page_queues[PQ_CACHE+1] = { 115 {NULL, &no_queue}, 116 { &vm_page_queue_free, &cnt.v_free_count}, 117 { &vm_page_queue_zero, &cnt.v_free_count}, 118 { &vm_page_queue_inactive, &cnt.v_inactive_count}, 119 { &vm_page_queue_active, &cnt.v_active_count}, 120 { &vm_page_queue_cache, &cnt.v_cache_count} 121}; 122 123vm_page_t vm_page_array; 124static int vm_page_array_size; 125long first_page; 126static long last_page; 127static vm_size_t page_mask; 128static int page_shift; 129int vm_page_zero_count; 130 131/* 132 * map of contiguous valid DEV_BSIZE chunks in a page 133 * (this list is valid for page sizes upto 16*DEV_BSIZE) 134 */ 135static u_short vm_page_dev_bsize_chunks[] = { 136 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff, 137 0x1ff, 0x3ff, 0x7ff, 0xfff, 0x1fff, 0x3fff, 0x7fff, 0xffff 138}; 139 140static inline __pure int 141 vm_page_hash __P((vm_object_t object, vm_pindex_t pindex)) 142 __pure2; 143 144/* 145 * vm_set_page_size: 146 * 147 * Sets the page size, perhaps based upon the memory 148 * size. Must be called before any use of page-size 149 * dependent functions. 150 * 151 * Sets page_shift and page_mask from cnt.v_page_size. 152 */ 153void 154vm_set_page_size() 155{ 156 157 if (cnt.v_page_size == 0) 158 cnt.v_page_size = DEFAULT_PAGE_SIZE; 159 page_mask = cnt.v_page_size - 1; 160 if ((page_mask & cnt.v_page_size) != 0) 161 panic("vm_set_page_size: page size not a power of two"); 162 for (page_shift = 0;; page_shift++) 163 if ((1 << page_shift) == cnt.v_page_size) 164 break; 165} 166 167/* 168 * vm_page_startup: 169 * 170 * Initializes the resident memory module. 171 * 172 * Allocates memory for the page cells, and 173 * for the object/offset-to-page hash table headers. 174 * Each page cell is initialized and placed on the free list. 175 */ 176 177vm_offset_t 178vm_page_startup(starta, enda, vaddr) 179 register vm_offset_t starta; 180 vm_offset_t enda; 181 register vm_offset_t vaddr; 182{ 183 register vm_offset_t mapped; 184 register vm_page_t m; 185 register struct pglist *bucket; 186 vm_size_t npages, page_range; 187 register vm_offset_t new_start; 188 int i; 189 vm_offset_t pa; 190 int nblocks; 191 vm_offset_t first_managed_page; 192 193 /* the biggest memory array is the second group of pages */ 194 vm_offset_t start; 195 vm_offset_t biggestone, biggestsize; 196 197 vm_offset_t total; 198 199 total = 0; 200 biggestsize = 0; 201 biggestone = 0; 202 nblocks = 0; 203 vaddr = round_page(vaddr); 204 205 for (i = 0; phys_avail[i + 1]; i += 2) { 206 phys_avail[i] = round_page(phys_avail[i]); 207 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); 208 } 209 210 for (i = 0; phys_avail[i + 1]; i += 2) { 211 int size = phys_avail[i + 1] - phys_avail[i]; 212 213 if (size > biggestsize) { 214 biggestone = i; 215 biggestsize = size; 216 } 217 ++nblocks; 218 total += size; 219 } 220 221 start = phys_avail[biggestone]; 222 223 /* 224 * Initialize the queue headers for the free queue, the active queue 225 * and the inactive queue. 226 */ 227 228 TAILQ_INIT(&vm_page_queue_free); 229 TAILQ_INIT(&vm_page_queue_zero); 230 TAILQ_INIT(&vm_page_queue_active); 231 TAILQ_INIT(&vm_page_queue_inactive); 232 TAILQ_INIT(&vm_page_queue_cache); 233 234 /* 235 * Allocate (and initialize) the hash table buckets. 236 * 237 * The number of buckets MUST BE a power of 2, and the actual value is 238 * the next power of 2 greater than the number of physical pages in 239 * the system. 240 * 241 * Note: This computation can be tweaked if desired. 242 */ 243 vm_page_buckets = (struct pglist *) vaddr; 244 bucket = vm_page_buckets; 245 if (vm_page_bucket_count == 0) { 246 vm_page_bucket_count = 1; 247 while (vm_page_bucket_count < atop(total)) 248 vm_page_bucket_count <<= 1; 249 } 250 vm_page_hash_mask = vm_page_bucket_count - 1; 251 252 /* 253 * Validate these addresses. 254 */ 255 256 new_start = start + vm_page_bucket_count * sizeof(struct pglist); 257 new_start = round_page(new_start); 258 mapped = vaddr; 259 vaddr = pmap_map(mapped, start, new_start, 260 VM_PROT_READ | VM_PROT_WRITE); 261 start = new_start; 262 bzero((caddr_t) mapped, vaddr - mapped); 263 mapped = vaddr; 264 265 for (i = 0; i < vm_page_bucket_count; i++) { 266 TAILQ_INIT(bucket); 267 bucket++; 268 } 269 270 /* 271 * round (or truncate) the addresses to our page size. 272 */ 273 274 /* 275 * Pre-allocate maps and map entries that cannot be dynamically 276 * allocated via malloc(). The maps include the kernel_map and 277 * kmem_map which must be initialized before malloc() will work 278 * (obviously). Also could include pager maps which would be 279 * allocated before kmeminit. 280 * 281 * Allow some kernel map entries... this should be plenty since people 282 * shouldn't be cluttering up the kernel map (they should use their 283 * own maps). 284 */ 285 286 kentry_data_size = MAX_KMAP * sizeof(struct vm_map) + 287 MAX_KMAPENT * sizeof(struct vm_map_entry); 288 kentry_data_size = round_page(kentry_data_size); 289 kentry_data = (vm_offset_t) vaddr; 290 vaddr += kentry_data_size; 291 292 /* 293 * Validate these zone addresses. 294 */ 295 296 new_start = start + (vaddr - mapped); 297 pmap_map(mapped, start, new_start, VM_PROT_READ | VM_PROT_WRITE); 298 bzero((caddr_t) mapped, (vaddr - mapped)); 299 start = round_page(new_start); 300 301 /* 302 * Compute the number of pages of memory that will be available for 303 * use (taking into account the overhead of a page structure per 304 * page). 305 */ 306 307 first_page = phys_avail[0] / PAGE_SIZE; 308 last_page = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE; 309 310 page_range = last_page - (phys_avail[0] / PAGE_SIZE); 311 npages = (total - (page_range * sizeof(struct vm_page)) - 312 (start - phys_avail[biggestone])) / PAGE_SIZE; 313 314 /* 315 * Initialize the mem entry structures now, and put them in the free 316 * queue. 317 */ 318 319 vm_page_array = (vm_page_t) vaddr; 320 mapped = vaddr; 321 322 /* 323 * Validate these addresses. 324 */ 325 326 new_start = round_page(start + page_range * sizeof(struct vm_page)); 327 mapped = pmap_map(mapped, start, new_start, 328 VM_PROT_READ | VM_PROT_WRITE); 329 start = new_start; 330 331 first_managed_page = start / PAGE_SIZE; 332 333 /* 334 * Clear all of the page structures 335 */ 336 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); 337 vm_page_array_size = page_range; 338 339 cnt.v_page_count = 0; 340 cnt.v_free_count = 0; 341 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) { 342 if (i == biggestone) 343 pa = ptoa(first_managed_page); 344 else 345 pa = phys_avail[i]; 346 while (pa < phys_avail[i + 1] && npages-- > 0) { 347 ++cnt.v_page_count; 348 ++cnt.v_free_count; 349 m = PHYS_TO_VM_PAGE(pa); 350 m->queue = PQ_FREE; 351 m->flags = 0; 352 m->phys_addr = pa; 353 TAILQ_INSERT_TAIL(&vm_page_queue_free, m, pageq); 354 pa += PAGE_SIZE; 355 } 356 } 357 358 return (mapped); 359} 360 361/* 362 * vm_page_hash: 363 * 364 * Distributes the object/offset key pair among hash buckets. 365 * 366 * NOTE: This macro depends on vm_page_bucket_count being a power of 2. 367 */ 368static inline __pure int 369vm_page_hash(object, pindex) 370 vm_object_t object; 371 vm_pindex_t pindex; 372{ 373 return ((unsigned) object + pindex) & vm_page_hash_mask; 374} 375 376/* 377 * vm_page_insert: [ internal use only ] 378 * 379 * Inserts the given mem entry into the object/object-page 380 * table and object list. 381 * 382 * The object and page must be locked, and must be splhigh. 383 */ 384 385__inline void 386vm_page_insert(m, object, pindex) 387 register vm_page_t m; 388 register vm_object_t object; 389 register vm_pindex_t pindex; 390{ 391 register struct pglist *bucket; 392 393 if (m->flags & PG_TABLED) 394 panic("vm_page_insert: already inserted"); 395 396 /* 397 * Record the object/offset pair in this page 398 */ 399 400 m->object = object; 401 m->pindex = pindex; 402 403 /* 404 * Insert it into the object_object/offset hash table 405 */ 406 407 bucket = &vm_page_buckets[vm_page_hash(object, pindex)]; 408 TAILQ_INSERT_TAIL(bucket, m, hashq); 409 410 /* 411 * Now link into the object's list of backed pages. 412 */ 413 414 TAILQ_INSERT_TAIL(&object->memq, m, listq); 415 m->flags |= PG_TABLED; 416 417 /* 418 * And show that the object has one more resident page. 419 */ 420 421 object->resident_page_count++; 422} 423 424/* 425 * vm_page_remove: [ internal use only ] 426 * NOTE: used by device pager as well -wfj 427 * 428 * Removes the given mem entry from the object/offset-page 429 * table and the object page list. 430 * 431 * The object and page must be locked, and at splhigh. 432 */ 433 434__inline void 435vm_page_remove(m) 436 register vm_page_t m; 437{ 438 register struct pglist *bucket; 439 440 if (!(m->flags & PG_TABLED)) 441 return; 442 443 /* 444 * Remove from the object_object/offset hash table 445 */ 446 447 bucket = &vm_page_buckets[vm_page_hash(m->object, m->pindex)]; 448 TAILQ_REMOVE(bucket, m, hashq); 449 450 /* 451 * Now remove from the object's list of backed pages. 452 */ 453 454 TAILQ_REMOVE(&m->object->memq, m, listq); 455 456 /* 457 * And show that the object has one fewer resident page. 458 */ 459 460 m->object->resident_page_count--; 461 462 m->flags &= ~PG_TABLED; 463} 464 465/* 466 * vm_page_lookup: 467 * 468 * Returns the page associated with the object/offset 469 * pair specified; if none is found, NULL is returned. 470 * 471 * The object must be locked. No side effects. 472 */ 473 474vm_page_t 475vm_page_lookup(object, pindex) 476 register vm_object_t object; 477 register vm_pindex_t pindex; 478{ 479 register vm_page_t m; 480 register struct pglist *bucket; 481 int s; 482 483 /* 484 * Search the hash table for this object/offset pair 485 */ 486 487 bucket = &vm_page_buckets[vm_page_hash(object, pindex)]; 488 489 s = splvm(); 490 for (m = TAILQ_FIRST(bucket); m != NULL; m = TAILQ_NEXT(m,hashq)) { 491 if ((m->object == object) && (m->pindex == pindex)) { 492 splx(s); 493 return (m); 494 } 495 } 496 splx(s); 497 return (NULL); 498} 499 500/* 501 * vm_page_rename: 502 * 503 * Move the given memory entry from its 504 * current object to the specified target object/offset. 505 * 506 * The object must be locked. 507 */ 508void 509vm_page_rename(m, new_object, new_pindex) 510 register vm_page_t m; 511 register vm_object_t new_object; 512 vm_pindex_t new_pindex; 513{ 514 int s; 515 516 s = splvm(); 517 vm_page_remove(m); 518 vm_page_insert(m, new_object, new_pindex); 519 splx(s); 520} 521 522/* 523 * vm_page_unqueue must be called at splhigh(); 524 */ 525__inline void 526vm_page_unqueue(vm_page_t m) 527{ 528 int queue = m->queue; 529 if (queue == PQ_NONE) 530 return; 531 m->queue = PQ_NONE; 532 TAILQ_REMOVE(vm_page_queues[queue].pl, m, pageq); 533 --(*vm_page_queues[queue].cnt); 534 if (queue == PQ_CACHE) { 535 if ((cnt.v_cache_count + cnt.v_free_count) < 536 (cnt.v_free_reserved + cnt.v_cache_min)) 537 pagedaemon_wakeup(); 538 } 539 return; 540} 541 542/* 543 * vm_page_alloc: 544 * 545 * Allocate and return a memory cell associated 546 * with this VM object/offset pair. 547 * 548 * page_req classes: 549 * VM_ALLOC_NORMAL normal process request 550 * VM_ALLOC_SYSTEM system *really* needs a page 551 * VM_ALLOC_INTERRUPT interrupt time request 552 * VM_ALLOC_ZERO zero page 553 * 554 * Object must be locked. 555 */ 556vm_page_t 557vm_page_alloc(object, pindex, page_req) 558 vm_object_t object; 559 vm_pindex_t pindex; 560 int page_req; 561{ 562 register vm_page_t m; 563 int queue; 564 int s; 565 566#ifdef DIAGNOSTIC 567 m = vm_page_lookup(object, pindex); 568 if (m) 569 panic("vm_page_alloc: page already allocated"); 570#endif 571 572 if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) { 573 page_req = VM_ALLOC_SYSTEM; 574 }; 575 576 s = splvm(); 577 578 switch (page_req) { 579 580 case VM_ALLOC_NORMAL: 581 if (cnt.v_free_count >= cnt.v_free_reserved) { 582 m = TAILQ_FIRST(&vm_page_queue_free); 583 if (m == NULL) { 584 --vm_page_zero_count; 585 m = TAILQ_FIRST(&vm_page_queue_zero); 586 } 587 } else { 588 m = TAILQ_FIRST(&vm_page_queue_cache); 589 if (m == NULL) { 590 splx(s); 591 pagedaemon_wakeup(); 592 return (NULL); 593 } 594 } 595 break; 596 597 case VM_ALLOC_ZERO: 598 if (cnt.v_free_count >= cnt.v_free_reserved) { 599 m = TAILQ_FIRST(&vm_page_queue_zero); 600 if (m) { 601 --vm_page_zero_count; 602 } else { 603 m = TAILQ_FIRST(&vm_page_queue_free); 604 } 605 } else { 606 m = TAILQ_FIRST(&vm_page_queue_cache); 607 if (m == NULL) { 608 splx(s); 609 pagedaemon_wakeup(); 610 return (NULL); 611 } 612 } 613 break; 614 615 case VM_ALLOC_SYSTEM: 616 if ((cnt.v_free_count >= cnt.v_free_reserved) || 617 ((cnt.v_cache_count == 0) && 618 (cnt.v_free_count >= cnt.v_interrupt_free_min))) { 619 m = TAILQ_FIRST(&vm_page_queue_free); 620 if (m == NULL) { 621 --vm_page_zero_count; 622 m = TAILQ_FIRST(&vm_page_queue_zero); 623 } 624 } else { 625 m = TAILQ_FIRST(&vm_page_queue_cache); 626 if (m == NULL) { 627 splx(s); 628 pagedaemon_wakeup(); 629 return (NULL); 630 } 631 } 632 break; 633 634 case VM_ALLOC_INTERRUPT: 635 if (cnt.v_free_count > 0) { 636 m = TAILQ_FIRST(&vm_page_queue_free); 637 if (m == NULL) { 638 --vm_page_zero_count; 639 m = TAILQ_FIRST(&vm_page_queue_zero); 640 } 641 } else { 642 splx(s); 643 pagedaemon_wakeup(); 644 return (NULL); 645 } 646 break; 647 648 default: 649 panic("vm_page_alloc: invalid allocation class"); 650 } 651 652 queue = m->queue; 653 TAILQ_REMOVE(vm_page_queues[queue].pl, m, pageq); 654 --(*vm_page_queues[queue].cnt); 655 if (queue == PQ_ZERO) { 656 m->flags = PG_ZERO|PG_BUSY; 657 } else if (queue == PQ_CACHE) { 658 vm_page_remove(m); 659 m->flags = PG_BUSY; 660 } else { 661 m->flags = PG_BUSY; 662 } 663 m->wire_count = 0; 664 m->act_count = 0; 665 m->hold_count = 0; 666 m->busy = 0; 667 m->valid = 0; 668 m->dirty = 0; 669 m->queue = PQ_NONE; 670 671 /* XXX before splx until vm_page_insert is safe */ 672 vm_page_insert(m, object, pindex); 673 674 splx(s); 675 676 /* 677 * Don't wakeup too often - wakeup the pageout daemon when 678 * we would be nearly out of memory. 679 */ 680 if (((cnt.v_free_count + cnt.v_cache_count) < 681 (cnt.v_free_reserved + cnt.v_cache_min)) || 682 (cnt.v_free_count < cnt.v_pageout_free_min)) 683 pagedaemon_wakeup(); 684 685 return (m); 686} 687 688/* 689 * vm_page_activate: 690 * 691 * Put the specified page on the active list (if appropriate). 692 * 693 * The page queues must be locked. 694 */ 695void 696vm_page_activate(m) 697 register vm_page_t m; 698{ 699 int s; 700 701 s = splvm(); 702 if (m->queue == PQ_ACTIVE) 703 panic("vm_page_activate: already active"); 704 705 if (m->queue == PQ_CACHE) 706 cnt.v_reactivated++; 707 708 vm_page_unqueue(m); 709 710 if (m->wire_count == 0) { 711 if (m->act_count < 5) 712 m->act_count = 5; 713 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 714 m->queue = PQ_ACTIVE; 715 cnt.v_active_count++; 716 } 717 splx(s); 718} 719 720/* 721 * vm_page_free: 722 * 723 * Returns the given page to the free list, 724 * disassociating it with any VM object. 725 * 726 * Object and page must be locked prior to entry. 727 */ 728void 729vm_page_free(m) 730 register vm_page_t m; 731{ 732 int s; 733 int flags = m->flags; 734 735 s = splvm(); 736 if (m->busy || (flags & PG_BUSY) || (m->queue == PQ_FREE)) { 737 printf("vm_page_free: pindex(%ld), busy(%d), PG_BUSY(%d)\n", 738 m->pindex, m->busy, (flags & PG_BUSY) ? 1 : 0); 739 if (m->queue == PQ_FREE) 740 panic("vm_page_free: freeing free page"); 741 else 742 panic("vm_page_free: freeing busy page"); 743 } 744 745 if (m->hold_count) { 746 panic("freeing held page, count=%d, pindex=%d(0x%x)", 747 m->hold_count, m->pindex, m->pindex); 748 } 749 750 vm_page_remove(m); 751 vm_page_unqueue(m); 752 753 if ((flags & PG_FICTITIOUS) == 0) { 754 if (m->wire_count) { 755 if (m->wire_count > 1) { 756 printf("vm_page_free: wire count > 1 (%d)", m->wire_count); 757 panic("vm_page_free: invalid wire count"); 758 } 759 cnt.v_wire_count--; 760 m->wire_count = 0; 761 } 762 m->queue = PQ_FREE; 763 764 /* 765 * If the pageout process is grabbing the page, it is likely 766 * that the page is NOT in the cache. It is more likely that 767 * the page will be partially in the cache if it is being 768 * explicitly freed. 769 */ 770 if (curproc == pageproc) { 771 TAILQ_INSERT_TAIL(&vm_page_queue_free, m, pageq); 772 } else { 773 TAILQ_INSERT_HEAD(&vm_page_queue_free, m, pageq); 774 } 775 776 splx(s); 777 /* 778 * if pageout daemon needs pages, then tell it that there are 779 * some free. 780 */ 781 if (vm_pageout_pages_needed) { 782 wakeup(&vm_pageout_pages_needed); 783 vm_pageout_pages_needed = 0; 784 } 785 786 cnt.v_free_count++; 787 /* 788 * wakeup processes that are waiting on memory if we hit a 789 * high water mark. And wakeup scheduler process if we have 790 * lots of memory. this process will swapin processes. 791 */ 792 if ((cnt.v_free_count + cnt.v_cache_count) == cnt.v_free_min) { 793 wakeup(&cnt.v_free_count); 794 } 795 } else { 796 splx(s); 797 } 798 cnt.v_tfree++; 799} 800 801 802/* 803 * vm_page_wire: 804 * 805 * Mark this page as wired down by yet 806 * another map, removing it from paging queues 807 * as necessary. 808 * 809 * The page queues must be locked. 810 */ 811void 812vm_page_wire(m) 813 register vm_page_t m; 814{ 815 int s; 816 817 if (m->wire_count == 0) { 818 s = splvm(); 819 vm_page_unqueue(m); 820 splx(s); 821 cnt.v_wire_count++; 822 } 823 m->wire_count++; 824 m->flags |= PG_MAPPED; 825} 826 827/* 828 * vm_page_unwire: 829 * 830 * Release one wiring of this page, potentially 831 * enabling it to be paged again. 832 * 833 * The page queues must be locked. 834 */ 835void 836vm_page_unwire(m) 837 register vm_page_t m; 838{ 839 int s; 840 841 s = splvm(); 842 843 if (m->wire_count > 0) 844 m->wire_count--; 845 846 if (m->wire_count == 0) { 847 cnt.v_wire_count--; 848 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 849 m->queue = PQ_ACTIVE; 850 if (m->act_count < 5) 851 m->act_count = 5; 852 cnt.v_active_count++; 853 } 854 splx(s); 855} 856 857 858/* 859 * vm_page_deactivate: 860 * 861 * Returns the given page to the inactive list, 862 * indicating that no physical maps have access 863 * to this page. [Used by the physical mapping system.] 864 * 865 * The page queues must be locked. 866 */ 867void 868vm_page_deactivate(m) 869 register vm_page_t m; 870{ 871 int spl; 872 873 /* 874 * Only move active pages -- ignore locked or already inactive ones. 875 * 876 * XXX: sometimes we get pages which aren't wired down or on any queue - 877 * we need to put them on the inactive queue also, otherwise we lose 878 * track of them. Paul Mackerras (paulus@cs.anu.edu.au) 9-Jan-93. 879 */ 880 if (m->queue == PQ_INACTIVE) 881 return; 882 883 spl = splvm(); 884 if (m->wire_count == 0 && m->hold_count == 0) { 885 if (m->queue == PQ_CACHE) 886 cnt.v_reactivated++; 887 vm_page_unqueue(m); 888 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); 889 m->queue = PQ_INACTIVE; 890 cnt.v_inactive_count++; 891 } 892 splx(spl); 893} 894 895/* 896 * vm_page_cache 897 * 898 * Put the specified page onto the page cache queue (if appropriate). 899 */ 900void 901vm_page_cache(m) 902 register vm_page_t m; 903{ 904 int s; 905 906 if ((m->flags & PG_BUSY) || m->busy || m->wire_count) { 907 printf("vm_page_cache: attempting to cache busy page\n"); 908 return; 909 } 910 if (m->queue == PQ_CACHE) 911 return; 912 913 vm_page_protect(m, VM_PROT_NONE); 914 if (m->dirty != 0) { 915 panic("vm_page_cache: caching a dirty page, pindex: %d", m->pindex); 916 } 917 s = splvm(); 918 vm_page_unqueue(m); 919 TAILQ_INSERT_TAIL(&vm_page_queue_cache, m, pageq); 920 m->queue = PQ_CACHE; 921 cnt.v_cache_count++; 922 if ((cnt.v_free_count + cnt.v_cache_count) == cnt.v_free_min) { 923 wakeup(&cnt.v_free_count); 924 wakeup(&proc0); 925 } 926 if (vm_pageout_pages_needed) { 927 wakeup(&vm_pageout_pages_needed); 928 vm_pageout_pages_needed = 0; 929 } 930 splx(s); 931} 932 933 934/* 935 * mapping function for valid bits or for dirty bits in 936 * a page 937 */ 938inline int 939vm_page_bits(int base, int size) 940{ 941 u_short chunk; 942 943 if ((base == 0) && (size >= PAGE_SIZE)) 944 return VM_PAGE_BITS_ALL; 945 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 946 base = (base % PAGE_SIZE) / DEV_BSIZE; 947 chunk = vm_page_dev_bsize_chunks[size / DEV_BSIZE]; 948 return (chunk << base) & VM_PAGE_BITS_ALL; 949} 950 951/* 952 * set a page valid and clean 953 */ 954void 955vm_page_set_validclean(m, base, size) 956 vm_page_t m; 957 int base; 958 int size; 959{ 960 int pagebits = vm_page_bits(base, size); 961 m->valid |= pagebits; 962 m->dirty &= ~pagebits; 963 if( base == 0 && size == PAGE_SIZE) 964 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 965} 966 967/* 968 * set a page (partially) invalid 969 */ 970void 971vm_page_set_invalid(m, base, size) 972 vm_page_t m; 973 int base; 974 int size; 975{ 976 int bits; 977 978 m->valid &= ~(bits = vm_page_bits(base, size)); 979 if (m->valid == 0) 980 m->dirty &= ~bits; 981} 982 983/* 984 * is (partial) page valid? 985 */ 986int 987vm_page_is_valid(m, base, size) 988 vm_page_t m; 989 int base; 990 int size; 991{ 992 int bits = vm_page_bits(base, size); 993 994 if (m->valid && ((m->valid & bits) == bits)) 995 return 1; 996 else 997 return 0; 998} 999 1000void 1001vm_page_test_dirty(m) 1002 vm_page_t m; 1003{ 1004 if ((m->dirty != VM_PAGE_BITS_ALL) && 1005 pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1006 m->dirty = VM_PAGE_BITS_ALL; 1007 } 1008} 1009 1010/* 1011 * This interface is for merging with malloc() someday. 1012 * Even if we never implement compaction so that contiguous allocation 1013 * works after initialization time, malloc()'s data structures are good 1014 * for statistics and for allocations of less than a page. 1015 */ 1016void * 1017contigmalloc(size, type, flags, low, high, alignment, boundary) 1018 unsigned long size; /* should be size_t here and for malloc() */ 1019 int type; 1020 int flags; 1021 unsigned long low; 1022 unsigned long high; 1023 unsigned long alignment; 1024 unsigned long boundary; 1025{ 1026 int i, s, start; 1027 vm_offset_t addr, phys, tmp_addr; 1028 vm_page_t pga = vm_page_array; 1029 1030 size = round_page(size); 1031 if (size == 0) 1032 panic("vm_page_alloc_contig: size must not be 0"); 1033 if ((alignment & (alignment - 1)) != 0) 1034 panic("vm_page_alloc_contig: alignment must be a power of 2"); 1035 if ((boundary & (boundary - 1)) != 0) 1036 panic("vm_page_alloc_contig: boundary must be a power of 2"); 1037 1038 start = 0; 1039 s = splvm(); 1040again: 1041 /* 1042 * Find first page in array that is free, within range, aligned, and 1043 * such that the boundary won't be crossed. 1044 */ 1045 for (i = start; i < cnt.v_page_count; i++) { 1046 phys = VM_PAGE_TO_PHYS(&pga[i]); 1047 if ((pga[i].queue == PQ_FREE) && 1048 (phys >= low) && (phys < high) && 1049 ((phys & (alignment - 1)) == 0) && 1050 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0)) 1051 break; 1052 } 1053 1054 /* 1055 * If the above failed or we will exceed the upper bound, fail. 1056 */ 1057 if ((i == cnt.v_page_count) || 1058 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 1059 splx(s); 1060 return (NULL); 1061 } 1062 start = i; 1063 1064 /* 1065 * Check successive pages for contiguous and free. 1066 */ 1067 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 1068 if ((VM_PAGE_TO_PHYS(&pga[i]) != 1069 (VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) || 1070 (pga[i].queue != PQ_FREE)) { 1071 start++; 1072 goto again; 1073 } 1074 } 1075 1076 /* 1077 * We've found a contiguous chunk that meets are requirements. 1078 * Allocate kernel VM, unfree and assign the physical pages to it and 1079 * return kernel VM pointer. 1080 */ 1081 tmp_addr = addr = kmem_alloc_pageable(kernel_map, size); 1082 if (addr == 0) { 1083 splx(s); 1084 return (NULL); 1085 } 1086 1087 for (i = start; i < (start + size / PAGE_SIZE); i++) { 1088 vm_page_t m = &pga[i]; 1089 1090 TAILQ_REMOVE(&vm_page_queue_free, m, pageq); 1091 cnt.v_free_count--; 1092 m->valid = VM_PAGE_BITS_ALL; 1093 m->flags = 0; 1094 m->dirty = 0; 1095 m->wire_count = 0; 1096 m->busy = 0; 1097 m->queue = PQ_NONE; 1098 vm_page_insert(m, kernel_object, 1099 OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS)); 1100 vm_page_wire(m); 1101 pmap_kenter(tmp_addr, VM_PAGE_TO_PHYS(m)); 1102 tmp_addr += PAGE_SIZE; 1103 } 1104 1105 splx(s); 1106 return ((void *)addr); 1107} 1108 1109vm_offset_t 1110vm_page_alloc_contig(size, low, high, alignment) 1111 vm_offset_t size; 1112 vm_offset_t low; 1113 vm_offset_t high; 1114 vm_offset_t alignment; 1115{ 1116 return ((vm_offset_t)contigmalloc(size, M_DEVBUF, M_NOWAIT, low, high, 1117 alignment, 0ul)); 1118} 1119#ifdef DDB 1120void 1121DDB_print_page_info(void) 1122{ 1123 printf("cnt.v_free_count: %d\n", cnt.v_free_count); 1124 printf("cnt.v_cache_count: %d\n", cnt.v_cache_count); 1125 printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count); 1126 printf("cnt.v_active_count: %d\n", cnt.v_active_count); 1127 printf("cnt.v_wire_count: %d\n", cnt.v_wire_count); 1128 printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved); 1129 printf("cnt.v_free_min: %d\n", cnt.v_free_min); 1130 printf("cnt.v_free_target: %d\n", cnt.v_free_target); 1131 printf("cnt.v_cache_min: %d\n", cnt.v_cache_min); 1132 printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target); 1133} 1134#endif 1135