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 1.35 1995/09/03 19:57:25 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 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/proc.h> 74 75 #include <vm/vm.h> 76 #include <vm/vm_kern.h> 77 #include <vm/vm_page.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_pageout.h> 80 81 /* 82 * Associated with page of user-allocatable memory is a 83 * page structure. 84 */ 85 86 struct pglist *vm_page_buckets; /* Array of buckets */ 87 int vm_page_bucket_count; /* How big is array? */ 88 int vm_page_hash_mask; /* Mask for hash function */ 89 90 struct pglist vm_page_queue_free; 91 struct pglist vm_page_queue_zero; 92 struct pglist vm_page_queue_active; 93 struct pglist vm_page_queue_inactive; 94 struct pglist vm_page_queue_cache; 95 96 /* has physical page allocation been initialized? */ 97 boolean_t vm_page_startup_initialized; 98 99 vm_page_t vm_page_array; 100 int vm_page_array_size; 101 long first_page; 102 long last_page; 103 vm_offset_t first_phys_addr; 104 vm_offset_t last_phys_addr; 105 vm_size_t page_mask; 106 int page_shift; 107 108 /* 109 * map of contiguous valid DEV_BSIZE chunks in a page 110 * (this list is valid for page sizes upto 16*DEV_BSIZE) 111 */ 112 static u_short vm_page_dev_bsize_chunks[] = { 113 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff, 114 0x1ff, 0x3ff, 0x7ff, 0xfff, 0x1fff, 0x3fff, 0x7fff, 0xffff 115 }; 116 117 118 /* 119 * vm_set_page_size: 120 * 121 * Sets the page size, perhaps based upon the memory 122 * size. Must be called before any use of page-size 123 * dependent functions. 124 * 125 * Sets page_shift and page_mask from cnt.v_page_size. 126 */ 127 void 128 vm_set_page_size() 129 { 130 131 if (cnt.v_page_size == 0) 132 cnt.v_page_size = DEFAULT_PAGE_SIZE; 133 page_mask = cnt.v_page_size - 1; 134 if ((page_mask & cnt.v_page_size) != 0) 135 panic("vm_set_page_size: page size not a power of two"); 136 for (page_shift = 0;; page_shift++) 137 if ((1 << page_shift) == cnt.v_page_size) 138 break; 139 } 140 141 /* 142 * vm_page_startup: 143 * 144 * Initializes the resident memory module. 145 * 146 * Allocates memory for the page cells, and 147 * for the object/offset-to-page hash table headers. 148 * Each page cell is initialized and placed on the free list. 149 */ 150 151 vm_offset_t 152 vm_page_startup(starta, enda, vaddr) 153 register vm_offset_t starta; 154 vm_offset_t enda; 155 register vm_offset_t vaddr; 156 { 157 register vm_offset_t mapped; 158 register vm_page_t m; 159 register struct pglist *bucket; 160 vm_size_t npages, page_range; 161 register vm_offset_t new_start; 162 int i; 163 vm_offset_t pa; 164 int nblocks; 165 vm_offset_t first_managed_page; 166 167 /* the biggest memory array is the second group of pages */ 168 vm_offset_t start; 169 vm_offset_t biggestone, biggestsize; 170 171 vm_offset_t total; 172 173 total = 0; 174 biggestsize = 0; 175 biggestone = 0; 176 nblocks = 0; 177 vaddr = round_page(vaddr); 178 179 for (i = 0; phys_avail[i + 1]; i += 2) { 180 phys_avail[i] = round_page(phys_avail[i]); 181 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); 182 } 183 184 for (i = 0; phys_avail[i + 1]; i += 2) { 185 int size = phys_avail[i + 1] - phys_avail[i]; 186 187 if (size > biggestsize) { 188 biggestone = i; 189 biggestsize = size; 190 } 191 ++nblocks; 192 total += size; 193 } 194 195 start = phys_avail[biggestone]; 196 197 /* 198 * Initialize the queue headers for the free queue, the active queue 199 * and the inactive queue. 200 */ 201 202 TAILQ_INIT(&vm_page_queue_free); 203 TAILQ_INIT(&vm_page_queue_zero); 204 TAILQ_INIT(&vm_page_queue_active); 205 TAILQ_INIT(&vm_page_queue_inactive); 206 TAILQ_INIT(&vm_page_queue_cache); 207 208 /* 209 * Allocate (and initialize) the hash table buckets. 210 * 211 * The number of buckets MUST BE a power of 2, and the actual value is 212 * the next power of 2 greater than the number of physical pages in 213 * the system. 214 * 215 * Note: This computation can be tweaked if desired. 216 */ 217 vm_page_buckets = (struct pglist *) vaddr; 218 bucket = vm_page_buckets; 219 if (vm_page_bucket_count == 0) { 220 vm_page_bucket_count = 1; 221 while (vm_page_bucket_count < atop(total)) 222 vm_page_bucket_count <<= 1; 223 } 224 vm_page_hash_mask = vm_page_bucket_count - 1; 225 226 /* 227 * Validate these addresses. 228 */ 229 230 new_start = start + vm_page_bucket_count * sizeof(struct pglist); 231 new_start = round_page(new_start); 232 mapped = vaddr; 233 vaddr = pmap_map(mapped, start, new_start, 234 VM_PROT_READ | VM_PROT_WRITE); 235 start = new_start; 236 bzero((caddr_t) mapped, vaddr - mapped); 237 mapped = vaddr; 238 239 for (i = 0; i < vm_page_bucket_count; i++) { 240 TAILQ_INIT(bucket); 241 bucket++; 242 } 243 244 /* 245 * round (or truncate) the addresses to our page size. 246 */ 247 248 /* 249 * Pre-allocate maps and map entries that cannot be dynamically 250 * allocated via malloc(). The maps include the kernel_map and 251 * kmem_map which must be initialized before malloc() will work 252 * (obviously). Also could include pager maps which would be 253 * allocated before kmeminit. 254 * 255 * Allow some kernel map entries... this should be plenty since people 256 * shouldn't be cluttering up the kernel map (they should use their 257 * own maps). 258 */ 259 260 kentry_data_size = MAX_KMAP * sizeof(struct vm_map) + 261 MAX_KMAPENT * sizeof(struct vm_map_entry); 262 kentry_data_size = round_page(kentry_data_size); 263 kentry_data = (vm_offset_t) vaddr; 264 vaddr += kentry_data_size; 265 266 /* 267 * Validate these zone addresses. 268 */ 269 270 new_start = start + (vaddr - mapped); 271 pmap_map(mapped, start, new_start, VM_PROT_READ | VM_PROT_WRITE); 272 bzero((caddr_t) mapped, (vaddr - mapped)); 273 start = round_page(new_start); 274 275 /* 276 * Compute the number of pages of memory that will be available for 277 * use (taking into account the overhead of a page structure per 278 * page). 279 */ 280 281 first_page = phys_avail[0] / PAGE_SIZE; 282 last_page = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE; 283 284 page_range = last_page - (phys_avail[0] / PAGE_SIZE); 285 npages = (total - (page_range * sizeof(struct vm_page)) - 286 (start - phys_avail[biggestone])) / PAGE_SIZE; 287 288 /* 289 * Initialize the mem entry structures now, and put them in the free 290 * queue. 291 */ 292 293 vm_page_array = (vm_page_t) vaddr; 294 mapped = vaddr; 295 296 297 /* 298 * Validate these addresses. 299 */ 300 301 new_start = round_page(start + page_range * sizeof(struct vm_page)); 302 mapped = pmap_map(mapped, start, new_start, 303 VM_PROT_READ | VM_PROT_WRITE); 304 start = new_start; 305 306 first_managed_page = start / PAGE_SIZE; 307 308 /* 309 * Clear all of the page structures 310 */ 311 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); 312 vm_page_array_size = page_range; 313 314 cnt.v_page_count = 0; 315 cnt.v_free_count = 0; 316 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) { 317 if (i == biggestone) 318 pa = ptoa(first_managed_page); 319 else 320 pa = phys_avail[i]; 321 while (pa < phys_avail[i + 1] && npages-- > 0) { 322 ++cnt.v_page_count; 323 ++cnt.v_free_count; 324 m = PHYS_TO_VM_PAGE(pa); 325 m->flags = PG_FREE; 326 m->phys_addr = pa; 327 TAILQ_INSERT_TAIL(&vm_page_queue_free, m, pageq); 328 pa += PAGE_SIZE; 329 } 330 } 331 332 return (mapped); 333 } 334 335 /* 336 * vm_page_hash: 337 * 338 * Distributes the object/offset key pair among hash buckets. 339 * 340 * NOTE: This macro depends on vm_page_bucket_count being a power of 2. 341 */ 342 inline const int 343 vm_page_hash(object, offset) 344 vm_object_t object; 345 vm_offset_t offset; 346 { 347 return ((unsigned) object + offset / NBPG) & vm_page_hash_mask; 348 } 349 350 /* 351 * vm_page_insert: [ internal use only ] 352 * 353 * Inserts the given mem entry into the object/object-page 354 * table and object list. 355 * 356 * The object and page must be locked, and must be splhigh. 357 */ 358 359 inline void 360 vm_page_insert(mem, object, offset) 361 register vm_page_t mem; 362 register vm_object_t object; 363 register vm_offset_t offset; 364 { 365 register struct pglist *bucket; 366 367 if (mem->flags & PG_TABLED) 368 panic("vm_page_insert: already inserted"); 369 370 /* 371 * Record the object/offset pair in this page 372 */ 373 374 mem->object = object; 375 mem->offset = offset; 376 377 /* 378 * Insert it into the object_object/offset hash table 379 */ 380 381 bucket = &vm_page_buckets[vm_page_hash(object, offset)]; 382 TAILQ_INSERT_TAIL(bucket, mem, hashq); 383 384 /* 385 * Now link into the object's list of backed pages. 386 */ 387 388 TAILQ_INSERT_TAIL(&object->memq, mem, listq); 389 mem->flags |= PG_TABLED; 390 391 /* 392 * And show that the object has one more resident page. 393 */ 394 395 object->resident_page_count++; 396 } 397 398 /* 399 * vm_page_remove: [ internal use only ] 400 * NOTE: used by device pager as well -wfj 401 * 402 * Removes the given mem entry from the object/offset-page 403 * table and the object page list. 404 * 405 * The object and page must be locked, and at splhigh. 406 */ 407 408 inline void 409 vm_page_remove(mem) 410 register vm_page_t mem; 411 { 412 register struct pglist *bucket; 413 414 if (!(mem->flags & PG_TABLED)) 415 return; 416 417 /* 418 * Remove from the object_object/offset hash table 419 */ 420 421 bucket = &vm_page_buckets[vm_page_hash(mem->object, mem->offset)]; 422 TAILQ_REMOVE(bucket, mem, hashq); 423 424 /* 425 * Now remove from the object's list of backed pages. 426 */ 427 428 TAILQ_REMOVE(&mem->object->memq, mem, listq); 429 430 /* 431 * And show that the object has one fewer resident page. 432 */ 433 434 mem->object->resident_page_count--; 435 436 mem->flags &= ~PG_TABLED; 437 } 438 439 /* 440 * vm_page_lookup: 441 * 442 * Returns the page associated with the object/offset 443 * pair specified; if none is found, NULL is returned. 444 * 445 * The object must be locked. No side effects. 446 */ 447 448 vm_page_t 449 vm_page_lookup(object, offset) 450 register vm_object_t object; 451 register vm_offset_t offset; 452 { 453 register vm_page_t mem; 454 register struct pglist *bucket; 455 int s; 456 457 /* 458 * Search the hash table for this object/offset pair 459 */ 460 461 bucket = &vm_page_buckets[vm_page_hash(object, offset)]; 462 463 s = splhigh(); 464 for (mem = bucket->tqh_first; mem != NULL; mem = mem->hashq.tqe_next) { 465 if ((mem->object == object) && (mem->offset == offset)) { 466 splx(s); 467 return (mem); 468 } 469 } 470 471 splx(s); 472 return (NULL); 473 } 474 475 /* 476 * vm_page_rename: 477 * 478 * Move the given memory entry from its 479 * current object to the specified target object/offset. 480 * 481 * The object must be locked. 482 */ 483 void 484 vm_page_rename(mem, new_object, new_offset) 485 register vm_page_t mem; 486 register vm_object_t new_object; 487 vm_offset_t new_offset; 488 { 489 int s; 490 491 if (mem->object == new_object) 492 return; 493 494 s = splhigh(); 495 vm_page_remove(mem); 496 vm_page_insert(mem, new_object, new_offset); 497 splx(s); 498 } 499 500 /* 501 * vm_page_unqueue must be called at splhigh(); 502 */ 503 inline void 504 vm_page_unqueue(vm_page_t mem) 505 { 506 int origflags; 507 508 origflags = mem->flags; 509 510 if ((origflags & (PG_ACTIVE|PG_INACTIVE|PG_CACHE)) == 0) 511 return; 512 513 if (origflags & PG_ACTIVE) { 514 TAILQ_REMOVE(&vm_page_queue_active, mem, pageq); 515 cnt.v_active_count--; 516 mem->flags &= ~PG_ACTIVE; 517 } else if (origflags & PG_INACTIVE) { 518 TAILQ_REMOVE(&vm_page_queue_inactive, mem, pageq); 519 cnt.v_inactive_count--; 520 mem->flags &= ~PG_INACTIVE; 521 } else if (origflags & PG_CACHE) { 522 TAILQ_REMOVE(&vm_page_queue_cache, mem, pageq); 523 cnt.v_cache_count--; 524 mem->flags &= ~PG_CACHE; 525 if (cnt.v_cache_count + cnt.v_free_count < cnt.v_free_reserved) 526 pagedaemon_wakeup(); 527 } 528 return; 529 } 530 531 /* 532 * vm_page_alloc: 533 * 534 * Allocate and return a memory cell associated 535 * with this VM object/offset pair. 536 * 537 * page_req classes: 538 * VM_ALLOC_NORMAL normal process request 539 * VM_ALLOC_SYSTEM system *really* needs a page 540 * VM_ALLOC_INTERRUPT interrupt time request 541 * or in: 542 * VM_ALLOC_ZERO zero page 543 * 544 * Object must be locked. 545 */ 546 vm_page_t 547 vm_page_alloc(object, offset, page_req) 548 vm_object_t object; 549 vm_offset_t offset; 550 int page_req; 551 { 552 register vm_page_t mem; 553 int s; 554 555 #ifdef DIAGNOSTIC 556 if (offset != trunc_page(offset)) 557 panic("vm_page_alloc: offset not page aligned"); 558 559 #if 0 560 mem = vm_page_lookup(object, offset); 561 if (mem) 562 panic("vm_page_alloc: page already allocated"); 563 #endif 564 #endif 565 566 if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) { 567 page_req = VM_ALLOC_SYSTEM; 568 }; 569 570 s = splhigh(); 571 572 switch ((page_req & ~(VM_ALLOC_ZERO))) { 573 case VM_ALLOC_NORMAL: 574 if (cnt.v_free_count >= cnt.v_free_reserved) { 575 if (page_req & VM_ALLOC_ZERO) { 576 mem = vm_page_queue_zero.tqh_first; 577 if (mem) { 578 TAILQ_REMOVE(&vm_page_queue_zero, mem, pageq); 579 mem->flags = PG_BUSY|PG_ZERO; 580 } else { 581 mem = vm_page_queue_free.tqh_first; 582 TAILQ_REMOVE(&vm_page_queue_free, mem, pageq); 583 mem->flags = PG_BUSY; 584 } 585 } else { 586 mem = vm_page_queue_free.tqh_first; 587 if (mem) { 588 TAILQ_REMOVE(&vm_page_queue_free, mem, pageq); 589 mem->flags = PG_BUSY; 590 } else { 591 mem = vm_page_queue_zero.tqh_first; 592 TAILQ_REMOVE(&vm_page_queue_zero, mem, pageq); 593 mem->flags = PG_BUSY|PG_ZERO; 594 } 595 } 596 cnt.v_free_count--; 597 } else { 598 mem = vm_page_queue_cache.tqh_first; 599 if (mem != NULL) { 600 TAILQ_REMOVE(&vm_page_queue_cache, mem, pageq); 601 vm_page_remove(mem); 602 mem->flags = PG_BUSY; 603 cnt.v_cache_count--; 604 } else { 605 splx(s); 606 pagedaemon_wakeup(); 607 return (NULL); 608 } 609 } 610 break; 611 612 case VM_ALLOC_SYSTEM: 613 if ((cnt.v_free_count >= cnt.v_free_reserved) || 614 ((cnt.v_cache_count == 0) && 615 (cnt.v_free_count >= cnt.v_interrupt_free_min))) { 616 if (page_req & VM_ALLOC_ZERO) { 617 mem = vm_page_queue_zero.tqh_first; 618 if (mem) { 619 TAILQ_REMOVE(&vm_page_queue_zero, mem, pageq); 620 mem->flags = PG_BUSY|PG_ZERO; 621 } else { 622 mem = vm_page_queue_free.tqh_first; 623 TAILQ_REMOVE(&vm_page_queue_free, mem, pageq); 624 mem->flags = PG_BUSY; 625 } 626 } else { 627 mem = vm_page_queue_free.tqh_first; 628 if (mem) { 629 TAILQ_REMOVE(&vm_page_queue_free, mem, pageq); 630 mem->flags = PG_BUSY; 631 } else { 632 mem = vm_page_queue_zero.tqh_first; 633 TAILQ_REMOVE(&vm_page_queue_zero, mem, pageq); 634 mem->flags = PG_BUSY|PG_ZERO; 635 } 636 } 637 cnt.v_free_count--; 638 } else { 639 mem = vm_page_queue_cache.tqh_first; 640 if (mem != NULL) { 641 TAILQ_REMOVE(&vm_page_queue_cache, mem, pageq); 642 vm_page_remove(mem); 643 mem->flags = PG_BUSY; 644 cnt.v_cache_count--; 645 } else { 646 splx(s); 647 pagedaemon_wakeup(); 648 return (NULL); 649 } 650 } 651 break; 652 653 case VM_ALLOC_INTERRUPT: 654 if (cnt.v_free_count > 0) { 655 mem = vm_page_queue_free.tqh_first; 656 if (mem) { 657 TAILQ_REMOVE(&vm_page_queue_free, mem, pageq); 658 mem->flags = PG_BUSY; 659 } else { 660 mem = vm_page_queue_zero.tqh_first; 661 TAILQ_REMOVE(&vm_page_queue_zero, mem, pageq); 662 mem->flags = PG_BUSY|PG_ZERO; 663 } 664 cnt.v_free_count--; 665 } else { 666 splx(s); 667 pagedaemon_wakeup(); 668 return NULL; 669 } 670 break; 671 672 default: 673 panic("vm_page_alloc: invalid allocation class"); 674 } 675 676 mem->wire_count = 0; 677 mem->hold_count = 0; 678 mem->act_count = 0; 679 mem->busy = 0; 680 mem->valid = 0; 681 mem->dirty = 0; 682 mem->bmapped = 0; 683 684 /* XXX before splx until vm_page_insert is safe */ 685 vm_page_insert(mem, object, offset); 686 687 splx(s); 688 689 /* 690 * Don't wakeup too often - wakeup the pageout daemon when 691 * we would be nearly out of memory. 692 */ 693 if (((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) || 694 (cnt.v_free_count < cnt.v_pageout_free_min)) 695 pagedaemon_wakeup(); 696 697 return (mem); 698 } 699 700 vm_offset_t 701 vm_page_alloc_contig(size, low, high, alignment) 702 vm_offset_t size; 703 vm_offset_t low; 704 vm_offset_t high; 705 vm_offset_t alignment; 706 { 707 int i, s, start; 708 vm_offset_t addr, phys, tmp_addr; 709 vm_page_t pga = vm_page_array; 710 711 if ((alignment & (alignment - 1)) != 0) 712 panic("vm_page_alloc_contig: alignment must be a power of 2"); 713 714 start = 0; 715 s = splhigh(); 716 again: 717 /* 718 * Find first page in array that is free, within range, and aligned. 719 */ 720 for (i = start; i < cnt.v_page_count; i++) { 721 phys = VM_PAGE_TO_PHYS(&pga[i]); 722 if (((pga[i].flags & PG_FREE) == PG_FREE) && 723 (phys >= low) && (phys < high) && 724 ((phys & (alignment - 1)) == 0)) 725 break; 726 } 727 728 /* 729 * If the above failed or we will exceed the upper bound, fail. 730 */ 731 if ((i == cnt.v_page_count) || ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 732 splx(s); 733 return (NULL); 734 } 735 start = i; 736 737 /* 738 * Check successive pages for contiguous and free. 739 */ 740 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 741 if ((VM_PAGE_TO_PHYS(&pga[i]) != 742 (VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) || 743 ((pga[i].flags & PG_FREE) != PG_FREE)) { 744 start++; 745 goto again; 746 } 747 } 748 749 /* 750 * We've found a contiguous chunk that meets are requirements. 751 * Allocate kernel VM, unfree and assign the physical pages to it and 752 * return kernel VM pointer. 753 */ 754 tmp_addr = addr = kmem_alloc_pageable(kernel_map, size); 755 756 for (i = start; i < (start + size / PAGE_SIZE); i++) { 757 vm_page_t m = &pga[i]; 758 759 TAILQ_REMOVE(&vm_page_queue_free, m, pageq); 760 cnt.v_free_count--; 761 m->valid = VM_PAGE_BITS_ALL; 762 m->flags = 0; 763 m->dirty = 0; 764 m->wire_count = 0; 765 m->act_count = 0; 766 m->bmapped = 0; 767 m->busy = 0; 768 vm_page_insert(m, kernel_object, tmp_addr - VM_MIN_KERNEL_ADDRESS); 769 vm_page_wire(m); 770 pmap_kenter(tmp_addr, VM_PAGE_TO_PHYS(m)); 771 tmp_addr += PAGE_SIZE; 772 } 773 774 splx(s); 775 return (addr); 776 } 777 778 /* 779 * vm_page_free: 780 * 781 * Returns the given page to the free list, 782 * disassociating it with any VM object. 783 * 784 * Object and page must be locked prior to entry. 785 */ 786 void 787 vm_page_free(mem) 788 register vm_page_t mem; 789 { 790 int s; 791 int flags; 792 793 s = splhigh(); 794 vm_page_remove(mem); 795 vm_page_unqueue(mem); 796 797 flags = mem->flags; 798 if (mem->bmapped || mem->busy || flags & (PG_BUSY|PG_FREE)) { 799 if (flags & PG_FREE) 800 panic("vm_page_free: freeing free page"); 801 printf("vm_page_free: offset(%d), bmapped(%d), busy(%d), PG_BUSY(%d)\n", 802 mem->offset, mem->bmapped, mem->busy, (flags & PG_BUSY) ? 1 : 0); 803 panic("vm_page_free: freeing busy page"); 804 } 805 806 if ((flags & PG_WANTED) != 0) 807 wakeup(mem); 808 if ((flags & PG_FICTITIOUS) == 0) { 809 if (mem->wire_count) { 810 if (mem->wire_count > 1) { 811 printf("vm_page_free: wire count > 1 (%d)", mem->wire_count); 812 panic("vm_page_free: invalid wire count"); 813 } 814 cnt.v_wire_count--; 815 mem->wire_count = 0; 816 } 817 mem->flags |= PG_FREE; 818 TAILQ_INSERT_TAIL(&vm_page_queue_free, mem, pageq); 819 splx(s); 820 /* 821 * if pageout daemon needs pages, then tell it that there are 822 * some free. 823 */ 824 if (vm_pageout_pages_needed) { 825 wakeup(&vm_pageout_pages_needed); 826 vm_pageout_pages_needed = 0; 827 } 828 829 cnt.v_free_count++; 830 /* 831 * wakeup processes that are waiting on memory if we hit a 832 * high water mark. And wakeup scheduler process if we have 833 * lots of memory. this process will swapin processes. 834 */ 835 if ((cnt.v_free_count + cnt.v_cache_count) == cnt.v_free_min) { 836 wakeup(&cnt.v_free_count); 837 wakeup(&proc0); 838 } 839 } else { 840 splx(s); 841 } 842 cnt.v_tfree++; 843 } 844 845 846 /* 847 * vm_page_wire: 848 * 849 * Mark this page as wired down by yet 850 * another map, removing it from paging queues 851 * as necessary. 852 * 853 * The page queues must be locked. 854 */ 855 void 856 vm_page_wire(mem) 857 register vm_page_t mem; 858 { 859 int s; 860 861 if (mem->wire_count == 0) { 862 s = splhigh(); 863 vm_page_unqueue(mem); 864 splx(s); 865 cnt.v_wire_count++; 866 } 867 mem->flags |= PG_WRITEABLE|PG_MAPPED; 868 mem->wire_count++; 869 } 870 871 /* 872 * vm_page_unwire: 873 * 874 * Release one wiring of this page, potentially 875 * enabling it to be paged again. 876 * 877 * The page queues must be locked. 878 */ 879 void 880 vm_page_unwire(mem) 881 register vm_page_t mem; 882 { 883 int s; 884 885 s = splhigh(); 886 887 if (mem->wire_count) 888 mem->wire_count--; 889 if (mem->wire_count == 0) { 890 TAILQ_INSERT_TAIL(&vm_page_queue_active, mem, pageq); 891 cnt.v_active_count++; 892 mem->flags |= PG_ACTIVE; 893 cnt.v_wire_count--; 894 } 895 splx(s); 896 } 897 898 /* 899 * vm_page_activate: 900 * 901 * Put the specified page on the active list (if appropriate). 902 * 903 * The page queues must be locked. 904 */ 905 void 906 vm_page_activate(m) 907 register vm_page_t m; 908 { 909 int s; 910 911 s = splhigh(); 912 if (m->flags & PG_ACTIVE) 913 panic("vm_page_activate: already active"); 914 915 if (m->flags & PG_CACHE) 916 cnt.v_reactivated++; 917 918 vm_page_unqueue(m); 919 920 if (m->wire_count == 0) { 921 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 922 m->flags |= PG_ACTIVE; 923 if (m->act_count < 5) 924 m->act_count = 5; 925 else if( m->act_count < ACT_MAX) 926 m->act_count += 1; 927 cnt.v_active_count++; 928 } 929 splx(s); 930 } 931 932 /* 933 * vm_page_deactivate: 934 * 935 * Returns the given page to the inactive list, 936 * indicating that no physical maps have access 937 * to this page. [Used by the physical mapping system.] 938 * 939 * The page queues must be locked. 940 */ 941 void 942 vm_page_deactivate(m) 943 register vm_page_t m; 944 { 945 int spl; 946 947 /* 948 * Only move active pages -- ignore locked or already inactive ones. 949 * 950 * XXX: sometimes we get pages which aren't wired down or on any queue - 951 * we need to put them on the inactive queue also, otherwise we lose 952 * track of them. Paul Mackerras (paulus@cs.anu.edu.au) 9-Jan-93. 953 */ 954 955 spl = splhigh(); 956 if (!(m->flags & PG_INACTIVE) && m->wire_count == 0 && 957 m->hold_count == 0) { 958 if (m->flags & PG_CACHE) 959 cnt.v_reactivated++; 960 vm_page_unqueue(m); 961 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); 962 m->flags |= PG_INACTIVE; 963 cnt.v_inactive_count++; 964 m->act_count = 0; 965 } 966 splx(spl); 967 } 968 969 /* 970 * vm_page_cache 971 * 972 * Put the specified page onto the page cache queue (if appropriate). 973 */ 974 void 975 vm_page_cache(m) 976 register vm_page_t m; 977 { 978 int s; 979 980 if ((m->flags & (PG_CACHE | PG_BUSY)) || m->busy || m->wire_count || 981 m->bmapped) 982 return; 983 984 s = splhigh(); 985 vm_page_unqueue(m); 986 vm_page_protect(m, VM_PROT_NONE); 987 988 TAILQ_INSERT_TAIL(&vm_page_queue_cache, m, pageq); 989 m->flags |= PG_CACHE; 990 cnt.v_cache_count++; 991 if ((cnt.v_free_count + cnt.v_cache_count) == cnt.v_free_min) { 992 wakeup(&cnt.v_free_count); 993 wakeup(&proc0); 994 } 995 if (vm_pageout_pages_needed) { 996 wakeup(&vm_pageout_pages_needed); 997 vm_pageout_pages_needed = 0; 998 } 999 1000 splx(s); 1001 } 1002 1003 /* 1004 * vm_page_zero_fill: 1005 * 1006 * Zero-fill the specified page. 1007 * Written as a standard pagein routine, to 1008 * be used by the zero-fill object. 1009 */ 1010 boolean_t 1011 vm_page_zero_fill(m) 1012 vm_page_t m; 1013 { 1014 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 1015 return (TRUE); 1016 } 1017 1018 /* 1019 * vm_page_copy: 1020 * 1021 * Copy one page to another 1022 */ 1023 void 1024 vm_page_copy(src_m, dest_m) 1025 vm_page_t src_m; 1026 vm_page_t dest_m; 1027 { 1028 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 1029 dest_m->valid = VM_PAGE_BITS_ALL; 1030 } 1031 1032 1033 /* 1034 * mapping function for valid bits or for dirty bits in 1035 * a page 1036 */ 1037 inline int 1038 vm_page_bits(int base, int size) 1039 { 1040 u_short chunk; 1041 1042 if ((base == 0) && (size >= PAGE_SIZE)) 1043 return VM_PAGE_BITS_ALL; 1044 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 1045 base = (base % PAGE_SIZE) / DEV_BSIZE; 1046 chunk = vm_page_dev_bsize_chunks[size / DEV_BSIZE]; 1047 return (chunk << base) & VM_PAGE_BITS_ALL; 1048 } 1049 1050 /* 1051 * set a page valid and clean 1052 */ 1053 void 1054 vm_page_set_validclean(m, base, size) 1055 vm_page_t m; 1056 int base; 1057 int size; 1058 { 1059 int pagebits = vm_page_bits(base, size); 1060 m->valid |= pagebits; 1061 m->dirty &= ~pagebits; 1062 if( base == 0 && size == PAGE_SIZE) 1063 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1064 } 1065 1066 /* 1067 * set a page (partially) valid 1068 */ 1069 void 1070 vm_page_set_valid(m, base, size) 1071 vm_page_t m; 1072 int base; 1073 int size; 1074 { 1075 m->valid |= vm_page_bits(base, size); 1076 } 1077 1078 /* 1079 * set a page (partially) invalid 1080 */ 1081 void 1082 vm_page_set_invalid(m, base, size) 1083 vm_page_t m; 1084 int base; 1085 int size; 1086 { 1087 int bits; 1088 1089 m->valid &= ~(bits = vm_page_bits(base, size)); 1090 if (m->valid == 0) 1091 m->dirty &= ~bits; 1092 } 1093 1094 /* 1095 * is (partial) page valid? 1096 */ 1097 int 1098 vm_page_is_valid(m, base, size) 1099 vm_page_t m; 1100 int base; 1101 int size; 1102 { 1103 int bits = vm_page_bits(base, size); 1104 1105 if (m->valid && ((m->valid & bits) == bits)) 1106 return 1; 1107 else 1108 return 0; 1109 } 1110 1111 1112 /* 1113 * set a page (partially) dirty 1114 */ 1115 void 1116 vm_page_set_dirty(m, base, size) 1117 vm_page_t m; 1118 int base; 1119 int size; 1120 { 1121 if ((base != 0) || (size != PAGE_SIZE)) { 1122 if (pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1123 m->dirty = VM_PAGE_BITS_ALL; 1124 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1125 return; 1126 } 1127 m->dirty |= vm_page_bits(base, size); 1128 } else { 1129 m->dirty = VM_PAGE_BITS_ALL; 1130 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1131 } 1132 } 1133 1134 void 1135 vm_page_test_dirty(m) 1136 vm_page_t m; 1137 { 1138 if ((m->dirty != VM_PAGE_BITS_ALL) && 1139 pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1140 m->dirty = VM_PAGE_BITS_ALL; 1141 } 1142 } 1143 1144 /* 1145 * set a page (partially) clean 1146 */ 1147 void 1148 vm_page_set_clean(m, base, size) 1149 vm_page_t m; 1150 int base; 1151 int size; 1152 { 1153 m->dirty &= ~vm_page_bits(base, size); 1154 if( base == 0 && size == PAGE_SIZE) 1155 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1156 } 1157 1158 /* 1159 * is (partial) page clean 1160 */ 1161 int 1162 vm_page_is_clean(m, base, size) 1163 vm_page_t m; 1164 int base; 1165 int size; 1166 { 1167 if (pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1168 m->dirty = VM_PAGE_BITS_ALL; 1169 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1170 } 1171 if ((m->dirty & m->valid & vm_page_bits(base, size)) == 0) 1172 return 1; 1173 else 1174 return 0; 1175 } 1176 1177 #ifdef DDB 1178 void 1179 print_page_info() 1180 { 1181 printf("cnt.v_free_count: %d\n", cnt.v_free_count); 1182 printf("cnt.v_cache_count: %d\n", cnt.v_cache_count); 1183 printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count); 1184 printf("cnt.v_active_count: %d\n", cnt.v_active_count); 1185 printf("cnt.v_wire_count: %d\n", cnt.v_wire_count); 1186 printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved); 1187 printf("cnt.v_free_min: %d\n", cnt.v_free_min); 1188 printf("cnt.v_free_target: %d\n", cnt.v_free_target); 1189 printf("cnt.v_cache_min: %d\n", cnt.v_cache_min); 1190 printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target); 1191 } 1192 #endif 1193