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 1.21 1995/02/22 10:16:21 davidg 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_page.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_pageout.h> 79 80 /* 81 * Associated with page of user-allocatable memory is a 82 * page structure. 83 */ 84 85 struct pglist *vm_page_buckets; /* Array of buckets */ 86 int vm_page_bucket_count = 0; /* How big is array? */ 87 int vm_page_hash_mask; /* Mask for hash function */ 88 simple_lock_data_t bucket_lock; /* lock for all buckets XXX */ 89 90 struct pglist vm_page_queue_free; 91 struct pglist vm_page_queue_active; 92 struct pglist vm_page_queue_inactive; 93 struct pglist vm_page_queue_cache; 94 simple_lock_data_t vm_page_queue_lock; 95 simple_lock_data_t vm_page_queue_free_lock; 96 97 /* has physical page allocation been initialized? */ 98 boolean_t vm_page_startup_initialized; 99 100 vm_page_t vm_page_array; 101 int vm_page_array_size; 102 long first_page; 103 long last_page; 104 vm_offset_t first_phys_addr; 105 vm_offset_t last_phys_addr; 106 vm_size_t page_mask; 107 int page_shift; 108 109 /* 110 * map of contiguous valid DEV_BSIZE chunks in a page 111 * (this list is valid for page sizes upto 16*DEV_BSIZE) 112 */ 113 static u_short vm_page_dev_bsize_chunks[] = { 114 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff, 115 0x1ff, 0x3ff, 0x7ff, 0xfff, 0x1fff, 0x3fff, 0x7fff, 0xffff 116 }; 117 118 119 /* 120 * vm_set_page_size: 121 * 122 * Sets the page size, perhaps based upon the memory 123 * size. Must be called before any use of page-size 124 * dependent functions. 125 * 126 * Sets page_shift and page_mask from cnt.v_page_size. 127 */ 128 void 129 vm_set_page_size() 130 { 131 132 if (cnt.v_page_size == 0) 133 cnt.v_page_size = DEFAULT_PAGE_SIZE; 134 page_mask = cnt.v_page_size - 1; 135 if ((page_mask & cnt.v_page_size) != 0) 136 panic("vm_set_page_size: page size not a power of two"); 137 for (page_shift = 0;; page_shift++) 138 if ((1 << page_shift) == cnt.v_page_size) 139 break; 140 } 141 142 /* 143 * vm_page_startup: 144 * 145 * Initializes the resident memory module. 146 * 147 * Allocates memory for the page cells, and 148 * for the object/offset-to-page hash table headers. 149 * Each page cell is initialized and placed on the free list. 150 */ 151 152 vm_offset_t 153 vm_page_startup(starta, enda, vaddr) 154 register vm_offset_t starta; 155 vm_offset_t enda; 156 register vm_offset_t vaddr; 157 { 158 register vm_offset_t mapped; 159 register vm_page_t m; 160 register struct pglist *bucket; 161 vm_size_t npages, page_range; 162 register vm_offset_t new_start; 163 int i; 164 vm_offset_t pa; 165 int nblocks; 166 vm_offset_t first_managed_page; 167 168 extern vm_offset_t kentry_data; 169 extern vm_size_t kentry_data_size; 170 extern vm_offset_t phys_avail[]; 171 172 /* the biggest memory array is the second group of pages */ 173 vm_offset_t start; 174 vm_offset_t biggestone, biggestsize; 175 176 vm_offset_t total; 177 178 total = 0; 179 biggestsize = 0; 180 biggestone = 0; 181 nblocks = 0; 182 vaddr = round_page(vaddr); 183 184 for (i = 0; phys_avail[i + 1]; i += 2) { 185 phys_avail[i] = round_page(phys_avail[i]); 186 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); 187 } 188 189 for (i = 0; phys_avail[i + 1]; i += 2) { 190 int size = phys_avail[i + 1] - phys_avail[i]; 191 192 if (size > biggestsize) { 193 biggestone = i; 194 biggestsize = size; 195 } 196 ++nblocks; 197 total += size; 198 } 199 200 start = phys_avail[biggestone]; 201 202 203 /* 204 * Initialize the locks 205 */ 206 207 simple_lock_init(&vm_page_queue_free_lock); 208 simple_lock_init(&vm_page_queue_lock); 209 210 /* 211 * Initialize the queue headers for the free queue, the active queue 212 * and the inactive queue. 213 */ 214 215 TAILQ_INIT(&vm_page_queue_free); 216 TAILQ_INIT(&vm_page_queue_active); 217 TAILQ_INIT(&vm_page_queue_inactive); 218 TAILQ_INIT(&vm_page_queue_cache); 219 220 /* 221 * Allocate (and initialize) the hash table buckets. 222 * 223 * The number of buckets MUST BE a power of 2, and the actual value is 224 * the next power of 2 greater than the number of physical pages in 225 * the system. 226 * 227 * Note: This computation can be tweaked if desired. 228 */ 229 vm_page_buckets = (struct pglist *) vaddr; 230 bucket = vm_page_buckets; 231 if (vm_page_bucket_count == 0) { 232 vm_page_bucket_count = 1; 233 while (vm_page_bucket_count < atop(total)) 234 vm_page_bucket_count <<= 1; 235 } 236 vm_page_hash_mask = vm_page_bucket_count - 1; 237 238 /* 239 * Validate these addresses. 240 */ 241 242 new_start = start + vm_page_bucket_count * sizeof(struct pglist); 243 new_start = round_page(new_start); 244 mapped = vaddr; 245 vaddr = pmap_map(mapped, start, new_start, 246 VM_PROT_READ | VM_PROT_WRITE); 247 start = new_start; 248 bzero((caddr_t) mapped, vaddr - mapped); 249 mapped = vaddr; 250 251 for (i = 0; i < vm_page_bucket_count; i++) { 252 TAILQ_INIT(bucket); 253 bucket++; 254 } 255 256 simple_lock_init(&bucket_lock); 257 258 /* 259 * round (or truncate) the addresses to our page size. 260 */ 261 262 /* 263 * Pre-allocate maps and map entries that cannot be dynamically 264 * allocated via malloc(). The maps include the kernel_map and 265 * kmem_map which must be initialized before malloc() will work 266 * (obviously). Also could include pager maps which would be 267 * allocated before kmeminit. 268 * 269 * Allow some kernel map entries... this should be plenty since people 270 * shouldn't be cluttering up the kernel map (they should use their 271 * own maps). 272 */ 273 274 kentry_data_size = MAX_KMAP * sizeof(struct vm_map) + 275 MAX_KMAPENT * sizeof(struct vm_map_entry); 276 kentry_data_size = round_page(kentry_data_size); 277 kentry_data = (vm_offset_t) vaddr; 278 vaddr += kentry_data_size; 279 280 /* 281 * Validate these zone addresses. 282 */ 283 284 new_start = start + (vaddr - mapped); 285 pmap_map(mapped, start, new_start, VM_PROT_READ | VM_PROT_WRITE); 286 bzero((caddr_t) mapped, (vaddr - mapped)); 287 start = round_page(new_start); 288 289 /* 290 * Compute the number of pages of memory that will be available for 291 * use (taking into account the overhead of a page structure per 292 * page). 293 */ 294 295 first_page = phys_avail[0] / PAGE_SIZE; 296 297 /* for VM_PAGE_CHECK() */ 298 last_page = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE; 299 300 page_range = last_page - (phys_avail[0] / PAGE_SIZE); 301 npages = (total - (page_range * sizeof(struct vm_page)) - 302 (start - phys_avail[biggestone])) / PAGE_SIZE; 303 304 /* 305 * Initialize the mem entry structures now, and put them in the free 306 * queue. 307 */ 308 309 vm_page_array = (vm_page_t) vaddr; 310 mapped = vaddr; 311 312 313 /* 314 * Validate these addresses. 315 */ 316 317 new_start = round_page(start + page_range * sizeof(struct vm_page)); 318 mapped = pmap_map(mapped, start, new_start, 319 VM_PROT_READ | VM_PROT_WRITE); 320 start = new_start; 321 322 first_managed_page = start / PAGE_SIZE; 323 324 /* 325 * Clear all of the page structures 326 */ 327 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); 328 vm_page_array_size = page_range; 329 330 cnt.v_page_count = 0; 331 cnt.v_free_count = 0; 332 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) { 333 if (i == biggestone) 334 pa = ptoa(first_managed_page); 335 else 336 pa = phys_avail[i]; 337 while (pa < phys_avail[i + 1] && npages-- > 0) { 338 ++cnt.v_page_count; 339 ++cnt.v_free_count; 340 m = PHYS_TO_VM_PAGE(pa); 341 m->flags = PG_FREE; 342 vm_page_set_clean(m, 0, PAGE_SIZE); 343 m->object = 0; 344 m->phys_addr = pa; 345 m->hold_count = 0; 346 TAILQ_INSERT_TAIL(&vm_page_queue_free, m, pageq); 347 pa += PAGE_SIZE; 348 } 349 } 350 351 /* 352 * Initialize vm_pages_needed lock here - don't wait for pageout 353 * daemon XXX 354 */ 355 simple_lock_init(&vm_pages_needed_lock); 356 357 return (mapped); 358 } 359 360 /* 361 * vm_page_hash: 362 * 363 * Distributes the object/offset key pair among hash buckets. 364 * 365 * NOTE: This macro depends on vm_page_bucket_count being a power of 2. 366 */ 367 inline const int 368 vm_page_hash(object, offset) 369 vm_object_t object; 370 vm_offset_t offset; 371 { 372 return ((unsigned) object + offset / NBPG) & vm_page_hash_mask; 373 } 374 375 /* 376 * vm_page_insert: [ internal use only ] 377 * 378 * Inserts the given mem entry into the object/object-page 379 * table and object list. 380 * 381 * The object and page must be locked. 382 */ 383 384 void 385 vm_page_insert(mem, object, offset) 386 register vm_page_t mem; 387 register vm_object_t object; 388 register vm_offset_t offset; 389 { 390 register struct pglist *bucket; 391 int s; 392 393 VM_PAGE_CHECK(mem); 394 395 if (mem->flags & PG_TABLED) 396 panic("vm_page_insert: already inserted"); 397 398 /* 399 * Record the object/offset pair in this page 400 */ 401 402 mem->object = object; 403 mem->offset = offset; 404 405 /* 406 * Insert it into the object_object/offset hash table 407 */ 408 409 bucket = &vm_page_buckets[vm_page_hash(object, offset)]; 410 s = splhigh(); 411 simple_lock(&bucket_lock); 412 TAILQ_INSERT_TAIL(bucket, mem, hashq); 413 simple_unlock(&bucket_lock); 414 415 /* 416 * Now link into the object's list of backed pages. 417 */ 418 419 TAILQ_INSERT_TAIL(&object->memq, mem, listq); 420 (void) splx(s); 421 mem->flags |= PG_TABLED; 422 423 /* 424 * And show that the object has one more resident page. 425 */ 426 427 object->resident_page_count++; 428 } 429 430 /* 431 * vm_page_remove: [ internal use only ] 432 * NOTE: used by device pager as well -wfj 433 * 434 * Removes the given mem entry from the object/offset-page 435 * table and the object page list. 436 * 437 * The object and page must be locked. 438 */ 439 440 void 441 vm_page_remove(mem) 442 register vm_page_t mem; 443 { 444 register struct pglist *bucket; 445 int s; 446 447 VM_PAGE_CHECK(mem); 448 449 450 if (!(mem->flags & PG_TABLED)) 451 return; 452 453 /* 454 * Remove from the object_object/offset hash table 455 */ 456 457 bucket = &vm_page_buckets[vm_page_hash(mem->object, mem->offset)]; 458 s = splhigh(); 459 simple_lock(&bucket_lock); 460 TAILQ_REMOVE(bucket, mem, hashq); 461 simple_unlock(&bucket_lock); 462 463 /* 464 * Now remove from the object's list of backed pages. 465 */ 466 467 TAILQ_REMOVE(&mem->object->memq, mem, listq); 468 (void) splx(s); 469 470 /* 471 * And show that the object has one fewer resident page. 472 */ 473 474 mem->object->resident_page_count--; 475 476 mem->flags &= ~PG_TABLED; 477 } 478 479 /* 480 * vm_page_lookup: 481 * 482 * Returns the page associated with the object/offset 483 * pair specified; if none is found, NULL is returned. 484 * 485 * The object must be locked. No side effects. 486 */ 487 488 vm_page_t 489 vm_page_lookup(object, offset) 490 register vm_object_t object; 491 register vm_offset_t offset; 492 { 493 register vm_page_t mem; 494 register struct pglist *bucket; 495 int s; 496 497 /* 498 * Search the hash table for this object/offset pair 499 */ 500 501 bucket = &vm_page_buckets[vm_page_hash(object, offset)]; 502 503 s = splhigh(); 504 simple_lock(&bucket_lock); 505 for (mem = bucket->tqh_first; mem != NULL; mem = mem->hashq.tqe_next) { 506 VM_PAGE_CHECK(mem); 507 if ((mem->object == object) && (mem->offset == offset)) { 508 simple_unlock(&bucket_lock); 509 splx(s); 510 return (mem); 511 } 512 } 513 514 simple_unlock(&bucket_lock); 515 splx(s); 516 return (NULL); 517 } 518 519 /* 520 * vm_page_rename: 521 * 522 * Move the given memory entry from its 523 * current object to the specified target object/offset. 524 * 525 * The object must be locked. 526 */ 527 void 528 vm_page_rename(mem, new_object, new_offset) 529 register vm_page_t mem; 530 register vm_object_t new_object; 531 vm_offset_t new_offset; 532 { 533 int s; 534 535 if (mem->object == new_object) 536 return; 537 538 vm_page_lock_queues(); /* keep page from moving out from under pageout daemon */ 539 s = splhigh(); 540 vm_page_remove(mem); 541 vm_page_insert(mem, new_object, new_offset); 542 splx(s); 543 vm_page_unlock_queues(); 544 } 545 546 int 547 vm_page_unqueue(vm_page_t mem) 548 { 549 int s, origflags; 550 551 origflags = mem->flags; 552 553 if ((origflags & (PG_ACTIVE|PG_INACTIVE|PG_CACHE)) == 0) 554 return origflags; 555 556 s = splhigh(); 557 if (mem->flags & PG_ACTIVE) { 558 TAILQ_REMOVE(&vm_page_queue_active, mem, pageq); 559 cnt.v_active_count--; 560 mem->flags &= ~PG_ACTIVE; 561 } else if (mem->flags & PG_INACTIVE) { 562 TAILQ_REMOVE(&vm_page_queue_inactive, mem, pageq); 563 cnt.v_inactive_count--; 564 mem->flags &= ~PG_INACTIVE; 565 } else if (mem->flags & PG_CACHE) { 566 TAILQ_REMOVE(&vm_page_queue_cache, mem, pageq); 567 cnt.v_cache_count--; 568 mem->flags &= ~PG_CACHE; 569 if (cnt.v_cache_count + cnt.v_free_count < cnt.v_free_reserved) 570 wakeup((caddr_t) &vm_pages_needed); 571 } 572 splx(s); 573 return origflags; 574 } 575 576 void 577 vm_page_requeue(vm_page_t mem, int flags) 578 { 579 int s; 580 581 if (mem->wire_count) 582 return; 583 s = splhigh(); 584 if (flags & PG_CACHE) { 585 TAILQ_INSERT_TAIL(&vm_page_queue_cache, mem, pageq); 586 mem->flags |= PG_CACHE; 587 cnt.v_cache_count++; 588 } else if (flags & PG_ACTIVE) { 589 TAILQ_INSERT_TAIL(&vm_page_queue_active, mem, pageq); 590 mem->flags |= PG_ACTIVE; 591 cnt.v_active_count++; 592 } else if (flags & PG_INACTIVE) { 593 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, mem, pageq); 594 mem->flags |= PG_INACTIVE; 595 cnt.v_inactive_count++; 596 } 597 TAILQ_REMOVE(&mem->object->memq, mem, listq); 598 TAILQ_INSERT_TAIL(&mem->object->memq, mem, listq); 599 splx(s); 600 } 601 602 /* 603 * vm_page_alloc: 604 * 605 * Allocate and return a memory cell associated 606 * with this VM object/offset pair. 607 * 608 * page_req -- 0 normal process request VM_ALLOC_NORMAL 609 * page_req -- 1 interrupt time request VM_ALLOC_INTERRUPT 610 * page_req -- 2 system *really* needs a page VM_ALLOC_SYSTEM 611 * but *cannot* be at interrupt time 612 * 613 * Object must be locked. 614 */ 615 vm_page_t 616 vm_page_alloc(object, offset, page_req) 617 vm_object_t object; 618 vm_offset_t offset; 619 int page_req; 620 { 621 register vm_page_t mem; 622 int s; 623 624 simple_lock(&vm_page_queue_free_lock); 625 626 s = splhigh(); 627 628 if (((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_reserved) && 629 (page_req == VM_ALLOC_NORMAL) && 630 (curproc != pageproc)) { 631 simple_unlock(&vm_page_queue_free_lock); 632 splx(s); 633 return (NULL); 634 } 635 636 if (page_req == VM_ALLOC_INTERRUPT) { 637 if ((mem = vm_page_queue_free.tqh_first) == 0) { 638 simple_unlock(&vm_page_queue_free_lock); 639 splx(s); 640 /* 641 * need to wakeup at interrupt time -- it doesn't do VM_WAIT 642 */ 643 wakeup((caddr_t) &vm_pages_needed); 644 return NULL; 645 } 646 } else { 647 if ((cnt.v_free_count < cnt.v_free_reserved) || 648 (mem = vm_page_queue_free.tqh_first) == 0) { 649 mem = vm_page_queue_cache.tqh_first; 650 if (mem) { 651 TAILQ_REMOVE(&vm_page_queue_cache, mem, pageq); 652 vm_page_remove(mem); 653 cnt.v_cache_count--; 654 goto gotpage; 655 } 656 657 if (page_req == VM_ALLOC_SYSTEM && 658 cnt.v_free_count > cnt.v_interrupt_free_min) { 659 mem = vm_page_queue_free.tqh_first; 660 } 661 662 if( !mem) { 663 simple_unlock(&vm_page_queue_free_lock); 664 splx(s); 665 wakeup((caddr_t) &vm_pages_needed); 666 return (NULL); 667 } 668 } 669 } 670 671 TAILQ_REMOVE(&vm_page_queue_free, mem, pageq); 672 cnt.v_free_count--; 673 674 gotpage: 675 simple_unlock(&vm_page_queue_free_lock); 676 677 mem->flags = PG_BUSY; 678 mem->wire_count = 0; 679 mem->hold_count = 0; 680 mem->act_count = 0; 681 mem->busy = 0; 682 mem->valid = 0; 683 mem->dirty = 0; 684 mem->bmapped = 0; 685 686 /* XXX before splx until vm_page_insert is safe */ 687 vm_page_insert(mem, object, offset); 688 689 splx(s); 690 691 /* 692 * don't wakeup too often, so we wakeup the pageout daemon when 693 * we would be nearly out of memory. 694 */ 695 if (curproc != pageproc && 696 ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) || 697 (cnt.v_free_count < cnt.v_pageout_free_min)) 698 wakeup((caddr_t) &vm_pages_needed); 699 700 return (mem); 701 } 702 703 vm_offset_t 704 vm_page_alloc_contig(size, low, high, alignment) 705 vm_offset_t size; 706 vm_offset_t low; 707 vm_offset_t high; 708 vm_offset_t alignment; 709 { 710 int i, s, start; 711 vm_offset_t addr, phys, tmp_addr; 712 vm_page_t pga = vm_page_array; 713 extern vm_map_t kernel_map; 714 715 if ((alignment & (alignment - 1)) != 0) 716 panic("vm_page_alloc_contig: alignment must be a power of 2"); 717 718 start = 0; 719 s = splhigh(); 720 again: 721 /* 722 * Find first page in array that is free, within range, and aligned. 723 */ 724 for (i = start; i < cnt.v_page_count; i++) { 725 phys = VM_PAGE_TO_PHYS(&pga[i]); 726 if (((pga[i].flags & PG_FREE) == PG_FREE) && 727 (phys >= low) && (phys < high) && 728 ((phys & (alignment - 1)) == 0)) 729 break; 730 } 731 732 /* 733 * If the above failed or we will exceed the upper bound, fail. 734 */ 735 if ((i == cnt.v_page_count) || ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 736 splx(s); 737 return (NULL); 738 } 739 start = i; 740 741 /* 742 * Check successive pages for contiguous and free. 743 */ 744 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 745 if ((VM_PAGE_TO_PHYS(&pga[i]) != 746 (VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) || 747 ((pga[i].flags & PG_FREE) != PG_FREE)) { 748 start++; 749 goto again; 750 } 751 } 752 753 /* 754 * We've found a contiguous chunk that meets are requirements. 755 * Allocate kernel VM, unfree and assign the physical pages to it and 756 * return kernel VM pointer. 757 */ 758 tmp_addr = addr = kmem_alloc_pageable(kernel_map, size); 759 760 for (i = start; i < (start + size / PAGE_SIZE); i++) { 761 vm_page_t m = &pga[i];; 762 763 TAILQ_REMOVE(&vm_page_queue_free, m, pageq); 764 cnt.v_free_count--; 765 m->valid = VM_PAGE_BITS_ALL; 766 m->flags = 0; 767 m->dirty = 0; 768 m->wire_count = 0; 769 m->act_count = 0; 770 m->bmapped = 0; 771 m->busy = 0; 772 vm_page_insert(m, kernel_object, tmp_addr - VM_MIN_KERNEL_ADDRESS); 773 vm_page_wire(m); 774 pmap_kenter(tmp_addr, VM_PAGE_TO_PHYS(&pga[i])); 775 tmp_addr += PAGE_SIZE; 776 } 777 778 splx(s); 779 return (addr); 780 } 781 782 /* 783 * vm_page_free: 784 * 785 * Returns the given page to the free list, 786 * disassociating it with any VM object. 787 * 788 * Object and page must be locked prior to entry. 789 */ 790 void 791 vm_page_free(mem) 792 register vm_page_t mem; 793 { 794 int s; 795 796 s = splhigh(); 797 vm_page_remove(mem); 798 vm_page_unqueue(mem); 799 800 if (mem->bmapped || mem->busy || mem->flags & PG_BUSY) { 801 printf("vm_page_free: offset(%d), bmapped(%d), busy(%d), PG_BUSY(%d)\n", 802 mem->offset, mem->bmapped, mem->busy, (mem->flags & PG_BUSY) ? 1 : 0); 803 panic("vm_page_free: freeing busy page\n"); 804 } 805 if (mem->flags & PG_FREE) 806 panic("vm_page_free: freeing free page"); 807 808 if (!(mem->flags & PG_FICTITIOUS)) { 809 810 simple_lock(&vm_page_queue_free_lock); 811 if (mem->wire_count) { 812 if (mem->wire_count > 1) { 813 printf("vm_page_free: wire count > 1 (%d)", mem->wire_count); 814 panic("vm_page_free: invalid wire count"); 815 } 816 cnt.v_wire_count--; 817 mem->wire_count = 0; 818 } 819 mem->flags |= PG_FREE; 820 TAILQ_INSERT_TAIL(&vm_page_queue_free, mem, pageq); 821 822 cnt.v_free_count++; 823 simple_unlock(&vm_page_queue_free_lock); 824 splx(s); 825 /* 826 * if pageout daemon needs pages, then tell it that there are 827 * some free. 828 */ 829 if (vm_pageout_pages_needed) { 830 wakeup((caddr_t) &vm_pageout_pages_needed); 831 vm_pageout_pages_needed = 0; 832 } 833 834 /* 835 * wakeup processes that are waiting on memory if we hit a 836 * high water mark. And wakeup scheduler process if we have 837 * lots of memory. this process will swapin processes. 838 */ 839 if ((cnt.v_free_count + cnt.v_cache_count) == cnt.v_free_min) { 840 wakeup((caddr_t) &cnt.v_free_count); 841 wakeup((caddr_t) &proc0); 842 } 843 } else { 844 splx(s); 845 } 846 if (mem->flags & PG_WANTED) 847 wakeup((caddr_t) mem); 848 cnt.v_tfree++; 849 } 850 851 852 /* 853 * vm_page_wire: 854 * 855 * Mark this page as wired down by yet 856 * another map, removing it from paging queues 857 * as necessary. 858 * 859 * The page queues must be locked. 860 */ 861 void 862 vm_page_wire(mem) 863 register vm_page_t mem; 864 { 865 int s; 866 867 VM_PAGE_CHECK(mem); 868 869 if (mem->wire_count == 0) { 870 vm_page_unqueue(mem); 871 cnt.v_wire_count++; 872 } 873 mem->wire_count++; 874 } 875 876 /* 877 * vm_page_unwire: 878 * 879 * Release one wiring of this page, potentially 880 * enabling it to be paged again. 881 * 882 * The page queues must be locked. 883 */ 884 void 885 vm_page_unwire(mem) 886 register vm_page_t mem; 887 { 888 int s; 889 890 VM_PAGE_CHECK(mem); 891 892 s = splhigh(); 893 894 if (mem->wire_count) 895 mem->wire_count--; 896 if (mem->wire_count == 0) { 897 TAILQ_INSERT_TAIL(&vm_page_queue_active, mem, pageq); 898 cnt.v_active_count++; 899 mem->flags |= PG_ACTIVE; 900 cnt.v_wire_count--; 901 } 902 splx(s); 903 } 904 905 /* 906 * vm_page_deactivate: 907 * 908 * Returns the given page to the inactive list, 909 * indicating that no physical maps have access 910 * to this page. [Used by the physical mapping system.] 911 * 912 * The page queues must be locked. 913 */ 914 void 915 vm_page_deactivate(m) 916 register vm_page_t m; 917 { 918 int spl; 919 920 VM_PAGE_CHECK(m); 921 922 /* 923 * Only move active pages -- ignore locked or already inactive ones. 924 * 925 * XXX: sometimes we get pages which aren't wired down or on any queue - 926 * we need to put them on the inactive queue also, otherwise we lose 927 * track of them. Paul Mackerras (paulus@cs.anu.edu.au) 9-Jan-93. 928 */ 929 930 spl = splhigh(); 931 if (!(m->flags & PG_INACTIVE) && m->wire_count == 0 && 932 m->hold_count == 0) { 933 pmap_clear_reference(VM_PAGE_TO_PHYS(m)); 934 vm_page_unqueue(m); 935 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); 936 m->flags |= PG_INACTIVE; 937 cnt.v_inactive_count++; 938 m->act_count = 0; 939 } 940 splx(spl); 941 } 942 943 /* 944 * vm_page_cache 945 * 946 * Put the specified page onto the page cache queue (if appropriate). 947 */ 948 949 void 950 vm_page_cache(m) 951 register vm_page_t m; 952 { 953 int s; 954 955 VM_PAGE_CHECK(m); 956 if ((m->flags & (PG_CACHE | PG_BUSY)) || m->busy || m->wire_count || 957 m->bmapped) 958 return; 959 960 s = splhigh(); 961 vm_page_unqueue(m); 962 pmap_page_protect(VM_PAGE_TO_PHYS(m), VM_PROT_NONE); 963 964 TAILQ_INSERT_TAIL(&vm_page_queue_cache, m, pageq); 965 m->flags |= PG_CACHE; 966 cnt.v_cache_count++; 967 if ((cnt.v_free_count + cnt.v_cache_count) == cnt.v_free_min) { 968 wakeup((caddr_t) &cnt.v_free_count); 969 wakeup((caddr_t) &proc0); 970 } 971 if (vm_pageout_pages_needed) { 972 wakeup((caddr_t) &vm_pageout_pages_needed); 973 vm_pageout_pages_needed = 0; 974 } 975 976 splx(s); 977 } 978 979 /* 980 * vm_page_activate: 981 * 982 * Put the specified page on the active list (if appropriate). 983 * 984 * The page queues must be locked. 985 */ 986 987 void 988 vm_page_activate(m) 989 register vm_page_t m; 990 { 991 int s; 992 993 VM_PAGE_CHECK(m); 994 995 s = splhigh(); 996 if (m->flags & PG_ACTIVE) 997 panic("vm_page_activate: already active"); 998 999 vm_page_unqueue(m); 1000 1001 if (m->wire_count == 0) { 1002 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 1003 m->flags |= PG_ACTIVE; 1004 TAILQ_REMOVE(&m->object->memq, m, listq); 1005 TAILQ_INSERT_TAIL(&m->object->memq, m, listq); 1006 if (m->act_count < 5) 1007 m->act_count = 5; 1008 else if( m->act_count < ACT_MAX) 1009 m->act_count += 1; 1010 cnt.v_active_count++; 1011 } 1012 splx(s); 1013 } 1014 1015 /* 1016 * vm_page_zero_fill: 1017 * 1018 * Zero-fill the specified page. 1019 * Written as a standard pagein routine, to 1020 * be used by the zero-fill object. 1021 */ 1022 1023 boolean_t 1024 vm_page_zero_fill(m) 1025 vm_page_t m; 1026 { 1027 VM_PAGE_CHECK(m); 1028 1029 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 1030 m->valid = VM_PAGE_BITS_ALL; 1031 return (TRUE); 1032 } 1033 1034 /* 1035 * vm_page_copy: 1036 * 1037 * Copy one page to another 1038 */ 1039 void 1040 vm_page_copy(src_m, dest_m) 1041 vm_page_t src_m; 1042 vm_page_t dest_m; 1043 { 1044 VM_PAGE_CHECK(src_m); 1045 VM_PAGE_CHECK(dest_m); 1046 1047 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m)); 1048 dest_m->valid = VM_PAGE_BITS_ALL; 1049 } 1050 1051 1052 /* 1053 * mapping function for valid bits or for dirty bits in 1054 * a page 1055 */ 1056 inline int 1057 vm_page_bits(int base, int size) 1058 { 1059 u_short chunk; 1060 1061 if ((base == 0) && (size >= PAGE_SIZE)) 1062 return VM_PAGE_BITS_ALL; 1063 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 1064 base = (base % PAGE_SIZE) / DEV_BSIZE; 1065 chunk = vm_page_dev_bsize_chunks[size / DEV_BSIZE]; 1066 return (chunk << base) & VM_PAGE_BITS_ALL; 1067 } 1068 1069 /* 1070 * set a page (partially) valid 1071 */ 1072 void 1073 vm_page_set_valid(m, base, size) 1074 vm_page_t m; 1075 int base; 1076 int size; 1077 { 1078 m->valid |= vm_page_bits(base, size); 1079 } 1080 1081 /* 1082 * set a page (partially) invalid 1083 */ 1084 void 1085 vm_page_set_invalid(m, base, size) 1086 vm_page_t m; 1087 int base; 1088 int size; 1089 { 1090 int bits; 1091 1092 m->valid &= ~(bits = vm_page_bits(base, size)); 1093 if (m->valid == 0) 1094 m->dirty &= ~bits; 1095 } 1096 1097 /* 1098 * is (partial) page valid? 1099 */ 1100 int 1101 vm_page_is_valid(m, base, size) 1102 vm_page_t m; 1103 int base; 1104 int size; 1105 { 1106 int bits; 1107 1108 if (m->valid && ((m->valid & (bits = vm_page_bits(base, size))) == bits)) 1109 return 1; 1110 else 1111 return 0; 1112 } 1113 1114 1115 /* 1116 * set a page (partially) dirty 1117 */ 1118 void 1119 vm_page_set_dirty(m, base, size) 1120 vm_page_t m; 1121 int base; 1122 int size; 1123 { 1124 if ((base != 0) || (size != PAGE_SIZE)) { 1125 if (pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1126 m->dirty = VM_PAGE_BITS_ALL; 1127 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1128 return; 1129 } 1130 m->dirty |= vm_page_bits(base, size); 1131 } else { 1132 m->dirty = VM_PAGE_BITS_ALL; 1133 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1134 } 1135 } 1136 1137 void 1138 vm_page_test_dirty(m) 1139 vm_page_t m; 1140 { 1141 if ((m->dirty != VM_PAGE_BITS_ALL) && 1142 pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1143 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1144 m->dirty = VM_PAGE_BITS_ALL; 1145 } 1146 } 1147 1148 /* 1149 * set a page (partially) clean 1150 */ 1151 void 1152 vm_page_set_clean(m, base, size) 1153 vm_page_t m; 1154 int base; 1155 int size; 1156 { 1157 m->dirty &= ~vm_page_bits(base, size); 1158 } 1159 1160 /* 1161 * is (partial) page clean 1162 */ 1163 int 1164 vm_page_is_clean(m, base, size) 1165 vm_page_t m; 1166 int base; 1167 int size; 1168 { 1169 if (pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1170 m->dirty = VM_PAGE_BITS_ALL; 1171 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1172 } 1173 if ((m->dirty & m->valid & vm_page_bits(base, size)) == 0) 1174 return 1; 1175 else 1176 return 0; 1177 } 1178 1179 void 1180 print_page_info() 1181 { 1182 printf("cnt.v_free_count: %d\n", cnt.v_free_count); 1183 printf("cnt.v_cache_count: %d\n", cnt.v_cache_count); 1184 printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count); 1185 printf("cnt.v_active_count: %d\n", cnt.v_active_count); 1186 printf("cnt.v_wire_count: %d\n", cnt.v_wire_count); 1187 printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved); 1188 printf("cnt.v_free_min: %d\n", cnt.v_free_min); 1189 printf("cnt.v_free_target: %d\n", cnt.v_free_target); 1190 printf("cnt.v_cache_min: %d\n", cnt.v_cache_min); 1191 printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target); 1192 } 1193