vm_page.c revision 33181
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 1.92 1998/02/06 12:14:27 eivind 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/malloc.h> 74#include <sys/proc.h> 75#include <sys/vmmeter.h> 76#include <sys/vnode.h> 77 78#include <vm/vm.h> 79#include <vm/vm_param.h> 80#include <vm/vm_prot.h> 81#include <sys/lock.h> 82#include <vm/vm_kern.h> 83#include <vm/vm_object.h> 84#include <vm/vm_page.h> 85#include <vm/vm_pageout.h> 86#include <vm/vm_extern.h> 87 88static void vm_page_queue_init __P((void)); 89static vm_page_t vm_page_select_free __P((vm_object_t object, 90 vm_pindex_t pindex, int prefqueue)); 91 92/* 93 * Associated with page of user-allocatable memory is a 94 * page structure. 95 */ 96 97static struct pglist *vm_page_buckets; /* Array of buckets */ 98static int vm_page_bucket_count; /* How big is array? */ 99static int vm_page_hash_mask; /* Mask for hash function */ 100static volatile int vm_page_bucket_generation; 101 102struct pglist vm_page_queue_free[PQ_L2_SIZE] = {0}; 103struct pglist vm_page_queue_zero[PQ_L2_SIZE] = {0}; 104struct pglist vm_page_queue_active = {0}; 105struct pglist vm_page_queue_inactive = {0}; 106struct pglist vm_page_queue_cache[PQ_L2_SIZE] = {0}; 107 108static int no_queue=0; 109 110struct vpgqueues vm_page_queues[PQ_COUNT] = {0}; 111static int pqcnt[PQ_COUNT] = {0}; 112 113static void 114vm_page_queue_init(void) { 115 int i; 116 117 vm_page_queues[PQ_NONE].pl = NULL; 118 vm_page_queues[PQ_NONE].cnt = &no_queue; 119 for(i=0;i<PQ_L2_SIZE;i++) { 120 vm_page_queues[PQ_FREE+i].pl = &vm_page_queue_free[i]; 121 vm_page_queues[PQ_FREE+i].cnt = &cnt.v_free_count; 122 } 123 for(i=0;i<PQ_L2_SIZE;i++) { 124 vm_page_queues[PQ_ZERO+i].pl = &vm_page_queue_zero[i]; 125 vm_page_queues[PQ_ZERO+i].cnt = &cnt.v_free_count; 126 } 127 vm_page_queues[PQ_INACTIVE].pl = &vm_page_queue_inactive; 128 vm_page_queues[PQ_INACTIVE].cnt = &cnt.v_inactive_count; 129 130 vm_page_queues[PQ_ACTIVE].pl = &vm_page_queue_active; 131 vm_page_queues[PQ_ACTIVE].cnt = &cnt.v_active_count; 132 for(i=0;i<PQ_L2_SIZE;i++) { 133 vm_page_queues[PQ_CACHE+i].pl = &vm_page_queue_cache[i]; 134 vm_page_queues[PQ_CACHE+i].cnt = &cnt.v_cache_count; 135 } 136 for(i=0;i<PQ_COUNT;i++) { 137 if (vm_page_queues[i].pl) { 138 TAILQ_INIT(vm_page_queues[i].pl); 139 } else if (i != 0) { 140 panic("vm_page_queue_init: queue %d is null", i); 141 } 142 vm_page_queues[i].lcnt = &pqcnt[i]; 143 } 144} 145 146vm_page_t vm_page_array = 0; 147static int vm_page_array_size = 0; 148long first_page = 0; 149static long last_page; 150static vm_size_t page_mask; 151static int page_shift; 152int vm_page_zero_count = 0; 153 154/* 155 * map of contiguous valid DEV_BSIZE chunks in a page 156 * (this list is valid for page sizes upto 16*DEV_BSIZE) 157 */ 158static u_short vm_page_dev_bsize_chunks[] = { 159 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff, 160 0x1ff, 0x3ff, 0x7ff, 0xfff, 0x1fff, 0x3fff, 0x7fff, 0xffff 161}; 162 163static inline int vm_page_hash __P((vm_object_t object, vm_pindex_t pindex)); 164static int vm_page_freechk_and_unqueue __P((vm_page_t m)); 165static void vm_page_free_wakeup __P((void)); 166 167/* 168 * vm_set_page_size: 169 * 170 * Sets the page size, perhaps based upon the memory 171 * size. Must be called before any use of page-size 172 * dependent functions. 173 * 174 * Sets page_shift and page_mask from cnt.v_page_size. 175 */ 176void 177vm_set_page_size() 178{ 179 180 if (cnt.v_page_size == 0) 181 cnt.v_page_size = DEFAULT_PAGE_SIZE; 182 page_mask = cnt.v_page_size - 1; 183 if ((page_mask & cnt.v_page_size) != 0) 184 panic("vm_set_page_size: page size not a power of two"); 185 for (page_shift = 0;; page_shift++) 186 if ((1 << page_shift) == cnt.v_page_size) 187 break; 188} 189 190/* 191 * vm_page_startup: 192 * 193 * Initializes the resident memory module. 194 * 195 * Allocates memory for the page cells, and 196 * for the object/offset-to-page hash table headers. 197 * Each page cell is initialized and placed on the free list. 198 */ 199 200vm_offset_t 201vm_page_startup(starta, enda, vaddr) 202 register vm_offset_t starta; 203 vm_offset_t enda; 204 register vm_offset_t vaddr; 205{ 206 register vm_offset_t mapped; 207 register vm_page_t m; 208 register struct pglist *bucket; 209 vm_size_t npages, page_range; 210 register vm_offset_t new_start; 211 int i; 212 vm_offset_t pa; 213 int nblocks; 214 vm_offset_t first_managed_page; 215 216 /* the biggest memory array is the second group of pages */ 217 vm_offset_t start; 218 vm_offset_t biggestone, biggestsize; 219 220 vm_offset_t total; 221 222 total = 0; 223 biggestsize = 0; 224 biggestone = 0; 225 nblocks = 0; 226 vaddr = round_page(vaddr); 227 228 for (i = 0; phys_avail[i + 1]; i += 2) { 229 phys_avail[i] = round_page(phys_avail[i]); 230 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); 231 } 232 233 for (i = 0; phys_avail[i + 1]; i += 2) { 234 int size = phys_avail[i + 1] - phys_avail[i]; 235 236 if (size > biggestsize) { 237 biggestone = i; 238 biggestsize = size; 239 } 240 ++nblocks; 241 total += size; 242 } 243 244 start = phys_avail[biggestone]; 245 246 /* 247 * Initialize the queue headers for the free queue, the active queue 248 * and the inactive queue. 249 */ 250 251 vm_page_queue_init(); 252 253 /* 254 * Allocate (and initialize) the hash table buckets. 255 * 256 * The number of buckets MUST BE a power of 2, and the actual value is 257 * the next power of 2 greater than the number of physical pages in 258 * the system. 259 * 260 * Note: This computation can be tweaked if desired. 261 */ 262 vm_page_buckets = (struct pglist *) vaddr; 263 bucket = vm_page_buckets; 264 if (vm_page_bucket_count == 0) { 265 vm_page_bucket_count = 1; 266 while (vm_page_bucket_count < atop(total)) 267 vm_page_bucket_count <<= 1; 268 } 269 vm_page_hash_mask = vm_page_bucket_count - 1; 270 271 /* 272 * Validate these addresses. 273 */ 274 275 new_start = start + vm_page_bucket_count * sizeof(struct pglist); 276 new_start = round_page(new_start); 277 mapped = vaddr; 278 vaddr = pmap_map(mapped, start, new_start, 279 VM_PROT_READ | VM_PROT_WRITE); 280 start = new_start; 281 bzero((caddr_t) mapped, vaddr - mapped); 282 mapped = vaddr; 283 284 for (i = 0; i < vm_page_bucket_count; i++) { 285 TAILQ_INIT(bucket); 286 bucket++; 287 } 288 289 /* 290 * Validate these zone addresses. 291 */ 292 293 new_start = start + (vaddr - mapped); 294 pmap_map(mapped, start, new_start, VM_PROT_READ | VM_PROT_WRITE); 295 bzero((caddr_t) mapped, (vaddr - mapped)); 296 start = round_page(new_start); 297 298 /* 299 * Compute the number of pages of memory that will be available for 300 * use (taking into account the overhead of a page structure per 301 * page). 302 */ 303 304 first_page = phys_avail[0] / PAGE_SIZE; 305 last_page = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE; 306 307 page_range = last_page - (phys_avail[0] / PAGE_SIZE); 308 npages = (total - (page_range * sizeof(struct vm_page)) - 309 (start - phys_avail[biggestone])) / PAGE_SIZE; 310 311 /* 312 * Initialize the mem entry structures now, and put them in the free 313 * queue. 314 */ 315 316 vm_page_array = (vm_page_t) vaddr; 317 mapped = vaddr; 318 319 /* 320 * Validate these addresses. 321 */ 322 323 new_start = round_page(start + page_range * sizeof(struct vm_page)); 324 mapped = pmap_map(mapped, start, new_start, 325 VM_PROT_READ | VM_PROT_WRITE); 326 start = new_start; 327 328 first_managed_page = start / PAGE_SIZE; 329 330 /* 331 * Clear all of the page structures 332 */ 333 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); 334 vm_page_array_size = page_range; 335 336 cnt.v_page_count = 0; 337 cnt.v_free_count = 0; 338 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) { 339 if (i == biggestone) 340 pa = ptoa(first_managed_page); 341 else 342 pa = phys_avail[i]; 343 while (pa < phys_avail[i + 1] && npages-- > 0) { 344 ++cnt.v_page_count; 345 ++cnt.v_free_count; 346 m = PHYS_TO_VM_PAGE(pa); 347 m->phys_addr = pa; 348 m->flags = 0; 349 m->pc = (pa >> PAGE_SHIFT) & PQ_L2_MASK; 350 m->queue = PQ_FREE + m->pc; 351 TAILQ_INSERT_TAIL(vm_page_queues[m->queue].pl, m, pageq); 352 ++(*vm_page_queues[m->queue].lcnt); 353 pa += PAGE_SIZE; 354 } 355 } 356 return (mapped); 357} 358 359/* 360 * vm_page_hash: 361 * 362 * Distributes the object/offset key pair among hash buckets. 363 * 364 * NOTE: This macro depends on vm_page_bucket_count being a power of 2. 365 */ 366static inline int 367vm_page_hash(object, pindex) 368 vm_object_t object; 369 vm_pindex_t pindex; 370{ 371 return ((((unsigned) object) >> 5) + (pindex >> 1)) & vm_page_hash_mask; 372} 373 374/* 375 * vm_page_insert: [ internal use only ] 376 * 377 * Inserts the given mem entry into the object/object-page 378 * table and object list. 379 * 380 * The object and page must be locked, and must be splhigh. 381 */ 382 383void 384vm_page_insert(m, object, pindex) 385 register vm_page_t m; 386 register vm_object_t object; 387 register vm_pindex_t pindex; 388{ 389 register struct pglist *bucket; 390 391#if !defined(MAX_PERF) 392 if (m->flags & PG_TABLED) 393 panic("vm_page_insert: already inserted"); 394#endif 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 vm_page_bucket_generation++; 410 411 /* 412 * Now link into the object's list of backed pages. 413 */ 414 415 TAILQ_INSERT_TAIL(&object->memq, m, listq); 416 m->flags |= PG_TABLED; 417 m->object->page_hint = m; 418 m->object->generation++; 419 420 if (m->wire_count) 421 object->wire_count++; 422 423 if ((m->queue - m->pc) == PQ_CACHE) 424 object->cache_count++; 425 426 /* 427 * And show that the object has one more resident page. 428 */ 429 430 object->resident_page_count++; 431} 432 433/* 434 * vm_page_remove: [ internal use only ] 435 * NOTE: used by device pager as well -wfj 436 * 437 * Removes the given mem entry from the object/offset-page 438 * table and the object page list. 439 * 440 * The object and page must be locked, and at splhigh. 441 */ 442 443void 444vm_page_remove(m) 445 register vm_page_t m; 446{ 447 register struct pglist *bucket; 448 vm_object_t object; 449 450 if (!(m->flags & PG_TABLED)) 451 return; 452 453#if !defined(MAX_PERF) 454 if ((m->flags & PG_BUSY) == 0) { 455 panic("vm_page_remove: page not busy"); 456 } 457#endif 458 459 m->flags &= ~PG_BUSY; 460 if (m->flags & PG_WANTED) { 461 m->flags &= ~PG_WANTED; 462 wakeup(m); 463 } 464 465 object = m->object; 466 if (object->page_hint == m) 467 object->page_hint = NULL; 468 469 if (m->wire_count) 470 object->wire_count--; 471 472 if ((m->queue - m->pc) == PQ_CACHE) 473 object->cache_count--; 474 475 /* 476 * Remove from the object_object/offset hash table 477 */ 478 479 bucket = &vm_page_buckets[vm_page_hash(m->object, m->pindex)]; 480 TAILQ_REMOVE(bucket, m, hashq); 481 vm_page_bucket_generation++; 482 483 /* 484 * Now remove from the object's list of backed pages. 485 */ 486 487 TAILQ_REMOVE(&object->memq, m, listq); 488 489 /* 490 * And show that the object has one fewer resident page. 491 */ 492 493 object->resident_page_count--; 494 object->generation++; 495 m->object = NULL; 496 497 m->flags &= ~PG_TABLED; 498} 499 500/* 501 * vm_page_lookup: 502 * 503 * Returns the page associated with the object/offset 504 * pair specified; if none is found, NULL is returned. 505 * 506 * The object must be locked. No side effects. 507 */ 508 509vm_page_t 510vm_page_lookup(object, pindex) 511 register vm_object_t object; 512 register vm_pindex_t pindex; 513{ 514 register vm_page_t m; 515 register struct pglist *bucket; 516 int generation; 517 int s; 518 519 /* 520 * Search the hash table for this object/offset pair 521 */ 522 523 if (object->page_hint && (object->page_hint->pindex == pindex) && 524 (object->page_hint->object == object)) 525 return object->page_hint; 526 527retry: 528 generation = vm_page_bucket_generation; 529 bucket = &vm_page_buckets[vm_page_hash(object, pindex)]; 530 for (m = TAILQ_FIRST(bucket); m != NULL; m = TAILQ_NEXT(m,hashq)) { 531 if ((m->object == object) && (m->pindex == pindex)) { 532 if (vm_page_bucket_generation != generation) 533 goto retry; 534 m->object->page_hint = m; 535 return (m); 536 } 537 } 538 if (vm_page_bucket_generation != generation) 539 goto retry; 540 return (NULL); 541} 542 543/* 544 * vm_page_rename: 545 * 546 * Move the given memory entry from its 547 * current object to the specified target object/offset. 548 * 549 * The object must be locked. 550 */ 551void 552vm_page_rename(m, new_object, new_pindex) 553 register vm_page_t m; 554 register vm_object_t new_object; 555 vm_pindex_t new_pindex; 556{ 557 int s; 558 559 s = splvm(); 560 vm_page_remove(m); 561 vm_page_insert(m, new_object, new_pindex); 562 splx(s); 563} 564 565/* 566 * vm_page_unqueue without any wakeup 567 */ 568void 569vm_page_unqueue_nowakeup(m) 570 vm_page_t m; 571{ 572 int queue = m->queue; 573 struct vpgqueues *pq; 574 if (queue != PQ_NONE) { 575 pq = &vm_page_queues[queue]; 576 m->queue = PQ_NONE; 577 TAILQ_REMOVE(pq->pl, m, pageq); 578 (*pq->cnt)--; 579 (*pq->lcnt)--; 580 if ((queue - m->pc) == PQ_CACHE) { 581 if (m->object) 582 m->object->cache_count--; 583 } 584 } 585} 586 587/* 588 * vm_page_unqueue must be called at splhigh(); 589 */ 590void 591vm_page_unqueue(m) 592 vm_page_t m; 593{ 594 int queue = m->queue; 595 struct vpgqueues *pq; 596 if (queue != PQ_NONE) { 597 m->queue = PQ_NONE; 598 pq = &vm_page_queues[queue]; 599 TAILQ_REMOVE(pq->pl, m, pageq); 600 (*pq->cnt)--; 601 (*pq->lcnt)--; 602 if ((queue - m->pc) == PQ_CACHE) { 603 if ((cnt.v_cache_count + cnt.v_free_count) < 604 (cnt.v_free_reserved + cnt.v_cache_min)) 605 pagedaemon_wakeup(); 606 if (m->object) 607 m->object->cache_count--; 608 } 609 } 610} 611 612/* 613 * Find a page on the specified queue with color optimization. 614 */ 615vm_page_t 616vm_page_list_find(basequeue, index) 617 int basequeue, index; 618{ 619#if PQ_L2_SIZE > 1 620 621 int i,j; 622 vm_page_t m; 623 int hindex; 624 struct vpgqueues *pq; 625 626 pq = &vm_page_queues[basequeue]; 627 628 m = TAILQ_FIRST(pq[index].pl); 629 if (m) 630 return m; 631 632 for(j = 0; j < PQ_L1_SIZE; j++) { 633 int ij; 634 for(i = (PQ_L2_SIZE / 2) - PQ_L1_SIZE; 635 (ij = i + j) > 0; 636 i -= PQ_L1_SIZE) { 637 638 hindex = index + ij; 639 if (hindex >= PQ_L2_SIZE) 640 hindex -= PQ_L2_SIZE; 641 if (m = TAILQ_FIRST(pq[hindex].pl)) 642 return m; 643 644 hindex = index - ij; 645 if (hindex < 0) 646 hindex += PQ_L2_SIZE; 647 if (m = TAILQ_FIRST(pq[hindex].pl)) 648 return m; 649 } 650 } 651 652 hindex = index + PQ_L2_SIZE / 2; 653 if (hindex >= PQ_L2_SIZE) 654 hindex -= PQ_L2_SIZE; 655 m = TAILQ_FIRST(pq[hindex].pl); 656 if (m) 657 return m; 658 659 return NULL; 660#else 661 return TAILQ_FIRST(vm_page_queues[basequeue].pl); 662#endif 663 664} 665 666/* 667 * Find a page on the specified queue with color optimization. 668 */ 669vm_page_t 670vm_page_select(object, pindex, basequeue) 671 vm_object_t object; 672 vm_pindex_t pindex; 673 int basequeue; 674{ 675 676#if PQ_L2_SIZE > 1 677 int index; 678 index = (pindex + object->pg_color) & PQ_L2_MASK; 679 return vm_page_list_find(basequeue, index); 680 681#else 682 return TAILQ_FIRST(vm_page_queues[basequeue].pl); 683#endif 684 685} 686 687/* 688 * Find a free or zero page, with specified preference. 689 */ 690static vm_page_t 691vm_page_select_free(object, pindex, prefqueue) 692 vm_object_t object; 693 vm_pindex_t pindex; 694 int prefqueue; 695{ 696#if PQ_L2_SIZE > 1 697 int i,j; 698 int index, hindex; 699#endif 700 vm_page_t m, mh; 701 int oqueuediff; 702 struct vpgqueues *pq; 703 704 if (prefqueue == PQ_ZERO) 705 oqueuediff = PQ_FREE - PQ_ZERO; 706 else 707 oqueuediff = PQ_ZERO - PQ_FREE; 708 709 if (mh = object->page_hint) { 710 if (mh->pindex == (pindex - 1)) { 711 if ((mh->flags & PG_FICTITIOUS) == 0) { 712 if ((mh < &vm_page_array[cnt.v_page_count-1]) && 713 (mh >= &vm_page_array[0])) { 714 int queue; 715 m = mh + 1; 716 if (VM_PAGE_TO_PHYS(m) == (VM_PAGE_TO_PHYS(mh) + PAGE_SIZE)) { 717 queue = m->queue - m->pc; 718 if (queue == PQ_FREE || queue == PQ_ZERO) { 719 return m; 720 } 721 } 722 } 723 } 724 } 725 } 726 727 pq = &vm_page_queues[prefqueue]; 728 729#if PQ_L2_SIZE > 1 730 731 index = (pindex + object->pg_color) & PQ_L2_MASK; 732 733 if (m = TAILQ_FIRST(pq[index].pl)) 734 return m; 735 if (m = TAILQ_FIRST(pq[index + oqueuediff].pl)) 736 return m; 737 738 for(j = 0; j < PQ_L1_SIZE; j++) { 739 int ij; 740 for(i = (PQ_L2_SIZE / 2) - PQ_L1_SIZE; 741 (ij = i + j) >= 0; 742 i -= PQ_L1_SIZE) { 743 744 hindex = index + ij; 745 if (hindex >= PQ_L2_SIZE) 746 hindex -= PQ_L2_SIZE; 747 if (m = TAILQ_FIRST(pq[hindex].pl)) 748 return m; 749 if (m = TAILQ_FIRST(pq[hindex + oqueuediff].pl)) 750 return m; 751 752 hindex = index - ij; 753 if (hindex < 0) 754 hindex += PQ_L2_SIZE; 755 if (m = TAILQ_FIRST(pq[hindex].pl)) 756 return m; 757 if (m = TAILQ_FIRST(pq[hindex + oqueuediff].pl)) 758 return m; 759 } 760 } 761 762 hindex = index + PQ_L2_SIZE / 2; 763 if (hindex >= PQ_L2_SIZE) 764 hindex -= PQ_L2_SIZE; 765 if (m = TAILQ_FIRST(pq[hindex].pl)) 766 return m; 767 if (m = TAILQ_FIRST(pq[hindex+oqueuediff].pl)) 768 return m; 769 770#else 771 if (m = TAILQ_FIRST(pq[0].pl)) 772 return m; 773 else 774 return TAILQ_FIRST(pq[oqueuediff].pl); 775#endif 776 777 return NULL; 778} 779 780/* 781 * vm_page_alloc: 782 * 783 * Allocate and return a memory cell associated 784 * with this VM object/offset pair. 785 * 786 * page_req classes: 787 * VM_ALLOC_NORMAL normal process request 788 * VM_ALLOC_SYSTEM system *really* needs a page 789 * VM_ALLOC_INTERRUPT interrupt time request 790 * VM_ALLOC_ZERO zero page 791 * 792 * Object must be locked. 793 */ 794vm_page_t 795vm_page_alloc(object, pindex, page_req) 796 vm_object_t object; 797 vm_pindex_t pindex; 798 int page_req; 799{ 800 register vm_page_t m; 801 struct vpgqueues *pq; 802 vm_object_t oldobject; 803 int queue, qtype; 804 int s; 805 806#ifdef DIAGNOSTIC 807 m = vm_page_lookup(object, pindex); 808 if (m) 809 panic("vm_page_alloc: page already allocated"); 810#endif 811 812 if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) { 813 page_req = VM_ALLOC_SYSTEM; 814 }; 815 816 s = splvm(); 817 818 switch (page_req) { 819 820 case VM_ALLOC_NORMAL: 821 if (cnt.v_free_count >= cnt.v_free_reserved) { 822 m = vm_page_select_free(object, pindex, PQ_FREE); 823#if defined(DIAGNOSTIC) 824 if (m == NULL) 825 panic("vm_page_alloc(NORMAL): missing page on free queue\n"); 826#endif 827 } else { 828 m = vm_page_select(object, pindex, PQ_CACHE); 829 if (m == NULL) { 830 splx(s); 831#if defined(DIAGNOSTIC) 832 if (cnt.v_cache_count > 0) 833 printf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", cnt.v_cache_count); 834#endif 835 vm_pageout_deficit++; 836 pagedaemon_wakeup(); 837 return (NULL); 838 } 839 } 840 break; 841 842 case VM_ALLOC_ZERO: 843 if (cnt.v_free_count >= cnt.v_free_reserved) { 844 m = vm_page_select_free(object, pindex, PQ_ZERO); 845#if defined(DIAGNOSTIC) 846 if (m == NULL) 847 panic("vm_page_alloc(ZERO): missing page on free queue\n"); 848#endif 849 } else { 850 m = vm_page_select(object, pindex, PQ_CACHE); 851 if (m == NULL) { 852 splx(s); 853#if defined(DIAGNOSTIC) 854 if (cnt.v_cache_count > 0) 855 printf("vm_page_alloc(ZERO): missing pages on cache queue: %d\n", cnt.v_cache_count); 856#endif 857 vm_pageout_deficit++; 858 pagedaemon_wakeup(); 859 return (NULL); 860 } 861 } 862 break; 863 864 case VM_ALLOC_SYSTEM: 865 if ((cnt.v_free_count >= cnt.v_free_reserved) || 866 ((cnt.v_cache_count == 0) && 867 (cnt.v_free_count >= cnt.v_interrupt_free_min))) { 868 m = vm_page_select_free(object, pindex, PQ_FREE); 869#if defined(DIAGNOSTIC) 870 if (m == NULL) 871 panic("vm_page_alloc(SYSTEM): missing page on free queue\n"); 872#endif 873 } else { 874 m = vm_page_select(object, pindex, PQ_CACHE); 875 if (m == NULL) { 876 splx(s); 877#if defined(DIAGNOSTIC) 878 if (cnt.v_cache_count > 0) 879 printf("vm_page_alloc(SYSTEM): missing pages on cache queue: %d\n", cnt.v_cache_count); 880#endif 881 vm_pageout_deficit++; 882 pagedaemon_wakeup(); 883 return (NULL); 884 } 885 } 886 break; 887 888 case VM_ALLOC_INTERRUPT: 889 if (cnt.v_free_count > 0) { 890 m = vm_page_select_free(object, pindex, PQ_FREE); 891#if defined(DIAGNOSTIC) 892 if (m == NULL) 893 panic("vm_page_alloc(INTERRUPT): missing page on free queue\n"); 894#endif 895 } else { 896 splx(s); 897 vm_pageout_deficit++; 898 pagedaemon_wakeup(); 899 return (NULL); 900 } 901 break; 902 903 default: 904 m = NULL; 905#if !defined(MAX_PERF) 906 panic("vm_page_alloc: invalid allocation class"); 907#endif 908 } 909 910 queue = m->queue; 911 qtype = queue - m->pc; 912 if (qtype == PQ_ZERO) 913 vm_page_zero_count--; 914 pq = &vm_page_queues[queue]; 915 TAILQ_REMOVE(pq->pl, m, pageq); 916 (*pq->cnt)--; 917 (*pq->lcnt)--; 918 oldobject = NULL; 919 if (qtype == PQ_ZERO) { 920 m->flags = PG_ZERO|PG_BUSY; 921 } else if (qtype == PQ_CACHE) { 922 oldobject = m->object; 923 m->flags |= PG_BUSY; 924 vm_page_remove(m); 925 m->flags = PG_BUSY; 926 } else { 927 m->flags = PG_BUSY; 928 } 929 m->wire_count = 0; 930 m->hold_count = 0; 931 m->act_count = 0; 932 m->busy = 0; 933 m->valid = 0; 934 m->dirty = 0; 935 m->queue = PQ_NONE; 936 937 /* XXX before splx until vm_page_insert is safe */ 938 vm_page_insert(m, object, pindex); 939 940 /* 941 * Don't wakeup too often - wakeup the pageout daemon when 942 * we would be nearly out of memory. 943 */ 944 if (((cnt.v_free_count + cnt.v_cache_count) < 945 (cnt.v_free_reserved + cnt.v_cache_min)) || 946 (cnt.v_free_count < cnt.v_pageout_free_min)) 947 pagedaemon_wakeup(); 948 949 if ((qtype == PQ_CACHE) && 950 ((page_req == VM_ALLOC_NORMAL) || (page_req == VM_ALLOC_ZERO)) && 951 oldobject && (oldobject->type == OBJT_VNODE) && 952 ((oldobject->flags & OBJ_DEAD) == 0)) { 953 struct vnode *vp; 954 vp = (struct vnode *) oldobject->handle; 955 if (vp && VSHOULDFREE(vp)) { 956 if ((vp->v_flag & (VFREE|VTBFREE|VDOOMED)) == 0) { 957 TAILQ_INSERT_TAIL(&vnode_tobefree_list, vp, v_freelist); 958 vp->v_flag |= VTBFREE; 959 } 960 } 961 } 962 splx(s); 963 964 return (m); 965} 966 967void 968vm_wait() 969{ 970 int s; 971 972 s = splvm(); 973 if (curproc == pageproc) { 974 vm_pageout_pages_needed = 1; 975 tsleep(&vm_pageout_pages_needed, PSWP, "vmwait", 0); 976 } else { 977 if (!vm_pages_needed) { 978 vm_pages_needed++; 979 wakeup(&vm_pages_needed); 980 } 981 tsleep(&cnt.v_free_count, PVM, "vmwait", 0); 982 } 983 splx(s); 984} 985 986 987/* 988 * vm_page_activate: 989 * 990 * Put the specified page on the active list (if appropriate). 991 * 992 * The page queues must be locked. 993 */ 994void 995vm_page_activate(m) 996 register vm_page_t m; 997{ 998 int s; 999 vm_page_t np; 1000 vm_object_t object; 1001 1002 s = splvm(); 1003 if (m->queue != PQ_ACTIVE) { 1004 if ((m->queue - m->pc) == PQ_CACHE) 1005 cnt.v_reactivated++; 1006 1007 vm_page_unqueue(m); 1008 1009 if (m->wire_count == 0) { 1010 m->queue = PQ_ACTIVE; 1011 ++(*vm_page_queues[PQ_ACTIVE].lcnt); 1012 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 1013 if (m->act_count < ACT_INIT) 1014 m->act_count = ACT_INIT; 1015 cnt.v_active_count++; 1016 } 1017 } else { 1018 if (m->act_count < ACT_INIT) 1019 m->act_count = ACT_INIT; 1020 } 1021 1022 object = m->object; 1023 TAILQ_REMOVE(&object->memq, m, listq); 1024 TAILQ_INSERT_TAIL(&object->memq, m, listq); 1025 object->generation++; 1026 1027 splx(s); 1028} 1029 1030/* 1031 * helper routine for vm_page_free and vm_page_free_zero 1032 */ 1033static int 1034vm_page_freechk_and_unqueue(m) 1035 vm_page_t m; 1036{ 1037 vm_object_t oldobject; 1038 1039 oldobject = m->object; 1040 1041#if !defined(MAX_PERF) 1042 if (m->busy || ((m->queue - m->pc) == PQ_FREE) || 1043 (m->hold_count != 0)) { 1044 printf("vm_page_free: pindex(%ld), busy(%d), PG_BUSY(%d), hold(%d)\n", 1045 m->pindex, m->busy, 1046 (m->flags & PG_BUSY) ? 1 : 0, m->hold_count); 1047 if ((m->queue - m->pc) == PQ_FREE) 1048 panic("vm_page_free: freeing free page"); 1049 else 1050 panic("vm_page_free: freeing busy page"); 1051 } 1052#endif 1053 1054 vm_page_unqueue_nowakeup(m); 1055 vm_page_remove(m); 1056 1057 if ((m->flags & PG_FICTITIOUS) != 0) { 1058 return 0; 1059 } 1060 1061 if (m->wire_count != 0) { 1062#if !defined(MAX_PERF) 1063 if (m->wire_count > 1) { 1064 panic("vm_page_free: invalid wire count (%d), pindex: 0x%x", 1065 m->wire_count, m->pindex); 1066 } 1067#endif 1068 m->wire_count = 0; 1069 if (m->object) 1070 m->object->wire_count--; 1071 cnt.v_wire_count--; 1072 } 1073 1074 if (oldobject && (oldobject->type == OBJT_VNODE) && 1075 ((oldobject->flags & OBJ_DEAD) == 0)) { 1076 struct vnode *vp; 1077 vp = (struct vnode *) oldobject->handle; 1078 if (vp && VSHOULDFREE(vp)) { 1079 if ((vp->v_flag & (VTBFREE|VDOOMED|VFREE)) == 0) { 1080 TAILQ_INSERT_TAIL(&vnode_tobefree_list, vp, v_freelist); 1081 vp->v_flag |= VTBFREE; 1082 } 1083 } 1084 } 1085 1086 return 1; 1087} 1088 1089/* 1090 * helper routine for vm_page_free and vm_page_free_zero 1091 */ 1092static __inline void 1093vm_page_free_wakeup() 1094{ 1095 1096/* 1097 * if pageout daemon needs pages, then tell it that there are 1098 * some free. 1099 */ 1100 if (vm_pageout_pages_needed) { 1101 wakeup(&vm_pageout_pages_needed); 1102 vm_pageout_pages_needed = 0; 1103 } 1104 /* 1105 * wakeup processes that are waiting on memory if we hit a 1106 * high water mark. And wakeup scheduler process if we have 1107 * lots of memory. this process will swapin processes. 1108 */ 1109 if (vm_pages_needed && 1110 ((cnt.v_free_count + cnt.v_cache_count) >= cnt.v_free_min)) { 1111 wakeup(&cnt.v_free_count); 1112 vm_pages_needed = 0; 1113 } 1114} 1115 1116/* 1117 * vm_page_free: 1118 * 1119 * Returns the given page to the free list, 1120 * disassociating it with any VM object. 1121 * 1122 * Object and page must be locked prior to entry. 1123 */ 1124void 1125vm_page_free(m) 1126 register vm_page_t m; 1127{ 1128 int s; 1129 struct vpgqueues *pq; 1130 1131 s = splvm(); 1132 1133 cnt.v_tfree++; 1134 1135 if (!vm_page_freechk_and_unqueue(m)) { 1136 splx(s); 1137 return; 1138 } 1139 1140 m->queue = PQ_FREE + m->pc; 1141 pq = &vm_page_queues[m->queue]; 1142 ++(*pq->lcnt); 1143 ++(*pq->cnt); 1144 /* 1145 * If the pageout process is grabbing the page, it is likely 1146 * that the page is NOT in the cache. It is more likely that 1147 * the page will be partially in the cache if it is being 1148 * explicitly freed. 1149 */ 1150 if (curproc == pageproc) { 1151 TAILQ_INSERT_TAIL(pq->pl, m, pageq); 1152 } else { 1153 TAILQ_INSERT_HEAD(pq->pl, m, pageq); 1154 } 1155 1156 vm_page_free_wakeup(); 1157 splx(s); 1158} 1159 1160void 1161vm_page_free_zero(m) 1162 register vm_page_t m; 1163{ 1164 int s; 1165 struct vpgqueues *pq; 1166 1167 s = splvm(); 1168 1169 cnt.v_tfree++; 1170 1171 if (!vm_page_freechk_and_unqueue(m)) { 1172 splx(s); 1173 return; 1174 } 1175 1176 m->queue = PQ_ZERO + m->pc; 1177 pq = &vm_page_queues[m->queue]; 1178 ++(*pq->lcnt); 1179 ++(*pq->cnt); 1180 1181 TAILQ_INSERT_HEAD(pq->pl, m, pageq); 1182 ++vm_page_zero_count; 1183 vm_page_free_wakeup(); 1184 splx(s); 1185} 1186 1187/* 1188 * vm_page_wire: 1189 * 1190 * Mark this page as wired down by yet 1191 * another map, removing it from paging queues 1192 * as necessary. 1193 * 1194 * The page queues must be locked. 1195 */ 1196void 1197vm_page_wire(m) 1198 register vm_page_t m; 1199{ 1200 int s; 1201 1202 if (m->wire_count == 0) { 1203 s = splvm(); 1204 vm_page_unqueue(m); 1205 splx(s); 1206 cnt.v_wire_count++; 1207 if (m->object) 1208 m->object->wire_count++; 1209 } 1210 (*vm_page_queues[PQ_NONE].lcnt)++; 1211 m->wire_count++; 1212 m->flags |= PG_MAPPED; 1213} 1214 1215/* 1216 * vm_page_unwire: 1217 * 1218 * Release one wiring of this page, potentially 1219 * enabling it to be paged again. 1220 * 1221 * The page queues must be locked. 1222 */ 1223void 1224vm_page_unwire(m) 1225 register vm_page_t m; 1226{ 1227 int s; 1228 1229 s = splvm(); 1230 1231 if (m->wire_count > 0) { 1232 m->wire_count--; 1233 if (m->wire_count == 0) { 1234 if (m->object) 1235 m->object->wire_count--; 1236 cnt.v_wire_count--; 1237 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 1238 m->queue = PQ_ACTIVE; 1239 (*vm_page_queues[PQ_ACTIVE].lcnt)++; 1240 cnt.v_active_count++; 1241 } 1242 } else { 1243#if !defined(MAX_PERF) 1244 panic("vm_page_unwire: invalid wire count: %d\n", m->wire_count); 1245#endif 1246 } 1247 splx(s); 1248} 1249 1250 1251/* 1252 * vm_page_deactivate: 1253 * 1254 * Returns the given page to the inactive list, 1255 * indicating that no physical maps have access 1256 * to this page. [Used by the physical mapping system.] 1257 * 1258 * The page queues must be locked. 1259 */ 1260void 1261vm_page_deactivate(m) 1262 register vm_page_t m; 1263{ 1264 int s; 1265 1266 /* 1267 * Only move active pages -- ignore locked or already inactive ones. 1268 * 1269 * XXX: sometimes we get pages which aren't wired down or on any queue - 1270 * we need to put them on the inactive queue also, otherwise we lose 1271 * track of them. Paul Mackerras (paulus@cs.anu.edu.au) 9-Jan-93. 1272 */ 1273 if (m->queue == PQ_INACTIVE) 1274 return; 1275 1276 s = splvm(); 1277 if (m->wire_count == 0 && m->hold_count == 0) { 1278 if ((m->queue - m->pc) == PQ_CACHE) 1279 cnt.v_reactivated++; 1280 vm_page_unqueue(m); 1281 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); 1282 m->queue = PQ_INACTIVE; 1283 ++(*vm_page_queues[PQ_INACTIVE].lcnt); 1284 cnt.v_inactive_count++; 1285 } 1286 splx(s); 1287} 1288 1289/* 1290 * vm_page_cache 1291 * 1292 * Put the specified page onto the page cache queue (if appropriate). 1293 */ 1294void 1295vm_page_cache(m) 1296 register vm_page_t m; 1297{ 1298 int s; 1299 1300#if !defined(MAX_PERF) 1301 if ((m->flags & PG_BUSY) || m->busy || m->wire_count) { 1302 printf("vm_page_cache: attempting to cache busy page\n"); 1303 return; 1304 } 1305#endif 1306 if ((m->queue - m->pc) == PQ_CACHE) 1307 return; 1308 1309 vm_page_protect(m, VM_PROT_NONE); 1310#if !defined(MAX_PERF) 1311 if (m->dirty != 0) { 1312 panic("vm_page_cache: caching a dirty page, pindex: %d", m->pindex); 1313 } 1314#endif 1315 s = splvm(); 1316 vm_page_unqueue_nowakeup(m); 1317 m->queue = PQ_CACHE + m->pc; 1318 (*vm_page_queues[m->queue].lcnt)++; 1319 TAILQ_INSERT_TAIL(vm_page_queues[m->queue].pl, m, pageq); 1320 cnt.v_cache_count++; 1321 m->object->cache_count++; 1322 vm_page_free_wakeup(); 1323 splx(s); 1324} 1325 1326/* 1327 * Grab a page, waiting until we are waken up due to the page 1328 * changing state. We keep on waiting, if the page continues 1329 * to be in the object. If the page doesn't exist, allocate it. 1330 */ 1331vm_page_t 1332vm_page_grab(object, pindex, allocflags) 1333 vm_object_t object; 1334 vm_pindex_t pindex; 1335 int allocflags; 1336{ 1337 1338 vm_page_t m; 1339 int s, generation; 1340 1341retrylookup: 1342 if ((m = vm_page_lookup(object, pindex)) != NULL) { 1343 if (m->busy || (m->flags & PG_BUSY)) { 1344 generation = object->generation; 1345 1346 s = splvm(); 1347 while ((object->generation == generation) && 1348 (m->busy || (m->flags & PG_BUSY))) { 1349 m->flags |= PG_WANTED | PG_REFERENCED; 1350 tsleep(m, PVM, "pgrbwt", 0); 1351 if ((allocflags & VM_ALLOC_RETRY) == 0) { 1352 splx(s); 1353 return NULL; 1354 } 1355 } 1356 splx(s); 1357 goto retrylookup; 1358 } else { 1359 m->flags |= PG_BUSY; 1360 return m; 1361 } 1362 } 1363 1364 m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY); 1365 if (m == NULL) { 1366 VM_WAIT; 1367 if ((allocflags & VM_ALLOC_RETRY) == 0) 1368 return NULL; 1369 goto retrylookup; 1370 } 1371 1372 return m; 1373} 1374 1375/* 1376 * mapping function for valid bits or for dirty bits in 1377 * a page 1378 */ 1379inline int 1380vm_page_bits(int base, int size) 1381{ 1382 u_short chunk; 1383 1384 if ((base == 0) && (size >= PAGE_SIZE)) 1385 return VM_PAGE_BITS_ALL; 1386 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 1387 base = (base % PAGE_SIZE) / DEV_BSIZE; 1388 chunk = vm_page_dev_bsize_chunks[size / DEV_BSIZE]; 1389 return (chunk << base) & VM_PAGE_BITS_ALL; 1390} 1391 1392/* 1393 * set a page valid and clean 1394 */ 1395void 1396vm_page_set_validclean(m, base, size) 1397 vm_page_t m; 1398 int base; 1399 int size; 1400{ 1401 int pagebits = vm_page_bits(base, size); 1402 m->valid |= pagebits; 1403 m->dirty &= ~pagebits; 1404 if( base == 0 && size == PAGE_SIZE) 1405 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1406} 1407 1408/* 1409 * set a page (partially) invalid 1410 */ 1411void 1412vm_page_set_invalid(m, base, size) 1413 vm_page_t m; 1414 int base; 1415 int size; 1416{ 1417 int bits; 1418 1419 m->valid &= ~(bits = vm_page_bits(base, size)); 1420 if (m->valid == 0) 1421 m->dirty &= ~bits; 1422} 1423 1424/* 1425 * is (partial) page valid? 1426 */ 1427int 1428vm_page_is_valid(m, base, size) 1429 vm_page_t m; 1430 int base; 1431 int size; 1432{ 1433 int bits = vm_page_bits(base, size); 1434 1435 if (m->valid && ((m->valid & bits) == bits)) 1436 return 1; 1437 else 1438 return 0; 1439} 1440 1441void 1442vm_page_test_dirty(m) 1443 vm_page_t m; 1444{ 1445 if ((m->dirty != VM_PAGE_BITS_ALL) && 1446 pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1447 m->dirty = VM_PAGE_BITS_ALL; 1448 } 1449} 1450 1451/* 1452 * This interface is for merging with malloc() someday. 1453 * Even if we never implement compaction so that contiguous allocation 1454 * works after initialization time, malloc()'s data structures are good 1455 * for statistics and for allocations of less than a page. 1456 */ 1457void * 1458contigmalloc1(size, type, flags, low, high, alignment, boundary, map) 1459 unsigned long size; /* should be size_t here and for malloc() */ 1460 struct malloc_type *type; 1461 int flags; 1462 unsigned long low; 1463 unsigned long high; 1464 unsigned long alignment; 1465 unsigned long boundary; 1466 vm_map_t map; 1467{ 1468 int i, s, start; 1469 vm_offset_t addr, phys, tmp_addr; 1470 int pass; 1471 vm_page_t pga = vm_page_array; 1472 1473 size = round_page(size); 1474#if !defined(MAX_PERF) 1475 if (size == 0) 1476 panic("contigmalloc1: size must not be 0"); 1477 if ((alignment & (alignment - 1)) != 0) 1478 panic("contigmalloc1: alignment must be a power of 2"); 1479 if ((boundary & (boundary - 1)) != 0) 1480 panic("contigmalloc1: boundary must be a power of 2"); 1481#endif 1482 1483 start = 0; 1484 for (pass = 0; pass <= 1; pass++) { 1485 s = splvm(); 1486again: 1487 /* 1488 * Find first page in array that is free, within range, aligned, and 1489 * such that the boundary won't be crossed. 1490 */ 1491 for (i = start; i < cnt.v_page_count; i++) { 1492 int pqtype; 1493 phys = VM_PAGE_TO_PHYS(&pga[i]); 1494 pqtype = pga[i].queue - pga[i].pc; 1495 if (((pqtype == PQ_ZERO) || (pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) && 1496 (phys >= low) && (phys < high) && 1497 ((phys & (alignment - 1)) == 0) && 1498 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0)) 1499 break; 1500 } 1501 1502 /* 1503 * If the above failed or we will exceed the upper bound, fail. 1504 */ 1505 if ((i == cnt.v_page_count) || 1506 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 1507 vm_page_t m, next; 1508 1509again1: 1510 for (m = TAILQ_FIRST(&vm_page_queue_inactive); 1511 m != NULL; 1512 m = next) { 1513 1514 if (m->queue != PQ_INACTIVE) { 1515 break; 1516 } 1517 1518 next = TAILQ_NEXT(m, pageq); 1519 if (m->flags & PG_BUSY) { 1520 m->flags |= PG_WANTED; 1521 tsleep(m, PVM, "vpctw0", 0); 1522 goto again1; 1523 } 1524 vm_page_test_dirty(m); 1525 if (m->dirty) { 1526 if (m->object->type == OBJT_VNODE) { 1527 vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); 1528 vm_object_page_clean(m->object, 0, 0, TRUE); 1529 VOP_UNLOCK(m->object->handle, 0, curproc); 1530 goto again1; 1531 } else if (m->object->type == OBJT_SWAP || 1532 m->object->type == OBJT_DEFAULT) { 1533 vm_page_protect(m, VM_PROT_NONE); 1534 vm_pageout_flush(&m, 1, 0); 1535 goto again1; 1536 } 1537 } 1538 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 1539 vm_page_cache(m); 1540 } 1541 1542 for (m = TAILQ_FIRST(&vm_page_queue_active); 1543 m != NULL; 1544 m = next) { 1545 1546 if (m->queue != PQ_ACTIVE) { 1547 break; 1548 } 1549 1550 next = TAILQ_NEXT(m, pageq); 1551 if (m->flags & PG_BUSY) { 1552 m->flags |= PG_WANTED; 1553 tsleep(m, PVM, "vpctw1", 0); 1554 goto again1; 1555 } 1556 vm_page_test_dirty(m); 1557 if (m->dirty) { 1558 if (m->object->type == OBJT_VNODE) { 1559 vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); 1560 vm_object_page_clean(m->object, 0, 0, TRUE); 1561 VOP_UNLOCK(m->object->handle, 0, curproc); 1562 goto again1; 1563 } else if (m->object->type == OBJT_SWAP || 1564 m->object->type == OBJT_DEFAULT) { 1565 vm_page_protect(m, VM_PROT_NONE); 1566 vm_pageout_flush(&m, 1, 0); 1567 goto again1; 1568 } 1569 } 1570 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 1571 vm_page_cache(m); 1572 } 1573 1574 splx(s); 1575 continue; 1576 } 1577 start = i; 1578 1579 /* 1580 * Check successive pages for contiguous and free. 1581 */ 1582 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 1583 int pqtype; 1584 pqtype = pga[i].queue - pga[i].pc; 1585 if ((VM_PAGE_TO_PHYS(&pga[i]) != 1586 (VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) || 1587 ((pqtype != PQ_ZERO) && (pqtype != PQ_FREE) && (pqtype != PQ_CACHE))) { 1588 start++; 1589 goto again; 1590 } 1591 } 1592 1593 for (i = start; i < (start + size / PAGE_SIZE); i++) { 1594 int pqtype; 1595 vm_page_t m = &pga[i]; 1596 1597 pqtype = m->queue - m->pc; 1598 if (pqtype == PQ_CACHE) { 1599 m->flags |= PG_BUSY; 1600 vm_page_free(m); 1601 } 1602 1603 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 1604 (*vm_page_queues[m->queue].lcnt)--; 1605 cnt.v_free_count--; 1606 m->valid = VM_PAGE_BITS_ALL; 1607 m->flags = 0; 1608 m->dirty = 0; 1609 m->wire_count = 0; 1610 m->busy = 0; 1611 m->queue = PQ_NONE; 1612 m->object = NULL; 1613 vm_page_wire(m); 1614 } 1615 1616 /* 1617 * We've found a contiguous chunk that meets are requirements. 1618 * Allocate kernel VM, unfree and assign the physical pages to it and 1619 * return kernel VM pointer. 1620 */ 1621 tmp_addr = addr = kmem_alloc_pageable(map, size); 1622 if (addr == 0) { 1623 /* 1624 * XXX We almost never run out of kernel virtual 1625 * space, so we don't make the allocated memory 1626 * above available. 1627 */ 1628 splx(s); 1629 return (NULL); 1630 } 1631 1632 for (i = start; i < (start + size / PAGE_SIZE); i++) { 1633 vm_page_t m = &pga[i]; 1634 vm_page_insert(m, kernel_object, 1635 OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS)); 1636 pmap_kenter(tmp_addr, VM_PAGE_TO_PHYS(m)); 1637 tmp_addr += PAGE_SIZE; 1638 } 1639 1640 splx(s); 1641 return ((void *)addr); 1642 } 1643 return NULL; 1644} 1645 1646void * 1647contigmalloc(size, type, flags, low, high, alignment, boundary) 1648 unsigned long size; /* should be size_t here and for malloc() */ 1649 struct malloc_type *type; 1650 int flags; 1651 unsigned long low; 1652 unsigned long high; 1653 unsigned long alignment; 1654 unsigned long boundary; 1655{ 1656 return contigmalloc1(size, type, flags, low, high, alignment, boundary, 1657 kernel_map); 1658} 1659 1660vm_offset_t 1661vm_page_alloc_contig(size, low, high, alignment) 1662 vm_offset_t size; 1663 vm_offset_t low; 1664 vm_offset_t high; 1665 vm_offset_t alignment; 1666{ 1667 return ((vm_offset_t)contigmalloc1(size, M_DEVBUF, M_NOWAIT, low, high, 1668 alignment, 0ul, kernel_map)); 1669} 1670 1671#include "opt_ddb.h" 1672#ifdef DDB 1673#include <sys/kernel.h> 1674 1675#include <ddb/ddb.h> 1676 1677DB_SHOW_COMMAND(page, vm_page_print_page_info) 1678{ 1679 db_printf("cnt.v_free_count: %d\n", cnt.v_free_count); 1680 db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count); 1681 db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count); 1682 db_printf("cnt.v_active_count: %d\n", cnt.v_active_count); 1683 db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count); 1684 db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved); 1685 db_printf("cnt.v_free_min: %d\n", cnt.v_free_min); 1686 db_printf("cnt.v_free_target: %d\n", cnt.v_free_target); 1687 db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min); 1688 db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target); 1689} 1690 1691DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) 1692{ 1693 int i; 1694 db_printf("PQ_FREE:"); 1695 for(i=0;i<PQ_L2_SIZE;i++) { 1696 db_printf(" %d", *vm_page_queues[PQ_FREE + i].lcnt); 1697 } 1698 db_printf("\n"); 1699 1700 db_printf("PQ_CACHE:"); 1701 for(i=0;i<PQ_L2_SIZE;i++) { 1702 db_printf(" %d", *vm_page_queues[PQ_CACHE + i].lcnt); 1703 } 1704 db_printf("\n"); 1705 1706 db_printf("PQ_ZERO:"); 1707 for(i=0;i<PQ_L2_SIZE;i++) { 1708 db_printf(" %d", *vm_page_queues[PQ_ZERO + i].lcnt); 1709 } 1710 db_printf("\n"); 1711 1712 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n", 1713 *vm_page_queues[PQ_ACTIVE].lcnt, 1714 *vm_page_queues[PQ_INACTIVE].lcnt); 1715} 1716#endif /* DDB */ 1717