vm_page.c revision 32946
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 1.88 1998/01/31 11:56:47 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/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 int vm_page_bucket_generation; /* generation id for buckets */ 98static struct pglist *vm_page_buckets; /* Array of buckets */ 99static int vm_page_bucket_count; /* How big is array? */ 100static int vm_page_hash_mask; /* Mask for hash function */ 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 108int no_queue=0; 109 110struct vpgqueues vm_page_queues[PQ_COUNT] = {0}; 111int 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; 147int 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 } 356 357 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 */ 367static inline int 368vm_page_hash(object, pindex) 369 vm_object_t object; 370 vm_pindex_t pindex; 371{ 372 return ((((unsigned) object) >> 5) + (pindex >> 1)) & 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, and must be splhigh. 382 */ 383 384void 385vm_page_insert(m, object, pindex) 386 register vm_page_t m; 387 register vm_object_t object; 388 register vm_pindex_t pindex; 389{ 390 register struct pglist *bucket; 391 392 if (m->flags & PG_TABLED) 393 panic("vm_page_insert: already inserted"); 394 395 /* 396 * Record the object/offset pair in this page 397 */ 398 399 m->object = object; 400 m->pindex = pindex; 401 402 /* 403 * Insert it into the object_object/offset hash table 404 */ 405 406 bucket = &vm_page_buckets[vm_page_hash(object, pindex)]; 407 TAILQ_INSERT_TAIL(bucket, m, hashq); 408 vm_page_bucket_generation++; 409 410 /* 411 * Now link into the object's list of backed pages. 412 */ 413 414 TAILQ_INSERT_TAIL(&object->memq, m, listq); 415 m->flags |= PG_TABLED; 416 m->object->page_hint = m; 417 m->object->generation++; 418 419 if (m->wire_count) 420 object->wire_count++; 421 422 if ((m->queue - m->pc) == PQ_CACHE) 423 object->cache_count++; 424 425 /* 426 * And show that the object has one more resident page. 427 */ 428 429 object->resident_page_count++; 430} 431 432/* 433 * vm_page_remove: [ internal use only ] 434 * NOTE: used by device pager as well -wfj 435 * 436 * Removes the given mem entry from the object/offset-page 437 * table and the object page list. 438 * 439 * The object and page must be locked, and at splhigh. 440 */ 441 442void 443vm_page_remove(m) 444 register vm_page_t m; 445{ 446 register struct pglist *bucket; 447 448 if (!(m->flags & PG_TABLED)) 449 return; 450 451 if ((m->flags & PG_BUSY) == 0) { 452 panic("vm_page_remove: page not busy"); 453 } 454 455 m->flags &= ~PG_BUSY; 456 if (m->flags & PG_WANTED) { 457 m->flags &= ~PG_WANTED; 458 wakeup(m); 459 } 460 461 if (m->object->page_hint == m) 462 m->object->page_hint = NULL; 463 464 if (m->wire_count) 465 m->object->wire_count--; 466 467 if ((m->queue - m->pc) == PQ_CACHE) 468 m->object->cache_count--; 469 470 /* 471 * Remove from the object_object/offset hash table 472 */ 473 474 bucket = &vm_page_buckets[vm_page_hash(m->object, m->pindex)]; 475 TAILQ_REMOVE(bucket, m, hashq); 476 vm_page_bucket_generation++; 477 478 /* 479 * Now remove from the object's list of backed pages. 480 */ 481 482 TAILQ_REMOVE(&m->object->memq, m, listq); 483 484 /* 485 * And show that the object has one fewer resident page. 486 */ 487 488 m->object->resident_page_count--; 489 m->object->generation++; 490 491 m->flags &= ~PG_TABLED; 492} 493 494/* 495 * vm_page_lookup: 496 * 497 * Returns the page associated with the object/offset 498 * pair specified; if none is found, NULL is returned. 499 * 500 * The object must be locked. No side effects. 501 */ 502 503vm_page_t 504vm_page_lookup(object, pindex) 505 register vm_object_t object; 506 register vm_pindex_t pindex; 507{ 508 register vm_page_t m; 509 register struct pglist *bucket; 510 int curgeneration; 511 int s; 512 513 /* 514 * Search the hash table for this object/offset pair 515 */ 516 517 bucket = &vm_page_buckets[vm_page_hash(object, pindex)]; 518 519restart: 520 curgeneration = vm_page_bucket_generation; 521 for (m = TAILQ_FIRST(bucket); m != NULL; m = TAILQ_NEXT(m,hashq)) { 522 if (curgeneration != vm_page_bucket_generation) 523 goto restart; 524 if ((m->object == object) && (m->pindex == pindex)) { 525 m->object->page_hint = m; 526 return (m); 527 } 528 } 529 return (NULL); 530} 531 532/* 533 * vm_page_rename: 534 * 535 * Move the given memory entry from its 536 * current object to the specified target object/offset. 537 * 538 * The object must be locked. 539 */ 540void 541vm_page_rename(m, new_object, new_pindex) 542 register vm_page_t m; 543 register vm_object_t new_object; 544 vm_pindex_t new_pindex; 545{ 546 int s; 547 548 s = splvm(); 549 vm_page_remove(m); 550 vm_page_insert(m, new_object, new_pindex); 551 splx(s); 552} 553 554/* 555 * vm_page_unqueue without any wakeup 556 */ 557void 558vm_page_unqueue_nowakeup(m) 559 vm_page_t m; 560{ 561 int queue = m->queue; 562 struct vpgqueues *pq; 563 if (queue != PQ_NONE) { 564 pq = &vm_page_queues[queue]; 565 m->queue = PQ_NONE; 566 TAILQ_REMOVE(pq->pl, m, pageq); 567 (*pq->cnt)--; 568 (*pq->lcnt)--; 569 if ((queue - m->pc) == PQ_CACHE) { 570 m->object->cache_count--; 571 } 572 } 573} 574 575/* 576 * vm_page_unqueue must be called at splhigh(); 577 */ 578void 579vm_page_unqueue(m) 580 vm_page_t m; 581{ 582 int queue = m->queue; 583 struct vpgqueues *pq; 584 if (queue != PQ_NONE) { 585 m->queue = PQ_NONE; 586 pq = &vm_page_queues[queue]; 587 TAILQ_REMOVE(pq->pl, m, pageq); 588 (*pq->cnt)--; 589 (*pq->lcnt)--; 590 if ((queue - m->pc) == PQ_CACHE) { 591 if ((cnt.v_cache_count + cnt.v_free_count) < 592 (cnt.v_free_reserved + cnt.v_cache_min)) 593 pagedaemon_wakeup(); 594 m->object->cache_count--; 595 } 596 } 597} 598 599/* 600 * Find a page on the specified queue with color optimization. 601 */ 602vm_page_t 603vm_page_list_find(basequeue, index) 604 int basequeue, index; 605{ 606#if PQ_L2_SIZE > 1 607 608 int i,j; 609 vm_page_t m; 610 int hindex; 611 struct vpgqueues *pq; 612 613 pq = &vm_page_queues[basequeue]; 614 615 m = TAILQ_FIRST(pq[index].pl); 616 if (m) 617 return m; 618 619 for(j = 0; j < PQ_L1_SIZE; j++) { 620 int ij; 621 for(i = (PQ_L2_SIZE / 2) - PQ_L1_SIZE; 622 (ij = i + j) > 0; 623 i -= PQ_L1_SIZE) { 624 625 hindex = index + ij; 626 if (hindex >= PQ_L2_SIZE) 627 hindex -= PQ_L2_SIZE; 628 if (m = TAILQ_FIRST(pq[hindex].pl)) 629 return m; 630 631 hindex = index - ij; 632 if (hindex < 0) 633 hindex += PQ_L2_SIZE; 634 if (m = TAILQ_FIRST(pq[hindex].pl)) 635 return m; 636 } 637 } 638 639 hindex = index + PQ_L2_SIZE / 2; 640 if (hindex >= PQ_L2_SIZE) 641 hindex -= PQ_L2_SIZE; 642 m = TAILQ_FIRST(pq[hindex].pl); 643 if (m) 644 return m; 645 646 return NULL; 647#else 648 return TAILQ_FIRST(vm_page_queues[basequeue].pl); 649#endif 650 651} 652 653/* 654 * Find a page on the specified queue with color optimization. 655 */ 656vm_page_t 657vm_page_select(object, pindex, basequeue) 658 vm_object_t object; 659 vm_pindex_t pindex; 660 int basequeue; 661{ 662 663#if PQ_L2_SIZE > 1 664 int index; 665 index = (pindex + object->pg_color) & PQ_L2_MASK; 666 return vm_page_list_find(basequeue, index); 667 668#else 669 return TAILQ_FIRST(vm_page_queues[basequeue].pl); 670#endif 671 672} 673 674/* 675 * Find a free or zero page, with specified preference. 676 */ 677static vm_page_t 678vm_page_select_free(object, pindex, prefqueue) 679 vm_object_t object; 680 vm_pindex_t pindex; 681 int prefqueue; 682{ 683#if PQ_L2_SIZE > 1 684 int i,j; 685 int index, hindex; 686#endif 687 vm_page_t m, mh; 688 int oqueuediff; 689 struct vpgqueues *pq; 690 691 if (prefqueue == PQ_ZERO) 692 oqueuediff = PQ_FREE - PQ_ZERO; 693 else 694 oqueuediff = PQ_ZERO - PQ_FREE; 695 696 if (mh = object->page_hint) { 697 if (mh->pindex == (pindex - 1)) { 698 if ((mh->flags & PG_FICTITIOUS) == 0) { 699 if ((mh < &vm_page_array[cnt.v_page_count-1]) && 700 (mh >= &vm_page_array[0])) { 701 int queue; 702 m = mh + 1; 703 if (VM_PAGE_TO_PHYS(m) == (VM_PAGE_TO_PHYS(mh) + PAGE_SIZE)) { 704 queue = m->queue - m->pc; 705 if (queue == PQ_FREE || queue == PQ_ZERO) { 706 return m; 707 } 708 } 709 } 710 } 711 } 712 } 713 714 pq = &vm_page_queues[prefqueue]; 715 716#if PQ_L2_SIZE > 1 717 718 index = (pindex + object->pg_color) & PQ_L2_MASK; 719 720 if (m = TAILQ_FIRST(pq[index].pl)) 721 return m; 722 if (m = TAILQ_FIRST(pq[index + oqueuediff].pl)) 723 return m; 724 725 for(j = 0; j < PQ_L1_SIZE; j++) { 726 int ij; 727 for(i = (PQ_L2_SIZE / 2) - PQ_L1_SIZE; 728 (ij = i + j) >= 0; 729 i -= PQ_L1_SIZE) { 730 731 hindex = index + ij; 732 if (hindex >= PQ_L2_SIZE) 733 hindex -= PQ_L2_SIZE; 734 if (m = TAILQ_FIRST(pq[hindex].pl)) 735 return m; 736 if (m = TAILQ_FIRST(pq[hindex + oqueuediff].pl)) 737 return m; 738 739 hindex = index - ij; 740 if (hindex < 0) 741 hindex += PQ_L2_SIZE; 742 if (m = TAILQ_FIRST(pq[hindex].pl)) 743 return m; 744 if (m = TAILQ_FIRST(pq[hindex + oqueuediff].pl)) 745 return m; 746 } 747 } 748 749 hindex = index + PQ_L2_SIZE / 2; 750 if (hindex >= PQ_L2_SIZE) 751 hindex -= PQ_L2_SIZE; 752 if (m = TAILQ_FIRST(pq[hindex].pl)) 753 return m; 754 if (m = TAILQ_FIRST(pq[hindex+oqueuediff].pl)) 755 return m; 756 757#else 758 if (m = TAILQ_FIRST(pq[0].pl)) 759 return m; 760 else 761 return TAILQ_FIRST(pq[oqueuediff].pl); 762#endif 763 764 return NULL; 765} 766 767/* 768 * vm_page_alloc: 769 * 770 * Allocate and return a memory cell associated 771 * with this VM object/offset pair. 772 * 773 * page_req classes: 774 * VM_ALLOC_NORMAL normal process request 775 * VM_ALLOC_SYSTEM system *really* needs a page 776 * VM_ALLOC_INTERRUPT interrupt time request 777 * VM_ALLOC_ZERO zero page 778 * 779 * Object must be locked. 780 */ 781vm_page_t 782vm_page_alloc(object, pindex, page_req) 783 vm_object_t object; 784 vm_pindex_t pindex; 785 int page_req; 786{ 787 register vm_page_t m; 788 struct vpgqueues *pq; 789 vm_object_t oldobject; 790 int queue, qtype; 791 int s; 792 793#ifdef DIAGNOSTIC 794 m = vm_page_lookup(object, pindex); 795 if (m) 796 panic("vm_page_alloc: page already allocated"); 797#endif 798 799 if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) { 800 page_req = VM_ALLOC_SYSTEM; 801 }; 802 803 s = splvm(); 804 805 switch (page_req) { 806 807 case VM_ALLOC_NORMAL: 808 if (cnt.v_free_count >= cnt.v_free_reserved) { 809 m = vm_page_select_free(object, pindex, PQ_FREE); 810#if defined(DIAGNOSTIC) 811 if (m == NULL) 812 panic("vm_page_alloc(NORMAL): missing page on free queue\n"); 813#endif 814 } else { 815 m = vm_page_select(object, pindex, PQ_CACHE); 816 if (m == NULL) { 817 splx(s); 818#if defined(DIAGNOSTIC) 819 if (cnt.v_cache_count > 0) 820 printf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", cnt.v_cache_count); 821#endif 822 vm_pageout_deficit++; 823 pagedaemon_wakeup(); 824 return (NULL); 825 } 826 } 827 break; 828 829 case VM_ALLOC_ZERO: 830 if (cnt.v_free_count >= cnt.v_free_reserved) { 831 m = vm_page_select_free(object, pindex, PQ_ZERO); 832#if defined(DIAGNOSTIC) 833 if (m == NULL) 834 panic("vm_page_alloc(ZERO): missing page on free queue\n"); 835#endif 836 } else { 837 m = vm_page_select(object, pindex, PQ_CACHE); 838 if (m == NULL) { 839 splx(s); 840#if defined(DIAGNOSTIC) 841 if (cnt.v_cache_count > 0) 842 printf("vm_page_alloc(ZERO): missing pages on cache queue: %d\n", cnt.v_cache_count); 843#endif 844 vm_pageout_deficit++; 845 pagedaemon_wakeup(); 846 return (NULL); 847 } 848 } 849 break; 850 851 case VM_ALLOC_SYSTEM: 852 if ((cnt.v_free_count >= cnt.v_free_reserved) || 853 ((cnt.v_cache_count == 0) && 854 (cnt.v_free_count >= cnt.v_interrupt_free_min))) { 855 m = vm_page_select_free(object, pindex, PQ_FREE); 856#if defined(DIAGNOSTIC) 857 if (m == NULL) 858 panic("vm_page_alloc(SYSTEM): missing page on free queue\n"); 859#endif 860 } else { 861 m = vm_page_select(object, pindex, PQ_CACHE); 862 if (m == NULL) { 863 splx(s); 864#if defined(DIAGNOSTIC) 865 if (cnt.v_cache_count > 0) 866 printf("vm_page_alloc(SYSTEM): missing pages on cache queue: %d\n", cnt.v_cache_count); 867#endif 868 vm_pageout_deficit++; 869 pagedaemon_wakeup(); 870 return (NULL); 871 } 872 } 873 break; 874 875 case VM_ALLOC_INTERRUPT: 876 if (cnt.v_free_count > 0) { 877 m = vm_page_select_free(object, pindex, PQ_FREE); 878#if defined(DIAGNOSTIC) 879 if (m == NULL) 880 panic("vm_page_alloc(INTERRUPT): missing page on free queue\n"); 881#endif 882 } else { 883 splx(s); 884 vm_pageout_deficit++; 885 pagedaemon_wakeup(); 886 return (NULL); 887 } 888 break; 889 890 default: 891 panic("vm_page_alloc: invalid allocation class"); 892 } 893 894 queue = m->queue; 895 qtype = queue - m->pc; 896 if (qtype == PQ_ZERO) 897 vm_page_zero_count--; 898 pq = &vm_page_queues[queue]; 899 TAILQ_REMOVE(pq->pl, m, pageq); 900 (*pq->cnt)--; 901 (*pq->lcnt)--; 902 oldobject = NULL; 903 if (qtype == PQ_ZERO) { 904 m->flags = PG_ZERO|PG_BUSY; 905 } else if (qtype == PQ_CACHE) { 906 oldobject = m->object; 907 m->flags |= PG_BUSY; 908 vm_page_remove(m); 909 m->flags = PG_BUSY; 910 } else { 911 m->flags = PG_BUSY; 912 } 913 m->wire_count = 0; 914 m->hold_count = 0; 915 m->act_count = 0; 916 m->busy = 0; 917 m->valid = 0; 918 m->dirty = 0; 919 m->queue = PQ_NONE; 920 921 /* XXX before splx until vm_page_insert is safe */ 922 vm_page_insert(m, object, pindex); 923 924 /* 925 * Don't wakeup too often - wakeup the pageout daemon when 926 * we would be nearly out of memory. 927 */ 928 if (((cnt.v_free_count + cnt.v_cache_count) < 929 (cnt.v_free_reserved + cnt.v_cache_min)) || 930 (cnt.v_free_count < cnt.v_pageout_free_min)) 931 pagedaemon_wakeup(); 932 933 if ((qtype == PQ_CACHE) && 934 ((page_req == VM_ALLOC_NORMAL) || (page_req == VM_ALLOC_ZERO)) && 935 oldobject && (oldobject->type == OBJT_VNODE) && 936 ((oldobject->flags & OBJ_DEAD) == 0)) { 937 struct vnode *vp; 938 vp = (struct vnode *) oldobject->handle; 939 if (vp && VSHOULDFREE(vp)) { 940 if ((vp->v_flag & (VFREE|VTBFREE|VDOOMED)) == 0) { 941 TAILQ_INSERT_TAIL(&vnode_tobefree_list, vp, v_freelist); 942 vp->v_flag |= VTBFREE; 943 } 944 } 945 } 946 splx(s); 947 948 return (m); 949} 950 951void 952vm_wait() 953{ 954 int s; 955 956 s = splvm(); 957 if (curproc == pageproc) { 958 vm_pageout_pages_needed = 1; 959 tsleep(&vm_pageout_pages_needed, PSWP, "vmwait", 0); 960 } else { 961 if (!vm_pages_needed) { 962 vm_pages_needed++; 963 wakeup(&vm_pages_needed); 964 } 965 tsleep(&cnt.v_free_count, PVM, "vmwait", 0); 966 } 967 splx(s); 968} 969 970 971/* 972 * vm_page_activate: 973 * 974 * Put the specified page on the active list (if appropriate). 975 * 976 * The page queues must be locked. 977 */ 978void 979vm_page_activate(m) 980 register vm_page_t m; 981{ 982 int s; 983 vm_page_t np; 984 vm_object_t object; 985 986 s = splvm(); 987 if (m->queue != PQ_ACTIVE) { 988 if ((m->queue - m->pc) == PQ_CACHE) 989 cnt.v_reactivated++; 990 991 vm_page_unqueue(m); 992 993 if (m->wire_count == 0) { 994 m->queue = PQ_ACTIVE; 995 ++(*vm_page_queues[PQ_ACTIVE].lcnt); 996 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 997 if (m->act_count < ACT_INIT) 998 m->act_count = ACT_INIT; 999 cnt.v_active_count++; 1000 } 1001 } else { 1002 if (m->act_count < ACT_INIT) 1003 m->act_count = ACT_INIT; 1004 } 1005 1006 object = m->object; 1007 TAILQ_REMOVE(&object->memq, m, listq); 1008 TAILQ_INSERT_TAIL(&object->memq, m, listq); 1009 object->generation++; 1010 1011 splx(s); 1012} 1013 1014/* 1015 * helper routine for vm_page_free and vm_page_free_zero 1016 */ 1017static int 1018vm_page_freechk_and_unqueue(m) 1019 vm_page_t m; 1020{ 1021 vm_object_t oldobject; 1022 1023 oldobject = m->object; 1024 1025#if !defined(MAX_PERF) 1026 if (m->busy || ((m->queue - m->pc) == PQ_FREE) || 1027 (m->hold_count != 0)) { 1028 printf("vm_page_free: pindex(%ld), busy(%d), PG_BUSY(%d), hold(%d)\n", 1029 m->pindex, m->busy, 1030 (m->flags & PG_BUSY) ? 1 : 0, m->hold_count); 1031 if ((m->queue - m->pc) == PQ_FREE) 1032 panic("vm_page_free: freeing free page"); 1033 else 1034 panic("vm_page_free: freeing busy page"); 1035 } 1036#endif 1037 1038 vm_page_unqueue_nowakeup(m); 1039 vm_page_remove(m); 1040 1041 if ((m->flags & PG_FICTITIOUS) != 0) { 1042 return 0; 1043 } 1044 1045 if (m->wire_count != 0) { 1046 if (m->wire_count > 1) { 1047 panic("vm_page_free: invalid wire count (%d), pindex: 0x%x", 1048 m->wire_count, m->pindex); 1049 } 1050 m->wire_count = 0; 1051 m->object->wire_count--; 1052 cnt.v_wire_count--; 1053 } 1054 1055 if (oldobject && (oldobject->type == OBJT_VNODE) && 1056 ((oldobject->flags & OBJ_DEAD) == 0)) { 1057 struct vnode *vp; 1058 vp = (struct vnode *) oldobject->handle; 1059 if (vp && VSHOULDFREE(vp)) { 1060 if ((vp->v_flag & (VTBFREE|VDOOMED|VFREE)) == 0) { 1061 TAILQ_INSERT_TAIL(&vnode_tobefree_list, vp, v_freelist); 1062 vp->v_flag |= VTBFREE; 1063 } 1064 } 1065 } 1066 1067 return 1; 1068} 1069 1070/* 1071 * helper routine for vm_page_free and vm_page_free_zero 1072 */ 1073static __inline void 1074vm_page_free_wakeup() 1075{ 1076 1077/* 1078 * if pageout daemon needs pages, then tell it that there are 1079 * some free. 1080 */ 1081 if (vm_pageout_pages_needed) { 1082 wakeup(&vm_pageout_pages_needed); 1083 vm_pageout_pages_needed = 0; 1084 } 1085 /* 1086 * wakeup processes that are waiting on memory if we hit a 1087 * high water mark. And wakeup scheduler process if we have 1088 * lots of memory. this process will swapin processes. 1089 */ 1090 if (vm_pages_needed && 1091 ((cnt.v_free_count + cnt.v_cache_count) >= cnt.v_free_min)) { 1092 wakeup(&cnt.v_free_count); 1093 vm_pages_needed = 0; 1094 } 1095} 1096 1097/* 1098 * vm_page_free: 1099 * 1100 * Returns the given page to the free list, 1101 * disassociating it with any VM object. 1102 * 1103 * Object and page must be locked prior to entry. 1104 */ 1105void 1106vm_page_free(m) 1107 register vm_page_t m; 1108{ 1109 int s; 1110 struct vpgqueues *pq; 1111 1112 s = splvm(); 1113 1114 cnt.v_tfree++; 1115 1116 if (!vm_page_freechk_and_unqueue(m)) { 1117 splx(s); 1118 return; 1119 } 1120 1121 m->queue = PQ_FREE + m->pc; 1122 pq = &vm_page_queues[m->queue]; 1123 ++(*pq->lcnt); 1124 ++(*pq->cnt); 1125 /* 1126 * If the pageout process is grabbing the page, it is likely 1127 * that the page is NOT in the cache. It is more likely that 1128 * the page will be partially in the cache if it is being 1129 * explicitly freed. 1130 */ 1131 if (curproc == pageproc) { 1132 TAILQ_INSERT_TAIL(pq->pl, m, pageq); 1133 } else { 1134 TAILQ_INSERT_HEAD(pq->pl, m, pageq); 1135 } 1136 1137 vm_page_free_wakeup(); 1138 splx(s); 1139} 1140 1141void 1142vm_page_free_zero(m) 1143 register vm_page_t m; 1144{ 1145 int s; 1146 struct vpgqueues *pq; 1147 1148 s = splvm(); 1149 1150 cnt.v_tfree++; 1151 1152 if (!vm_page_freechk_and_unqueue(m)) { 1153 splx(s); 1154 return; 1155 } 1156 1157 m->queue = PQ_ZERO + m->pc; 1158 pq = &vm_page_queues[m->queue]; 1159 ++(*pq->lcnt); 1160 ++(*pq->cnt); 1161 1162 TAILQ_INSERT_HEAD(pq->pl, m, pageq); 1163 ++vm_page_zero_count; 1164 vm_page_free_wakeup(); 1165 splx(s); 1166} 1167 1168/* 1169 * vm_page_wire: 1170 * 1171 * Mark this page as wired down by yet 1172 * another map, removing it from paging queues 1173 * as necessary. 1174 * 1175 * The page queues must be locked. 1176 */ 1177void 1178vm_page_wire(m) 1179 register vm_page_t m; 1180{ 1181 int s; 1182 1183 if (m->wire_count == 0) { 1184 s = splvm(); 1185 vm_page_unqueue(m); 1186 splx(s); 1187 cnt.v_wire_count++; 1188 if (m->object) 1189 m->object->wire_count++; 1190 } 1191 (*vm_page_queues[PQ_NONE].lcnt)++; 1192 m->wire_count++; 1193 m->flags |= PG_MAPPED; 1194} 1195 1196/* 1197 * vm_page_unwire: 1198 * 1199 * Release one wiring of this page, potentially 1200 * enabling it to be paged again. 1201 * 1202 * The page queues must be locked. 1203 */ 1204void 1205vm_page_unwire(m) 1206 register vm_page_t m; 1207{ 1208 int s; 1209 1210 s = splvm(); 1211 1212 if (m->wire_count > 0) { 1213 m->wire_count--; 1214 if (m->wire_count == 0) { 1215 if (m->object) 1216 m->object->wire_count--; 1217 cnt.v_wire_count--; 1218 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 1219 m->queue = PQ_ACTIVE; 1220 (*vm_page_queues[PQ_ACTIVE].lcnt)++; 1221 cnt.v_active_count++; 1222 } 1223 } else { 1224 panic("vm_page_unwire: invalid wire count: %d\n", m->wire_count); 1225 } 1226 splx(s); 1227} 1228 1229 1230/* 1231 * vm_page_deactivate: 1232 * 1233 * Returns the given page to the inactive list, 1234 * indicating that no physical maps have access 1235 * to this page. [Used by the physical mapping system.] 1236 * 1237 * The page queues must be locked. 1238 */ 1239void 1240vm_page_deactivate(m) 1241 register vm_page_t m; 1242{ 1243 int s; 1244 1245 /* 1246 * Only move active pages -- ignore locked or already inactive ones. 1247 * 1248 * XXX: sometimes we get pages which aren't wired down or on any queue - 1249 * we need to put them on the inactive queue also, otherwise we lose 1250 * track of them. Paul Mackerras (paulus@cs.anu.edu.au) 9-Jan-93. 1251 */ 1252 if (m->queue == PQ_INACTIVE) 1253 return; 1254 1255 s = splvm(); 1256 if (m->wire_count == 0 && m->hold_count == 0) { 1257 if ((m->queue - m->pc) == PQ_CACHE) 1258 cnt.v_reactivated++; 1259 vm_page_unqueue(m); 1260 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); 1261 m->queue = PQ_INACTIVE; 1262 ++(*vm_page_queues[PQ_INACTIVE].lcnt); 1263 cnt.v_inactive_count++; 1264 } 1265 splx(s); 1266} 1267 1268/* 1269 * vm_page_cache 1270 * 1271 * Put the specified page onto the page cache queue (if appropriate). 1272 */ 1273void 1274vm_page_cache(m) 1275 register vm_page_t m; 1276{ 1277 int s; 1278 1279 if ((m->flags & PG_BUSY) || m->busy || m->wire_count) { 1280 printf("vm_page_cache: attempting to cache busy page\n"); 1281 return; 1282 } 1283 if ((m->queue - m->pc) == PQ_CACHE) 1284 return; 1285 1286 vm_page_protect(m, VM_PROT_NONE); 1287 if (m->dirty != 0) { 1288 panic("vm_page_cache: caching a dirty page, pindex: %d", m->pindex); 1289 } 1290 s = splvm(); 1291 vm_page_unqueue_nowakeup(m); 1292 m->queue = PQ_CACHE + m->pc; 1293 (*vm_page_queues[m->queue].lcnt)++; 1294 TAILQ_INSERT_TAIL(vm_page_queues[m->queue].pl, m, pageq); 1295 cnt.v_cache_count++; 1296 m->object->cache_count++; 1297 vm_page_free_wakeup(); 1298 splx(s); 1299} 1300 1301 1302/* 1303 * mapping function for valid bits or for dirty bits in 1304 * a page 1305 */ 1306inline int 1307vm_page_bits(int base, int size) 1308{ 1309 u_short chunk; 1310 1311 if ((base == 0) && (size >= PAGE_SIZE)) 1312 return VM_PAGE_BITS_ALL; 1313 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 1314 base = (base % PAGE_SIZE) / DEV_BSIZE; 1315 chunk = vm_page_dev_bsize_chunks[size / DEV_BSIZE]; 1316 return (chunk << base) & VM_PAGE_BITS_ALL; 1317} 1318 1319/* 1320 * set a page valid and clean 1321 */ 1322void 1323vm_page_set_validclean(m, base, size) 1324 vm_page_t m; 1325 int base; 1326 int size; 1327{ 1328 int pagebits = vm_page_bits(base, size); 1329 m->valid |= pagebits; 1330 m->dirty &= ~pagebits; 1331 if( base == 0 && size == PAGE_SIZE) 1332 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1333} 1334 1335/* 1336 * set a page (partially) invalid 1337 */ 1338void 1339vm_page_set_invalid(m, base, size) 1340 vm_page_t m; 1341 int base; 1342 int size; 1343{ 1344 int bits; 1345 1346 m->valid &= ~(bits = vm_page_bits(base, size)); 1347 if (m->valid == 0) 1348 m->dirty &= ~bits; 1349} 1350 1351/* 1352 * is (partial) page valid? 1353 */ 1354int 1355vm_page_is_valid(m, base, size) 1356 vm_page_t m; 1357 int base; 1358 int size; 1359{ 1360 int bits = vm_page_bits(base, size); 1361 1362 if (m->valid && ((m->valid & bits) == bits)) 1363 return 1; 1364 else 1365 return 0; 1366} 1367 1368void 1369vm_page_test_dirty(m) 1370 vm_page_t m; 1371{ 1372 if ((m->dirty != VM_PAGE_BITS_ALL) && 1373 pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1374 m->dirty = VM_PAGE_BITS_ALL; 1375 } 1376} 1377 1378/* 1379 * This interface is for merging with malloc() someday. 1380 * Even if we never implement compaction so that contiguous allocation 1381 * works after initialization time, malloc()'s data structures are good 1382 * for statistics and for allocations of less than a page. 1383 */ 1384void * 1385contigmalloc1(size, type, flags, low, high, alignment, boundary, map) 1386 unsigned long size; /* should be size_t here and for malloc() */ 1387 struct malloc_type *type; 1388 int flags; 1389 unsigned long low; 1390 unsigned long high; 1391 unsigned long alignment; 1392 unsigned long boundary; 1393 vm_map_t map; 1394{ 1395 int i, s, start; 1396 vm_offset_t addr, phys, tmp_addr; 1397 int pass; 1398 vm_page_t pga = vm_page_array; 1399 1400 size = round_page(size); 1401 if (size == 0) 1402 panic("contigmalloc1: size must not be 0"); 1403 if ((alignment & (alignment - 1)) != 0) 1404 panic("contigmalloc1: alignment must be a power of 2"); 1405 if ((boundary & (boundary - 1)) != 0) 1406 panic("contigmalloc1: boundary must be a power of 2"); 1407 1408 start = 0; 1409 for (pass = 0; pass <= 1; pass++) { 1410 s = splvm(); 1411again: 1412 /* 1413 * Find first page in array that is free, within range, aligned, and 1414 * such that the boundary won't be crossed. 1415 */ 1416 for (i = start; i < cnt.v_page_count; i++) { 1417 int pqtype; 1418 phys = VM_PAGE_TO_PHYS(&pga[i]); 1419 pqtype = pga[i].queue - pga[i].pc; 1420 if (((pqtype == PQ_ZERO) || (pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) && 1421 (phys >= low) && (phys < high) && 1422 ((phys & (alignment - 1)) == 0) && 1423 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0)) 1424 break; 1425 } 1426 1427 /* 1428 * If the above failed or we will exceed the upper bound, fail. 1429 */ 1430 if ((i == cnt.v_page_count) || 1431 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 1432 vm_page_t m, next; 1433 1434again1: 1435 for (m = TAILQ_FIRST(&vm_page_queue_inactive); 1436 m != NULL; 1437 m = next) { 1438 1439 if (m->queue != PQ_INACTIVE) { 1440 break; 1441 } 1442 1443 next = TAILQ_NEXT(m, pageq); 1444 if (m->flags & PG_BUSY) { 1445 m->flags |= PG_WANTED; 1446 tsleep(m, PVM, "vpctw0", 0); 1447 goto again1; 1448 } 1449 vm_page_test_dirty(m); 1450 if (m->dirty) { 1451 if (m->object->type == OBJT_VNODE) { 1452 vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); 1453 vm_object_page_clean(m->object, 0, 0, TRUE); 1454 VOP_UNLOCK(m->object->handle, 0, curproc); 1455 goto again1; 1456 } else if (m->object->type == OBJT_SWAP || 1457 m->object->type == OBJT_DEFAULT) { 1458 vm_page_protect(m, VM_PROT_NONE); 1459 vm_pageout_flush(&m, 1, 0); 1460 goto again1; 1461 } 1462 } 1463 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 1464 vm_page_cache(m); 1465 } 1466 1467 for (m = TAILQ_FIRST(&vm_page_queue_active); 1468 m != NULL; 1469 m = next) { 1470 1471 if (m->queue != PQ_ACTIVE) { 1472 break; 1473 } 1474 1475 next = TAILQ_NEXT(m, pageq); 1476 if (m->flags & PG_BUSY) { 1477 m->flags |= PG_WANTED; 1478 tsleep(m, PVM, "vpctw1", 0); 1479 goto again1; 1480 } 1481 vm_page_test_dirty(m); 1482 if (m->dirty) { 1483 if (m->object->type == OBJT_VNODE) { 1484 vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); 1485 vm_object_page_clean(m->object, 0, 0, TRUE); 1486 VOP_UNLOCK(m->object->handle, 0, curproc); 1487 goto again1; 1488 } else if (m->object->type == OBJT_SWAP || 1489 m->object->type == OBJT_DEFAULT) { 1490 vm_page_protect(m, VM_PROT_NONE); 1491 vm_pageout_flush(&m, 1, 0); 1492 goto again1; 1493 } 1494 } 1495 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 1496 vm_page_cache(m); 1497 } 1498 1499 splx(s); 1500 continue; 1501 } 1502 start = i; 1503 1504 /* 1505 * Check successive pages for contiguous and free. 1506 */ 1507 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 1508 int pqtype; 1509 pqtype = pga[i].queue - pga[i].pc; 1510 if ((VM_PAGE_TO_PHYS(&pga[i]) != 1511 (VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) || 1512 ((pqtype != PQ_ZERO) && (pqtype != PQ_FREE) && (pqtype != PQ_CACHE))) { 1513 start++; 1514 goto again; 1515 } 1516 } 1517 1518 for (i = start; i < (start + size / PAGE_SIZE); i++) { 1519 int pqtype; 1520 vm_page_t m = &pga[i]; 1521 1522 pqtype = m->queue - m->pc; 1523 if (pqtype == PQ_CACHE) { 1524 m->flags |= PG_BUSY; 1525 vm_page_free(m); 1526 } 1527 1528 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 1529 (*vm_page_queues[m->queue].lcnt)--; 1530 cnt.v_free_count--; 1531 m->valid = VM_PAGE_BITS_ALL; 1532 m->flags = 0; 1533 m->dirty = 0; 1534 m->wire_count = 0; 1535 m->busy = 0; 1536 m->queue = PQ_NONE; 1537 m->object = NULL; 1538 vm_page_wire(m); 1539 } 1540 1541 /* 1542 * We've found a contiguous chunk that meets are requirements. 1543 * Allocate kernel VM, unfree and assign the physical pages to it and 1544 * return kernel VM pointer. 1545 */ 1546 tmp_addr = addr = kmem_alloc_pageable(map, size); 1547 if (addr == 0) { 1548 /* 1549 * XXX We almost never run out of kernel virtual 1550 * space, so we don't make the allocated memory 1551 * above available. 1552 */ 1553 splx(s); 1554 return (NULL); 1555 } 1556 1557 for (i = start; i < (start + size / PAGE_SIZE); i++) { 1558 vm_page_t m = &pga[i]; 1559 vm_page_insert(m, kernel_object, 1560 OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS)); 1561 pmap_kenter(tmp_addr, VM_PAGE_TO_PHYS(m)); 1562 tmp_addr += PAGE_SIZE; 1563 } 1564 1565 splx(s); 1566 return ((void *)addr); 1567 } 1568 return NULL; 1569} 1570 1571void * 1572contigmalloc(size, type, flags, low, high, alignment, boundary) 1573 unsigned long size; /* should be size_t here and for malloc() */ 1574 struct malloc_type *type; 1575 int flags; 1576 unsigned long low; 1577 unsigned long high; 1578 unsigned long alignment; 1579 unsigned long boundary; 1580{ 1581 return contigmalloc1(size, type, flags, low, high, alignment, boundary, 1582 kernel_map); 1583} 1584 1585vm_offset_t 1586vm_page_alloc_contig(size, low, high, alignment) 1587 vm_offset_t size; 1588 vm_offset_t low; 1589 vm_offset_t high; 1590 vm_offset_t alignment; 1591{ 1592 return ((vm_offset_t)contigmalloc1(size, M_DEVBUF, M_NOWAIT, low, high, 1593 alignment, 0ul, kernel_map)); 1594} 1595 1596#include "opt_ddb.h" 1597#ifdef DDB 1598#include <sys/kernel.h> 1599 1600#include <ddb/ddb.h> 1601 1602DB_SHOW_COMMAND(page, vm_page_print_page_info) 1603{ 1604 db_printf("cnt.v_free_count: %d\n", cnt.v_free_count); 1605 db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count); 1606 db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count); 1607 db_printf("cnt.v_active_count: %d\n", cnt.v_active_count); 1608 db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count); 1609 db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved); 1610 db_printf("cnt.v_free_min: %d\n", cnt.v_free_min); 1611 db_printf("cnt.v_free_target: %d\n", cnt.v_free_target); 1612 db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min); 1613 db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target); 1614} 1615 1616DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) 1617{ 1618 int i; 1619 db_printf("PQ_FREE:"); 1620 for(i=0;i<PQ_L2_SIZE;i++) { 1621 db_printf(" %d", *vm_page_queues[PQ_FREE + i].lcnt); 1622 } 1623 db_printf("\n"); 1624 1625 db_printf("PQ_CACHE:"); 1626 for(i=0;i<PQ_L2_SIZE;i++) { 1627 db_printf(" %d", *vm_page_queues[PQ_CACHE + i].lcnt); 1628 } 1629 db_printf("\n"); 1630 1631 db_printf("PQ_ZERO:"); 1632 for(i=0;i<PQ_L2_SIZE;i++) { 1633 db_printf(" %d", *vm_page_queues[PQ_ZERO + i].lcnt); 1634 } 1635 db_printf("\n"); 1636 1637 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n", 1638 *vm_page_queues[PQ_ACTIVE].lcnt, 1639 *vm_page_queues[PQ_INACTIVE].lcnt); 1640} 1641#endif /* DDB */ 1642