Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_page.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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 * $FreeBSD$ 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 88 static void vm_page_queue_init __P((void)); 89 static vm_page_t vm_page_select_free __P((vm_object_t object, 90 vm_pindex_t pindex, int prefqueue)); 91 static vm_page_t vm_page_select_cache __P((vm_object_t, vm_pindex_t)); 92 93 /* 94 * Associated with page of user-allocatable memory is a 95 * page structure. 96 */ 97 98 static struct pglist *vm_page_buckets; /* Array of buckets */ 99 static int vm_page_bucket_count; /* How big is array? */ 100 static int vm_page_hash_mask; /* Mask for hash function */ 101 static volatile int vm_page_bucket_generation; 102 103 struct pglist vm_page_queue_free[PQ_L2_SIZE] = {0}; 104 struct pglist vm_page_queue_zero[PQ_L2_SIZE] = {0}; 105 struct pglist vm_page_queue_active = {0}; 106 struct pglist vm_page_queue_inactive = {0}; 107 struct pglist vm_page_queue_cache[PQ_L2_SIZE] = {0}; 108 109 static int no_queue=0; 110 111 struct vpgqueues vm_page_queues[PQ_COUNT] = {0}; 112 static int pqcnt[PQ_COUNT] = {0}; 113 114 static void 115 vm_page_queue_init(void) { 116 int i; 117 118 vm_page_queues[PQ_NONE].pl = NULL; 119 vm_page_queues[PQ_NONE].cnt = &no_queue; 120 for(i=0;i<PQ_L2_SIZE;i++) { 121 vm_page_queues[PQ_FREE+i].pl = &vm_page_queue_free[i]; 122 vm_page_queues[PQ_FREE+i].cnt = &cnt.v_free_count; 123 } 124 for(i=0;i<PQ_L2_SIZE;i++) { 125 vm_page_queues[PQ_ZERO+i].pl = &vm_page_queue_zero[i]; 126 vm_page_queues[PQ_ZERO+i].cnt = &cnt.v_free_count; 127 } 128 vm_page_queues[PQ_INACTIVE].pl = &vm_page_queue_inactive; 129 vm_page_queues[PQ_INACTIVE].cnt = &cnt.v_inactive_count; 130 131 vm_page_queues[PQ_ACTIVE].pl = &vm_page_queue_active; 132 vm_page_queues[PQ_ACTIVE].cnt = &cnt.v_active_count; 133 for(i=0;i<PQ_L2_SIZE;i++) { 134 vm_page_queues[PQ_CACHE+i].pl = &vm_page_queue_cache[i]; 135 vm_page_queues[PQ_CACHE+i].cnt = &cnt.v_cache_count; 136 } 137 for(i=0;i<PQ_COUNT;i++) { 138 if (vm_page_queues[i].pl) { 139 TAILQ_INIT(vm_page_queues[i].pl); 140 } else if (i != 0) { 141 panic("vm_page_queue_init: queue %d is null", i); 142 } 143 vm_page_queues[i].lcnt = &pqcnt[i]; 144 } 145 } 146 147 vm_page_t vm_page_array = 0; 148 static int vm_page_array_size = 0; 149 long first_page = 0; 150 static long last_page; 151 static vm_size_t page_mask; 152 static int page_shift; 153 int vm_page_zero_count = 0; 154 155 /* 156 * map of contiguous valid DEV_BSIZE chunks in a page 157 * (this list is valid for page sizes upto 16*DEV_BSIZE) 158 */ 159 static u_short vm_page_dev_bsize_chunks[] = { 160 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff, 161 0x1ff, 0x3ff, 0x7ff, 0xfff, 0x1fff, 0x3fff, 0x7fff, 0xffff 162 }; 163 164 static __inline int vm_page_hash __P((vm_object_t object, vm_pindex_t pindex)); 165 static int vm_page_freechk_and_unqueue __P((vm_page_t m)); 166 static void vm_page_free_wakeup __P((void)); 167 168 /* 169 * vm_set_page_size: 170 * 171 * Sets the page size, perhaps based upon the memory 172 * size. Must be called before any use of page-size 173 * dependent functions. 174 * 175 * Sets page_shift and page_mask from cnt.v_page_size. 176 */ 177 void 178 vm_set_page_size() 179 { 180 181 if (cnt.v_page_size == 0) 182 cnt.v_page_size = PAGE_SIZE; 183 page_mask = cnt.v_page_size - 1; 184 if ((page_mask & cnt.v_page_size) != 0) 185 panic("vm_set_page_size: page size not a power of two"); 186 for (page_shift = 0;; page_shift++) 187 if ((1 << page_shift) == cnt.v_page_size) 188 break; 189 } 190 191 /* 192 * vm_page_startup: 193 * 194 * Initializes the resident memory module. 195 * 196 * Allocates memory for the page cells, and 197 * for the object/offset-to-page hash table headers. 198 * Each page cell is initialized and placed on the free list. 199 */ 200 201 vm_offset_t 202 vm_page_startup(starta, enda, vaddr) 203 register vm_offset_t starta; 204 vm_offset_t enda; 205 register vm_offset_t vaddr; 206 { 207 register vm_offset_t mapped; 208 register vm_page_t m; 209 register struct pglist *bucket; 210 vm_size_t npages, page_range; 211 register vm_offset_t new_start; 212 int i; 213 vm_offset_t pa; 214 int nblocks; 215 vm_offset_t first_managed_page; 216 217 /* the biggest memory array is the second group of pages */ 218 vm_offset_t start; 219 vm_offset_t biggestone, biggestsize; 220 221 vm_offset_t total; 222 223 total = 0; 224 biggestsize = 0; 225 biggestone = 0; 226 nblocks = 0; 227 vaddr = round_page(vaddr); 228 229 for (i = 0; phys_avail[i + 1]; i += 2) { 230 phys_avail[i] = round_page(phys_avail[i]); 231 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); 232 } 233 234 for (i = 0; phys_avail[i + 1]; i += 2) { 235 int size = phys_avail[i + 1] - phys_avail[i]; 236 237 if (size > biggestsize) { 238 biggestone = i; 239 biggestsize = size; 240 } 241 ++nblocks; 242 total += size; 243 } 244 245 start = phys_avail[biggestone]; 246 247 /* 248 * Initialize the queue headers for the free queue, the active queue 249 * and the inactive queue. 250 */ 251 252 vm_page_queue_init(); 253 254 /* 255 * Allocate (and initialize) the hash table buckets. 256 * 257 * The number of buckets MUST BE a power of 2, and the actual value is 258 * the next power of 2 greater than the number of physical pages in 259 * the system. 260 * 261 * Note: This computation can be tweaked if desired. 262 */ 263 vm_page_buckets = (struct pglist *) vaddr; 264 bucket = vm_page_buckets; 265 if (vm_page_bucket_count == 0) { 266 vm_page_bucket_count = 1; 267 while (vm_page_bucket_count < atop(total)) 268 vm_page_bucket_count <<= 1; 269 } 270 vm_page_hash_mask = vm_page_bucket_count - 1; 271 272 /* 273 * Validate these addresses. 274 */ 275 276 new_start = start + vm_page_bucket_count * sizeof(struct pglist); 277 new_start = round_page(new_start); 278 mapped = round_page(vaddr); 279 vaddr = pmap_map(mapped, start, new_start, 280 VM_PROT_READ | VM_PROT_WRITE); 281 start = new_start; 282 vaddr = round_page(vaddr); 283 bzero((caddr_t) mapped, vaddr - mapped); 284 285 for (i = 0; i < vm_page_bucket_count; i++) { 286 TAILQ_INIT(bucket); 287 bucket++; 288 } 289 290 /* 291 * Compute the number of pages of memory that will be available for 292 * use (taking into account the overhead of a page structure per 293 * page). 294 */ 295 296 first_page = phys_avail[0] / PAGE_SIZE; 297 last_page = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE; 298 299 page_range = last_page - (phys_avail[0] / PAGE_SIZE); 300 npages = (total - (page_range * sizeof(struct vm_page)) - 301 (start - phys_avail[biggestone])) / PAGE_SIZE; 302 303 /* 304 * Initialize the mem entry structures now, and put them in the free 305 * queue. 306 */ 307 vm_page_array = (vm_page_t) vaddr; 308 mapped = vaddr; 309 310 /* 311 * Validate these addresses. 312 */ 313 new_start = round_page(start + page_range * sizeof(struct vm_page)); 314 mapped = pmap_map(mapped, start, new_start, 315 VM_PROT_READ | VM_PROT_WRITE); 316 start = new_start; 317 318 first_managed_page = start / PAGE_SIZE; 319 320 /* 321 * Clear all of the page structures 322 */ 323 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); 324 vm_page_array_size = page_range; 325 326 /* 327 * Construct the free queue(s) in descending order (by physical 328 * address) so that the first 16MB of physical memory is allocated 329 * last rather than first. On large-memory machines, this avoids 330 * the exhaustion of low physical memory before isa_dmainit has run. 331 */ 332 cnt.v_page_count = 0; 333 cnt.v_free_count = 0; 334 for (i = 0; phys_avail[i + 1] && npages > 0; i += 2) { 335 if (i == biggestone) 336 pa = ptoa(first_managed_page); 337 else 338 pa = phys_avail[i]; 339 while (pa < phys_avail[i + 1] && npages-- > 0) { 340 ++cnt.v_page_count; 341 ++cnt.v_free_count; 342 m = PHYS_TO_VM_PAGE(pa); 343 m->phys_addr = pa; 344 m->flags = 0; 345 m->pc = (pa >> PAGE_SHIFT) & PQ_L2_MASK; 346 m->queue = m->pc + PQ_FREE; 347 TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq); 348 ++(*vm_page_queues[m->queue].lcnt); 349 pa += PAGE_SIZE; 350 } 351 } 352 return (mapped); 353 } 354 355 /* 356 * vm_page_hash: 357 * 358 * Distributes the object/offset key pair among hash buckets. 359 * 360 * NOTE: This macro depends on vm_page_bucket_count being a power of 2. 361 * This routine may not block. 362 */ 363 static __inline int 364 vm_page_hash(object, pindex) 365 vm_object_t object; 366 vm_pindex_t pindex; 367 { 368 return ((((uintptr_t) object) >> 5) + (pindex >> 1)) & vm_page_hash_mask; 369 } 370 371 /* 372 * vm_page_insert: [ internal use only ] 373 * 374 * Inserts the given mem entry into the object and object list. 375 * 376 * The pagetables are not updated but will presumably fault the page 377 * in if necessary, or if a kernel page the caller will at some point 378 * enter the page into the kernel's pmap. We are not allowed to block 379 * here so we *can't* do this anyway. 380 * 381 * The object and page must be locked, and must be splhigh. 382 * This routine may not block. 383 */ 384 385 void 386 vm_page_insert(m, object, pindex) 387 register vm_page_t m; 388 register vm_object_t object; 389 register vm_pindex_t pindex; 390 { 391 register struct pglist *bucket; 392 393 if (m->object != NULL) 394 panic("vm_page_insert: already inserted"); 395 396 /* 397 * Record the object/offset pair in this page 398 */ 399 400 m->object = object; 401 m->pindex = pindex; 402 403 /* 404 * Insert it into the object_object/offset hash table 405 */ 406 407 bucket = &vm_page_buckets[vm_page_hash(object, pindex)]; 408 TAILQ_INSERT_TAIL(bucket, m, hashq); 409 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->object->page_hint = m; 417 m->object->generation++; 418 419 /* 420 * And show that the object has one more resident page. 421 */ 422 423 object->resident_page_count++; 424 } 425 426 /* 427 * vm_page_remove: [ internal use only ] 428 * NOTE: used by device pager as well -wfj 429 * 430 * Removes the given mem entry from the object/offset-page 431 * table and the object page list. 432 * 433 * The object and page must be locked, and at splhigh. 434 * This routine may not block. 435 * 436 * I do not think the underlying pmap entry (if any) is removed here. 437 */ 438 439 void 440 vm_page_remove(m) 441 register vm_page_t m; 442 { 443 register struct pglist *bucket; 444 vm_object_t object; 445 446 if (m->object == NULL) 447 return; 448 449 #if !defined(MAX_PERF) 450 if ((m->flags & PG_BUSY) == 0) { 451 panic("vm_page_remove: page not busy"); 452 } 453 #endif 454 455 vm_page_flag_clear(m, PG_BUSY); 456 if (m->flags & PG_WANTED) { 457 vm_page_flag_clear(m, PG_WANTED); 458 wakeup(m); 459 } 460 461 object = m->object; 462 if (object->page_hint == m) 463 object->page_hint = NULL; 464 465 /* 466 * Remove from the object_object/offset hash table 467 */ 468 469 bucket = &vm_page_buckets[vm_page_hash(m->object, m->pindex)]; 470 TAILQ_REMOVE(bucket, m, hashq); 471 vm_page_bucket_generation++; 472 473 /* 474 * Now remove from the object's list of backed pages. 475 */ 476 477 TAILQ_REMOVE(&object->memq, m, listq); 478 479 /* 480 * And show that the object has one fewer resident page. 481 */ 482 483 object->resident_page_count--; 484 object->generation++; 485 486 m->object = NULL; 487 } 488 489 /* 490 * vm_page_lookup: 491 * 492 * Returns the page associated with the object/offset 493 * pair specified; if none is found, NULL is returned. 494 * 495 * The object must be locked. No side effects. 496 * This routine may not block. 497 */ 498 499 vm_page_t 500 vm_page_lookup(object, pindex) 501 register vm_object_t object; 502 register vm_pindex_t pindex; 503 { 504 register vm_page_t m; 505 register struct pglist *bucket; 506 int generation; 507 508 /* 509 * Search the hash table for this object/offset pair 510 */ 511 512 if (object->page_hint && (object->page_hint->pindex == pindex) && 513 (object->page_hint->object == object)) 514 return object->page_hint; 515 516 retry: 517 generation = vm_page_bucket_generation; 518 bucket = &vm_page_buckets[vm_page_hash(object, pindex)]; 519 for (m = TAILQ_FIRST(bucket); m != NULL; m = TAILQ_NEXT(m,hashq)) { 520 if ((m->object == object) && (m->pindex == pindex)) { 521 if (vm_page_bucket_generation != generation) 522 goto retry; 523 m->object->page_hint = m; 524 return (m); 525 } 526 } 527 if (vm_page_bucket_generation != generation) 528 goto retry; 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 * This routine may not block. 540 * 541 * Note: this routine will raise itself to splvm(), the caller need not. 542 */ 543 544 void 545 vm_page_rename(m, new_object, new_pindex) 546 register vm_page_t m; 547 register vm_object_t new_object; 548 vm_pindex_t new_pindex; 549 { 550 int s; 551 552 s = splvm(); 553 vm_page_remove(m); 554 vm_page_insert(m, new_object, new_pindex); 555 splx(s); 556 } 557 558 /* 559 * vm_page_unqueue_nowakeup: 560 * 561 * vm_page_unqueue() without any wakeup 562 * 563 * This routine must be called at splhigh(). 564 * This routine may not block. 565 */ 566 567 void 568 vm_page_unqueue_nowakeup(m) 569 vm_page_t m; 570 { 571 int queue = m->queue; 572 struct vpgqueues *pq; 573 if (queue != PQ_NONE) { 574 pq = &vm_page_queues[queue]; 575 m->queue = PQ_NONE; 576 TAILQ_REMOVE(pq->pl, m, pageq); 577 (*pq->cnt)--; 578 (*pq->lcnt)--; 579 } 580 } 581 582 /* 583 * vm_page_unqueue: 584 * 585 * Remove a page from its queue. 586 * 587 * This routine must be called at splhigh(). 588 * This routine may not block. 589 */ 590 591 void 592 vm_page_unqueue(m) 593 vm_page_t m; 594 { 595 int queue = m->queue; 596 struct vpgqueues *pq; 597 if (queue != PQ_NONE) { 598 m->queue = PQ_NONE; 599 pq = &vm_page_queues[queue]; 600 TAILQ_REMOVE(pq->pl, m, pageq); 601 (*pq->cnt)--; 602 (*pq->lcnt)--; 603 if ((queue - m->pc) == PQ_CACHE) { 604 if ((cnt.v_cache_count + cnt.v_free_count) < 605 (cnt.v_free_reserved + cnt.v_cache_min)) 606 pagedaemon_wakeup(); 607 } 608 } 609 } 610 611 /* 612 * vm_page_list_find: 613 * 614 * Find a page on the specified queue with color optimization. 615 * 616 * This routine must be called at splvm(). 617 * This routine may not block. 618 */ 619 vm_page_t 620 vm_page_list_find(basequeue, index) 621 int basequeue, index; 622 { 623 #if PQ_L2_SIZE > 1 624 625 int i,j; 626 vm_page_t m; 627 int hindex; 628 struct vpgqueues *pq; 629 630 pq = &vm_page_queues[basequeue]; 631 632 m = TAILQ_FIRST(pq[index].pl); 633 if (m) 634 return m; 635 636 for(j = 0; j < PQ_L1_SIZE; j++) { 637 int ij; 638 for(i = (PQ_L2_SIZE / 2) - PQ_L1_SIZE; 639 (ij = i + j) > 0; 640 i -= PQ_L1_SIZE) { 641 642 hindex = index + ij; 643 if (hindex >= PQ_L2_SIZE) 644 hindex -= PQ_L2_SIZE; 645 if (m = TAILQ_FIRST(pq[hindex].pl)) 646 return m; 647 648 hindex = index - ij; 649 if (hindex < 0) 650 hindex += PQ_L2_SIZE; 651 if (m = TAILQ_FIRST(pq[hindex].pl)) 652 return m; 653 } 654 } 655 656 hindex = index + PQ_L2_SIZE / 2; 657 if (hindex >= PQ_L2_SIZE) 658 hindex -= PQ_L2_SIZE; 659 m = TAILQ_FIRST(pq[hindex].pl); 660 if (m) 661 return m; 662 663 return NULL; 664 #else 665 return TAILQ_FIRST(vm_page_queues[basequeue].pl); 666 #endif 667 668 } 669 670 /* 671 * vm_page_select: 672 * 673 * Find a page on the specified queue with color optimization. 674 * 675 * This routine must be called at splvm(). 676 * This routine may not block. 677 */ 678 vm_page_t 679 vm_page_select(object, pindex, basequeue) 680 vm_object_t object; 681 vm_pindex_t pindex; 682 int basequeue; 683 { 684 685 #if PQ_L2_SIZE > 1 686 int index; 687 index = (pindex + object->pg_color) & PQ_L2_MASK; 688 return vm_page_list_find(basequeue, index); 689 690 #else 691 return TAILQ_FIRST(vm_page_queues[basequeue].pl); 692 #endif 693 694 } 695 696 /* 697 * vm_page_select_cache: 698 * 699 * Find a page on the cache queue with color optimization. As pages 700 * might be found, but not applicable, they are deactivated. This 701 * keeps us from using potentially busy cached pages. 702 * 703 * This routine must be called at splvm(). 704 * This routine may not block. 705 */ 706 vm_page_t 707 vm_page_select_cache(object, pindex) 708 vm_object_t object; 709 vm_pindex_t pindex; 710 { 711 vm_page_t m; 712 713 while (TRUE) { 714 #if PQ_L2_SIZE > 1 715 int index; 716 index = (pindex + object->pg_color) & PQ_L2_MASK; 717 m = vm_page_list_find(PQ_CACHE, index); 718 719 #else 720 m = TAILQ_FIRST(vm_page_queues[PQ_CACHE].pl); 721 #endif 722 if (m && ((m->flags & PG_BUSY) || m->busy || 723 m->hold_count || m->wire_count)) { 724 vm_page_deactivate(m); 725 continue; 726 } 727 return m; 728 } 729 } 730 731 /* 732 * vm_page_select_free: 733 * 734 * Find a free or zero page, with specified preference. 735 * 736 * This routine must be called at splvm(). 737 * This routine may not block. 738 */ 739 740 static vm_page_t 741 vm_page_select_free(object, pindex, prefqueue) 742 vm_object_t object; 743 vm_pindex_t pindex; 744 int prefqueue; 745 { 746 #if PQ_L2_SIZE > 1 747 int i,j; 748 int index, hindex; 749 #endif 750 vm_page_t m, mh; 751 int oqueuediff; 752 struct vpgqueues *pq; 753 754 if (prefqueue == PQ_ZERO) 755 oqueuediff = PQ_FREE - PQ_ZERO; 756 else 757 oqueuediff = PQ_ZERO - PQ_FREE; 758 759 if (mh = object->page_hint) { 760 if (mh->pindex == (pindex - 1)) { 761 if ((mh->flags & PG_FICTITIOUS) == 0) { 762 if ((mh < &vm_page_array[cnt.v_page_count-1]) && 763 (mh >= &vm_page_array[0])) { 764 int queue; 765 m = mh + 1; 766 if (VM_PAGE_TO_PHYS(m) == (VM_PAGE_TO_PHYS(mh) + PAGE_SIZE)) { 767 queue = m->queue - m->pc; 768 if (queue == PQ_FREE || queue == PQ_ZERO) { 769 return m; 770 } 771 } 772 } 773 } 774 } 775 } 776 777 pq = &vm_page_queues[prefqueue]; 778 779 #if PQ_L2_SIZE > 1 780 781 index = (pindex + object->pg_color) & PQ_L2_MASK; 782 783 if (m = TAILQ_FIRST(pq[index].pl)) 784 return m; 785 if (m = TAILQ_FIRST(pq[index + oqueuediff].pl)) 786 return m; 787 788 for(j = 0; j < PQ_L1_SIZE; j++) { 789 int ij; 790 for(i = (PQ_L2_SIZE / 2) - PQ_L1_SIZE; 791 (ij = i + j) >= 0; 792 i -= PQ_L1_SIZE) { 793 794 hindex = index + ij; 795 if (hindex >= PQ_L2_SIZE) 796 hindex -= PQ_L2_SIZE; 797 if (m = TAILQ_FIRST(pq[hindex].pl)) 798 return m; 799 if (m = TAILQ_FIRST(pq[hindex + oqueuediff].pl)) 800 return m; 801 802 hindex = index - ij; 803 if (hindex < 0) 804 hindex += PQ_L2_SIZE; 805 if (m = TAILQ_FIRST(pq[hindex].pl)) 806 return m; 807 if (m = TAILQ_FIRST(pq[hindex + oqueuediff].pl)) 808 return m; 809 } 810 } 811 812 hindex = index + PQ_L2_SIZE / 2; 813 if (hindex >= PQ_L2_SIZE) 814 hindex -= PQ_L2_SIZE; 815 if (m = TAILQ_FIRST(pq[hindex].pl)) 816 return m; 817 if (m = TAILQ_FIRST(pq[hindex+oqueuediff].pl)) 818 return m; 819 820 #else 821 if (m = TAILQ_FIRST(pq[0].pl)) 822 return m; 823 else 824 return TAILQ_FIRST(pq[oqueuediff].pl); 825 #endif 826 827 return NULL; 828 } 829 830 /* 831 * vm_page_alloc: 832 * 833 * Allocate and return a memory cell associated 834 * with this VM object/offset pair. 835 * 836 * page_req classes: 837 * VM_ALLOC_NORMAL normal process request 838 * VM_ALLOC_SYSTEM system *really* needs a page 839 * VM_ALLOC_INTERRUPT interrupt time request 840 * VM_ALLOC_ZERO zero page 841 * 842 * Object must be locked. 843 * This routine may not block. 844 * 845 * Additional special handling is required when called from an 846 * interrupt (VM_ALLOC_INTERRUPT). We are not allowed to mess with 847 * the page cache in this case. 848 */ 849 vm_page_t 850 vm_page_alloc(object, pindex, page_req) 851 vm_object_t object; 852 vm_pindex_t pindex; 853 int page_req; 854 { 855 register vm_page_t m; 856 struct vpgqueues *pq; 857 vm_object_t oldobject; 858 int queue, qtype; 859 int s; 860 861 KASSERT(!vm_page_lookup(object, pindex), 862 ("vm_page_alloc: page already allocated")); 863 864 if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) { 865 page_req = VM_ALLOC_SYSTEM; 866 }; 867 868 s = splvm(); 869 870 switch (page_req) { 871 872 case VM_ALLOC_NORMAL: 873 if (cnt.v_free_count >= cnt.v_free_reserved) { 874 m = vm_page_select_free(object, pindex, PQ_FREE); 875 KASSERT(m != NULL, ("vm_page_alloc(NORMAL): missing page on free queue\n")); 876 } else { 877 m = vm_page_select_cache(object, pindex); 878 if (m == NULL) { 879 splx(s); 880 #if defined(DIAGNOSTIC) 881 if (cnt.v_cache_count > 0) 882 printf("vm_page_alloc(NORMAL): missing pages on cache queue: %d\n", cnt.v_cache_count); 883 #endif 884 vm_pageout_deficit++; 885 pagedaemon_wakeup(); 886 return (NULL); 887 } 888 } 889 break; 890 891 case VM_ALLOC_ZERO: 892 if (cnt.v_free_count >= cnt.v_free_reserved) { 893 m = vm_page_select_free(object, pindex, PQ_ZERO); 894 KASSERT(m != NULL, ("vm_page_alloc(ZERO): missing page on free queue\n")); 895 } else { 896 m = vm_page_select_cache(object, pindex); 897 if (m == NULL) { 898 splx(s); 899 #if defined(DIAGNOSTIC) 900 if (cnt.v_cache_count > 0) 901 printf("vm_page_alloc(ZERO): missing pages on cache queue: %d\n", cnt.v_cache_count); 902 #endif 903 vm_pageout_deficit++; 904 pagedaemon_wakeup(); 905 return (NULL); 906 } 907 } 908 break; 909 910 case VM_ALLOC_SYSTEM: 911 if ((cnt.v_free_count >= cnt.v_free_reserved) || 912 ((cnt.v_cache_count == 0) && 913 (cnt.v_free_count >= cnt.v_interrupt_free_min))) { 914 m = vm_page_select_free(object, pindex, PQ_FREE); 915 KASSERT(m != NULL, ("vm_page_alloc(SYSTEM): missing page on free queue\n")); 916 } else { 917 m = vm_page_select_cache(object, pindex); 918 if (m == NULL) { 919 splx(s); 920 #if defined(DIAGNOSTIC) 921 if (cnt.v_cache_count > 0) 922 printf("vm_page_alloc(SYSTEM): missing pages on cache queue: %d\n", cnt.v_cache_count); 923 #endif 924 vm_pageout_deficit++; 925 pagedaemon_wakeup(); 926 return (NULL); 927 } 928 } 929 break; 930 931 case VM_ALLOC_INTERRUPT: 932 if (cnt.v_free_count > 0) { 933 m = vm_page_select_free(object, pindex, PQ_FREE); 934 KASSERT(m != NULL, ("vm_page_alloc(INTERRUPT): missing page on free queue\n")); 935 } else { 936 splx(s); 937 vm_pageout_deficit++; 938 pagedaemon_wakeup(); 939 return (NULL); 940 } 941 break; 942 943 default: 944 m = NULL; 945 #if !defined(MAX_PERF) 946 panic("vm_page_alloc: invalid allocation class"); 947 #endif 948 } 949 950 queue = m->queue; 951 qtype = queue - m->pc; 952 if (qtype == PQ_ZERO) 953 vm_page_zero_count--; 954 pq = &vm_page_queues[queue]; 955 TAILQ_REMOVE(pq->pl, m, pageq); 956 (*pq->cnt)--; 957 (*pq->lcnt)--; 958 oldobject = NULL; 959 if (qtype == PQ_ZERO) { 960 m->flags = PG_ZERO | PG_BUSY; 961 } else if (qtype == PQ_CACHE) { 962 oldobject = m->object; 963 vm_page_busy(m); 964 vm_page_remove(m); 965 m->flags = PG_BUSY; 966 } else { 967 m->flags = PG_BUSY; 968 } 969 m->wire_count = 0; 970 m->hold_count = 0; 971 m->act_count = 0; 972 m->busy = 0; 973 m->valid = 0; 974 m->dirty = 0; 975 m->queue = PQ_NONE; 976 977 /* 978 * vm_page_insert() is safe prior to the splx(). Note also that 979 * inserting a page here does not insert it into the pmap (which 980 * could cause us to block allocating memory). We cannot block 981 * anywhere. 982 */ 983 984 vm_page_insert(m, object, pindex); 985 986 /* 987 * Don't wakeup too often - wakeup the pageout daemon when 988 * we would be nearly out of memory. 989 */ 990 if (((cnt.v_free_count + cnt.v_cache_count) < 991 (cnt.v_free_reserved + cnt.v_cache_min)) || 992 (cnt.v_free_count < cnt.v_pageout_free_min)) 993 pagedaemon_wakeup(); 994 995 if ((qtype == PQ_CACHE) && 996 ((page_req == VM_ALLOC_NORMAL) || (page_req == VM_ALLOC_ZERO)) && 997 oldobject && (oldobject->type == OBJT_VNODE) && 998 ((oldobject->flags & OBJ_DEAD) == 0)) { 999 struct vnode *vp; 1000 vp = (struct vnode *) oldobject->handle; 1001 if (vp && VSHOULDFREE(vp)) { 1002 if ((vp->v_flag & (VFREE|VTBFREE|VDOOMED)) == 0) { 1003 TAILQ_INSERT_TAIL(&vnode_tobefree_list, vp, v_freelist); 1004 vp->v_flag |= VTBFREE; 1005 } 1006 } 1007 } 1008 splx(s); 1009 1010 return (m); 1011 } 1012 1013 /* 1014 * vm_wait: (also see VM_WAIT macro) 1015 * 1016 * Block until free pages are available for allocation 1017 */ 1018 1019 void 1020 vm_wait() 1021 { 1022 int s; 1023 1024 s = splvm(); 1025 if (curproc == pageproc) { 1026 vm_pageout_pages_needed = 1; 1027 tsleep(&vm_pageout_pages_needed, PSWP, "vmwait", 0); 1028 } else { 1029 if (!vm_pages_needed) { 1030 vm_pages_needed++; 1031 wakeup(&vm_pages_needed); 1032 } 1033 tsleep(&cnt.v_free_count, PVM, "vmwait", 0); 1034 } 1035 splx(s); 1036 } 1037 1038 /* 1039 * vm_page_sleep: 1040 * 1041 * Block until page is no longer busy. 1042 */ 1043 1044 int 1045 vm_page_sleep(vm_page_t m, char *msg, char *busy) { 1046 int slept = 0; 1047 if ((busy && *busy) || (m->flags & PG_BUSY)) { 1048 int s; 1049 s = splvm(); 1050 if ((busy && *busy) || (m->flags & PG_BUSY)) { 1051 vm_page_flag_set(m, PG_WANTED); 1052 tsleep(m, PVM, msg, 0); 1053 slept = 1; 1054 } 1055 splx(s); 1056 } 1057 return slept; 1058 } 1059 1060 /* 1061 * vm_page_activate: 1062 * 1063 * Put the specified page on the active list (if appropriate). 1064 * 1065 * The page queues must be locked. 1066 * This routine may not block. 1067 */ 1068 void 1069 vm_page_activate(m) 1070 register vm_page_t m; 1071 { 1072 int s; 1073 1074 s = splvm(); 1075 if (m->queue != PQ_ACTIVE) { 1076 if ((m->queue - m->pc) == PQ_CACHE) 1077 cnt.v_reactivated++; 1078 1079 vm_page_unqueue(m); 1080 1081 if (m->wire_count == 0) { 1082 m->queue = PQ_ACTIVE; 1083 ++(*vm_page_queues[PQ_ACTIVE].lcnt); 1084 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 1085 if (m->act_count < ACT_INIT) 1086 m->act_count = ACT_INIT; 1087 cnt.v_active_count++; 1088 } 1089 } else { 1090 if (m->act_count < ACT_INIT) 1091 m->act_count = ACT_INIT; 1092 } 1093 1094 splx(s); 1095 } 1096 1097 /* 1098 * helper routine for vm_page_free and vm_page_free_zero. 1099 * 1100 * This routine may not block. 1101 */ 1102 static int 1103 vm_page_freechk_and_unqueue(m) 1104 vm_page_t m; 1105 { 1106 vm_object_t oldobject; 1107 1108 oldobject = m->object; 1109 1110 #if !defined(MAX_PERF) 1111 if (m->busy || ((m->queue - m->pc) == PQ_FREE) || 1112 (m->hold_count != 0)) { 1113 printf( 1114 "vm_page_free: pindex(%lu), busy(%d), PG_BUSY(%d), hold(%d)\n", 1115 (u_long)m->pindex, m->busy, (m->flags & PG_BUSY) ? 1 : 0, 1116 m->hold_count); 1117 if ((m->queue - m->pc) == PQ_FREE) 1118 panic("vm_page_free: freeing free page"); 1119 else 1120 panic("vm_page_free: freeing busy page"); 1121 } 1122 #endif 1123 1124 vm_page_unqueue_nowakeup(m); 1125 vm_page_remove(m); 1126 1127 if ((m->flags & PG_FICTITIOUS) != 0) { 1128 return 0; 1129 } 1130 1131 m->valid = 0; 1132 1133 if (m->wire_count != 0) { 1134 #if !defined(MAX_PERF) 1135 if (m->wire_count > 1) { 1136 panic("vm_page_free: invalid wire count (%d), pindex: 0x%x", 1137 m->wire_count, m->pindex); 1138 } 1139 #endif 1140 printf("vm_page_free: freeing wired page\n"); 1141 m->wire_count = 0; 1142 cnt.v_wire_count--; 1143 } 1144 1145 if (oldobject && (oldobject->type == OBJT_VNODE) && 1146 ((oldobject->flags & OBJ_DEAD) == 0)) { 1147 struct vnode *vp; 1148 vp = (struct vnode *) oldobject->handle; 1149 if (vp && VSHOULDFREE(vp)) { 1150 if ((vp->v_flag & (VTBFREE|VDOOMED|VFREE)) == 0) { 1151 TAILQ_INSERT_TAIL(&vnode_tobefree_list, vp, v_freelist); 1152 vp->v_flag |= VTBFREE; 1153 } 1154 } 1155 } 1156 1157 #ifdef __alpha__ 1158 pmap_page_is_free(m); 1159 #endif 1160 1161 return 1; 1162 } 1163 1164 /* 1165 * helper routine for vm_page_free and vm_page_free_zero. 1166 * 1167 * This routine may not block. 1168 */ 1169 static __inline void 1170 vm_page_free_wakeup() 1171 { 1172 1173 /* 1174 * if pageout daemon needs pages, then tell it that there are 1175 * some free. 1176 */ 1177 if (vm_pageout_pages_needed) { 1178 wakeup(&vm_pageout_pages_needed); 1179 vm_pageout_pages_needed = 0; 1180 } 1181 /* 1182 * wakeup processes that are waiting on memory if we hit a 1183 * high water mark. And wakeup scheduler process if we have 1184 * lots of memory. this process will swapin processes. 1185 */ 1186 if (vm_pages_needed && 1187 ((cnt.v_free_count + cnt.v_cache_count) >= cnt.v_free_min)) { 1188 wakeup(&cnt.v_free_count); 1189 vm_pages_needed = 0; 1190 } 1191 } 1192 1193 /* 1194 * vm_page_free: 1195 * 1196 * Returns the given page to the free list, 1197 * disassociating it with any VM object. 1198 * 1199 * Object and page must be locked prior to entry. 1200 * This routine may not block. 1201 */ 1202 void 1203 vm_page_free(m) 1204 register vm_page_t m; 1205 { 1206 int s; 1207 struct vpgqueues *pq; 1208 1209 s = splvm(); 1210 1211 cnt.v_tfree++; 1212 1213 if (!vm_page_freechk_and_unqueue(m)) { 1214 splx(s); 1215 return; 1216 } 1217 1218 m->queue = PQ_FREE + m->pc; 1219 pq = &vm_page_queues[m->queue]; 1220 ++(*pq->lcnt); 1221 ++(*pq->cnt); 1222 /* 1223 * If the pageout process is grabbing the page, it is likely 1224 * that the page is NOT in the cache. It is more likely that 1225 * the page will be partially in the cache if it is being 1226 * explicitly freed. 1227 */ 1228 if (curproc == pageproc) { 1229 TAILQ_INSERT_TAIL(pq->pl, m, pageq); 1230 } else { 1231 TAILQ_INSERT_HEAD(pq->pl, m, pageq); 1232 } 1233 1234 vm_page_free_wakeup(); 1235 splx(s); 1236 } 1237 1238 void 1239 vm_page_free_zero(m) 1240 register vm_page_t m; 1241 { 1242 int s; 1243 struct vpgqueues *pq; 1244 1245 s = splvm(); 1246 1247 cnt.v_tfree++; 1248 1249 if (!vm_page_freechk_and_unqueue(m)) { 1250 splx(s); 1251 return; 1252 } 1253 1254 m->queue = PQ_ZERO + m->pc; 1255 pq = &vm_page_queues[m->queue]; 1256 ++(*pq->lcnt); 1257 ++(*pq->cnt); 1258 1259 TAILQ_INSERT_HEAD(pq->pl, m, pageq); 1260 ++vm_page_zero_count; 1261 vm_page_free_wakeup(); 1262 splx(s); 1263 } 1264 1265 /* 1266 * vm_page_wire: 1267 * 1268 * Mark this page as wired down by yet 1269 * another map, removing it from paging queues 1270 * as necessary. 1271 * 1272 * The page queues must be locked. 1273 * This routine may not block. 1274 */ 1275 void 1276 vm_page_wire(m) 1277 register vm_page_t m; 1278 { 1279 int s; 1280 1281 s = splvm(); 1282 if (m->wire_count == 0) { 1283 vm_page_unqueue(m); 1284 cnt.v_wire_count++; 1285 } 1286 m->wire_count++; 1287 splx(s); 1288 (*vm_page_queues[PQ_NONE].lcnt)++; 1289 vm_page_flag_set(m, PG_MAPPED); 1290 } 1291 1292 /* 1293 * vm_page_unwire: 1294 * 1295 * Release one wiring of this page, potentially 1296 * enabling it to be paged again. 1297 * 1298 * The page queues must be locked. 1299 * This routine may not block. 1300 */ 1301 void 1302 vm_page_unwire(m, activate) 1303 register vm_page_t m; 1304 int activate; 1305 { 1306 int s; 1307 1308 s = splvm(); 1309 1310 if (m->wire_count > 0) { 1311 m->wire_count--; 1312 if (m->wire_count == 0) { 1313 cnt.v_wire_count--; 1314 if (activate) { 1315 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); 1316 m->queue = PQ_ACTIVE; 1317 (*vm_page_queues[PQ_ACTIVE].lcnt)++; 1318 cnt.v_active_count++; 1319 } else { 1320 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); 1321 m->queue = PQ_INACTIVE; 1322 (*vm_page_queues[PQ_INACTIVE].lcnt)++; 1323 cnt.v_inactive_count++; 1324 } 1325 } 1326 } else { 1327 #if !defined(MAX_PERF) 1328 panic("vm_page_unwire: invalid wire count: %d\n", m->wire_count); 1329 #endif 1330 } 1331 splx(s); 1332 } 1333 1334 1335 /* 1336 * Move the specified page to the inactive queue. 1337 * 1338 * This routine may not block. 1339 */ 1340 void 1341 vm_page_deactivate(m) 1342 register vm_page_t m; 1343 { 1344 int s; 1345 1346 /* 1347 * Ignore if already inactive. 1348 */ 1349 if (m->queue == PQ_INACTIVE) 1350 return; 1351 1352 s = splvm(); 1353 if (m->wire_count == 0) { 1354 if ((m->queue - m->pc) == PQ_CACHE) 1355 cnt.v_reactivated++; 1356 vm_page_unqueue(m); 1357 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); 1358 m->queue = PQ_INACTIVE; 1359 ++(*vm_page_queues[PQ_INACTIVE].lcnt); 1360 cnt.v_inactive_count++; 1361 } 1362 splx(s); 1363 } 1364 1365 /* 1366 * vm_page_cache 1367 * 1368 * Put the specified page onto the page cache queue (if appropriate). 1369 * This routine may not block. 1370 */ 1371 void 1372 vm_page_cache(m) 1373 register vm_page_t m; 1374 { 1375 int s; 1376 1377 #if !defined(MAX_PERF) 1378 if ((m->flags & PG_BUSY) || m->busy || m->wire_count) { 1379 printf("vm_page_cache: attempting to cache busy page\n"); 1380 return; 1381 } 1382 #endif 1383 if ((m->queue - m->pc) == PQ_CACHE) 1384 return; 1385 1386 vm_page_protect(m, VM_PROT_NONE); 1387 #if !defined(MAX_PERF) 1388 if (m->dirty != 0) { 1389 panic("vm_page_cache: caching a dirty page, pindex: %d", m->pindex); 1390 } 1391 #endif 1392 s = splvm(); 1393 vm_page_unqueue_nowakeup(m); 1394 m->queue = PQ_CACHE + m->pc; 1395 (*vm_page_queues[m->queue].lcnt)++; 1396 TAILQ_INSERT_TAIL(vm_page_queues[m->queue].pl, m, pageq); 1397 cnt.v_cache_count++; 1398 vm_page_free_wakeup(); 1399 splx(s); 1400 } 1401 1402 /* 1403 * Grab a page, waiting until we are waken up due to the page 1404 * changing state. We keep on waiting, if the page continues 1405 * to be in the object. If the page doesn't exist, allocate it. 1406 * 1407 * This routine may block. 1408 */ 1409 vm_page_t 1410 vm_page_grab(object, pindex, allocflags) 1411 vm_object_t object; 1412 vm_pindex_t pindex; 1413 int allocflags; 1414 { 1415 1416 vm_page_t m; 1417 int s, generation; 1418 1419 retrylookup: 1420 if ((m = vm_page_lookup(object, pindex)) != NULL) { 1421 if (m->busy || (m->flags & PG_BUSY)) { 1422 generation = object->generation; 1423 1424 s = splvm(); 1425 while ((object->generation == generation) && 1426 (m->busy || (m->flags & PG_BUSY))) { 1427 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED); 1428 tsleep(m, PVM, "pgrbwt", 0); 1429 if ((allocflags & VM_ALLOC_RETRY) == 0) { 1430 splx(s); 1431 return NULL; 1432 } 1433 } 1434 splx(s); 1435 goto retrylookup; 1436 } else { 1437 vm_page_busy(m); 1438 return m; 1439 } 1440 } 1441 1442 m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY); 1443 if (m == NULL) { 1444 VM_WAIT; 1445 if ((allocflags & VM_ALLOC_RETRY) == 0) 1446 return NULL; 1447 goto retrylookup; 1448 } 1449 1450 return m; 1451 } 1452 1453 /* 1454 * mapping function for valid bits or for dirty bits in 1455 * a page. May not block. 1456 */ 1457 __inline int 1458 vm_page_bits(int base, int size) 1459 { 1460 u_short chunk; 1461 1462 if ((base == 0) && (size >= PAGE_SIZE)) 1463 return VM_PAGE_BITS_ALL; 1464 1465 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 1466 base &= PAGE_MASK; 1467 if (size > PAGE_SIZE - base) { 1468 size = PAGE_SIZE - base; 1469 } 1470 1471 base = base / DEV_BSIZE; 1472 chunk = vm_page_dev_bsize_chunks[size / DEV_BSIZE]; 1473 return (chunk << base) & VM_PAGE_BITS_ALL; 1474 } 1475 1476 /* 1477 * set a page valid and clean. May not block. 1478 */ 1479 void 1480 vm_page_set_validclean(m, base, size) 1481 vm_page_t m; 1482 int base; 1483 int size; 1484 { 1485 int pagebits = vm_page_bits(base, size); 1486 m->valid |= pagebits; 1487 m->dirty &= ~pagebits; 1488 if( base == 0 && size == PAGE_SIZE) 1489 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 1490 } 1491 1492 /* 1493 * set a page (partially) invalid. May not block. 1494 */ 1495 void 1496 vm_page_set_invalid(m, base, size) 1497 vm_page_t m; 1498 int base; 1499 int size; 1500 { 1501 int bits; 1502 1503 m->valid &= ~(bits = vm_page_bits(base, size)); 1504 if (m->valid == 0) 1505 m->dirty &= ~bits; 1506 m->object->generation++; 1507 } 1508 1509 /* 1510 * is (partial) page valid? May not block. 1511 */ 1512 int 1513 vm_page_is_valid(m, base, size) 1514 vm_page_t m; 1515 int base; 1516 int size; 1517 { 1518 int bits = vm_page_bits(base, size); 1519 1520 if (m->valid && ((m->valid & bits) == bits)) 1521 return 1; 1522 else 1523 return 0; 1524 } 1525 1526 /* 1527 * update dirty bits from pmap/mmu. May not block. 1528 */ 1529 1530 void 1531 vm_page_test_dirty(m) 1532 vm_page_t m; 1533 { 1534 if ((m->dirty != VM_PAGE_BITS_ALL) && 1535 pmap_is_modified(VM_PAGE_TO_PHYS(m))) { 1536 m->dirty = VM_PAGE_BITS_ALL; 1537 } 1538 } 1539 1540 /* 1541 * This interface is for merging with malloc() someday. 1542 * Even if we never implement compaction so that contiguous allocation 1543 * works after initialization time, malloc()'s data structures are good 1544 * for statistics and for allocations of less than a page. 1545 */ 1546 void * 1547 contigmalloc1(size, type, flags, low, high, alignment, boundary, map) 1548 unsigned long size; /* should be size_t here and for malloc() */ 1549 struct malloc_type *type; 1550 int flags; 1551 unsigned long low; 1552 unsigned long high; 1553 unsigned long alignment; 1554 unsigned long boundary; 1555 vm_map_t map; 1556 { 1557 int i, s, start; 1558 vm_offset_t addr, phys, tmp_addr; 1559 int pass; 1560 vm_page_t pga = vm_page_array; 1561 1562 size = round_page(size); 1563 #if !defined(MAX_PERF) 1564 if (size == 0) 1565 panic("contigmalloc1: size must not be 0"); 1566 if ((alignment & (alignment - 1)) != 0) 1567 panic("contigmalloc1: alignment must be a power of 2"); 1568 if ((boundary & (boundary - 1)) != 0) 1569 panic("contigmalloc1: boundary must be a power of 2"); 1570 #endif 1571 1572 start = 0; 1573 for (pass = 0; pass <= 1; pass++) { 1574 s = splvm(); 1575 again: 1576 /* 1577 * Find first page in array that is free, within range, aligned, and 1578 * such that the boundary won't be crossed. 1579 */ 1580 for (i = start; i < cnt.v_page_count; i++) { 1581 int pqtype; 1582 phys = VM_PAGE_TO_PHYS(&pga[i]); 1583 pqtype = pga[i].queue - pga[i].pc; 1584 if (((pqtype == PQ_ZERO) || (pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) && 1585 (phys >= low) && (phys < high) && 1586 ((phys & (alignment - 1)) == 0) && 1587 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0)) 1588 break; 1589 } 1590 1591 /* 1592 * If the above failed or we will exceed the upper bound, fail. 1593 */ 1594 if ((i == cnt.v_page_count) || 1595 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) { 1596 vm_page_t m, next; 1597 1598 again1: 1599 for (m = TAILQ_FIRST(&vm_page_queue_inactive); 1600 m != NULL; 1601 m = next) { 1602 1603 if (m->queue != PQ_INACTIVE) { 1604 break; 1605 } 1606 1607 next = TAILQ_NEXT(m, pageq); 1608 if (vm_page_sleep(m, "vpctw0", &m->busy)) 1609 goto again1; 1610 vm_page_test_dirty(m); 1611 if (m->dirty) { 1612 if (m->object->type == OBJT_VNODE) { 1613 vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); 1614 vm_object_page_clean(m->object, 0, 0, OBJPC_SYNC); 1615 VOP_UNLOCK(m->object->handle, 0, curproc); 1616 goto again1; 1617 } else if (m->object->type == OBJT_SWAP || 1618 m->object->type == OBJT_DEFAULT) { 1619 vm_pageout_flush(&m, 1, 0); 1620 goto again1; 1621 } 1622 } 1623 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 1624 vm_page_cache(m); 1625 } 1626 1627 for (m = TAILQ_FIRST(&vm_page_queue_active); 1628 m != NULL; 1629 m = next) { 1630 1631 if (m->queue != PQ_ACTIVE) { 1632 break; 1633 } 1634 1635 next = TAILQ_NEXT(m, pageq); 1636 if (vm_page_sleep(m, "vpctw1", &m->busy)) 1637 goto again1; 1638 vm_page_test_dirty(m); 1639 if (m->dirty) { 1640 if (m->object->type == OBJT_VNODE) { 1641 vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); 1642 vm_object_page_clean(m->object, 0, 0, OBJPC_SYNC); 1643 VOP_UNLOCK(m->object->handle, 0, curproc); 1644 goto again1; 1645 } else if (m->object->type == OBJT_SWAP || 1646 m->object->type == OBJT_DEFAULT) { 1647 vm_pageout_flush(&m, 1, 0); 1648 goto again1; 1649 } 1650 } 1651 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0)) 1652 vm_page_cache(m); 1653 } 1654 1655 splx(s); 1656 continue; 1657 } 1658 start = i; 1659 1660 /* 1661 * Check successive pages for contiguous and free. 1662 */ 1663 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) { 1664 int pqtype; 1665 pqtype = pga[i].queue - pga[i].pc; 1666 if ((VM_PAGE_TO_PHYS(&pga[i]) != 1667 (VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) || 1668 ((pqtype != PQ_ZERO) && (pqtype != PQ_FREE) && (pqtype != PQ_CACHE))) { 1669 start++; 1670 goto again; 1671 } 1672 } 1673 1674 for (i = start; i < (start + size / PAGE_SIZE); i++) { 1675 int pqtype; 1676 vm_page_t m = &pga[i]; 1677 1678 pqtype = m->queue - m->pc; 1679 if (pqtype == PQ_CACHE) { 1680 vm_page_busy(m); 1681 vm_page_free(m); 1682 } 1683 1684 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 1685 (*vm_page_queues[m->queue].lcnt)--; 1686 cnt.v_free_count--; 1687 m->valid = VM_PAGE_BITS_ALL; 1688 m->flags = 0; 1689 m->dirty = 0; 1690 m->wire_count = 0; 1691 m->busy = 0; 1692 m->queue = PQ_NONE; 1693 m->object = NULL; 1694 vm_page_wire(m); 1695 } 1696 1697 /* 1698 * We've found a contiguous chunk that meets are requirements. 1699 * Allocate kernel VM, unfree and assign the physical pages to it and 1700 * return kernel VM pointer. 1701 */ 1702 tmp_addr = addr = kmem_alloc_pageable(map, size); 1703 if (addr == 0) { 1704 /* 1705 * XXX We almost never run out of kernel virtual 1706 * space, so we don't make the allocated memory 1707 * above available. 1708 */ 1709 splx(s); 1710 return (NULL); 1711 } 1712 1713 for (i = start; i < (start + size / PAGE_SIZE); i++) { 1714 vm_page_t m = &pga[i]; 1715 vm_page_insert(m, kernel_object, 1716 OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS)); 1717 pmap_kenter(tmp_addr, VM_PAGE_TO_PHYS(m)); 1718 tmp_addr += PAGE_SIZE; 1719 } 1720 1721 splx(s); 1722 return ((void *)addr); 1723 } 1724 return NULL; 1725 } 1726 1727 void * 1728 contigmalloc(size, type, flags, low, high, alignment, boundary) 1729 unsigned long size; /* should be size_t here and for malloc() */ 1730 struct malloc_type *type; 1731 int flags; 1732 unsigned long low; 1733 unsigned long high; 1734 unsigned long alignment; 1735 unsigned long boundary; 1736 { 1737 return contigmalloc1(size, type, flags, low, high, alignment, boundary, 1738 kernel_map); 1739 } 1740 1741 vm_offset_t 1742 vm_page_alloc_contig(size, low, high, alignment) 1743 vm_offset_t size; 1744 vm_offset_t low; 1745 vm_offset_t high; 1746 vm_offset_t alignment; 1747 { 1748 return ((vm_offset_t)contigmalloc1(size, M_DEVBUF, M_NOWAIT, low, high, 1749 alignment, 0ul, kernel_map)); 1750 } 1751 1752 #include "opt_ddb.h" 1753 #ifdef DDB 1754 #include <sys/kernel.h> 1755 1756 #include <ddb/ddb.h> 1757 1758 DB_SHOW_COMMAND(page, vm_page_print_page_info) 1759 { 1760 db_printf("cnt.v_free_count: %d\n", cnt.v_free_count); 1761 db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count); 1762 db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count); 1763 db_printf("cnt.v_active_count: %d\n", cnt.v_active_count); 1764 db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count); 1765 db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved); 1766 db_printf("cnt.v_free_min: %d\n", cnt.v_free_min); 1767 db_printf("cnt.v_free_target: %d\n", cnt.v_free_target); 1768 db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min); 1769 db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target); 1770 } 1771 1772 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) 1773 { 1774 int i; 1775 db_printf("PQ_FREE:"); 1776 for(i=0;i<PQ_L2_SIZE;i++) { 1777 db_printf(" %d", *vm_page_queues[PQ_FREE + i].lcnt); 1778 } 1779 db_printf("\n"); 1780 1781 db_printf("PQ_CACHE:"); 1782 for(i=0;i<PQ_L2_SIZE;i++) { 1783 db_printf(" %d", *vm_page_queues[PQ_CACHE + i].lcnt); 1784 } 1785 db_printf("\n"); 1786 1787 db_printf("PQ_ZERO:"); 1788 for(i=0;i<PQ_L2_SIZE;i++) { 1789 db_printf(" %d", *vm_page_queues[PQ_ZERO + i].lcnt); 1790 } 1791 db_printf("\n"); 1792 1793 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n", 1794 *vm_page_queues[PQ_ACTIVE].lcnt, 1795 *vm_page_queues[PQ_INACTIVE].lcnt); 1796 } 1797 #endif /* DDB */
Cache object: 6437b16b5cef5316756d1f2ee748cf62
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]