1 /*- 2 * Copyright (c) 1991 Regents of the University of California. 3 * All rights reserved. 4 * Copyright (c) 1998 Matthew Dillon. All Rights Reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * The Mach Operating System project at Carnegie-Mellon University. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91 34 */ 35 36 /*- 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 */ 62 63 /* 64 * GENERAL RULES ON VM_PAGE MANIPULATION 65 * 66 * - A page queue lock is required when adding or removing a page from a 67 * page queue regardless of other locks or the busy state of a page. 68 * 69 * * In general, no thread besides the page daemon can acquire or 70 * hold more than one page queue lock at a time. 71 * 72 * * The page daemon can acquire and hold any pair of page queue 73 * locks in any order. 74 * 75 * - The object lock is required when inserting or removing 76 * pages from an object (vm_page_insert() or vm_page_remove()). 77 * 78 */ 79 80 /* 81 * Resident memory management module. 82 */ 83 84 #include <sys/cdefs.h> 85 __FBSDID("$FreeBSD$"); 86 87 #include "opt_vm.h" 88 89 #include <sys/param.h> 90 #include <sys/systm.h> 91 #include <sys/lock.h> 92 #include <sys/kernel.h> 93 #include <sys/limits.h> 94 #include <sys/linker.h> 95 #include <sys/malloc.h> 96 #include <sys/mman.h> 97 #include <sys/msgbuf.h> 98 #include <sys/mutex.h> 99 #include <sys/proc.h> 100 #include <sys/rwlock.h> 101 #include <sys/sbuf.h> 102 #include <sys/smp.h> 103 #include <sys/sysctl.h> 104 #include <sys/vmmeter.h> 105 #include <sys/vnode.h> 106 107 #include <vm/vm.h> 108 #include <vm/pmap.h> 109 #include <vm/vm_param.h> 110 #include <vm/vm_kern.h> 111 #include <vm/vm_object.h> 112 #include <vm/vm_page.h> 113 #include <vm/vm_pageout.h> 114 #include <vm/vm_pager.h> 115 #include <vm/vm_phys.h> 116 #include <vm/vm_radix.h> 117 #include <vm/vm_reserv.h> 118 #include <vm/vm_extern.h> 119 #include <vm/uma.h> 120 #include <vm/uma_int.h> 121 122 #include <machine/md_var.h> 123 124 /* 125 * Associated with page of user-allocatable memory is a 126 * page structure. 127 */ 128 129 struct vm_domain vm_dom[MAXMEMDOM]; 130 struct mtx_padalign vm_page_queue_free_mtx; 131 132 struct mtx_padalign pa_lock[PA_LOCK_COUNT]; 133 134 vm_page_t vm_page_array; 135 long vm_page_array_size; 136 long first_page; 137 138 static int boot_pages = UMA_BOOT_PAGES; 139 SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 140 &boot_pages, 0, 141 "number of pages allocated for bootstrapping the VM system"); 142 143 static int pa_tryrelock_restart; 144 SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD, 145 &pa_tryrelock_restart, 0, "Number of tryrelock restarts"); 146 147 static TAILQ_HEAD(, vm_page) blacklist_head; 148 static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS); 149 SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD | 150 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages"); 151 152 /* Is the page daemon waiting for free pages? */ 153 static int vm_pageout_pages_needed; 154 155 static uma_zone_t fakepg_zone; 156 157 static struct vnode *vm_page_alloc_init(vm_page_t m); 158 static void vm_page_cache_turn_free(vm_page_t m); 159 static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits); 160 static void vm_page_enqueue(uint8_t queue, vm_page_t m); 161 static void vm_page_free_wakeup(void); 162 static void vm_page_init_fakepg(void *dummy); 163 static int vm_page_insert_after(vm_page_t m, vm_object_t object, 164 vm_pindex_t pindex, vm_page_t mpred); 165 static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object, 166 vm_page_t mpred); 167 static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run, 168 vm_paddr_t high); 169 170 SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL); 171 172 static void 173 vm_page_init_fakepg(void *dummy) 174 { 175 176 fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL, 177 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); 178 } 179 180 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */ 181 #if PAGE_SIZE == 32768 182 #ifdef CTASSERT 183 CTASSERT(sizeof(u_long) >= 8); 184 #endif 185 #endif 186 187 /* 188 * Try to acquire a physical address lock while a pmap is locked. If we 189 * fail to trylock we unlock and lock the pmap directly and cache the 190 * locked pa in *locked. The caller should then restart their loop in case 191 * the virtual to physical mapping has changed. 192 */ 193 int 194 vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked) 195 { 196 vm_paddr_t lockpa; 197 198 lockpa = *locked; 199 *locked = pa; 200 if (lockpa) { 201 PA_LOCK_ASSERT(lockpa, MA_OWNED); 202 if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa)) 203 return (0); 204 PA_UNLOCK(lockpa); 205 } 206 if (PA_TRYLOCK(pa)) 207 return (0); 208 PMAP_UNLOCK(pmap); 209 atomic_add_int(&pa_tryrelock_restart, 1); 210 PA_LOCK(pa); 211 PMAP_LOCK(pmap); 212 return (EAGAIN); 213 } 214 215 /* 216 * vm_set_page_size: 217 * 218 * Sets the page size, perhaps based upon the memory 219 * size. Must be called before any use of page-size 220 * dependent functions. 221 */ 222 void 223 vm_set_page_size(void) 224 { 225 if (vm_cnt.v_page_size == 0) 226 vm_cnt.v_page_size = PAGE_SIZE; 227 if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0) 228 panic("vm_set_page_size: page size not a power of two"); 229 } 230 231 /* 232 * vm_page_blacklist_next: 233 * 234 * Find the next entry in the provided string of blacklist 235 * addresses. Entries are separated by space, comma, or newline. 236 * If an invalid integer is encountered then the rest of the 237 * string is skipped. Updates the list pointer to the next 238 * character, or NULL if the string is exhausted or invalid. 239 */ 240 static vm_paddr_t 241 vm_page_blacklist_next(char **list, char *end) 242 { 243 vm_paddr_t bad; 244 char *cp, *pos; 245 246 if (list == NULL || *list == NULL) 247 return (0); 248 if (**list =='\0') { 249 *list = NULL; 250 return (0); 251 } 252 253 /* 254 * If there's no end pointer then the buffer is coming from 255 * the kenv and we know it's null-terminated. 256 */ 257 if (end == NULL) 258 end = *list + strlen(*list); 259 260 /* Ensure that strtoq() won't walk off the end */ 261 if (*end != '\0') { 262 if (*end == '\n' || *end == ' ' || *end == ',') 263 *end = '\0'; 264 else { 265 printf("Blacklist not terminated, skipping\n"); 266 *list = NULL; 267 return (0); 268 } 269 } 270 271 for (pos = *list; *pos != '\0'; pos = cp) { 272 bad = strtoq(pos, &cp, 0); 273 if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') { 274 if (bad == 0) { 275 if (++cp < end) 276 continue; 277 else 278 break; 279 } 280 } else 281 break; 282 if (*cp == '\0' || ++cp >= end) 283 *list = NULL; 284 else 285 *list = cp; 286 return (trunc_page(bad)); 287 } 288 printf("Garbage in RAM blacklist, skipping\n"); 289 *list = NULL; 290 return (0); 291 } 292 293 /* 294 * vm_page_blacklist_check: 295 * 296 * Iterate through the provided string of blacklist addresses, pulling 297 * each entry out of the physical allocator free list and putting it 298 * onto a list for reporting via the vm.page_blacklist sysctl. 299 */ 300 static void 301 vm_page_blacklist_check(char *list, char *end) 302 { 303 vm_paddr_t pa; 304 vm_page_t m; 305 char *next; 306 int ret; 307 308 next = list; 309 while (next != NULL) { 310 if ((pa = vm_page_blacklist_next(&next, end)) == 0) 311 continue; 312 m = vm_phys_paddr_to_vm_page(pa); 313 if (m == NULL) 314 continue; 315 mtx_lock(&vm_page_queue_free_mtx); 316 ret = vm_phys_unfree_page(m); 317 mtx_unlock(&vm_page_queue_free_mtx); 318 if (ret == TRUE) { 319 TAILQ_INSERT_TAIL(&blacklist_head, m, listq); 320 if (bootverbose) 321 printf("Skipping page with pa 0x%jx\n", 322 (uintmax_t)pa); 323 } 324 } 325 } 326 327 /* 328 * vm_page_blacklist_load: 329 * 330 * Search for a special module named "ram_blacklist". It'll be a 331 * plain text file provided by the user via the loader directive 332 * of the same name. 333 */ 334 static void 335 vm_page_blacklist_load(char **list, char **end) 336 { 337 void *mod; 338 u_char *ptr; 339 u_int len; 340 341 mod = NULL; 342 ptr = NULL; 343 344 mod = preload_search_by_type("ram_blacklist"); 345 if (mod != NULL) { 346 ptr = preload_fetch_addr(mod); 347 len = preload_fetch_size(mod); 348 } 349 *list = ptr; 350 if (ptr != NULL) 351 *end = ptr + len; 352 else 353 *end = NULL; 354 return; 355 } 356 357 static int 358 sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS) 359 { 360 vm_page_t m; 361 struct sbuf sbuf; 362 int error, first; 363 364 first = 1; 365 error = sysctl_wire_old_buffer(req, 0); 366 if (error != 0) 367 return (error); 368 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 369 TAILQ_FOREACH(m, &blacklist_head, listq) { 370 sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",", 371 (uintmax_t)m->phys_addr); 372 first = 0; 373 } 374 error = sbuf_finish(&sbuf); 375 sbuf_delete(&sbuf); 376 return (error); 377 } 378 379 static void 380 vm_page_domain_init(struct vm_domain *vmd) 381 { 382 struct vm_pagequeue *pq; 383 int i; 384 385 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) = 386 "vm inactive pagequeue"; 387 *__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) = 388 &vm_cnt.v_inactive_count; 389 *__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) = 390 "vm active pagequeue"; 391 *__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) = 392 &vm_cnt.v_active_count; 393 vmd->vmd_page_count = 0; 394 vmd->vmd_free_count = 0; 395 vmd->vmd_segs = 0; 396 vmd->vmd_oom = FALSE; 397 for (i = 0; i < PQ_COUNT; i++) { 398 pq = &vmd->vmd_pagequeues[i]; 399 TAILQ_INIT(&pq->pq_pl); 400 mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue", 401 MTX_DEF | MTX_DUPOK); 402 } 403 } 404 405 /* 406 * vm_page_startup: 407 * 408 * Initializes the resident memory module. 409 * 410 * Allocates memory for the page cells, and 411 * for the object/offset-to-page hash table headers. 412 * Each page cell is initialized and placed on the free list. 413 */ 414 vm_offset_t 415 vm_page_startup(vm_offset_t vaddr) 416 { 417 vm_offset_t mapped; 418 vm_paddr_t page_range; 419 vm_paddr_t new_end; 420 int i; 421 vm_paddr_t pa; 422 vm_paddr_t last_pa; 423 char *list, *listend; 424 vm_paddr_t end; 425 vm_paddr_t biggestsize; 426 vm_paddr_t low_water, high_water; 427 int biggestone; 428 int pages_per_zone; 429 430 biggestsize = 0; 431 biggestone = 0; 432 vaddr = round_page(vaddr); 433 434 for (i = 0; phys_avail[i + 1]; i += 2) { 435 phys_avail[i] = round_page(phys_avail[i]); 436 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); 437 } 438 439 low_water = phys_avail[0]; 440 high_water = phys_avail[1]; 441 442 for (i = 0; i < vm_phys_nsegs; i++) { 443 if (vm_phys_segs[i].start < low_water) 444 low_water = vm_phys_segs[i].start; 445 if (vm_phys_segs[i].end > high_water) 446 high_water = vm_phys_segs[i].end; 447 } 448 for (i = 0; phys_avail[i + 1]; i += 2) { 449 vm_paddr_t size = phys_avail[i + 1] - phys_avail[i]; 450 451 if (size > biggestsize) { 452 biggestone = i; 453 biggestsize = size; 454 } 455 if (phys_avail[i] < low_water) 456 low_water = phys_avail[i]; 457 if (phys_avail[i + 1] > high_water) 458 high_water = phys_avail[i + 1]; 459 } 460 461 end = phys_avail[biggestone+1]; 462 463 /* 464 * Initialize the page and queue locks. 465 */ 466 mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF); 467 for (i = 0; i < PA_LOCK_COUNT; i++) 468 mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF); 469 for (i = 0; i < vm_ndomains; i++) 470 vm_page_domain_init(&vm_dom[i]); 471 472 /* 473 * Almost all of the pages needed for boot strapping UMA are used 474 * for zone structures, so if the number of CPUs results in those 475 * structures taking more than one page each, we set aside more pages 476 * in proportion to the zone structure size. 477 */ 478 pages_per_zone = howmany(sizeof(struct uma_zone) + 479 sizeof(struct uma_cache) * (mp_maxid + 1), UMA_SLAB_SIZE); 480 if (pages_per_zone > 1) { 481 /* Reserve more pages so that we don't run out. */ 482 boot_pages = UMA_BOOT_PAGES_ZONES * pages_per_zone; 483 } 484 485 /* 486 * Allocate memory for use when boot strapping the kernel memory 487 * allocator. 488 * 489 * CTFLAG_RDTUN doesn't work during the early boot process, so we must 490 * manually fetch the value. 491 */ 492 TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages); 493 new_end = end - (boot_pages * UMA_SLAB_SIZE); 494 new_end = trunc_page(new_end); 495 mapped = pmap_map(&vaddr, new_end, end, 496 VM_PROT_READ | VM_PROT_WRITE); 497 bzero((void *)mapped, end - new_end); 498 uma_startup((void *)mapped, boot_pages); 499 500 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \ 501 defined(__i386__) || defined(__mips__) 502 /* 503 * Allocate a bitmap to indicate that a random physical page 504 * needs to be included in a minidump. 505 * 506 * The amd64 port needs this to indicate which direct map pages 507 * need to be dumped, via calls to dump_add_page()/dump_drop_page(). 508 * 509 * However, i386 still needs this workspace internally within the 510 * minidump code. In theory, they are not needed on i386, but are 511 * included should the sf_buf code decide to use them. 512 */ 513 last_pa = 0; 514 for (i = 0; dump_avail[i + 1] != 0; i += 2) 515 if (dump_avail[i + 1] > last_pa) 516 last_pa = dump_avail[i + 1]; 517 page_range = last_pa / PAGE_SIZE; 518 vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY); 519 new_end -= vm_page_dump_size; 520 vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end, 521 new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE); 522 bzero((void *)vm_page_dump, vm_page_dump_size); 523 #endif 524 #ifdef __amd64__ 525 /* 526 * Request that the physical pages underlying the message buffer be 527 * included in a crash dump. Since the message buffer is accessed 528 * through the direct map, they are not automatically included. 529 */ 530 pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr); 531 last_pa = pa + round_page(msgbufsize); 532 while (pa < last_pa) { 533 dump_add_page(pa); 534 pa += PAGE_SIZE; 535 } 536 #endif 537 /* 538 * Compute the number of pages of memory that will be available for 539 * use (taking into account the overhead of a page structure per 540 * page). 541 */ 542 first_page = low_water / PAGE_SIZE; 543 #ifdef VM_PHYSSEG_SPARSE 544 page_range = 0; 545 for (i = 0; i < vm_phys_nsegs; i++) { 546 page_range += atop(vm_phys_segs[i].end - 547 vm_phys_segs[i].start); 548 } 549 for (i = 0; phys_avail[i + 1] != 0; i += 2) 550 page_range += atop(phys_avail[i + 1] - phys_avail[i]); 551 #elif defined(VM_PHYSSEG_DENSE) 552 page_range = high_water / PAGE_SIZE - first_page; 553 #else 554 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." 555 #endif 556 end = new_end; 557 558 /* 559 * Reserve an unmapped guard page to trap access to vm_page_array[-1]. 560 */ 561 vaddr += PAGE_SIZE; 562 563 /* 564 * Initialize the mem entry structures now, and put them in the free 565 * queue. 566 */ 567 new_end = trunc_page(end - page_range * sizeof(struct vm_page)); 568 mapped = pmap_map(&vaddr, new_end, end, 569 VM_PROT_READ | VM_PROT_WRITE); 570 vm_page_array = (vm_page_t) mapped; 571 #if VM_NRESERVLEVEL > 0 572 /* 573 * Allocate memory for the reservation management system's data 574 * structures. 575 */ 576 new_end = vm_reserv_startup(&vaddr, new_end, high_water); 577 #endif 578 #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) 579 /* 580 * pmap_map on arm64, amd64, and mips can come out of the direct-map, 581 * not kvm like i386, so the pages must be tracked for a crashdump to 582 * include this data. This includes the vm_page_array and the early 583 * UMA bootstrap pages. 584 */ 585 for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE) 586 dump_add_page(pa); 587 #endif 588 phys_avail[biggestone + 1] = new_end; 589 590 /* 591 * Add physical memory segments corresponding to the available 592 * physical pages. 593 */ 594 for (i = 0; phys_avail[i + 1] != 0; i += 2) 595 vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]); 596 597 /* 598 * Clear all of the page structures 599 */ 600 bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); 601 for (i = 0; i < page_range; i++) 602 vm_page_array[i].order = VM_NFREEORDER; 603 vm_page_array_size = page_range; 604 605 /* 606 * Initialize the physical memory allocator. 607 */ 608 vm_phys_init(); 609 610 /* 611 * Add every available physical page that is not blacklisted to 612 * the free lists. 613 */ 614 vm_cnt.v_page_count = 0; 615 vm_cnt.v_free_count = 0; 616 for (i = 0; phys_avail[i + 1] != 0; i += 2) { 617 pa = phys_avail[i]; 618 last_pa = phys_avail[i + 1]; 619 while (pa < last_pa) { 620 vm_phys_add_page(pa); 621 pa += PAGE_SIZE; 622 } 623 } 624 625 TAILQ_INIT(&blacklist_head); 626 vm_page_blacklist_load(&list, &listend); 627 vm_page_blacklist_check(list, listend); 628 629 list = kern_getenv("vm.blacklist"); 630 vm_page_blacklist_check(list, NULL); 631 632 freeenv(list); 633 #if VM_NRESERVLEVEL > 0 634 /* 635 * Initialize the reservation management system. 636 */ 637 vm_reserv_init(); 638 #endif 639 return (vaddr); 640 } 641 642 void 643 vm_page_reference(vm_page_t m) 644 { 645 646 vm_page_aflag_set(m, PGA_REFERENCED); 647 } 648 649 /* 650 * vm_page_busy_downgrade: 651 * 652 * Downgrade an exclusive busy page into a single shared busy page. 653 */ 654 void 655 vm_page_busy_downgrade(vm_page_t m) 656 { 657 u_int x; 658 659 vm_page_assert_xbusied(m); 660 661 for (;;) { 662 x = m->busy_lock; 663 x &= VPB_BIT_WAITERS; 664 if (atomic_cmpset_rel_int(&m->busy_lock, 665 VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x)) 666 break; 667 } 668 } 669 670 /* 671 * vm_page_sbusied: 672 * 673 * Return a positive value if the page is shared busied, 0 otherwise. 674 */ 675 int 676 vm_page_sbusied(vm_page_t m) 677 { 678 u_int x; 679 680 x = m->busy_lock; 681 return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED); 682 } 683 684 /* 685 * vm_page_sunbusy: 686 * 687 * Shared unbusy a page. 688 */ 689 void 690 vm_page_sunbusy(vm_page_t m) 691 { 692 u_int x; 693 694 vm_page_assert_sbusied(m); 695 696 for (;;) { 697 x = m->busy_lock; 698 if (VPB_SHARERS(x) > 1) { 699 if (atomic_cmpset_int(&m->busy_lock, x, 700 x - VPB_ONE_SHARER)) 701 break; 702 continue; 703 } 704 if ((x & VPB_BIT_WAITERS) == 0) { 705 KASSERT(x == VPB_SHARERS_WORD(1), 706 ("vm_page_sunbusy: invalid lock state")); 707 if (atomic_cmpset_int(&m->busy_lock, 708 VPB_SHARERS_WORD(1), VPB_UNBUSIED)) 709 break; 710 continue; 711 } 712 KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS), 713 ("vm_page_sunbusy: invalid lock state for waiters")); 714 715 vm_page_lock(m); 716 if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) { 717 vm_page_unlock(m); 718 continue; 719 } 720 wakeup(m); 721 vm_page_unlock(m); 722 break; 723 } 724 } 725 726 /* 727 * vm_page_busy_sleep: 728 * 729 * Sleep and release the page lock, using the page pointer as wchan. 730 * This is used to implement the hard-path of busying mechanism. 731 * 732 * The given page must be locked. 733 */ 734 void 735 vm_page_busy_sleep(vm_page_t m, const char *wmesg) 736 { 737 u_int x; 738 739 vm_page_lock_assert(m, MA_OWNED); 740 741 x = m->busy_lock; 742 if (x == VPB_UNBUSIED) { 743 vm_page_unlock(m); 744 return; 745 } 746 if ((x & VPB_BIT_WAITERS) == 0 && 747 !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) { 748 vm_page_unlock(m); 749 return; 750 } 751 msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0); 752 } 753 754 /* 755 * vm_page_trysbusy: 756 * 757 * Try to shared busy a page. 758 * If the operation succeeds 1 is returned otherwise 0. 759 * The operation never sleeps. 760 */ 761 int 762 vm_page_trysbusy(vm_page_t m) 763 { 764 u_int x; 765 766 for (;;) { 767 x = m->busy_lock; 768 if ((x & VPB_BIT_SHARED) == 0) 769 return (0); 770 if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER)) 771 return (1); 772 } 773 } 774 775 static void 776 vm_page_xunbusy_locked(vm_page_t m) 777 { 778 779 vm_page_assert_xbusied(m); 780 vm_page_assert_locked(m); 781 782 atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED); 783 /* There is a waiter, do wakeup() instead of vm_page_flash(). */ 784 wakeup(m); 785 } 786 787 static void 788 vm_page_xunbusy_maybelocked(vm_page_t m) 789 { 790 bool lockacq; 791 792 vm_page_assert_xbusied(m); 793 794 /* 795 * Fast path for unbusy. If it succeeds, we know that there 796 * are no waiters, so we do not need a wakeup. 797 */ 798 if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER, 799 VPB_UNBUSIED)) 800 return; 801 802 lockacq = !mtx_owned(vm_page_lockptr(m)); 803 if (lockacq) 804 vm_page_lock(m); 805 vm_page_xunbusy_locked(m); 806 if (lockacq) 807 vm_page_unlock(m); 808 } 809 810 /* 811 * vm_page_xunbusy_hard: 812 * 813 * Called after the first try the exclusive unbusy of a page failed. 814 * It is assumed that the waiters bit is on. 815 */ 816 void 817 vm_page_xunbusy_hard(vm_page_t m) 818 { 819 820 vm_page_assert_xbusied(m); 821 822 vm_page_lock(m); 823 vm_page_xunbusy_locked(m); 824 vm_page_unlock(m); 825 } 826 827 /* 828 * vm_page_flash: 829 * 830 * Wakeup anyone waiting for the page. 831 * The ownership bits do not change. 832 * 833 * The given page must be locked. 834 */ 835 void 836 vm_page_flash(vm_page_t m) 837 { 838 u_int x; 839 840 vm_page_lock_assert(m, MA_OWNED); 841 842 for (;;) { 843 x = m->busy_lock; 844 if ((x & VPB_BIT_WAITERS) == 0) 845 return; 846 if (atomic_cmpset_int(&m->busy_lock, x, 847 x & (~VPB_BIT_WAITERS))) 848 break; 849 } 850 wakeup(m); 851 } 852 853 /* 854 * Keep page from being freed by the page daemon 855 * much of the same effect as wiring, except much lower 856 * overhead and should be used only for *very* temporary 857 * holding ("wiring"). 858 */ 859 void 860 vm_page_hold(vm_page_t mem) 861 { 862 863 vm_page_lock_assert(mem, MA_OWNED); 864 mem->hold_count++; 865 } 866 867 void 868 vm_page_unhold(vm_page_t mem) 869 { 870 871 vm_page_lock_assert(mem, MA_OWNED); 872 KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!")); 873 --mem->hold_count; 874 if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0) 875 vm_page_free_toq(mem); 876 } 877 878 /* 879 * vm_page_unhold_pages: 880 * 881 * Unhold each of the pages that is referenced by the given array. 882 */ 883 void 884 vm_page_unhold_pages(vm_page_t *ma, int count) 885 { 886 struct mtx *mtx, *new_mtx; 887 888 mtx = NULL; 889 for (; count != 0; count--) { 890 /* 891 * Avoid releasing and reacquiring the same page lock. 892 */ 893 new_mtx = vm_page_lockptr(*ma); 894 if (mtx != new_mtx) { 895 if (mtx != NULL) 896 mtx_unlock(mtx); 897 mtx = new_mtx; 898 mtx_lock(mtx); 899 } 900 vm_page_unhold(*ma); 901 ma++; 902 } 903 if (mtx != NULL) 904 mtx_unlock(mtx); 905 } 906 907 vm_page_t 908 PHYS_TO_VM_PAGE(vm_paddr_t pa) 909 { 910 vm_page_t m; 911 912 #ifdef VM_PHYSSEG_SPARSE 913 m = vm_phys_paddr_to_vm_page(pa); 914 if (m == NULL) 915 m = vm_phys_fictitious_to_vm_page(pa); 916 return (m); 917 #elif defined(VM_PHYSSEG_DENSE) 918 long pi; 919 920 pi = atop(pa); 921 if (pi >= first_page && (pi - first_page) < vm_page_array_size) { 922 m = &vm_page_array[pi - first_page]; 923 return (m); 924 } 925 return (vm_phys_fictitious_to_vm_page(pa)); 926 #else 927 #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." 928 #endif 929 } 930 931 /* 932 * vm_page_getfake: 933 * 934 * Create a fictitious page with the specified physical address and 935 * memory attribute. The memory attribute is the only the machine- 936 * dependent aspect of a fictitious page that must be initialized. 937 */ 938 vm_page_t 939 vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr) 940 { 941 vm_page_t m; 942 943 m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO); 944 vm_page_initfake(m, paddr, memattr); 945 return (m); 946 } 947 948 void 949 vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) 950 { 951 952 if ((m->flags & PG_FICTITIOUS) != 0) { 953 /* 954 * The page's memattr might have changed since the 955 * previous initialization. Update the pmap to the 956 * new memattr. 957 */ 958 goto memattr; 959 } 960 m->phys_addr = paddr; 961 m->queue = PQ_NONE; 962 /* Fictitious pages don't use "segind". */ 963 m->flags = PG_FICTITIOUS; 964 /* Fictitious pages don't use "order" or "pool". */ 965 m->oflags = VPO_UNMANAGED; 966 m->busy_lock = VPB_SINGLE_EXCLUSIVER; 967 m->wire_count = 1; 968 pmap_page_init(m); 969 memattr: 970 pmap_page_set_memattr(m, memattr); 971 } 972 973 /* 974 * vm_page_putfake: 975 * 976 * Release a fictitious page. 977 */ 978 void 979 vm_page_putfake(vm_page_t m) 980 { 981 982 KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m)); 983 KASSERT((m->flags & PG_FICTITIOUS) != 0, 984 ("vm_page_putfake: bad page %p", m)); 985 uma_zfree(fakepg_zone, m); 986 } 987 988 /* 989 * vm_page_updatefake: 990 * 991 * Update the given fictitious page to the specified physical address and 992 * memory attribute. 993 */ 994 void 995 vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) 996 { 997 998 KASSERT((m->flags & PG_FICTITIOUS) != 0, 999 ("vm_page_updatefake: bad page %p", m)); 1000 m->phys_addr = paddr; 1001 pmap_page_set_memattr(m, memattr); 1002 } 1003 1004 /* 1005 * vm_page_free: 1006 * 1007 * Free a page. 1008 */ 1009 void 1010 vm_page_free(vm_page_t m) 1011 { 1012 1013 m->flags &= ~PG_ZERO; 1014 vm_page_free_toq(m); 1015 } 1016 1017 /* 1018 * vm_page_free_zero: 1019 * 1020 * Free a page to the zerod-pages queue 1021 */ 1022 void 1023 vm_page_free_zero(vm_page_t m) 1024 { 1025 1026 m->flags |= PG_ZERO; 1027 vm_page_free_toq(m); 1028 } 1029 1030 /* 1031 * Unbusy and handle the page queueing for a page from a getpages request that 1032 * was optionally read ahead or behind. 1033 */ 1034 void 1035 vm_page_readahead_finish(vm_page_t m) 1036 { 1037 1038 /* We shouldn't put invalid pages on queues. */ 1039 KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m)); 1040 1041 /* 1042 * Since the page is not the actually needed one, whether it should 1043 * be activated or deactivated is not obvious. Empirical results 1044 * have shown that deactivating the page is usually the best choice, 1045 * unless the page is wanted by another thread. 1046 */ 1047 vm_page_lock(m); 1048 if ((m->busy_lock & VPB_BIT_WAITERS) != 0) 1049 vm_page_activate(m); 1050 else 1051 vm_page_deactivate(m); 1052 vm_page_unlock(m); 1053 vm_page_xunbusy(m); 1054 } 1055 1056 /* 1057 * vm_page_sleep_if_busy: 1058 * 1059 * Sleep and release the page queues lock if the page is busied. 1060 * Returns TRUE if the thread slept. 1061 * 1062 * The given page must be unlocked and object containing it must 1063 * be locked. 1064 */ 1065 int 1066 vm_page_sleep_if_busy(vm_page_t m, const char *msg) 1067 { 1068 vm_object_t obj; 1069 1070 vm_page_lock_assert(m, MA_NOTOWNED); 1071 VM_OBJECT_ASSERT_WLOCKED(m->object); 1072 1073 if (vm_page_busied(m)) { 1074 /* 1075 * The page-specific object must be cached because page 1076 * identity can change during the sleep, causing the 1077 * re-lock of a different object. 1078 * It is assumed that a reference to the object is already 1079 * held by the callers. 1080 */ 1081 obj = m->object; 1082 vm_page_lock(m); 1083 VM_OBJECT_WUNLOCK(obj); 1084 vm_page_busy_sleep(m, msg); 1085 VM_OBJECT_WLOCK(obj); 1086 return (TRUE); 1087 } 1088 return (FALSE); 1089 } 1090 1091 /* 1092 * vm_page_dirty_KBI: [ internal use only ] 1093 * 1094 * Set all bits in the page's dirty field. 1095 * 1096 * The object containing the specified page must be locked if the 1097 * call is made from the machine-independent layer. 1098 * 1099 * See vm_page_clear_dirty_mask(). 1100 * 1101 * This function should only be called by vm_page_dirty(). 1102 */ 1103 void 1104 vm_page_dirty_KBI(vm_page_t m) 1105 { 1106 1107 /* These assertions refer to this operation by its public name. */ 1108 KASSERT((m->flags & PG_CACHED) == 0, 1109 ("vm_page_dirty: page in cache!")); 1110 KASSERT(m->valid == VM_PAGE_BITS_ALL, 1111 ("vm_page_dirty: page is invalid!")); 1112 m->dirty = VM_PAGE_BITS_ALL; 1113 } 1114 1115 /* 1116 * vm_page_insert: [ internal use only ] 1117 * 1118 * Inserts the given mem entry into the object and object list. 1119 * 1120 * The object must be locked. 1121 */ 1122 int 1123 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex) 1124 { 1125 vm_page_t mpred; 1126 1127 VM_OBJECT_ASSERT_WLOCKED(object); 1128 mpred = vm_radix_lookup_le(&object->rtree, pindex); 1129 return (vm_page_insert_after(m, object, pindex, mpred)); 1130 } 1131 1132 /* 1133 * vm_page_insert_after: 1134 * 1135 * Inserts the page "m" into the specified object at offset "pindex". 1136 * 1137 * The page "mpred" must immediately precede the offset "pindex" within 1138 * the specified object. 1139 * 1140 * The object must be locked. 1141 */ 1142 static int 1143 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, 1144 vm_page_t mpred) 1145 { 1146 vm_page_t msucc; 1147 1148 VM_OBJECT_ASSERT_WLOCKED(object); 1149 KASSERT(m->object == NULL, 1150 ("vm_page_insert_after: page already inserted")); 1151 if (mpred != NULL) { 1152 KASSERT(mpred->object == object, 1153 ("vm_page_insert_after: object doesn't contain mpred")); 1154 KASSERT(mpred->pindex < pindex, 1155 ("vm_page_insert_after: mpred doesn't precede pindex")); 1156 msucc = TAILQ_NEXT(mpred, listq); 1157 } else 1158 msucc = TAILQ_FIRST(&object->memq); 1159 if (msucc != NULL) 1160 KASSERT(msucc->pindex > pindex, 1161 ("vm_page_insert_after: msucc doesn't succeed pindex")); 1162 1163 /* 1164 * Record the object/offset pair in this page 1165 */ 1166 m->object = object; 1167 m->pindex = pindex; 1168 1169 /* 1170 * Now link into the object's ordered list of backed pages. 1171 */ 1172 if (vm_radix_insert(&object->rtree, m)) { 1173 m->object = NULL; 1174 m->pindex = 0; 1175 return (1); 1176 } 1177 vm_page_insert_radixdone(m, object, mpred); 1178 return (0); 1179 } 1180 1181 /* 1182 * vm_page_insert_radixdone: 1183 * 1184 * Complete page "m" insertion into the specified object after the 1185 * radix trie hooking. 1186 * 1187 * The page "mpred" must precede the offset "m->pindex" within the 1188 * specified object. 1189 * 1190 * The object must be locked. 1191 */ 1192 static void 1193 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred) 1194 { 1195 1196 VM_OBJECT_ASSERT_WLOCKED(object); 1197 KASSERT(object != NULL && m->object == object, 1198 ("vm_page_insert_radixdone: page %p has inconsistent object", m)); 1199 if (mpred != NULL) { 1200 KASSERT(mpred->object == object, 1201 ("vm_page_insert_after: object doesn't contain mpred")); 1202 KASSERT(mpred->pindex < m->pindex, 1203 ("vm_page_insert_after: mpred doesn't precede pindex")); 1204 } 1205 1206 if (mpred != NULL) 1207 TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq); 1208 else 1209 TAILQ_INSERT_HEAD(&object->memq, m, listq); 1210 1211 /* 1212 * Show that the object has one more resident page. 1213 */ 1214 object->resident_page_count++; 1215 1216 /* 1217 * Hold the vnode until the last page is released. 1218 */ 1219 if (object->resident_page_count == 1 && object->type == OBJT_VNODE) 1220 vhold(object->handle); 1221 1222 /* 1223 * Since we are inserting a new and possibly dirty page, 1224 * update the object's OBJ_MIGHTBEDIRTY flag. 1225 */ 1226 if (pmap_page_is_write_mapped(m)) 1227 vm_object_set_writeable_dirty(object); 1228 } 1229 1230 /* 1231 * vm_page_remove: 1232 * 1233 * Removes the given mem entry from the object/offset-page 1234 * table and the object page list, but do not invalidate/terminate 1235 * the backing store. 1236 * 1237 * The object must be locked. The page must be locked if it is managed. 1238 */ 1239 void 1240 vm_page_remove(vm_page_t m) 1241 { 1242 vm_object_t object; 1243 1244 if ((m->oflags & VPO_UNMANAGED) == 0) 1245 vm_page_assert_locked(m); 1246 if ((object = m->object) == NULL) 1247 return; 1248 VM_OBJECT_ASSERT_WLOCKED(object); 1249 if (vm_page_xbusied(m)) 1250 vm_page_xunbusy_maybelocked(m); 1251 1252 /* 1253 * Now remove from the object's list of backed pages. 1254 */ 1255 vm_radix_remove(&object->rtree, m->pindex); 1256 TAILQ_REMOVE(&object->memq, m, listq); 1257 1258 /* 1259 * And show that the object has one fewer resident page. 1260 */ 1261 object->resident_page_count--; 1262 1263 /* 1264 * The vnode may now be recycled. 1265 */ 1266 if (object->resident_page_count == 0 && object->type == OBJT_VNODE) 1267 vdrop(object->handle); 1268 1269 m->object = NULL; 1270 } 1271 1272 /* 1273 * vm_page_lookup: 1274 * 1275 * Returns the page associated with the object/offset 1276 * pair specified; if none is found, NULL is returned. 1277 * 1278 * The object must be locked. 1279 */ 1280 vm_page_t 1281 vm_page_lookup(vm_object_t object, vm_pindex_t pindex) 1282 { 1283 1284 VM_OBJECT_ASSERT_LOCKED(object); 1285 return (vm_radix_lookup(&object->rtree, pindex)); 1286 } 1287 1288 /* 1289 * vm_page_find_least: 1290 * 1291 * Returns the page associated with the object with least pindex 1292 * greater than or equal to the parameter pindex, or NULL. 1293 * 1294 * The object must be locked. 1295 */ 1296 vm_page_t 1297 vm_page_find_least(vm_object_t object, vm_pindex_t pindex) 1298 { 1299 vm_page_t m; 1300 1301 VM_OBJECT_ASSERT_LOCKED(object); 1302 if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex) 1303 m = vm_radix_lookup_ge(&object->rtree, pindex); 1304 return (m); 1305 } 1306 1307 /* 1308 * Returns the given page's successor (by pindex) within the object if it is 1309 * resident; if none is found, NULL is returned. 1310 * 1311 * The object must be locked. 1312 */ 1313 vm_page_t 1314 vm_page_next(vm_page_t m) 1315 { 1316 vm_page_t next; 1317 1318 VM_OBJECT_ASSERT_LOCKED(m->object); 1319 if ((next = TAILQ_NEXT(m, listq)) != NULL && 1320 next->pindex != m->pindex + 1) 1321 next = NULL; 1322 return (next); 1323 } 1324 1325 /* 1326 * Returns the given page's predecessor (by pindex) within the object if it is 1327 * resident; if none is found, NULL is returned. 1328 * 1329 * The object must be locked. 1330 */ 1331 vm_page_t 1332 vm_page_prev(vm_page_t m) 1333 { 1334 vm_page_t prev; 1335 1336 VM_OBJECT_ASSERT_LOCKED(m->object); 1337 if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL && 1338 prev->pindex != m->pindex - 1) 1339 prev = NULL; 1340 return (prev); 1341 } 1342 1343 /* 1344 * Uses the page mnew as a replacement for an existing page at index 1345 * pindex which must be already present in the object. 1346 * 1347 * The existing page must not be on a paging queue. 1348 */ 1349 vm_page_t 1350 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex) 1351 { 1352 vm_page_t mold; 1353 1354 VM_OBJECT_ASSERT_WLOCKED(object); 1355 KASSERT(mnew->object == NULL, 1356 ("vm_page_replace: page already in object")); 1357 1358 /* 1359 * This function mostly follows vm_page_insert() and 1360 * vm_page_remove() without the radix, object count and vnode 1361 * dance. Double check such functions for more comments. 1362 */ 1363 1364 mnew->object = object; 1365 mnew->pindex = pindex; 1366 mold = vm_radix_replace(&object->rtree, mnew); 1367 KASSERT(mold->queue == PQ_NONE, 1368 ("vm_page_replace: mold is on a paging queue")); 1369 1370 /* Keep the resident page list in sorted order. */ 1371 TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq); 1372 TAILQ_REMOVE(&object->memq, mold, listq); 1373 1374 mold->object = NULL; 1375 vm_page_xunbusy_maybelocked(mold); 1376 1377 /* 1378 * The object's resident_page_count does not change because we have 1379 * swapped one page for another, but OBJ_MIGHTBEDIRTY. 1380 */ 1381 if (pmap_page_is_write_mapped(mnew)) 1382 vm_object_set_writeable_dirty(object); 1383 return (mold); 1384 } 1385 1386 /* 1387 * vm_page_rename: 1388 * 1389 * Move the given memory entry from its 1390 * current object to the specified target object/offset. 1391 * 1392 * Note: swap associated with the page must be invalidated by the move. We 1393 * have to do this for several reasons: (1) we aren't freeing the 1394 * page, (2) we are dirtying the page, (3) the VM system is probably 1395 * moving the page from object A to B, and will then later move 1396 * the backing store from A to B and we can't have a conflict. 1397 * 1398 * Note: we *always* dirty the page. It is necessary both for the 1399 * fact that we moved it, and because we may be invalidating 1400 * swap. If the page is on the cache, we have to deactivate it 1401 * or vm_page_dirty() will panic. Dirty pages are not allowed 1402 * on the cache. 1403 * 1404 * The objects must be locked. 1405 */ 1406 int 1407 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex) 1408 { 1409 vm_page_t mpred; 1410 vm_pindex_t opidx; 1411 1412 VM_OBJECT_ASSERT_WLOCKED(new_object); 1413 1414 mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex); 1415 KASSERT(mpred == NULL || mpred->pindex != new_pindex, 1416 ("vm_page_rename: pindex already renamed")); 1417 1418 /* 1419 * Create a custom version of vm_page_insert() which does not depend 1420 * by m_prev and can cheat on the implementation aspects of the 1421 * function. 1422 */ 1423 opidx = m->pindex; 1424 m->pindex = new_pindex; 1425 if (vm_radix_insert(&new_object->rtree, m)) { 1426 m->pindex = opidx; 1427 return (1); 1428 } 1429 1430 /* 1431 * The operation cannot fail anymore. The removal must happen before 1432 * the listq iterator is tainted. 1433 */ 1434 m->pindex = opidx; 1435 vm_page_lock(m); 1436 vm_page_remove(m); 1437 1438 /* Return back to the new pindex to complete vm_page_insert(). */ 1439 m->pindex = new_pindex; 1440 m->object = new_object; 1441 vm_page_unlock(m); 1442 vm_page_insert_radixdone(m, new_object, mpred); 1443 vm_page_dirty(m); 1444 return (0); 1445 } 1446 1447 /* 1448 * Convert all of the given object's cached pages that have a 1449 * pindex within the given range into free pages. If the value 1450 * zero is given for "end", then the range's upper bound is 1451 * infinity. If the given object is backed by a vnode and it 1452 * transitions from having one or more cached pages to none, the 1453 * vnode's hold count is reduced. 1454 */ 1455 void 1456 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 1457 { 1458 vm_page_t m; 1459 boolean_t empty; 1460 1461 mtx_lock(&vm_page_queue_free_mtx); 1462 if (__predict_false(vm_radix_is_empty(&object->cache))) { 1463 mtx_unlock(&vm_page_queue_free_mtx); 1464 return; 1465 } 1466 while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) { 1467 if (end != 0 && m->pindex >= end) 1468 break; 1469 vm_radix_remove(&object->cache, m->pindex); 1470 vm_page_cache_turn_free(m); 1471 } 1472 empty = vm_radix_is_empty(&object->cache); 1473 mtx_unlock(&vm_page_queue_free_mtx); 1474 if (object->type == OBJT_VNODE && empty) 1475 vdrop(object->handle); 1476 } 1477 1478 /* 1479 * Returns the cached page that is associated with the given 1480 * object and offset. If, however, none exists, returns NULL. 1481 * 1482 * The free page queue must be locked. 1483 */ 1484 static inline vm_page_t 1485 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex) 1486 { 1487 1488 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1489 return (vm_radix_lookup(&object->cache, pindex)); 1490 } 1491 1492 /* 1493 * Remove the given cached page from its containing object's 1494 * collection of cached pages. 1495 * 1496 * The free page queue must be locked. 1497 */ 1498 static void 1499 vm_page_cache_remove(vm_page_t m) 1500 { 1501 1502 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1503 KASSERT((m->flags & PG_CACHED) != 0, 1504 ("vm_page_cache_remove: page %p is not cached", m)); 1505 vm_radix_remove(&m->object->cache, m->pindex); 1506 m->object = NULL; 1507 vm_cnt.v_cache_count--; 1508 } 1509 1510 /* 1511 * Transfer all of the cached pages with offset greater than or 1512 * equal to 'offidxstart' from the original object's cache to the 1513 * new object's cache. However, any cached pages with offset 1514 * greater than or equal to the new object's size are kept in the 1515 * original object. Initially, the new object's cache must be 1516 * empty. Offset 'offidxstart' in the original object must 1517 * correspond to offset zero in the new object. 1518 * 1519 * The new object must be locked. 1520 */ 1521 void 1522 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart, 1523 vm_object_t new_object) 1524 { 1525 vm_page_t m; 1526 1527 /* 1528 * Insertion into an object's collection of cached pages 1529 * requires the object to be locked. In contrast, removal does 1530 * not. 1531 */ 1532 VM_OBJECT_ASSERT_WLOCKED(new_object); 1533 KASSERT(vm_radix_is_empty(&new_object->cache), 1534 ("vm_page_cache_transfer: object %p has cached pages", 1535 new_object)); 1536 mtx_lock(&vm_page_queue_free_mtx); 1537 while ((m = vm_radix_lookup_ge(&orig_object->cache, 1538 offidxstart)) != NULL) { 1539 /* 1540 * Transfer all of the pages with offset greater than or 1541 * equal to 'offidxstart' from the original object's 1542 * cache to the new object's cache. 1543 */ 1544 if ((m->pindex - offidxstart) >= new_object->size) 1545 break; 1546 vm_radix_remove(&orig_object->cache, m->pindex); 1547 /* Update the page's object and offset. */ 1548 m->object = new_object; 1549 m->pindex -= offidxstart; 1550 if (vm_radix_insert(&new_object->cache, m)) 1551 vm_page_cache_turn_free(m); 1552 } 1553 mtx_unlock(&vm_page_queue_free_mtx); 1554 } 1555 1556 /* 1557 * Returns TRUE if a cached page is associated with the given object and 1558 * offset, and FALSE otherwise. 1559 * 1560 * The object must be locked. 1561 */ 1562 boolean_t 1563 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex) 1564 { 1565 vm_page_t m; 1566 1567 /* 1568 * Insertion into an object's collection of cached pages requires the 1569 * object to be locked. Therefore, if the object is locked and the 1570 * object's collection is empty, there is no need to acquire the free 1571 * page queues lock in order to prove that the specified page doesn't 1572 * exist. 1573 */ 1574 VM_OBJECT_ASSERT_WLOCKED(object); 1575 if (__predict_true(vm_object_cache_is_empty(object))) 1576 return (FALSE); 1577 mtx_lock(&vm_page_queue_free_mtx); 1578 m = vm_page_cache_lookup(object, pindex); 1579 mtx_unlock(&vm_page_queue_free_mtx); 1580 return (m != NULL); 1581 } 1582 1583 /* 1584 * vm_page_alloc: 1585 * 1586 * Allocate and return a page that is associated with the specified 1587 * object and offset pair. By default, this page is exclusive busied. 1588 * 1589 * The caller must always specify an allocation class. 1590 * 1591 * allocation classes: 1592 * VM_ALLOC_NORMAL normal process request 1593 * VM_ALLOC_SYSTEM system *really* needs a page 1594 * VM_ALLOC_INTERRUPT interrupt time request 1595 * 1596 * optional allocation flags: 1597 * VM_ALLOC_COUNT(number) the number of additional pages that the caller 1598 * intends to allocate 1599 * VM_ALLOC_IFCACHED return page only if it is cached 1600 * VM_ALLOC_IFNOTCACHED return NULL, do not reactivate if the page 1601 * is cached 1602 * VM_ALLOC_NOBUSY do not exclusive busy the page 1603 * VM_ALLOC_NODUMP do not include the page in a kernel core dump 1604 * VM_ALLOC_NOOBJ page is not associated with an object and 1605 * should not be exclusive busy 1606 * VM_ALLOC_SBUSY shared busy the allocated page 1607 * VM_ALLOC_WIRED wire the allocated page 1608 * VM_ALLOC_ZERO prefer a zeroed page 1609 * 1610 * This routine may not sleep. 1611 */ 1612 vm_page_t 1613 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req) 1614 { 1615 struct vnode *vp = NULL; 1616 vm_object_t m_object; 1617 vm_page_t m, mpred; 1618 int flags, req_class; 1619 1620 mpred = 0; /* XXX: pacify gcc */ 1621 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && 1622 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && 1623 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != 1624 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), 1625 ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object, 1626 req)); 1627 if (object != NULL) 1628 VM_OBJECT_ASSERT_WLOCKED(object); 1629 1630 req_class = req & VM_ALLOC_CLASS_MASK; 1631 1632 /* 1633 * The page daemon is allowed to dig deeper into the free page list. 1634 */ 1635 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) 1636 req_class = VM_ALLOC_SYSTEM; 1637 1638 if (object != NULL) { 1639 mpred = vm_radix_lookup_le(&object->rtree, pindex); 1640 KASSERT(mpred == NULL || mpred->pindex != pindex, 1641 ("vm_page_alloc: pindex already allocated")); 1642 } 1643 1644 /* 1645 * The page allocation request can came from consumers which already 1646 * hold the free page queue mutex, like vm_page_insert() in 1647 * vm_page_cache(). 1648 */ 1649 mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE); 1650 if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved || 1651 (req_class == VM_ALLOC_SYSTEM && 1652 vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) || 1653 (req_class == VM_ALLOC_INTERRUPT && 1654 vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) { 1655 /* 1656 * Allocate from the free queue if the number of free pages 1657 * exceeds the minimum for the request class. 1658 */ 1659 if (object != NULL && 1660 (m = vm_page_cache_lookup(object, pindex)) != NULL) { 1661 if ((req & VM_ALLOC_IFNOTCACHED) != 0) { 1662 mtx_unlock(&vm_page_queue_free_mtx); 1663 return (NULL); 1664 } 1665 if (vm_phys_unfree_page(m)) 1666 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0); 1667 #if VM_NRESERVLEVEL > 0 1668 else if (!vm_reserv_reactivate_page(m)) 1669 #else 1670 else 1671 #endif 1672 panic("vm_page_alloc: cache page %p is missing" 1673 " from the free queue", m); 1674 } else if ((req & VM_ALLOC_IFCACHED) != 0) { 1675 mtx_unlock(&vm_page_queue_free_mtx); 1676 return (NULL); 1677 #if VM_NRESERVLEVEL > 0 1678 } else if (object == NULL || (object->flags & (OBJ_COLORED | 1679 OBJ_FICTITIOUS)) != OBJ_COLORED || (m = 1680 vm_reserv_alloc_page(object, pindex, mpred)) == NULL) { 1681 #else 1682 } else { 1683 #endif 1684 m = vm_phys_alloc_pages(object != NULL ? 1685 VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0); 1686 #if VM_NRESERVLEVEL > 0 1687 if (m == NULL && vm_reserv_reclaim_inactive()) { 1688 m = vm_phys_alloc_pages(object != NULL ? 1689 VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 1690 0); 1691 } 1692 #endif 1693 } 1694 } else { 1695 /* 1696 * Not allocatable, give up. 1697 */ 1698 mtx_unlock(&vm_page_queue_free_mtx); 1699 atomic_add_int(&vm_pageout_deficit, 1700 max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); 1701 pagedaemon_wakeup(); 1702 return (NULL); 1703 } 1704 1705 /* 1706 * At this point we had better have found a good page. 1707 */ 1708 KASSERT(m != NULL, ("vm_page_alloc: missing page")); 1709 KASSERT(m->queue == PQ_NONE, 1710 ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue)); 1711 KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m)); 1712 KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m)); 1713 KASSERT(!vm_page_busied(m), ("vm_page_alloc: page %p is busy", m)); 1714 KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m)); 1715 KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, 1716 ("vm_page_alloc: page %p has unexpected memattr %d", m, 1717 pmap_page_get_memattr(m))); 1718 if ((m->flags & PG_CACHED) != 0) { 1719 KASSERT((m->flags & PG_ZERO) == 0, 1720 ("vm_page_alloc: cached page %p is PG_ZERO", m)); 1721 KASSERT(m->valid != 0, 1722 ("vm_page_alloc: cached page %p is invalid", m)); 1723 if (m->object == object && m->pindex == pindex) 1724 vm_cnt.v_reactivated++; 1725 else 1726 m->valid = 0; 1727 m_object = m->object; 1728 vm_page_cache_remove(m); 1729 if (m_object->type == OBJT_VNODE && 1730 vm_object_cache_is_empty(m_object)) 1731 vp = m_object->handle; 1732 } else { 1733 KASSERT(m->valid == 0, 1734 ("vm_page_alloc: free page %p is valid", m)); 1735 vm_phys_freecnt_adj(m, -1); 1736 } 1737 mtx_unlock(&vm_page_queue_free_mtx); 1738 1739 /* 1740 * Initialize the page. Only the PG_ZERO flag is inherited. 1741 */ 1742 flags = 0; 1743 if ((req & VM_ALLOC_ZERO) != 0) 1744 flags = PG_ZERO; 1745 flags &= m->flags; 1746 if ((req & VM_ALLOC_NODUMP) != 0) 1747 flags |= PG_NODUMP; 1748 m->flags = flags; 1749 m->aflags = 0; 1750 m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ? 1751 VPO_UNMANAGED : 0; 1752 m->busy_lock = VPB_UNBUSIED; 1753 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0) 1754 m->busy_lock = VPB_SINGLE_EXCLUSIVER; 1755 if ((req & VM_ALLOC_SBUSY) != 0) 1756 m->busy_lock = VPB_SHARERS_WORD(1); 1757 if (req & VM_ALLOC_WIRED) { 1758 /* 1759 * The page lock is not required for wiring a page until that 1760 * page is inserted into the object. 1761 */ 1762 atomic_add_int(&vm_cnt.v_wire_count, 1); 1763 m->wire_count = 1; 1764 } 1765 m->act_count = 0; 1766 1767 if (object != NULL) { 1768 if (vm_page_insert_after(m, object, pindex, mpred)) { 1769 /* See the comment below about hold count. */ 1770 if (vp != NULL) 1771 vdrop(vp); 1772 pagedaemon_wakeup(); 1773 if (req & VM_ALLOC_WIRED) { 1774 atomic_subtract_int(&vm_cnt.v_wire_count, 1); 1775 m->wire_count = 0; 1776 } 1777 m->object = NULL; 1778 m->oflags = VPO_UNMANAGED; 1779 m->busy_lock = VPB_UNBUSIED; 1780 vm_page_free(m); 1781 return (NULL); 1782 } 1783 1784 /* Ignore device objects; the pager sets "memattr" for them. */ 1785 if (object->memattr != VM_MEMATTR_DEFAULT && 1786 (object->flags & OBJ_FICTITIOUS) == 0) 1787 pmap_page_set_memattr(m, object->memattr); 1788 } else 1789 m->pindex = pindex; 1790 1791 /* 1792 * The following call to vdrop() must come after the above call 1793 * to vm_page_insert() in case both affect the same object and 1794 * vnode. Otherwise, the affected vnode's hold count could 1795 * temporarily become zero. 1796 */ 1797 if (vp != NULL) 1798 vdrop(vp); 1799 1800 /* 1801 * Don't wakeup too often - wakeup the pageout daemon when 1802 * we would be nearly out of memory. 1803 */ 1804 if (vm_paging_needed()) 1805 pagedaemon_wakeup(); 1806 1807 return (m); 1808 } 1809 1810 static void 1811 vm_page_alloc_contig_vdrop(struct spglist *lst) 1812 { 1813 1814 while (!SLIST_EMPTY(lst)) { 1815 vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv); 1816 SLIST_REMOVE_HEAD(lst, plinks.s.ss); 1817 } 1818 } 1819 1820 /* 1821 * vm_page_alloc_contig: 1822 * 1823 * Allocate a contiguous set of physical pages of the given size "npages" 1824 * from the free lists. All of the physical pages must be at or above 1825 * the given physical address "low" and below the given physical address 1826 * "high". The given value "alignment" determines the alignment of the 1827 * first physical page in the set. If the given value "boundary" is 1828 * non-zero, then the set of physical pages cannot cross any physical 1829 * address boundary that is a multiple of that value. Both "alignment" 1830 * and "boundary" must be a power of two. 1831 * 1832 * If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT, 1833 * then the memory attribute setting for the physical pages is configured 1834 * to the object's memory attribute setting. Otherwise, the memory 1835 * attribute setting for the physical pages is configured to "memattr", 1836 * overriding the object's memory attribute setting. However, if the 1837 * object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the 1838 * memory attribute setting for the physical pages cannot be configured 1839 * to VM_MEMATTR_DEFAULT. 1840 * 1841 * The caller must always specify an allocation class. 1842 * 1843 * allocation classes: 1844 * VM_ALLOC_NORMAL normal process request 1845 * VM_ALLOC_SYSTEM system *really* needs a page 1846 * VM_ALLOC_INTERRUPT interrupt time request 1847 * 1848 * optional allocation flags: 1849 * VM_ALLOC_NOBUSY do not exclusive busy the page 1850 * VM_ALLOC_NODUMP do not include the page in a kernel core dump 1851 * VM_ALLOC_NOOBJ page is not associated with an object and 1852 * should not be exclusive busy 1853 * VM_ALLOC_SBUSY shared busy the allocated page 1854 * VM_ALLOC_WIRED wire the allocated page 1855 * VM_ALLOC_ZERO prefer a zeroed page 1856 * 1857 * This routine may not sleep. 1858 */ 1859 vm_page_t 1860 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, 1861 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, 1862 vm_paddr_t boundary, vm_memattr_t memattr) 1863 { 1864 struct vnode *drop; 1865 struct spglist deferred_vdrop_list; 1866 vm_page_t m, m_tmp, m_ret; 1867 u_int flags; 1868 int req_class; 1869 1870 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && 1871 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && 1872 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != 1873 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), 1874 ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object, 1875 req)); 1876 if (object != NULL) { 1877 VM_OBJECT_ASSERT_WLOCKED(object); 1878 KASSERT(object->type == OBJT_PHYS, 1879 ("vm_page_alloc_contig: object %p isn't OBJT_PHYS", 1880 object)); 1881 } 1882 KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero")); 1883 req_class = req & VM_ALLOC_CLASS_MASK; 1884 1885 /* 1886 * The page daemon is allowed to dig deeper into the free page list. 1887 */ 1888 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) 1889 req_class = VM_ALLOC_SYSTEM; 1890 1891 SLIST_INIT(&deferred_vdrop_list); 1892 mtx_lock(&vm_page_queue_free_mtx); 1893 if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages + 1894 vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && 1895 vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages + 1896 vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && 1897 vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) { 1898 #if VM_NRESERVLEVEL > 0 1899 retry: 1900 if (object == NULL || (object->flags & OBJ_COLORED) == 0 || 1901 (m_ret = vm_reserv_alloc_contig(object, pindex, npages, 1902 low, high, alignment, boundary)) == NULL) 1903 #endif 1904 m_ret = vm_phys_alloc_contig(npages, low, high, 1905 alignment, boundary); 1906 } else { 1907 mtx_unlock(&vm_page_queue_free_mtx); 1908 atomic_add_int(&vm_pageout_deficit, npages); 1909 pagedaemon_wakeup(); 1910 return (NULL); 1911 } 1912 if (m_ret != NULL) 1913 for (m = m_ret; m < &m_ret[npages]; m++) { 1914 drop = vm_page_alloc_init(m); 1915 if (drop != NULL) { 1916 /* 1917 * Enqueue the vnode for deferred vdrop(). 1918 */ 1919 m->plinks.s.pv = drop; 1920 SLIST_INSERT_HEAD(&deferred_vdrop_list, m, 1921 plinks.s.ss); 1922 } 1923 } 1924 else { 1925 #if VM_NRESERVLEVEL > 0 1926 if (vm_reserv_reclaim_contig(npages, low, high, alignment, 1927 boundary)) 1928 goto retry; 1929 #endif 1930 } 1931 mtx_unlock(&vm_page_queue_free_mtx); 1932 if (m_ret == NULL) 1933 return (NULL); 1934 1935 /* 1936 * Initialize the pages. Only the PG_ZERO flag is inherited. 1937 */ 1938 flags = 0; 1939 if ((req & VM_ALLOC_ZERO) != 0) 1940 flags = PG_ZERO; 1941 if ((req & VM_ALLOC_NODUMP) != 0) 1942 flags |= PG_NODUMP; 1943 if ((req & VM_ALLOC_WIRED) != 0) 1944 atomic_add_int(&vm_cnt.v_wire_count, npages); 1945 if (object != NULL) { 1946 if (object->memattr != VM_MEMATTR_DEFAULT && 1947 memattr == VM_MEMATTR_DEFAULT) 1948 memattr = object->memattr; 1949 } 1950 for (m = m_ret; m < &m_ret[npages]; m++) { 1951 m->aflags = 0; 1952 m->flags = (m->flags | PG_NODUMP) & flags; 1953 m->busy_lock = VPB_UNBUSIED; 1954 if (object != NULL) { 1955 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) 1956 m->busy_lock = VPB_SINGLE_EXCLUSIVER; 1957 if ((req & VM_ALLOC_SBUSY) != 0) 1958 m->busy_lock = VPB_SHARERS_WORD(1); 1959 } 1960 if ((req & VM_ALLOC_WIRED) != 0) 1961 m->wire_count = 1; 1962 /* Unmanaged pages don't use "act_count". */ 1963 m->oflags = VPO_UNMANAGED; 1964 if (object != NULL) { 1965 if (vm_page_insert(m, object, pindex)) { 1966 vm_page_alloc_contig_vdrop( 1967 &deferred_vdrop_list); 1968 if (vm_paging_needed()) 1969 pagedaemon_wakeup(); 1970 if ((req & VM_ALLOC_WIRED) != 0) 1971 atomic_subtract_int(&vm_cnt.v_wire_count, 1972 npages); 1973 for (m_tmp = m, m = m_ret; 1974 m < &m_ret[npages]; m++) { 1975 if ((req & VM_ALLOC_WIRED) != 0) 1976 m->wire_count = 0; 1977 if (m >= m_tmp) { 1978 m->object = NULL; 1979 m->oflags |= VPO_UNMANAGED; 1980 } 1981 m->busy_lock = VPB_UNBUSIED; 1982 vm_page_free(m); 1983 } 1984 return (NULL); 1985 } 1986 } else 1987 m->pindex = pindex; 1988 if (memattr != VM_MEMATTR_DEFAULT) 1989 pmap_page_set_memattr(m, memattr); 1990 pindex++; 1991 } 1992 vm_page_alloc_contig_vdrop(&deferred_vdrop_list); 1993 if (vm_paging_needed()) 1994 pagedaemon_wakeup(); 1995 return (m_ret); 1996 } 1997 1998 /* 1999 * Initialize a page that has been freshly dequeued from a freelist. 2000 * The caller has to drop the vnode returned, if it is not NULL. 2001 * 2002 * This function may only be used to initialize unmanaged pages. 2003 * 2004 * To be called with vm_page_queue_free_mtx held. 2005 */ 2006 static struct vnode * 2007 vm_page_alloc_init(vm_page_t m) 2008 { 2009 struct vnode *drop; 2010 vm_object_t m_object; 2011 2012 KASSERT(m->queue == PQ_NONE, 2013 ("vm_page_alloc_init: page %p has unexpected queue %d", 2014 m, m->queue)); 2015 KASSERT(m->wire_count == 0, 2016 ("vm_page_alloc_init: page %p is wired", m)); 2017 KASSERT(m->hold_count == 0, 2018 ("vm_page_alloc_init: page %p is held", m)); 2019 KASSERT(!vm_page_busied(m), 2020 ("vm_page_alloc_init: page %p is busy", m)); 2021 KASSERT(m->dirty == 0, 2022 ("vm_page_alloc_init: page %p is dirty", m)); 2023 KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, 2024 ("vm_page_alloc_init: page %p has unexpected memattr %d", 2025 m, pmap_page_get_memattr(m))); 2026 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 2027 drop = NULL; 2028 if ((m->flags & PG_CACHED) != 0) { 2029 KASSERT((m->flags & PG_ZERO) == 0, 2030 ("vm_page_alloc_init: cached page %p is PG_ZERO", m)); 2031 m->valid = 0; 2032 m_object = m->object; 2033 vm_page_cache_remove(m); 2034 if (m_object->type == OBJT_VNODE && 2035 vm_object_cache_is_empty(m_object)) 2036 drop = m_object->handle; 2037 } else { 2038 KASSERT(m->valid == 0, 2039 ("vm_page_alloc_init: free page %p is valid", m)); 2040 vm_phys_freecnt_adj(m, -1); 2041 } 2042 return (drop); 2043 } 2044 2045 /* 2046 * vm_page_alloc_freelist: 2047 * 2048 * Allocate a physical page from the specified free page list. 2049 * 2050 * The caller must always specify an allocation class. 2051 * 2052 * allocation classes: 2053 * VM_ALLOC_NORMAL normal process request 2054 * VM_ALLOC_SYSTEM system *really* needs a page 2055 * VM_ALLOC_INTERRUPT interrupt time request 2056 * 2057 * optional allocation flags: 2058 * VM_ALLOC_COUNT(number) the number of additional pages that the caller 2059 * intends to allocate 2060 * VM_ALLOC_WIRED wire the allocated page 2061 * VM_ALLOC_ZERO prefer a zeroed page 2062 * 2063 * This routine may not sleep. 2064 */ 2065 vm_page_t 2066 vm_page_alloc_freelist(int flind, int req) 2067 { 2068 struct vnode *drop; 2069 vm_page_t m; 2070 u_int flags; 2071 int req_class; 2072 2073 req_class = req & VM_ALLOC_CLASS_MASK; 2074 2075 /* 2076 * The page daemon is allowed to dig deeper into the free page list. 2077 */ 2078 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) 2079 req_class = VM_ALLOC_SYSTEM; 2080 2081 /* 2082 * Do not allocate reserved pages unless the req has asked for it. 2083 */ 2084 mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE); 2085 if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved || 2086 (req_class == VM_ALLOC_SYSTEM && 2087 vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) || 2088 (req_class == VM_ALLOC_INTERRUPT && 2089 vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) 2090 m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0); 2091 else { 2092 mtx_unlock(&vm_page_queue_free_mtx); 2093 atomic_add_int(&vm_pageout_deficit, 2094 max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); 2095 pagedaemon_wakeup(); 2096 return (NULL); 2097 } 2098 if (m == NULL) { 2099 mtx_unlock(&vm_page_queue_free_mtx); 2100 return (NULL); 2101 } 2102 drop = vm_page_alloc_init(m); 2103 mtx_unlock(&vm_page_queue_free_mtx); 2104 2105 /* 2106 * Initialize the page. Only the PG_ZERO flag is inherited. 2107 */ 2108 m->aflags = 0; 2109 flags = 0; 2110 if ((req & VM_ALLOC_ZERO) != 0) 2111 flags = PG_ZERO; 2112 m->flags &= flags; 2113 if ((req & VM_ALLOC_WIRED) != 0) { 2114 /* 2115 * The page lock is not required for wiring a page that does 2116 * not belong to an object. 2117 */ 2118 atomic_add_int(&vm_cnt.v_wire_count, 1); 2119 m->wire_count = 1; 2120 } 2121 /* Unmanaged pages don't use "act_count". */ 2122 m->oflags = VPO_UNMANAGED; 2123 if (drop != NULL) 2124 vdrop(drop); 2125 if (vm_paging_needed()) 2126 pagedaemon_wakeup(); 2127 return (m); 2128 } 2129 2130 #define VPSC_ANY 0 /* No restrictions. */ 2131 #define VPSC_NORESERV 1 /* Skip reservations; implies VPSC_NOSUPER. */ 2132 #define VPSC_NOSUPER 2 /* Skip superpages. */ 2133 2134 /* 2135 * vm_page_scan_contig: 2136 * 2137 * Scan vm_page_array[] between the specified entries "m_start" and 2138 * "m_end" for a run of contiguous physical pages that satisfy the 2139 * specified conditions, and return the lowest page in the run. The 2140 * specified "alignment" determines the alignment of the lowest physical 2141 * page in the run. If the specified "boundary" is non-zero, then the 2142 * run of physical pages cannot span a physical address that is a 2143 * multiple of "boundary". 2144 * 2145 * "m_end" is never dereferenced, so it need not point to a vm_page 2146 * structure within vm_page_array[]. 2147 * 2148 * "npages" must be greater than zero. "m_start" and "m_end" must not 2149 * span a hole (or discontiguity) in the physical address space. Both 2150 * "alignment" and "boundary" must be a power of two. 2151 */ 2152 vm_page_t 2153 vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end, 2154 u_long alignment, vm_paddr_t boundary, int options) 2155 { 2156 struct mtx *m_mtx, *new_mtx; 2157 vm_object_t object; 2158 vm_paddr_t pa; 2159 vm_page_t m, m_run; 2160 #if VM_NRESERVLEVEL > 0 2161 int level; 2162 #endif 2163 int m_inc, order, run_ext, run_len; 2164 2165 KASSERT(npages > 0, ("npages is 0")); 2166 KASSERT(powerof2(alignment), ("alignment is not a power of 2")); 2167 KASSERT(powerof2(boundary), ("boundary is not a power of 2")); 2168 m_run = NULL; 2169 run_len = 0; 2170 m_mtx = NULL; 2171 for (m = m_start; m < m_end && run_len < npages; m += m_inc) { 2172 KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0, 2173 ("page %p is PG_FICTITIOUS or PG_MARKER", m)); 2174 2175 /* 2176 * If the current page would be the start of a run, check its 2177 * physical address against the end, alignment, and boundary 2178 * conditions. If it doesn't satisfy these conditions, either 2179 * terminate the scan or advance to the next page that 2180 * satisfies the failed condition. 2181 */ 2182 if (run_len == 0) { 2183 KASSERT(m_run == NULL, ("m_run != NULL")); 2184 if (m + npages > m_end) 2185 break; 2186 pa = VM_PAGE_TO_PHYS(m); 2187 if ((pa & (alignment - 1)) != 0) { 2188 m_inc = atop(roundup2(pa, alignment) - pa); 2189 continue; 2190 } 2191 if (rounddown2(pa ^ (pa + ptoa(npages) - 1), 2192 boundary) != 0) { 2193 m_inc = atop(roundup2(pa, boundary) - pa); 2194 continue; 2195 } 2196 } else 2197 KASSERT(m_run != NULL, ("m_run == NULL")); 2198 2199 /* 2200 * Avoid releasing and reacquiring the same page lock. 2201 */ 2202 new_mtx = vm_page_lockptr(m); 2203 if (m_mtx != new_mtx) { 2204 if (m_mtx != NULL) 2205 mtx_unlock(m_mtx); 2206 m_mtx = new_mtx; 2207 mtx_lock(m_mtx); 2208 } 2209 m_inc = 1; 2210 retry: 2211 if (m->wire_count != 0 || m->hold_count != 0) 2212 run_ext = 0; 2213 #if VM_NRESERVLEVEL > 0 2214 else if ((level = vm_reserv_level(m)) >= 0 && 2215 (options & VPSC_NORESERV) != 0) { 2216 run_ext = 0; 2217 /* Advance to the end of the reservation. */ 2218 pa = VM_PAGE_TO_PHYS(m); 2219 m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) - 2220 pa); 2221 } 2222 #endif 2223 else if ((object = m->object) != NULL) { 2224 /* 2225 * The page is considered eligible for relocation if 2226 * and only if it could be laundered or reclaimed by 2227 * the page daemon. 2228 */ 2229 if (!VM_OBJECT_TRYRLOCK(object)) { 2230 mtx_unlock(m_mtx); 2231 VM_OBJECT_RLOCK(object); 2232 mtx_lock(m_mtx); 2233 if (m->object != object) { 2234 /* 2235 * The page may have been freed. 2236 */ 2237 VM_OBJECT_RUNLOCK(object); 2238 goto retry; 2239 } else if (m->wire_count != 0 || 2240 m->hold_count != 0) { 2241 run_ext = 0; 2242 goto unlock; 2243 } 2244 } 2245 KASSERT((m->flags & PG_UNHOLDFREE) == 0, 2246 ("page %p is PG_UNHOLDFREE", m)); 2247 /* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */ 2248 if (object->type != OBJT_DEFAULT && 2249 object->type != OBJT_SWAP && 2250 object->type != OBJT_VNODE) 2251 run_ext = 0; 2252 else if ((m->flags & PG_CACHED) != 0 || 2253 m != vm_page_lookup(object, m->pindex)) { 2254 /* 2255 * The page is cached or recently converted 2256 * from cached to free. 2257 */ 2258 #if VM_NRESERVLEVEL > 0 2259 if (level >= 0) { 2260 /* 2261 * The page is reserved. Extend the 2262 * current run by one page. 2263 */ 2264 run_ext = 1; 2265 } else 2266 #endif 2267 if ((order = m->order) < VM_NFREEORDER) { 2268 /* 2269 * The page is enqueued in the 2270 * physical memory allocator's cache/ 2271 * free page queues. Moreover, it is 2272 * the first page in a power-of-two- 2273 * sized run of contiguous cache/free 2274 * pages. Add these pages to the end 2275 * of the current run, and jump 2276 * ahead. 2277 */ 2278 run_ext = 1 << order; 2279 m_inc = 1 << order; 2280 } else 2281 run_ext = 0; 2282 #if VM_NRESERVLEVEL > 0 2283 } else if ((options & VPSC_NOSUPER) != 0 && 2284 (level = vm_reserv_level_iffullpop(m)) >= 0) { 2285 run_ext = 0; 2286 /* Advance to the end of the superpage. */ 2287 pa = VM_PAGE_TO_PHYS(m); 2288 m_inc = atop(roundup2(pa + 1, 2289 vm_reserv_size(level)) - pa); 2290 #endif 2291 } else if (object->memattr == VM_MEMATTR_DEFAULT && 2292 m->queue != PQ_NONE && !vm_page_busied(m)) { 2293 /* 2294 * The page is allocated but eligible for 2295 * relocation. Extend the current run by one 2296 * page. 2297 */ 2298 KASSERT(pmap_page_get_memattr(m) == 2299 VM_MEMATTR_DEFAULT, 2300 ("page %p has an unexpected memattr", m)); 2301 KASSERT((m->oflags & (VPO_SWAPINPROG | 2302 VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0, 2303 ("page %p has unexpected oflags", m)); 2304 /* Don't care: VPO_NOSYNC. */ 2305 run_ext = 1; 2306 } else 2307 run_ext = 0; 2308 unlock: 2309 VM_OBJECT_RUNLOCK(object); 2310 #if VM_NRESERVLEVEL > 0 2311 } else if (level >= 0) { 2312 /* 2313 * The page is reserved but not yet allocated. In 2314 * other words, it is still cached or free. Extend 2315 * the current run by one page. 2316 */ 2317 run_ext = 1; 2318 #endif 2319 } else if ((order = m->order) < VM_NFREEORDER) { 2320 /* 2321 * The page is enqueued in the physical memory 2322 * allocator's cache/free page queues. Moreover, it 2323 * is the first page in a power-of-two-sized run of 2324 * contiguous cache/free pages. Add these pages to 2325 * the end of the current run, and jump ahead. 2326 */ 2327 run_ext = 1 << order; 2328 m_inc = 1 << order; 2329 } else { 2330 /* 2331 * Skip the page for one of the following reasons: (1) 2332 * It is enqueued in the physical memory allocator's 2333 * cache/free page queues. However, it is not the 2334 * first page in a run of contiguous cache/free pages. 2335 * (This case rarely occurs because the scan is 2336 * performed in ascending order.) (2) It is not 2337 * reserved, and it is transitioning from free to 2338 * allocated. (Conversely, the transition from 2339 * allocated to free for managed pages is blocked by 2340 * the page lock.) (3) It is allocated but not 2341 * contained by an object and not wired, e.g., 2342 * allocated by Xen's balloon driver. 2343 */ 2344 run_ext = 0; 2345 } 2346 2347 /* 2348 * Extend or reset the current run of pages. 2349 */ 2350 if (run_ext > 0) { 2351 if (run_len == 0) 2352 m_run = m; 2353 run_len += run_ext; 2354 } else { 2355 if (run_len > 0) { 2356 m_run = NULL; 2357 run_len = 0; 2358 } 2359 } 2360 } 2361 if (m_mtx != NULL) 2362 mtx_unlock(m_mtx); 2363 if (run_len >= npages) 2364 return (m_run); 2365 return (NULL); 2366 } 2367 2368 /* 2369 * vm_page_reclaim_run: 2370 * 2371 * Try to relocate each of the allocated virtual pages within the 2372 * specified run of physical pages to a new physical address. Free the 2373 * physical pages underlying the relocated virtual pages. A virtual page 2374 * is relocatable if and only if it could be laundered or reclaimed by 2375 * the page daemon. Whenever possible, a virtual page is relocated to a 2376 * physical address above "high". 2377 * 2378 * Returns 0 if every physical page within the run was already free or 2379 * just freed by a successful relocation. Otherwise, returns a non-zero 2380 * value indicating why the last attempt to relocate a virtual page was 2381 * unsuccessful. 2382 * 2383 * "req_class" must be an allocation class. 2384 */ 2385 static int 2386 vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run, 2387 vm_paddr_t high) 2388 { 2389 struct mtx *m_mtx, *new_mtx; 2390 struct spglist free; 2391 vm_object_t object; 2392 vm_paddr_t pa; 2393 vm_page_t m, m_end, m_new; 2394 int error, order, req; 2395 2396 KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class, 2397 ("req_class is not an allocation class")); 2398 SLIST_INIT(&free); 2399 error = 0; 2400 m = m_run; 2401 m_end = m_run + npages; 2402 m_mtx = NULL; 2403 for (; error == 0 && m < m_end; m++) { 2404 KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0, 2405 ("page %p is PG_FICTITIOUS or PG_MARKER", m)); 2406 2407 /* 2408 * Avoid releasing and reacquiring the same page lock. 2409 */ 2410 new_mtx = vm_page_lockptr(m); 2411 if (m_mtx != new_mtx) { 2412 if (m_mtx != NULL) 2413 mtx_unlock(m_mtx); 2414 m_mtx = new_mtx; 2415 mtx_lock(m_mtx); 2416 } 2417 retry: 2418 if (m->wire_count != 0 || m->hold_count != 0) 2419 error = EBUSY; 2420 else if ((object = m->object) != NULL) { 2421 /* 2422 * The page is relocated if and only if it could be 2423 * laundered or reclaimed by the page daemon. 2424 */ 2425 if (!VM_OBJECT_TRYWLOCK(object)) { 2426 mtx_unlock(m_mtx); 2427 VM_OBJECT_WLOCK(object); 2428 mtx_lock(m_mtx); 2429 if (m->object != object) { 2430 /* 2431 * The page may have been freed. 2432 */ 2433 VM_OBJECT_WUNLOCK(object); 2434 goto retry; 2435 } else if (m->wire_count != 0 || 2436 m->hold_count != 0) { 2437 error = EBUSY; 2438 goto unlock; 2439 } 2440 } 2441 KASSERT((m->flags & PG_UNHOLDFREE) == 0, 2442 ("page %p is PG_UNHOLDFREE", m)); 2443 /* Don't care: PG_NODUMP, PG_WINATCFLS, PG_ZERO. */ 2444 if (object->type != OBJT_DEFAULT && 2445 object->type != OBJT_SWAP && 2446 object->type != OBJT_VNODE) 2447 error = EINVAL; 2448 else if ((m->flags & PG_CACHED) != 0 || 2449 m != vm_page_lookup(object, m->pindex)) { 2450 /* 2451 * The page is cached or recently converted 2452 * from cached to free. 2453 */ 2454 VM_OBJECT_WUNLOCK(object); 2455 goto cached; 2456 } else if (object->memattr != VM_MEMATTR_DEFAULT) 2457 error = EINVAL; 2458 else if (m->queue != PQ_NONE && !vm_page_busied(m)) { 2459 KASSERT(pmap_page_get_memattr(m) == 2460 VM_MEMATTR_DEFAULT, 2461 ("page %p has an unexpected memattr", m)); 2462 KASSERT((m->oflags & (VPO_SWAPINPROG | 2463 VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0, 2464 ("page %p has unexpected oflags", m)); 2465 /* Don't care: VPO_NOSYNC. */ 2466 if (m->valid != 0) { 2467 /* 2468 * First, try to allocate a new page 2469 * that is above "high". Failing 2470 * that, try to allocate a new page 2471 * that is below "m_run". Allocate 2472 * the new page between the end of 2473 * "m_run" and "high" only as a last 2474 * resort. 2475 */ 2476 req = req_class | VM_ALLOC_NOOBJ; 2477 if ((m->flags & PG_NODUMP) != 0) 2478 req |= VM_ALLOC_NODUMP; 2479 if (trunc_page(high) != 2480 ~(vm_paddr_t)PAGE_MASK) { 2481 m_new = vm_page_alloc_contig( 2482 NULL, 0, req, 1, 2483 round_page(high), 2484 ~(vm_paddr_t)0, 2485 PAGE_SIZE, 0, 2486 VM_MEMATTR_DEFAULT); 2487 } else 2488 m_new = NULL; 2489 if (m_new == NULL) { 2490 pa = VM_PAGE_TO_PHYS(m_run); 2491 m_new = vm_page_alloc_contig( 2492 NULL, 0, req, 1, 2493 0, pa - 1, PAGE_SIZE, 0, 2494 VM_MEMATTR_DEFAULT); 2495 } 2496 if (m_new == NULL) { 2497 pa += ptoa(npages); 2498 m_new = vm_page_alloc_contig( 2499 NULL, 0, req, 1, 2500 pa, high, PAGE_SIZE, 0, 2501 VM_MEMATTR_DEFAULT); 2502 } 2503 if (m_new == NULL) { 2504 error = ENOMEM; 2505 goto unlock; 2506 } 2507 KASSERT(m_new->wire_count == 0, 2508 ("page %p is wired", m)); 2509 2510 /* 2511 * Replace "m" with the new page. For 2512 * vm_page_replace(), "m" must be busy 2513 * and dequeued. Finally, change "m" 2514 * as if vm_page_free() was called. 2515 */ 2516 if (object->ref_count != 0) 2517 pmap_remove_all(m); 2518 m_new->aflags = m->aflags; 2519 KASSERT(m_new->oflags == VPO_UNMANAGED, 2520 ("page %p is managed", m)); 2521 m_new->oflags = m->oflags & VPO_NOSYNC; 2522 pmap_copy_page(m, m_new); 2523 m_new->valid = m->valid; 2524 m_new->dirty = m->dirty; 2525 m->flags &= ~PG_ZERO; 2526 vm_page_xbusy(m); 2527 vm_page_remque(m); 2528 vm_page_replace_checked(m_new, object, 2529 m->pindex, m); 2530 m->valid = 0; 2531 vm_page_undirty(m); 2532 2533 /* 2534 * The new page must be deactivated 2535 * before the object is unlocked. 2536 */ 2537 new_mtx = vm_page_lockptr(m_new); 2538 if (m_mtx != new_mtx) { 2539 mtx_unlock(m_mtx); 2540 m_mtx = new_mtx; 2541 mtx_lock(m_mtx); 2542 } 2543 vm_page_deactivate(m_new); 2544 } else { 2545 m->flags &= ~PG_ZERO; 2546 vm_page_remque(m); 2547 vm_page_remove(m); 2548 KASSERT(m->dirty == 0, 2549 ("page %p is dirty", m)); 2550 } 2551 SLIST_INSERT_HEAD(&free, m, plinks.s.ss); 2552 } else 2553 error = EBUSY; 2554 unlock: 2555 VM_OBJECT_WUNLOCK(object); 2556 } else { 2557 cached: 2558 mtx_lock(&vm_page_queue_free_mtx); 2559 order = m->order; 2560 if (order < VM_NFREEORDER) { 2561 /* 2562 * The page is enqueued in the physical memory 2563 * allocator's cache/free page queues. 2564 * Moreover, it is the first page in a power- 2565 * of-two-sized run of contiguous cache/free 2566 * pages. Jump ahead to the last page within 2567 * that run, and continue from there. 2568 */ 2569 m += (1 << order) - 1; 2570 } 2571 #if VM_NRESERVLEVEL > 0 2572 else if (vm_reserv_is_page_free(m)) 2573 order = 0; 2574 #endif 2575 mtx_unlock(&vm_page_queue_free_mtx); 2576 if (order == VM_NFREEORDER) 2577 error = EINVAL; 2578 } 2579 } 2580 if (m_mtx != NULL) 2581 mtx_unlock(m_mtx); 2582 if ((m = SLIST_FIRST(&free)) != NULL) { 2583 mtx_lock(&vm_page_queue_free_mtx); 2584 do { 2585 SLIST_REMOVE_HEAD(&free, plinks.s.ss); 2586 vm_phys_freecnt_adj(m, 1); 2587 #if VM_NRESERVLEVEL > 0 2588 if (!vm_reserv_free_page(m)) 2589 #else 2590 if (true) 2591 #endif 2592 vm_phys_free_pages(m, 0); 2593 } while ((m = SLIST_FIRST(&free)) != NULL); 2594 vm_page_free_wakeup(); 2595 mtx_unlock(&vm_page_queue_free_mtx); 2596 } 2597 return (error); 2598 } 2599 2600 #define NRUNS 16 2601 2602 CTASSERT(powerof2(NRUNS)); 2603 2604 #define RUN_INDEX(count) ((count) & (NRUNS - 1)) 2605 2606 #define MIN_RECLAIM 8 2607 2608 /* 2609 * vm_page_reclaim_contig: 2610 * 2611 * Reclaim allocated, contiguous physical memory satisfying the specified 2612 * conditions by relocating the virtual pages using that physical memory. 2613 * Returns true if reclamation is successful and false otherwise. Since 2614 * relocation requires the allocation of physical pages, reclamation may 2615 * fail due to a shortage of cache/free pages. When reclamation fails, 2616 * callers are expected to perform VM_WAIT before retrying a failed 2617 * allocation operation, e.g., vm_page_alloc_contig(). 2618 * 2619 * The caller must always specify an allocation class through "req". 2620 * 2621 * allocation classes: 2622 * VM_ALLOC_NORMAL normal process request 2623 * VM_ALLOC_SYSTEM system *really* needs a page 2624 * VM_ALLOC_INTERRUPT interrupt time request 2625 * 2626 * The optional allocation flags are ignored. 2627 * 2628 * "npages" must be greater than zero. Both "alignment" and "boundary" 2629 * must be a power of two. 2630 */ 2631 bool 2632 vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high, 2633 u_long alignment, vm_paddr_t boundary) 2634 { 2635 vm_paddr_t curr_low; 2636 vm_page_t m_run, m_runs[NRUNS]; 2637 u_long count, reclaimed; 2638 int error, i, options, req_class; 2639 2640 KASSERT(npages > 0, ("npages is 0")); 2641 KASSERT(powerof2(alignment), ("alignment is not a power of 2")); 2642 KASSERT(powerof2(boundary), ("boundary is not a power of 2")); 2643 req_class = req & VM_ALLOC_CLASS_MASK; 2644 2645 /* 2646 * The page daemon is allowed to dig deeper into the free page list. 2647 */ 2648 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) 2649 req_class = VM_ALLOC_SYSTEM; 2650 2651 /* 2652 * Return if the number of cached and free pages cannot satisfy the 2653 * requested allocation. 2654 */ 2655 count = vm_cnt.v_free_count + vm_cnt.v_cache_count; 2656 if (count < npages + vm_cnt.v_free_reserved || (count < npages + 2657 vm_cnt.v_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) || 2658 (count < npages && req_class == VM_ALLOC_INTERRUPT)) 2659 return (false); 2660 2661 /* 2662 * Scan up to three times, relaxing the restrictions ("options") on 2663 * the reclamation of reservations and superpages each time. 2664 */ 2665 for (options = VPSC_NORESERV;;) { 2666 /* 2667 * Find the highest runs that satisfy the given constraints 2668 * and restrictions, and record them in "m_runs". 2669 */ 2670 curr_low = low; 2671 count = 0; 2672 for (;;) { 2673 m_run = vm_phys_scan_contig(npages, curr_low, high, 2674 alignment, boundary, options); 2675 if (m_run == NULL) 2676 break; 2677 curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages); 2678 m_runs[RUN_INDEX(count)] = m_run; 2679 count++; 2680 } 2681 2682 /* 2683 * Reclaim the highest runs in LIFO (descending) order until 2684 * the number of reclaimed pages, "reclaimed", is at least 2685 * MIN_RECLAIM. Reset "reclaimed" each time because each 2686 * reclamation is idempotent, and runs will (likely) recur 2687 * from one scan to the next as restrictions are relaxed. 2688 */ 2689 reclaimed = 0; 2690 for (i = 0; count > 0 && i < NRUNS; i++) { 2691 count--; 2692 m_run = m_runs[RUN_INDEX(count)]; 2693 error = vm_page_reclaim_run(req_class, npages, m_run, 2694 high); 2695 if (error == 0) { 2696 reclaimed += npages; 2697 if (reclaimed >= MIN_RECLAIM) 2698 return (true); 2699 } 2700 } 2701 2702 /* 2703 * Either relax the restrictions on the next scan or return if 2704 * the last scan had no restrictions. 2705 */ 2706 if (options == VPSC_NORESERV) 2707 options = VPSC_NOSUPER; 2708 else if (options == VPSC_NOSUPER) 2709 options = VPSC_ANY; 2710 else if (options == VPSC_ANY) 2711 return (reclaimed != 0); 2712 } 2713 } 2714 2715 /* 2716 * vm_wait: (also see VM_WAIT macro) 2717 * 2718 * Sleep until free pages are available for allocation. 2719 * - Called in various places before memory allocations. 2720 */ 2721 void 2722 vm_wait(void) 2723 { 2724 2725 mtx_lock(&vm_page_queue_free_mtx); 2726 if (curproc == pageproc) { 2727 vm_pageout_pages_needed = 1; 2728 msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx, 2729 PDROP | PSWP, "VMWait", 0); 2730 } else { 2731 if (!vm_pageout_wanted) { 2732 vm_pageout_wanted = true; 2733 wakeup(&vm_pageout_wanted); 2734 } 2735 vm_pages_needed = true; 2736 msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM, 2737 "vmwait", 0); 2738 } 2739 } 2740 2741 /* 2742 * vm_waitpfault: (also see VM_WAITPFAULT macro) 2743 * 2744 * Sleep until free pages are available for allocation. 2745 * - Called only in vm_fault so that processes page faulting 2746 * can be easily tracked. 2747 * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing 2748 * processes will be able to grab memory first. Do not change 2749 * this balance without careful testing first. 2750 */ 2751 void 2752 vm_waitpfault(void) 2753 { 2754 2755 mtx_lock(&vm_page_queue_free_mtx); 2756 if (!vm_pageout_wanted) { 2757 vm_pageout_wanted = true; 2758 wakeup(&vm_pageout_wanted); 2759 } 2760 vm_pages_needed = true; 2761 msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER, 2762 "pfault", 0); 2763 } 2764 2765 struct vm_pagequeue * 2766 vm_page_pagequeue(vm_page_t m) 2767 { 2768 2769 return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]); 2770 } 2771 2772 /* 2773 * vm_page_dequeue: 2774 * 2775 * Remove the given page from its current page queue. 2776 * 2777 * The page must be locked. 2778 */ 2779 void 2780 vm_page_dequeue(vm_page_t m) 2781 { 2782 struct vm_pagequeue *pq; 2783 2784 vm_page_assert_locked(m); 2785 KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued", 2786 m)); 2787 pq = vm_page_pagequeue(m); 2788 vm_pagequeue_lock(pq); 2789 m->queue = PQ_NONE; 2790 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2791 vm_pagequeue_cnt_dec(pq); 2792 vm_pagequeue_unlock(pq); 2793 } 2794 2795 /* 2796 * vm_page_dequeue_locked: 2797 * 2798 * Remove the given page from its current page queue. 2799 * 2800 * The page and page queue must be locked. 2801 */ 2802 void 2803 vm_page_dequeue_locked(vm_page_t m) 2804 { 2805 struct vm_pagequeue *pq; 2806 2807 vm_page_lock_assert(m, MA_OWNED); 2808 pq = vm_page_pagequeue(m); 2809 vm_pagequeue_assert_locked(pq); 2810 m->queue = PQ_NONE; 2811 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2812 vm_pagequeue_cnt_dec(pq); 2813 } 2814 2815 /* 2816 * vm_page_enqueue: 2817 * 2818 * Add the given page to the specified page queue. 2819 * 2820 * The page must be locked. 2821 */ 2822 static void 2823 vm_page_enqueue(uint8_t queue, vm_page_t m) 2824 { 2825 struct vm_pagequeue *pq; 2826 2827 vm_page_lock_assert(m, MA_OWNED); 2828 KASSERT(queue < PQ_COUNT, 2829 ("vm_page_enqueue: invalid queue %u request for page %p", 2830 queue, m)); 2831 pq = &vm_phys_domain(m)->vmd_pagequeues[queue]; 2832 vm_pagequeue_lock(pq); 2833 m->queue = queue; 2834 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 2835 vm_pagequeue_cnt_inc(pq); 2836 vm_pagequeue_unlock(pq); 2837 } 2838 2839 /* 2840 * vm_page_requeue: 2841 * 2842 * Move the given page to the tail of its current page queue. 2843 * 2844 * The page must be locked. 2845 */ 2846 void 2847 vm_page_requeue(vm_page_t m) 2848 { 2849 struct vm_pagequeue *pq; 2850 2851 vm_page_lock_assert(m, MA_OWNED); 2852 KASSERT(m->queue != PQ_NONE, 2853 ("vm_page_requeue: page %p is not queued", m)); 2854 pq = vm_page_pagequeue(m); 2855 vm_pagequeue_lock(pq); 2856 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2857 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 2858 vm_pagequeue_unlock(pq); 2859 } 2860 2861 /* 2862 * vm_page_requeue_locked: 2863 * 2864 * Move the given page to the tail of its current page queue. 2865 * 2866 * The page queue must be locked. 2867 */ 2868 void 2869 vm_page_requeue_locked(vm_page_t m) 2870 { 2871 struct vm_pagequeue *pq; 2872 2873 KASSERT(m->queue != PQ_NONE, 2874 ("vm_page_requeue_locked: page %p is not queued", m)); 2875 pq = vm_page_pagequeue(m); 2876 vm_pagequeue_assert_locked(pq); 2877 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2878 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 2879 } 2880 2881 /* 2882 * vm_page_activate: 2883 * 2884 * Put the specified page on the active list (if appropriate). 2885 * Ensure that act_count is at least ACT_INIT but do not otherwise 2886 * mess with it. 2887 * 2888 * The page must be locked. 2889 */ 2890 void 2891 vm_page_activate(vm_page_t m) 2892 { 2893 int queue; 2894 2895 vm_page_lock_assert(m, MA_OWNED); 2896 if ((queue = m->queue) != PQ_ACTIVE) { 2897 if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { 2898 if (m->act_count < ACT_INIT) 2899 m->act_count = ACT_INIT; 2900 if (queue != PQ_NONE) 2901 vm_page_dequeue(m); 2902 vm_page_enqueue(PQ_ACTIVE, m); 2903 } else 2904 KASSERT(queue == PQ_NONE, 2905 ("vm_page_activate: wired page %p is queued", m)); 2906 } else { 2907 if (m->act_count < ACT_INIT) 2908 m->act_count = ACT_INIT; 2909 } 2910 } 2911 2912 /* 2913 * vm_page_free_wakeup: 2914 * 2915 * Helper routine for vm_page_free_toq() and vm_page_cache(). This 2916 * routine is called when a page has been added to the cache or free 2917 * queues. 2918 * 2919 * The page queues must be locked. 2920 */ 2921 static inline void 2922 vm_page_free_wakeup(void) 2923 { 2924 2925 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 2926 /* 2927 * if pageout daemon needs pages, then tell it that there are 2928 * some free. 2929 */ 2930 if (vm_pageout_pages_needed && 2931 vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) { 2932 wakeup(&vm_pageout_pages_needed); 2933 vm_pageout_pages_needed = 0; 2934 } 2935 /* 2936 * wakeup processes that are waiting on memory if we hit a 2937 * high water mark. And wakeup scheduler process if we have 2938 * lots of memory. this process will swapin processes. 2939 */ 2940 if (vm_pages_needed && !vm_page_count_min()) { 2941 vm_pages_needed = false; 2942 wakeup(&vm_cnt.v_free_count); 2943 } 2944 } 2945 2946 /* 2947 * Turn a cached page into a free page, by changing its attributes. 2948 * Keep the statistics up-to-date. 2949 * 2950 * The free page queue must be locked. 2951 */ 2952 static void 2953 vm_page_cache_turn_free(vm_page_t m) 2954 { 2955 2956 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 2957 2958 m->object = NULL; 2959 m->valid = 0; 2960 KASSERT((m->flags & PG_CACHED) != 0, 2961 ("vm_page_cache_turn_free: page %p is not cached", m)); 2962 m->flags &= ~PG_CACHED; 2963 vm_cnt.v_cache_count--; 2964 vm_phys_freecnt_adj(m, 1); 2965 } 2966 2967 /* 2968 * vm_page_free_toq: 2969 * 2970 * Returns the given page to the free list, 2971 * disassociating it with any VM object. 2972 * 2973 * The object must be locked. The page must be locked if it is managed. 2974 */ 2975 void 2976 vm_page_free_toq(vm_page_t m) 2977 { 2978 2979 if ((m->oflags & VPO_UNMANAGED) == 0) { 2980 vm_page_lock_assert(m, MA_OWNED); 2981 KASSERT(!pmap_page_is_mapped(m), 2982 ("vm_page_free_toq: freeing mapped page %p", m)); 2983 } else 2984 KASSERT(m->queue == PQ_NONE, 2985 ("vm_page_free_toq: unmanaged page %p is queued", m)); 2986 PCPU_INC(cnt.v_tfree); 2987 2988 if (vm_page_sbusied(m)) 2989 panic("vm_page_free: freeing busy page %p", m); 2990 2991 /* 2992 * Unqueue, then remove page. Note that we cannot destroy 2993 * the page here because we do not want to call the pager's 2994 * callback routine until after we've put the page on the 2995 * appropriate free queue. 2996 */ 2997 vm_page_remque(m); 2998 vm_page_remove(m); 2999 3000 /* 3001 * If fictitious remove object association and 3002 * return, otherwise delay object association removal. 3003 */ 3004 if ((m->flags & PG_FICTITIOUS) != 0) { 3005 return; 3006 } 3007 3008 m->valid = 0; 3009 vm_page_undirty(m); 3010 3011 if (m->wire_count != 0) 3012 panic("vm_page_free: freeing wired page %p", m); 3013 if (m->hold_count != 0) { 3014 m->flags &= ~PG_ZERO; 3015 KASSERT((m->flags & PG_UNHOLDFREE) == 0, 3016 ("vm_page_free: freeing PG_UNHOLDFREE page %p", m)); 3017 m->flags |= PG_UNHOLDFREE; 3018 } else { 3019 /* 3020 * Restore the default memory attribute to the page. 3021 */ 3022 if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) 3023 pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); 3024 3025 /* 3026 * Insert the page into the physical memory allocator's 3027 * cache/free page queues. 3028 */ 3029 mtx_lock(&vm_page_queue_free_mtx); 3030 vm_phys_freecnt_adj(m, 1); 3031 #if VM_NRESERVLEVEL > 0 3032 if (!vm_reserv_free_page(m)) 3033 #else 3034 if (TRUE) 3035 #endif 3036 vm_phys_free_pages(m, 0); 3037 vm_page_free_wakeup(); 3038 mtx_unlock(&vm_page_queue_free_mtx); 3039 } 3040 } 3041 3042 /* 3043 * vm_page_wire: 3044 * 3045 * Mark this page as wired down by yet 3046 * another map, removing it from paging queues 3047 * as necessary. 3048 * 3049 * If the page is fictitious, then its wire count must remain one. 3050 * 3051 * The page must be locked. 3052 */ 3053 void 3054 vm_page_wire(vm_page_t m) 3055 { 3056 3057 /* 3058 * Only bump the wire statistics if the page is not already wired, 3059 * and only unqueue the page if it is on some queue (if it is unmanaged 3060 * it is already off the queues). 3061 */ 3062 vm_page_lock_assert(m, MA_OWNED); 3063 if ((m->flags & PG_FICTITIOUS) != 0) { 3064 KASSERT(m->wire_count == 1, 3065 ("vm_page_wire: fictitious page %p's wire count isn't one", 3066 m)); 3067 return; 3068 } 3069 if (m->wire_count == 0) { 3070 KASSERT((m->oflags & VPO_UNMANAGED) == 0 || 3071 m->queue == PQ_NONE, 3072 ("vm_page_wire: unmanaged page %p is queued", m)); 3073 vm_page_remque(m); 3074 atomic_add_int(&vm_cnt.v_wire_count, 1); 3075 } 3076 m->wire_count++; 3077 KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m)); 3078 } 3079 3080 /* 3081 * vm_page_unwire: 3082 * 3083 * Release one wiring of the specified page, potentially allowing it to be 3084 * paged out. Returns TRUE if the number of wirings transitions to zero and 3085 * FALSE otherwise. 3086 * 3087 * Only managed pages belonging to an object can be paged out. If the number 3088 * of wirings transitions to zero and the page is eligible for page out, then 3089 * the page is added to the specified paging queue (unless PQ_NONE is 3090 * specified). 3091 * 3092 * If a page is fictitious, then its wire count must always be one. 3093 * 3094 * A managed page must be locked. 3095 */ 3096 boolean_t 3097 vm_page_unwire(vm_page_t m, uint8_t queue) 3098 { 3099 3100 KASSERT(queue < PQ_COUNT || queue == PQ_NONE, 3101 ("vm_page_unwire: invalid queue %u request for page %p", 3102 queue, m)); 3103 if ((m->oflags & VPO_UNMANAGED) == 0) 3104 vm_page_assert_locked(m); 3105 if ((m->flags & PG_FICTITIOUS) != 0) { 3106 KASSERT(m->wire_count == 1, 3107 ("vm_page_unwire: fictitious page %p's wire count isn't one", m)); 3108 return (FALSE); 3109 } 3110 if (m->wire_count > 0) { 3111 m->wire_count--; 3112 if (m->wire_count == 0) { 3113 atomic_subtract_int(&vm_cnt.v_wire_count, 1); 3114 if ((m->oflags & VPO_UNMANAGED) == 0 && 3115 m->object != NULL && queue != PQ_NONE) { 3116 if (queue == PQ_INACTIVE) 3117 m->flags &= ~PG_WINATCFLS; 3118 vm_page_enqueue(queue, m); 3119 } 3120 return (TRUE); 3121 } else 3122 return (FALSE); 3123 } else 3124 panic("vm_page_unwire: page %p's wire count is zero", m); 3125 } 3126 3127 /* 3128 * Move the specified page to the inactive queue. 3129 * 3130 * Many pages placed on the inactive queue should actually go 3131 * into the cache, but it is difficult to figure out which. What 3132 * we do instead, if the inactive target is well met, is to put 3133 * clean pages at the head of the inactive queue instead of the tail. 3134 * This will cause them to be moved to the cache more quickly and 3135 * if not actively re-referenced, reclaimed more quickly. If we just 3136 * stick these pages at the end of the inactive queue, heavy filesystem 3137 * meta-data accesses can cause an unnecessary paging load on memory bound 3138 * processes. This optimization causes one-time-use metadata to be 3139 * reused more quickly. 3140 * 3141 * Normally noreuse is FALSE, resulting in LRU operation. noreuse is set 3142 * to TRUE if we want this page to be 'as if it were placed in the cache', 3143 * except without unmapping it from the process address space. In 3144 * practice this is implemented by inserting the page at the head of the 3145 * queue, using a marker page to guide FIFO insertion ordering. 3146 * 3147 * The page must be locked. 3148 */ 3149 static inline void 3150 _vm_page_deactivate(vm_page_t m, boolean_t noreuse) 3151 { 3152 struct vm_pagequeue *pq; 3153 int queue; 3154 3155 vm_page_assert_locked(m); 3156 3157 /* 3158 * Ignore if the page is already inactive, unless it is unlikely to be 3159 * reactivated. 3160 */ 3161 if ((queue = m->queue) == PQ_INACTIVE && !noreuse) 3162 return; 3163 if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { 3164 pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE]; 3165 /* Avoid multiple acquisitions of the inactive queue lock. */ 3166 if (queue == PQ_INACTIVE) { 3167 vm_pagequeue_lock(pq); 3168 vm_page_dequeue_locked(m); 3169 } else { 3170 if (queue != PQ_NONE) 3171 vm_page_dequeue(m); 3172 m->flags &= ~PG_WINATCFLS; 3173 vm_pagequeue_lock(pq); 3174 } 3175 m->queue = PQ_INACTIVE; 3176 if (noreuse) 3177 TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead, 3178 m, plinks.q); 3179 else 3180 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 3181 vm_pagequeue_cnt_inc(pq); 3182 vm_pagequeue_unlock(pq); 3183 } 3184 } 3185 3186 /* 3187 * Move the specified page to the inactive queue. 3188 * 3189 * The page must be locked. 3190 */ 3191 void 3192 vm_page_deactivate(vm_page_t m) 3193 { 3194 3195 _vm_page_deactivate(m, FALSE); 3196 } 3197 3198 /* 3199 * Move the specified page to the inactive queue with the expectation 3200 * that it is unlikely to be reused. 3201 * 3202 * The page must be locked. 3203 */ 3204 void 3205 vm_page_deactivate_noreuse(vm_page_t m) 3206 { 3207 3208 _vm_page_deactivate(m, TRUE); 3209 } 3210 3211 /* 3212 * vm_page_try_to_cache: 3213 * 3214 * Returns 0 on failure, 1 on success 3215 */ 3216 int 3217 vm_page_try_to_cache(vm_page_t m) 3218 { 3219 3220 vm_page_lock_assert(m, MA_OWNED); 3221 VM_OBJECT_ASSERT_WLOCKED(m->object); 3222 if (m->dirty || m->hold_count || m->wire_count || 3223 (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) 3224 return (0); 3225 pmap_remove_all(m); 3226 if (m->dirty) 3227 return (0); 3228 vm_page_cache(m); 3229 return (1); 3230 } 3231 3232 /* 3233 * vm_page_try_to_free() 3234 * 3235 * Attempt to free the page. If we cannot free it, we do nothing. 3236 * 1 is returned on success, 0 on failure. 3237 */ 3238 int 3239 vm_page_try_to_free(vm_page_t m) 3240 { 3241 3242 vm_page_lock_assert(m, MA_OWNED); 3243 if (m->object != NULL) 3244 VM_OBJECT_ASSERT_WLOCKED(m->object); 3245 if (m->dirty || m->hold_count || m->wire_count || 3246 (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) 3247 return (0); 3248 pmap_remove_all(m); 3249 if (m->dirty) 3250 return (0); 3251 vm_page_free(m); 3252 return (1); 3253 } 3254 3255 /* 3256 * vm_page_cache 3257 * 3258 * Put the specified page onto the page cache queue (if appropriate). 3259 * 3260 * The object and page must be locked. 3261 */ 3262 void 3263 vm_page_cache(vm_page_t m) 3264 { 3265 vm_object_t object; 3266 boolean_t cache_was_empty; 3267 3268 vm_page_lock_assert(m, MA_OWNED); 3269 object = m->object; 3270 VM_OBJECT_ASSERT_WLOCKED(object); 3271 if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) || 3272 m->hold_count || m->wire_count) 3273 panic("vm_page_cache: attempting to cache busy page"); 3274 KASSERT(!pmap_page_is_mapped(m), 3275 ("vm_page_cache: page %p is mapped", m)); 3276 KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m)); 3277 if (m->valid == 0 || object->type == OBJT_DEFAULT || 3278 (object->type == OBJT_SWAP && 3279 !vm_pager_has_page(object, m->pindex, NULL, NULL))) { 3280 /* 3281 * Hypothesis: A cache-eligible page belonging to a 3282 * default object or swap object but without a backing 3283 * store must be zero filled. 3284 */ 3285 vm_page_free(m); 3286 return; 3287 } 3288 KASSERT((m->flags & PG_CACHED) == 0, 3289 ("vm_page_cache: page %p is already cached", m)); 3290 3291 /* 3292 * Remove the page from the paging queues. 3293 */ 3294 vm_page_remque(m); 3295 3296 /* 3297 * Remove the page from the object's collection of resident 3298 * pages. 3299 */ 3300 vm_radix_remove(&object->rtree, m->pindex); 3301 TAILQ_REMOVE(&object->memq, m, listq); 3302 object->resident_page_count--; 3303 3304 /* 3305 * Restore the default memory attribute to the page. 3306 */ 3307 if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) 3308 pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); 3309 3310 /* 3311 * Insert the page into the object's collection of cached pages 3312 * and the physical memory allocator's cache/free page queues. 3313 */ 3314 m->flags &= ~PG_ZERO; 3315 mtx_lock(&vm_page_queue_free_mtx); 3316 cache_was_empty = vm_radix_is_empty(&object->cache); 3317 if (vm_radix_insert(&object->cache, m)) { 3318 mtx_unlock(&vm_page_queue_free_mtx); 3319 if (object->type == OBJT_VNODE && 3320 object->resident_page_count == 0) 3321 vdrop(object->handle); 3322 m->object = NULL; 3323 vm_page_free(m); 3324 return; 3325 } 3326 3327 /* 3328 * The above call to vm_radix_insert() could reclaim the one pre- 3329 * existing cached page from this object, resulting in a call to 3330 * vdrop(). 3331 */ 3332 if (!cache_was_empty) 3333 cache_was_empty = vm_radix_is_singleton(&object->cache); 3334 3335 m->flags |= PG_CACHED; 3336 vm_cnt.v_cache_count++; 3337 PCPU_INC(cnt.v_tcached); 3338 #if VM_NRESERVLEVEL > 0 3339 if (!vm_reserv_free_page(m)) { 3340 #else 3341 if (TRUE) { 3342 #endif 3343 vm_phys_free_pages(m, 0); 3344 } 3345 vm_page_free_wakeup(); 3346 mtx_unlock(&vm_page_queue_free_mtx); 3347 3348 /* 3349 * Increment the vnode's hold count if this is the object's only 3350 * cached page. Decrement the vnode's hold count if this was 3351 * the object's only resident page. 3352 */ 3353 if (object->type == OBJT_VNODE) { 3354 if (cache_was_empty && object->resident_page_count != 0) 3355 vhold(object->handle); 3356 else if (!cache_was_empty && object->resident_page_count == 0) 3357 vdrop(object->handle); 3358 } 3359 } 3360 3361 /* 3362 * vm_page_advise 3363 * 3364 * Deactivate or do nothing, as appropriate. 3365 * 3366 * The object and page must be locked. 3367 */ 3368 void 3369 vm_page_advise(vm_page_t m, int advice) 3370 { 3371 3372 vm_page_assert_locked(m); 3373 VM_OBJECT_ASSERT_WLOCKED(m->object); 3374 if (advice == MADV_FREE) 3375 /* 3376 * Mark the page clean. This will allow the page to be freed 3377 * up by the system. However, such pages are often reused 3378 * quickly by malloc() so we do not do anything that would 3379 * cause a page fault if we can help it. 3380 * 3381 * Specifically, we do not try to actually free the page now 3382 * nor do we try to put it in the cache (which would cause a 3383 * page fault on reuse). 3384 * 3385 * But we do make the page as freeable as we can without 3386 * actually taking the step of unmapping it. 3387 */ 3388 vm_page_undirty(m); 3389 else if (advice != MADV_DONTNEED) 3390 return; 3391 3392 /* 3393 * Clear any references to the page. Otherwise, the page daemon will 3394 * immediately reactivate the page. 3395 */ 3396 vm_page_aflag_clear(m, PGA_REFERENCED); 3397 3398 if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m)) 3399 vm_page_dirty(m); 3400 3401 /* 3402 * Place clean pages near the head of the inactive queue rather than 3403 * the tail, thus defeating the queue's LRU operation and ensuring that 3404 * the page will be reused quickly. Dirty pages are given a chance to 3405 * cycle once through the inactive queue before becoming eligible for 3406 * laundering. 3407 */ 3408 _vm_page_deactivate(m, m->dirty == 0); 3409 } 3410 3411 /* 3412 * Grab a page, waiting until we are waken up due to the page 3413 * changing state. We keep on waiting, if the page continues 3414 * to be in the object. If the page doesn't exist, first allocate it 3415 * and then conditionally zero it. 3416 * 3417 * This routine may sleep. 3418 * 3419 * The object must be locked on entry. The lock will, however, be released 3420 * and reacquired if the routine sleeps. 3421 */ 3422 vm_page_t 3423 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags) 3424 { 3425 vm_page_t m; 3426 int sleep; 3427 3428 VM_OBJECT_ASSERT_WLOCKED(object); 3429 KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || 3430 (allocflags & VM_ALLOC_IGN_SBUSY) != 0, 3431 ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch")); 3432 retrylookup: 3433 if ((m = vm_page_lookup(object, pindex)) != NULL) { 3434 sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ? 3435 vm_page_xbusied(m) : vm_page_busied(m); 3436 if (sleep) { 3437 if ((allocflags & VM_ALLOC_NOWAIT) != 0) 3438 return (NULL); 3439 /* 3440 * Reference the page before unlocking and 3441 * sleeping so that the page daemon is less 3442 * likely to reclaim it. 3443 */ 3444 vm_page_aflag_set(m, PGA_REFERENCED); 3445 vm_page_lock(m); 3446 VM_OBJECT_WUNLOCK(object); 3447 vm_page_busy_sleep(m, "pgrbwt"); 3448 VM_OBJECT_WLOCK(object); 3449 goto retrylookup; 3450 } else { 3451 if ((allocflags & VM_ALLOC_WIRED) != 0) { 3452 vm_page_lock(m); 3453 vm_page_wire(m); 3454 vm_page_unlock(m); 3455 } 3456 if ((allocflags & 3457 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) 3458 vm_page_xbusy(m); 3459 if ((allocflags & VM_ALLOC_SBUSY) != 0) 3460 vm_page_sbusy(m); 3461 return (m); 3462 } 3463 } 3464 m = vm_page_alloc(object, pindex, allocflags); 3465 if (m == NULL) { 3466 if ((allocflags & VM_ALLOC_NOWAIT) != 0) 3467 return (NULL); 3468 VM_OBJECT_WUNLOCK(object); 3469 VM_WAIT; 3470 VM_OBJECT_WLOCK(object); 3471 goto retrylookup; 3472 } else if (m->valid != 0) 3473 return (m); 3474 if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0) 3475 pmap_zero_page(m); 3476 return (m); 3477 } 3478 3479 /* 3480 * Mapping function for valid or dirty bits in a page. 3481 * 3482 * Inputs are required to range within a page. 3483 */ 3484 vm_page_bits_t 3485 vm_page_bits(int base, int size) 3486 { 3487 int first_bit; 3488 int last_bit; 3489 3490 KASSERT( 3491 base + size <= PAGE_SIZE, 3492 ("vm_page_bits: illegal base/size %d/%d", base, size) 3493 ); 3494 3495 if (size == 0) /* handle degenerate case */ 3496 return (0); 3497 3498 first_bit = base >> DEV_BSHIFT; 3499 last_bit = (base + size - 1) >> DEV_BSHIFT; 3500 3501 return (((vm_page_bits_t)2 << last_bit) - 3502 ((vm_page_bits_t)1 << first_bit)); 3503 } 3504 3505 /* 3506 * vm_page_set_valid_range: 3507 * 3508 * Sets portions of a page valid. The arguments are expected 3509 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive 3510 * of any partial chunks touched by the range. The invalid portion of 3511 * such chunks will be zeroed. 3512 * 3513 * (base + size) must be less then or equal to PAGE_SIZE. 3514 */ 3515 void 3516 vm_page_set_valid_range(vm_page_t m, int base, int size) 3517 { 3518 int endoff, frag; 3519 3520 VM_OBJECT_ASSERT_WLOCKED(m->object); 3521 if (size == 0) /* handle degenerate case */ 3522 return; 3523 3524 /* 3525 * If the base is not DEV_BSIZE aligned and the valid 3526 * bit is clear, we have to zero out a portion of the 3527 * first block. 3528 */ 3529 if ((frag = rounddown2(base, DEV_BSIZE)) != base && 3530 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0) 3531 pmap_zero_page_area(m, frag, base - frag); 3532 3533 /* 3534 * If the ending offset is not DEV_BSIZE aligned and the 3535 * valid bit is clear, we have to zero out a portion of 3536 * the last block. 3537 */ 3538 endoff = base + size; 3539 if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff && 3540 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0) 3541 pmap_zero_page_area(m, endoff, 3542 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); 3543 3544 /* 3545 * Assert that no previously invalid block that is now being validated 3546 * is already dirty. 3547 */ 3548 KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0, 3549 ("vm_page_set_valid_range: page %p is dirty", m)); 3550 3551 /* 3552 * Set valid bits inclusive of any overlap. 3553 */ 3554 m->valid |= vm_page_bits(base, size); 3555 } 3556 3557 /* 3558 * Clear the given bits from the specified page's dirty field. 3559 */ 3560 static __inline void 3561 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits) 3562 { 3563 uintptr_t addr; 3564 #if PAGE_SIZE < 16384 3565 int shift; 3566 #endif 3567 3568 /* 3569 * If the object is locked and the page is neither exclusive busy nor 3570 * write mapped, then the page's dirty field cannot possibly be 3571 * set by a concurrent pmap operation. 3572 */ 3573 VM_OBJECT_ASSERT_WLOCKED(m->object); 3574 if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m)) 3575 m->dirty &= ~pagebits; 3576 else { 3577 /* 3578 * The pmap layer can call vm_page_dirty() without 3579 * holding a distinguished lock. The combination of 3580 * the object's lock and an atomic operation suffice 3581 * to guarantee consistency of the page dirty field. 3582 * 3583 * For PAGE_SIZE == 32768 case, compiler already 3584 * properly aligns the dirty field, so no forcible 3585 * alignment is needed. Only require existence of 3586 * atomic_clear_64 when page size is 32768. 3587 */ 3588 addr = (uintptr_t)&m->dirty; 3589 #if PAGE_SIZE == 32768 3590 atomic_clear_64((uint64_t *)addr, pagebits); 3591 #elif PAGE_SIZE == 16384 3592 atomic_clear_32((uint32_t *)addr, pagebits); 3593 #else /* PAGE_SIZE <= 8192 */ 3594 /* 3595 * Use a trick to perform a 32-bit atomic on the 3596 * containing aligned word, to not depend on the existence 3597 * of atomic_clear_{8, 16}. 3598 */ 3599 shift = addr & (sizeof(uint32_t) - 1); 3600 #if BYTE_ORDER == BIG_ENDIAN 3601 shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY; 3602 #else 3603 shift *= NBBY; 3604 #endif 3605 addr &= ~(sizeof(uint32_t) - 1); 3606 atomic_clear_32((uint32_t *)addr, pagebits << shift); 3607 #endif /* PAGE_SIZE */ 3608 } 3609 } 3610 3611 /* 3612 * vm_page_set_validclean: 3613 * 3614 * Sets portions of a page valid and clean. The arguments are expected 3615 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive 3616 * of any partial chunks touched by the range. The invalid portion of 3617 * such chunks will be zero'd. 3618 * 3619 * (base + size) must be less then or equal to PAGE_SIZE. 3620 */ 3621 void 3622 vm_page_set_validclean(vm_page_t m, int base, int size) 3623 { 3624 vm_page_bits_t oldvalid, pagebits; 3625 int endoff, frag; 3626 3627 VM_OBJECT_ASSERT_WLOCKED(m->object); 3628 if (size == 0) /* handle degenerate case */ 3629 return; 3630 3631 /* 3632 * If the base is not DEV_BSIZE aligned and the valid 3633 * bit is clear, we have to zero out a portion of the 3634 * first block. 3635 */ 3636 if ((frag = rounddown2(base, DEV_BSIZE)) != base && 3637 (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0) 3638 pmap_zero_page_area(m, frag, base - frag); 3639 3640 /* 3641 * If the ending offset is not DEV_BSIZE aligned and the 3642 * valid bit is clear, we have to zero out a portion of 3643 * the last block. 3644 */ 3645 endoff = base + size; 3646 if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff && 3647 (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0) 3648 pmap_zero_page_area(m, endoff, 3649 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); 3650 3651 /* 3652 * Set valid, clear dirty bits. If validating the entire 3653 * page we can safely clear the pmap modify bit. We also 3654 * use this opportunity to clear the VPO_NOSYNC flag. If a process 3655 * takes a write fault on a MAP_NOSYNC memory area the flag will 3656 * be set again. 3657 * 3658 * We set valid bits inclusive of any overlap, but we can only 3659 * clear dirty bits for DEV_BSIZE chunks that are fully within 3660 * the range. 3661 */ 3662 oldvalid = m->valid; 3663 pagebits = vm_page_bits(base, size); 3664 m->valid |= pagebits; 3665 #if 0 /* NOT YET */ 3666 if ((frag = base & (DEV_BSIZE - 1)) != 0) { 3667 frag = DEV_BSIZE - frag; 3668 base += frag; 3669 size -= frag; 3670 if (size < 0) 3671 size = 0; 3672 } 3673 pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1)); 3674 #endif 3675 if (base == 0 && size == PAGE_SIZE) { 3676 /* 3677 * The page can only be modified within the pmap if it is 3678 * mapped, and it can only be mapped if it was previously 3679 * fully valid. 3680 */ 3681 if (oldvalid == VM_PAGE_BITS_ALL) 3682 /* 3683 * Perform the pmap_clear_modify() first. Otherwise, 3684 * a concurrent pmap operation, such as 3685 * pmap_protect(), could clear a modification in the 3686 * pmap and set the dirty field on the page before 3687 * pmap_clear_modify() had begun and after the dirty 3688 * field was cleared here. 3689 */ 3690 pmap_clear_modify(m); 3691 m->dirty = 0; 3692 m->oflags &= ~VPO_NOSYNC; 3693 } else if (oldvalid != VM_PAGE_BITS_ALL) 3694 m->dirty &= ~pagebits; 3695 else 3696 vm_page_clear_dirty_mask(m, pagebits); 3697 } 3698 3699 void 3700 vm_page_clear_dirty(vm_page_t m, int base, int size) 3701 { 3702 3703 vm_page_clear_dirty_mask(m, vm_page_bits(base, size)); 3704 } 3705 3706 /* 3707 * vm_page_set_invalid: 3708 * 3709 * Invalidates DEV_BSIZE'd chunks within a page. Both the 3710 * valid and dirty bits for the effected areas are cleared. 3711 */ 3712 void 3713 vm_page_set_invalid(vm_page_t m, int base, int size) 3714 { 3715 vm_page_bits_t bits; 3716 vm_object_t object; 3717 3718 object = m->object; 3719 VM_OBJECT_ASSERT_WLOCKED(object); 3720 if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) + 3721 size >= object->un_pager.vnp.vnp_size) 3722 bits = VM_PAGE_BITS_ALL; 3723 else 3724 bits = vm_page_bits(base, size); 3725 if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL && 3726 bits != 0) 3727 pmap_remove_all(m); 3728 KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) || 3729 !pmap_page_is_mapped(m), 3730 ("vm_page_set_invalid: page %p is mapped", m)); 3731 m->valid &= ~bits; 3732 m->dirty &= ~bits; 3733 } 3734 3735 /* 3736 * vm_page_zero_invalid() 3737 * 3738 * The kernel assumes that the invalid portions of a page contain 3739 * garbage, but such pages can be mapped into memory by user code. 3740 * When this occurs, we must zero out the non-valid portions of the 3741 * page so user code sees what it expects. 3742 * 3743 * Pages are most often semi-valid when the end of a file is mapped 3744 * into memory and the file's size is not page aligned. 3745 */ 3746 void 3747 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid) 3748 { 3749 int b; 3750 int i; 3751 3752 VM_OBJECT_ASSERT_WLOCKED(m->object); 3753 /* 3754 * Scan the valid bits looking for invalid sections that 3755 * must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the 3756 * valid bit may be set ) have already been zeroed by 3757 * vm_page_set_validclean(). 3758 */ 3759 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) { 3760 if (i == (PAGE_SIZE / DEV_BSIZE) || 3761 (m->valid & ((vm_page_bits_t)1 << i))) { 3762 if (i > b) { 3763 pmap_zero_page_area(m, 3764 b << DEV_BSHIFT, (i - b) << DEV_BSHIFT); 3765 } 3766 b = i + 1; 3767 } 3768 } 3769 3770 /* 3771 * setvalid is TRUE when we can safely set the zero'd areas 3772 * as being valid. We can do this if there are no cache consistancy 3773 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS. 3774 */ 3775 if (setvalid) 3776 m->valid = VM_PAGE_BITS_ALL; 3777 } 3778 3779 /* 3780 * vm_page_is_valid: 3781 * 3782 * Is (partial) page valid? Note that the case where size == 0 3783 * will return FALSE in the degenerate case where the page is 3784 * entirely invalid, and TRUE otherwise. 3785 */ 3786 int 3787 vm_page_is_valid(vm_page_t m, int base, int size) 3788 { 3789 vm_page_bits_t bits; 3790 3791 VM_OBJECT_ASSERT_LOCKED(m->object); 3792 bits = vm_page_bits(base, size); 3793 return (m->valid != 0 && (m->valid & bits) == bits); 3794 } 3795 3796 /* 3797 * vm_page_ps_is_valid: 3798 * 3799 * Returns TRUE if the entire (super)page is valid and FALSE otherwise. 3800 */ 3801 boolean_t 3802 vm_page_ps_is_valid(vm_page_t m) 3803 { 3804 int i, npages; 3805 3806 VM_OBJECT_ASSERT_LOCKED(m->object); 3807 npages = atop(pagesizes[m->psind]); 3808 3809 /* 3810 * The physically contiguous pages that make up a superpage, i.e., a 3811 * page with a page size index ("psind") greater than zero, will 3812 * occupy adjacent entries in vm_page_array[]. 3813 */ 3814 for (i = 0; i < npages; i++) { 3815 if (m[i].valid != VM_PAGE_BITS_ALL) 3816 return (FALSE); 3817 } 3818 return (TRUE); 3819 } 3820 3821 /* 3822 * Set the page's dirty bits if the page is modified. 3823 */ 3824 void 3825 vm_page_test_dirty(vm_page_t m) 3826 { 3827 3828 VM_OBJECT_ASSERT_WLOCKED(m->object); 3829 if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m)) 3830 vm_page_dirty(m); 3831 } 3832 3833 void 3834 vm_page_lock_KBI(vm_page_t m, const char *file, int line) 3835 { 3836 3837 mtx_lock_flags_(vm_page_lockptr(m), 0, file, line); 3838 } 3839 3840 void 3841 vm_page_unlock_KBI(vm_page_t m, const char *file, int line) 3842 { 3843 3844 mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line); 3845 } 3846 3847 int 3848 vm_page_trylock_KBI(vm_page_t m, const char *file, int line) 3849 { 3850 3851 return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line)); 3852 } 3853 3854 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 3855 void 3856 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line) 3857 { 3858 3859 vm_page_lock_assert_KBI(m, MA_OWNED, file, line); 3860 } 3861 3862 void 3863 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line) 3864 { 3865 3866 mtx_assert_(vm_page_lockptr(m), a, file, line); 3867 } 3868 #endif 3869 3870 #ifdef INVARIANTS 3871 void 3872 vm_page_object_lock_assert(vm_page_t m) 3873 { 3874 3875 /* 3876 * Certain of the page's fields may only be modified by the 3877 * holder of the containing object's lock or the exclusive busy. 3878 * holder. Unfortunately, the holder of the write busy is 3879 * not recorded, and thus cannot be checked here. 3880 */ 3881 if (m->object != NULL && !vm_page_xbusied(m)) 3882 VM_OBJECT_ASSERT_WLOCKED(m->object); 3883 } 3884 3885 void 3886 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits) 3887 { 3888 3889 if ((bits & PGA_WRITEABLE) == 0) 3890 return; 3891 3892 /* 3893 * The PGA_WRITEABLE flag can only be set if the page is 3894 * managed, is exclusively busied or the object is locked. 3895 * Currently, this flag is only set by pmap_enter(). 3896 */ 3897 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 3898 ("PGA_WRITEABLE on unmanaged page")); 3899 if (!vm_page_xbusied(m)) 3900 VM_OBJECT_ASSERT_LOCKED(m->object); 3901 } 3902 #endif 3903 3904 #include "opt_ddb.h" 3905 #ifdef DDB 3906 #include <sys/kernel.h> 3907 3908 #include <ddb/ddb.h> 3909 3910 DB_SHOW_COMMAND(page, vm_page_print_page_info) 3911 { 3912 db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count); 3913 db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count); 3914 db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count); 3915 db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count); 3916 db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count); 3917 db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved); 3918 db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min); 3919 db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target); 3920 db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target); 3921 } 3922 3923 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) 3924 { 3925 int dom; 3926 3927 db_printf("pq_free %d pq_cache %d\n", 3928 vm_cnt.v_free_count, vm_cnt.v_cache_count); 3929 for (dom = 0; dom < vm_ndomains; dom++) { 3930 db_printf("dom %d page_cnt %d free %d pq_act %d pq_inact %d\n", 3931 dom, 3932 vm_dom[dom].vmd_page_count, 3933 vm_dom[dom].vmd_free_count, 3934 vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt, 3935 vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt); 3936 } 3937 } 3938 3939 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo) 3940 { 3941 vm_page_t m; 3942 boolean_t phys; 3943 3944 if (!have_addr) { 3945 db_printf("show pginfo addr\n"); 3946 return; 3947 } 3948 3949 phys = strchr(modif, 'p') != NULL; 3950 if (phys) 3951 m = PHYS_TO_VM_PAGE(addr); 3952 else 3953 m = (vm_page_t)addr; 3954 db_printf( 3955 "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n" 3956 " af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n", 3957 m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr, 3958 m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags, 3959 m->flags, m->act_count, m->busy_lock, m->valid, m->dirty); 3960 } 3961 #endif /* DDB */ 3962