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