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 was optionally read ahead or behind. 983 */ 984 void 985 vm_page_readahead_finish(vm_page_t m) 986 { 987 988 /* We shouldn't put invalid pages on queues. */ 989 KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m)); 990 991 /* 992 * Since the page is not the actually needed one, whether it should 993 * be activated or deactivated is not obvious. Empirical results 994 * have shown that deactivating the page is usually the best choice, 995 * unless the page is wanted by another thread. 996 */ 997 vm_page_lock(m); 998 if ((m->busy_lock & VPB_BIT_WAITERS) != 0) 999 vm_page_activate(m); 1000 else 1001 vm_page_deactivate(m); 1002 vm_page_unlock(m); 1003 vm_page_xunbusy(m); 1004 } 1005 1006 /* 1007 * vm_page_sleep_if_busy: 1008 * 1009 * Sleep and release the page queues lock if the page is busied. 1010 * Returns TRUE if the thread slept. 1011 * 1012 * The given page must be unlocked and object containing it must 1013 * be locked. 1014 */ 1015 int 1016 vm_page_sleep_if_busy(vm_page_t m, const char *msg) 1017 { 1018 vm_object_t obj; 1019 1020 vm_page_lock_assert(m, MA_NOTOWNED); 1021 VM_OBJECT_ASSERT_WLOCKED(m->object); 1022 1023 if (vm_page_busied(m)) { 1024 /* 1025 * The page-specific object must be cached because page 1026 * identity can change during the sleep, causing the 1027 * re-lock of a different object. 1028 * It is assumed that a reference to the object is already 1029 * held by the callers. 1030 */ 1031 obj = m->object; 1032 vm_page_lock(m); 1033 VM_OBJECT_WUNLOCK(obj); 1034 vm_page_busy_sleep(m, msg); 1035 VM_OBJECT_WLOCK(obj); 1036 return (TRUE); 1037 } 1038 return (FALSE); 1039 } 1040 1041 /* 1042 * vm_page_dirty_KBI: [ internal use only ] 1043 * 1044 * Set all bits in the page's dirty field. 1045 * 1046 * The object containing the specified page must be locked if the 1047 * call is made from the machine-independent layer. 1048 * 1049 * See vm_page_clear_dirty_mask(). 1050 * 1051 * This function should only be called by vm_page_dirty(). 1052 */ 1053 void 1054 vm_page_dirty_KBI(vm_page_t m) 1055 { 1056 1057 /* These assertions refer to this operation by its public name. */ 1058 KASSERT((m->flags & PG_CACHED) == 0, 1059 ("vm_page_dirty: page in cache!")); 1060 KASSERT(m->valid == VM_PAGE_BITS_ALL, 1061 ("vm_page_dirty: page is invalid!")); 1062 m->dirty = VM_PAGE_BITS_ALL; 1063 } 1064 1065 /* 1066 * vm_page_insert: [ internal use only ] 1067 * 1068 * Inserts the given mem entry into the object and object list. 1069 * 1070 * The object must be locked. 1071 */ 1072 int 1073 vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex) 1074 { 1075 vm_page_t mpred; 1076 1077 VM_OBJECT_ASSERT_WLOCKED(object); 1078 mpred = vm_radix_lookup_le(&object->rtree, pindex); 1079 return (vm_page_insert_after(m, object, pindex, mpred)); 1080 } 1081 1082 /* 1083 * vm_page_insert_after: 1084 * 1085 * Inserts the page "m" into the specified object at offset "pindex". 1086 * 1087 * The page "mpred" must immediately precede the offset "pindex" within 1088 * the specified object. 1089 * 1090 * The object must be locked. 1091 */ 1092 static int 1093 vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, 1094 vm_page_t mpred) 1095 { 1096 vm_pindex_t sidx; 1097 vm_object_t sobj; 1098 vm_page_t msucc; 1099 1100 VM_OBJECT_ASSERT_WLOCKED(object); 1101 KASSERT(m->object == NULL, 1102 ("vm_page_insert_after: page already inserted")); 1103 if (mpred != NULL) { 1104 KASSERT(mpred->object == object, 1105 ("vm_page_insert_after: object doesn't contain mpred")); 1106 KASSERT(mpred->pindex < pindex, 1107 ("vm_page_insert_after: mpred doesn't precede pindex")); 1108 msucc = TAILQ_NEXT(mpred, listq); 1109 } else 1110 msucc = TAILQ_FIRST(&object->memq); 1111 if (msucc != NULL) 1112 KASSERT(msucc->pindex > pindex, 1113 ("vm_page_insert_after: msucc doesn't succeed pindex")); 1114 1115 /* 1116 * Record the object/offset pair in this page 1117 */ 1118 sobj = m->object; 1119 sidx = m->pindex; 1120 m->object = object; 1121 m->pindex = pindex; 1122 1123 /* 1124 * Now link into the object's ordered list of backed pages. 1125 */ 1126 if (vm_radix_insert(&object->rtree, m)) { 1127 m->object = sobj; 1128 m->pindex = sidx; 1129 return (1); 1130 } 1131 vm_page_insert_radixdone(m, object, mpred); 1132 return (0); 1133 } 1134 1135 /* 1136 * vm_page_insert_radixdone: 1137 * 1138 * Complete page "m" insertion into the specified object after the 1139 * radix trie hooking. 1140 * 1141 * The page "mpred" must precede the offset "m->pindex" within the 1142 * specified object. 1143 * 1144 * The object must be locked. 1145 */ 1146 static void 1147 vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred) 1148 { 1149 1150 VM_OBJECT_ASSERT_WLOCKED(object); 1151 KASSERT(object != NULL && m->object == object, 1152 ("vm_page_insert_radixdone: page %p has inconsistent object", m)); 1153 if (mpred != NULL) { 1154 KASSERT(mpred->object == object, 1155 ("vm_page_insert_after: object doesn't contain mpred")); 1156 KASSERT(mpred->pindex < m->pindex, 1157 ("vm_page_insert_after: mpred doesn't precede pindex")); 1158 } 1159 1160 if (mpred != NULL) 1161 TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq); 1162 else 1163 TAILQ_INSERT_HEAD(&object->memq, m, listq); 1164 1165 /* 1166 * Show that the object has one more resident page. 1167 */ 1168 object->resident_page_count++; 1169 1170 /* 1171 * Hold the vnode until the last page is released. 1172 */ 1173 if (object->resident_page_count == 1 && object->type == OBJT_VNODE) 1174 vhold(object->handle); 1175 1176 /* 1177 * Since we are inserting a new and possibly dirty page, 1178 * update the object's OBJ_MIGHTBEDIRTY flag. 1179 */ 1180 if (pmap_page_is_write_mapped(m)) 1181 vm_object_set_writeable_dirty(object); 1182 } 1183 1184 /* 1185 * vm_page_remove: 1186 * 1187 * Removes the given mem entry from the object/offset-page 1188 * table and the object page list, but do not invalidate/terminate 1189 * the backing store. 1190 * 1191 * The object must be locked. The page must be locked if it is managed. 1192 */ 1193 void 1194 vm_page_remove(vm_page_t m) 1195 { 1196 vm_object_t object; 1197 boolean_t lockacq; 1198 1199 if ((m->oflags & VPO_UNMANAGED) == 0) 1200 vm_page_lock_assert(m, MA_OWNED); 1201 if ((object = m->object) == NULL) 1202 return; 1203 VM_OBJECT_ASSERT_WLOCKED(object); 1204 if (vm_page_xbusied(m)) { 1205 lockacq = FALSE; 1206 if ((m->oflags & VPO_UNMANAGED) != 0 && 1207 !mtx_owned(vm_page_lockptr(m))) { 1208 lockacq = TRUE; 1209 vm_page_lock(m); 1210 } 1211 vm_page_flash(m); 1212 atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED); 1213 if (lockacq) 1214 vm_page_unlock(m); 1215 } 1216 1217 /* 1218 * Now remove from the object's list of backed pages. 1219 */ 1220 vm_radix_remove(&object->rtree, m->pindex); 1221 TAILQ_REMOVE(&object->memq, m, listq); 1222 1223 /* 1224 * And show that the object has one fewer resident page. 1225 */ 1226 object->resident_page_count--; 1227 1228 /* 1229 * The vnode may now be recycled. 1230 */ 1231 if (object->resident_page_count == 0 && object->type == OBJT_VNODE) 1232 vdrop(object->handle); 1233 1234 m->object = NULL; 1235 } 1236 1237 /* 1238 * vm_page_lookup: 1239 * 1240 * Returns the page associated with the object/offset 1241 * pair specified; if none is found, NULL is returned. 1242 * 1243 * The object must be locked. 1244 */ 1245 vm_page_t 1246 vm_page_lookup(vm_object_t object, vm_pindex_t pindex) 1247 { 1248 1249 VM_OBJECT_ASSERT_LOCKED(object); 1250 return (vm_radix_lookup(&object->rtree, pindex)); 1251 } 1252 1253 /* 1254 * vm_page_find_least: 1255 * 1256 * Returns the page associated with the object with least pindex 1257 * greater than or equal to the parameter pindex, or NULL. 1258 * 1259 * The object must be locked. 1260 */ 1261 vm_page_t 1262 vm_page_find_least(vm_object_t object, vm_pindex_t pindex) 1263 { 1264 vm_page_t m; 1265 1266 VM_OBJECT_ASSERT_LOCKED(object); 1267 if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex) 1268 m = vm_radix_lookup_ge(&object->rtree, pindex); 1269 return (m); 1270 } 1271 1272 /* 1273 * Returns the given page's successor (by pindex) within the object if it is 1274 * resident; if none is found, NULL is returned. 1275 * 1276 * The object must be locked. 1277 */ 1278 vm_page_t 1279 vm_page_next(vm_page_t m) 1280 { 1281 vm_page_t next; 1282 1283 VM_OBJECT_ASSERT_WLOCKED(m->object); 1284 if ((next = TAILQ_NEXT(m, listq)) != NULL && 1285 next->pindex != m->pindex + 1) 1286 next = NULL; 1287 return (next); 1288 } 1289 1290 /* 1291 * Returns the given page's predecessor (by pindex) within the object if it is 1292 * resident; if none is found, NULL is returned. 1293 * 1294 * The object must be locked. 1295 */ 1296 vm_page_t 1297 vm_page_prev(vm_page_t m) 1298 { 1299 vm_page_t prev; 1300 1301 VM_OBJECT_ASSERT_WLOCKED(m->object); 1302 if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL && 1303 prev->pindex != m->pindex - 1) 1304 prev = NULL; 1305 return (prev); 1306 } 1307 1308 /* 1309 * Uses the page mnew as a replacement for an existing page at index 1310 * pindex which must be already present in the object. 1311 * 1312 * The existing page must not be on a paging queue. 1313 */ 1314 vm_page_t 1315 vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex) 1316 { 1317 vm_page_t mold; 1318 1319 VM_OBJECT_ASSERT_WLOCKED(object); 1320 KASSERT(mnew->object == NULL, 1321 ("vm_page_replace: page already in object")); 1322 1323 /* 1324 * This function mostly follows vm_page_insert() and 1325 * vm_page_remove() without the radix, object count and vnode 1326 * dance. Double check such functions for more comments. 1327 */ 1328 1329 mnew->object = object; 1330 mnew->pindex = pindex; 1331 mold = vm_radix_replace(&object->rtree, mnew); 1332 KASSERT(mold->queue == PQ_NONE, 1333 ("vm_page_replace: mold is on a paging queue")); 1334 1335 /* Keep the resident page list in sorted order. */ 1336 TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq); 1337 TAILQ_REMOVE(&object->memq, mold, listq); 1338 1339 mold->object = NULL; 1340 vm_page_xunbusy(mold); 1341 1342 /* 1343 * The object's resident_page_count does not change because we have 1344 * swapped one page for another, but OBJ_MIGHTBEDIRTY. 1345 */ 1346 if (pmap_page_is_write_mapped(mnew)) 1347 vm_object_set_writeable_dirty(object); 1348 return (mold); 1349 } 1350 1351 /* 1352 * vm_page_rename: 1353 * 1354 * Move the given memory entry from its 1355 * current object to the specified target object/offset. 1356 * 1357 * Note: swap associated with the page must be invalidated by the move. We 1358 * have to do this for several reasons: (1) we aren't freeing the 1359 * page, (2) we are dirtying the page, (3) the VM system is probably 1360 * moving the page from object A to B, and will then later move 1361 * the backing store from A to B and we can't have a conflict. 1362 * 1363 * Note: we *always* dirty the page. It is necessary both for the 1364 * fact that we moved it, and because we may be invalidating 1365 * swap. If the page is on the cache, we have to deactivate it 1366 * or vm_page_dirty() will panic. Dirty pages are not allowed 1367 * on the cache. 1368 * 1369 * The objects must be locked. 1370 */ 1371 int 1372 vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex) 1373 { 1374 vm_page_t mpred; 1375 vm_pindex_t opidx; 1376 1377 VM_OBJECT_ASSERT_WLOCKED(new_object); 1378 1379 mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex); 1380 KASSERT(mpred == NULL || mpred->pindex != new_pindex, 1381 ("vm_page_rename: pindex already renamed")); 1382 1383 /* 1384 * Create a custom version of vm_page_insert() which does not depend 1385 * by m_prev and can cheat on the implementation aspects of the 1386 * function. 1387 */ 1388 opidx = m->pindex; 1389 m->pindex = new_pindex; 1390 if (vm_radix_insert(&new_object->rtree, m)) { 1391 m->pindex = opidx; 1392 return (1); 1393 } 1394 1395 /* 1396 * The operation cannot fail anymore. The removal must happen before 1397 * the listq iterator is tainted. 1398 */ 1399 m->pindex = opidx; 1400 vm_page_lock(m); 1401 vm_page_remove(m); 1402 1403 /* Return back to the new pindex to complete vm_page_insert(). */ 1404 m->pindex = new_pindex; 1405 m->object = new_object; 1406 vm_page_unlock(m); 1407 vm_page_insert_radixdone(m, new_object, mpred); 1408 vm_page_dirty(m); 1409 return (0); 1410 } 1411 1412 /* 1413 * Convert all of the given object's cached pages that have a 1414 * pindex within the given range into free pages. If the value 1415 * zero is given for "end", then the range's upper bound is 1416 * infinity. If the given object is backed by a vnode and it 1417 * transitions from having one or more cached pages to none, the 1418 * vnode's hold count is reduced. 1419 */ 1420 void 1421 vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 1422 { 1423 vm_page_t m; 1424 boolean_t empty; 1425 1426 mtx_lock(&vm_page_queue_free_mtx); 1427 if (__predict_false(vm_radix_is_empty(&object->cache))) { 1428 mtx_unlock(&vm_page_queue_free_mtx); 1429 return; 1430 } 1431 while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) { 1432 if (end != 0 && m->pindex >= end) 1433 break; 1434 vm_radix_remove(&object->cache, m->pindex); 1435 vm_page_cache_turn_free(m); 1436 } 1437 empty = vm_radix_is_empty(&object->cache); 1438 mtx_unlock(&vm_page_queue_free_mtx); 1439 if (object->type == OBJT_VNODE && empty) 1440 vdrop(object->handle); 1441 } 1442 1443 /* 1444 * Returns the cached page that is associated with the given 1445 * object and offset. If, however, none exists, returns NULL. 1446 * 1447 * The free page queue must be locked. 1448 */ 1449 static inline vm_page_t 1450 vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex) 1451 { 1452 1453 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1454 return (vm_radix_lookup(&object->cache, pindex)); 1455 } 1456 1457 /* 1458 * Remove the given cached page from its containing object's 1459 * collection of cached pages. 1460 * 1461 * The free page queue must be locked. 1462 */ 1463 static void 1464 vm_page_cache_remove(vm_page_t m) 1465 { 1466 1467 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1468 KASSERT((m->flags & PG_CACHED) != 0, 1469 ("vm_page_cache_remove: page %p is not cached", m)); 1470 vm_radix_remove(&m->object->cache, m->pindex); 1471 m->object = NULL; 1472 vm_cnt.v_cache_count--; 1473 } 1474 1475 /* 1476 * Transfer all of the cached pages with offset greater than or 1477 * equal to 'offidxstart' from the original object's cache to the 1478 * new object's cache. However, any cached pages with offset 1479 * greater than or equal to the new object's size are kept in the 1480 * original object. Initially, the new object's cache must be 1481 * empty. Offset 'offidxstart' in the original object must 1482 * correspond to offset zero in the new object. 1483 * 1484 * The new object must be locked. 1485 */ 1486 void 1487 vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart, 1488 vm_object_t new_object) 1489 { 1490 vm_page_t m; 1491 1492 /* 1493 * Insertion into an object's collection of cached pages 1494 * requires the object to be locked. In contrast, removal does 1495 * not. 1496 */ 1497 VM_OBJECT_ASSERT_WLOCKED(new_object); 1498 KASSERT(vm_radix_is_empty(&new_object->cache), 1499 ("vm_page_cache_transfer: object %p has cached pages", 1500 new_object)); 1501 mtx_lock(&vm_page_queue_free_mtx); 1502 while ((m = vm_radix_lookup_ge(&orig_object->cache, 1503 offidxstart)) != NULL) { 1504 /* 1505 * Transfer all of the pages with offset greater than or 1506 * equal to 'offidxstart' from the original object's 1507 * cache to the new object's cache. 1508 */ 1509 if ((m->pindex - offidxstart) >= new_object->size) 1510 break; 1511 vm_radix_remove(&orig_object->cache, m->pindex); 1512 /* Update the page's object and offset. */ 1513 m->object = new_object; 1514 m->pindex -= offidxstart; 1515 if (vm_radix_insert(&new_object->cache, m)) 1516 vm_page_cache_turn_free(m); 1517 } 1518 mtx_unlock(&vm_page_queue_free_mtx); 1519 } 1520 1521 /* 1522 * Returns TRUE if a cached page is associated with the given object and 1523 * offset, and FALSE otherwise. 1524 * 1525 * The object must be locked. 1526 */ 1527 boolean_t 1528 vm_page_is_cached(vm_object_t object, vm_pindex_t pindex) 1529 { 1530 vm_page_t m; 1531 1532 /* 1533 * Insertion into an object's collection of cached pages requires the 1534 * object to be locked. Therefore, if the object is locked and the 1535 * object's collection is empty, there is no need to acquire the free 1536 * page queues lock in order to prove that the specified page doesn't 1537 * exist. 1538 */ 1539 VM_OBJECT_ASSERT_WLOCKED(object); 1540 if (__predict_true(vm_object_cache_is_empty(object))) 1541 return (FALSE); 1542 mtx_lock(&vm_page_queue_free_mtx); 1543 m = vm_page_cache_lookup(object, pindex); 1544 mtx_unlock(&vm_page_queue_free_mtx); 1545 return (m != NULL); 1546 } 1547 1548 /* 1549 * vm_page_alloc: 1550 * 1551 * Allocate and return a page that is associated with the specified 1552 * object and offset pair. By default, this page is exclusive busied. 1553 * 1554 * The caller must always specify an allocation class. 1555 * 1556 * allocation classes: 1557 * VM_ALLOC_NORMAL normal process request 1558 * VM_ALLOC_SYSTEM system *really* needs a page 1559 * VM_ALLOC_INTERRUPT interrupt time request 1560 * 1561 * optional allocation flags: 1562 * VM_ALLOC_COUNT(number) the number of additional pages that the caller 1563 * intends to allocate 1564 * VM_ALLOC_IFCACHED return page only if it is cached 1565 * VM_ALLOC_IFNOTCACHED return NULL, do not reactivate if the page 1566 * is cached 1567 * VM_ALLOC_NOBUSY do not exclusive busy the page 1568 * VM_ALLOC_NODUMP do not include the page in a kernel core dump 1569 * VM_ALLOC_NOOBJ page is not associated with an object and 1570 * should not be exclusive busy 1571 * VM_ALLOC_SBUSY shared busy the allocated page 1572 * VM_ALLOC_WIRED wire the allocated page 1573 * VM_ALLOC_ZERO prefer a zeroed page 1574 * 1575 * This routine may not sleep. 1576 */ 1577 vm_page_t 1578 vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req) 1579 { 1580 struct vnode *vp = NULL; 1581 vm_object_t m_object; 1582 vm_page_t m, mpred; 1583 int flags, req_class; 1584 1585 mpred = 0; /* XXX: pacify gcc */ 1586 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && 1587 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && 1588 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != 1589 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), 1590 ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object, 1591 req)); 1592 if (object != NULL) 1593 VM_OBJECT_ASSERT_WLOCKED(object); 1594 1595 req_class = req & VM_ALLOC_CLASS_MASK; 1596 1597 /* 1598 * The page daemon is allowed to dig deeper into the free page list. 1599 */ 1600 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) 1601 req_class = VM_ALLOC_SYSTEM; 1602 1603 if (object != NULL) { 1604 mpred = vm_radix_lookup_le(&object->rtree, pindex); 1605 KASSERT(mpred == NULL || mpred->pindex != pindex, 1606 ("vm_page_alloc: pindex already allocated")); 1607 } 1608 1609 /* 1610 * The page allocation request can came from consumers which already 1611 * hold the free page queue mutex, like vm_page_insert() in 1612 * vm_page_cache(). 1613 */ 1614 mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE); 1615 if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved || 1616 (req_class == VM_ALLOC_SYSTEM && 1617 vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) || 1618 (req_class == VM_ALLOC_INTERRUPT && 1619 vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) { 1620 /* 1621 * Allocate from the free queue if the number of free pages 1622 * exceeds the minimum for the request class. 1623 */ 1624 if (object != NULL && 1625 (m = vm_page_cache_lookup(object, pindex)) != NULL) { 1626 if ((req & VM_ALLOC_IFNOTCACHED) != 0) { 1627 mtx_unlock(&vm_page_queue_free_mtx); 1628 return (NULL); 1629 } 1630 if (vm_phys_unfree_page(m)) 1631 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0); 1632 #if VM_NRESERVLEVEL > 0 1633 else if (!vm_reserv_reactivate_page(m)) 1634 #else 1635 else 1636 #endif 1637 panic("vm_page_alloc: cache page %p is missing" 1638 " from the free queue", m); 1639 } else if ((req & VM_ALLOC_IFCACHED) != 0) { 1640 mtx_unlock(&vm_page_queue_free_mtx); 1641 return (NULL); 1642 #if VM_NRESERVLEVEL > 0 1643 } else if (object == NULL || (object->flags & (OBJ_COLORED | 1644 OBJ_FICTITIOUS)) != OBJ_COLORED || (m = 1645 vm_reserv_alloc_page(object, pindex, mpred)) == NULL) { 1646 #else 1647 } else { 1648 #endif 1649 m = vm_phys_alloc_pages(object != NULL ? 1650 VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0); 1651 #if VM_NRESERVLEVEL > 0 1652 if (m == NULL && vm_reserv_reclaim_inactive()) { 1653 m = vm_phys_alloc_pages(object != NULL ? 1654 VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 1655 0); 1656 } 1657 #endif 1658 } 1659 } else { 1660 /* 1661 * Not allocatable, give up. 1662 */ 1663 mtx_unlock(&vm_page_queue_free_mtx); 1664 atomic_add_int(&vm_pageout_deficit, 1665 max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); 1666 pagedaemon_wakeup(); 1667 return (NULL); 1668 } 1669 1670 /* 1671 * At this point we had better have found a good page. 1672 */ 1673 KASSERT(m != NULL, ("vm_page_alloc: missing page")); 1674 KASSERT(m->queue == PQ_NONE, 1675 ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue)); 1676 KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m)); 1677 KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m)); 1678 KASSERT(!vm_page_sbusied(m), 1679 ("vm_page_alloc: page %p is busy", m)); 1680 KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m)); 1681 KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, 1682 ("vm_page_alloc: page %p has unexpected memattr %d", m, 1683 pmap_page_get_memattr(m))); 1684 if ((m->flags & PG_CACHED) != 0) { 1685 KASSERT((m->flags & PG_ZERO) == 0, 1686 ("vm_page_alloc: cached page %p is PG_ZERO", m)); 1687 KASSERT(m->valid != 0, 1688 ("vm_page_alloc: cached page %p is invalid", m)); 1689 if (m->object == object && m->pindex == pindex) 1690 vm_cnt.v_reactivated++; 1691 else 1692 m->valid = 0; 1693 m_object = m->object; 1694 vm_page_cache_remove(m); 1695 if (m_object->type == OBJT_VNODE && 1696 vm_object_cache_is_empty(m_object)) 1697 vp = m_object->handle; 1698 } else { 1699 KASSERT(m->valid == 0, 1700 ("vm_page_alloc: free page %p is valid", m)); 1701 vm_phys_freecnt_adj(m, -1); 1702 if ((m->flags & PG_ZERO) != 0) 1703 vm_page_zero_count--; 1704 } 1705 mtx_unlock(&vm_page_queue_free_mtx); 1706 1707 /* 1708 * Initialize the page. Only the PG_ZERO flag is inherited. 1709 */ 1710 flags = 0; 1711 if ((req & VM_ALLOC_ZERO) != 0) 1712 flags = PG_ZERO; 1713 flags &= m->flags; 1714 if ((req & VM_ALLOC_NODUMP) != 0) 1715 flags |= PG_NODUMP; 1716 m->flags = flags; 1717 m->aflags = 0; 1718 m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ? 1719 VPO_UNMANAGED : 0; 1720 m->busy_lock = VPB_UNBUSIED; 1721 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0) 1722 m->busy_lock = VPB_SINGLE_EXCLUSIVER; 1723 if ((req & VM_ALLOC_SBUSY) != 0) 1724 m->busy_lock = VPB_SHARERS_WORD(1); 1725 if (req & VM_ALLOC_WIRED) { 1726 /* 1727 * The page lock is not required for wiring a page until that 1728 * page is inserted into the object. 1729 */ 1730 atomic_add_int(&vm_cnt.v_wire_count, 1); 1731 m->wire_count = 1; 1732 } 1733 m->act_count = 0; 1734 1735 if (object != NULL) { 1736 if (vm_page_insert_after(m, object, pindex, mpred)) { 1737 /* See the comment below about hold count. */ 1738 if (vp != NULL) 1739 vdrop(vp); 1740 pagedaemon_wakeup(); 1741 if (req & VM_ALLOC_WIRED) { 1742 atomic_subtract_int(&vm_cnt.v_wire_count, 1); 1743 m->wire_count = 0; 1744 } 1745 m->object = NULL; 1746 m->oflags = VPO_UNMANAGED; 1747 vm_page_free(m); 1748 return (NULL); 1749 } 1750 1751 /* Ignore device objects; the pager sets "memattr" for them. */ 1752 if (object->memattr != VM_MEMATTR_DEFAULT && 1753 (object->flags & OBJ_FICTITIOUS) == 0) 1754 pmap_page_set_memattr(m, object->memattr); 1755 } else 1756 m->pindex = pindex; 1757 1758 /* 1759 * The following call to vdrop() must come after the above call 1760 * to vm_page_insert() in case both affect the same object and 1761 * vnode. Otherwise, the affected vnode's hold count could 1762 * temporarily become zero. 1763 */ 1764 if (vp != NULL) 1765 vdrop(vp); 1766 1767 /* 1768 * Don't wakeup too often - wakeup the pageout daemon when 1769 * we would be nearly out of memory. 1770 */ 1771 if (vm_paging_needed()) 1772 pagedaemon_wakeup(); 1773 1774 return (m); 1775 } 1776 1777 static void 1778 vm_page_alloc_contig_vdrop(struct spglist *lst) 1779 { 1780 1781 while (!SLIST_EMPTY(lst)) { 1782 vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv); 1783 SLIST_REMOVE_HEAD(lst, plinks.s.ss); 1784 } 1785 } 1786 1787 /* 1788 * vm_page_alloc_contig: 1789 * 1790 * Allocate a contiguous set of physical pages of the given size "npages" 1791 * from the free lists. All of the physical pages must be at or above 1792 * the given physical address "low" and below the given physical address 1793 * "high". The given value "alignment" determines the alignment of the 1794 * first physical page in the set. If the given value "boundary" is 1795 * non-zero, then the set of physical pages cannot cross any physical 1796 * address boundary that is a multiple of that value. Both "alignment" 1797 * and "boundary" must be a power of two. 1798 * 1799 * If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT, 1800 * then the memory attribute setting for the physical pages is configured 1801 * to the object's memory attribute setting. Otherwise, the memory 1802 * attribute setting for the physical pages is configured to "memattr", 1803 * overriding the object's memory attribute setting. However, if the 1804 * object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the 1805 * memory attribute setting for the physical pages cannot be configured 1806 * to VM_MEMATTR_DEFAULT. 1807 * 1808 * The caller must always specify an allocation class. 1809 * 1810 * allocation classes: 1811 * VM_ALLOC_NORMAL normal process request 1812 * VM_ALLOC_SYSTEM system *really* needs a page 1813 * VM_ALLOC_INTERRUPT interrupt time request 1814 * 1815 * optional allocation flags: 1816 * VM_ALLOC_NOBUSY do not exclusive busy the page 1817 * VM_ALLOC_NODUMP do not include the page in a kernel core dump 1818 * VM_ALLOC_NOOBJ page is not associated with an object and 1819 * should not be exclusive busy 1820 * VM_ALLOC_SBUSY shared busy the allocated page 1821 * VM_ALLOC_WIRED wire the allocated page 1822 * VM_ALLOC_ZERO prefer a zeroed page 1823 * 1824 * This routine may not sleep. 1825 */ 1826 vm_page_t 1827 vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, 1828 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, 1829 vm_paddr_t boundary, vm_memattr_t memattr) 1830 { 1831 struct vnode *drop; 1832 struct spglist deferred_vdrop_list; 1833 vm_page_t m, m_tmp, m_ret; 1834 u_int flags; 1835 int req_class; 1836 1837 KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && 1838 (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && 1839 ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != 1840 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), 1841 ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object, 1842 req)); 1843 if (object != NULL) { 1844 VM_OBJECT_ASSERT_WLOCKED(object); 1845 KASSERT(object->type == OBJT_PHYS, 1846 ("vm_page_alloc_contig: object %p isn't OBJT_PHYS", 1847 object)); 1848 } 1849 KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero")); 1850 req_class = req & VM_ALLOC_CLASS_MASK; 1851 1852 /* 1853 * The page daemon is allowed to dig deeper into the free page list. 1854 */ 1855 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) 1856 req_class = VM_ALLOC_SYSTEM; 1857 1858 SLIST_INIT(&deferred_vdrop_list); 1859 mtx_lock(&vm_page_queue_free_mtx); 1860 if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages + 1861 vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && 1862 vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages + 1863 vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && 1864 vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) { 1865 #if VM_NRESERVLEVEL > 0 1866 retry: 1867 if (object == NULL || (object->flags & OBJ_COLORED) == 0 || 1868 (m_ret = vm_reserv_alloc_contig(object, pindex, npages, 1869 low, high, alignment, boundary)) == NULL) 1870 #endif 1871 m_ret = vm_phys_alloc_contig(npages, low, high, 1872 alignment, boundary); 1873 } else { 1874 mtx_unlock(&vm_page_queue_free_mtx); 1875 atomic_add_int(&vm_pageout_deficit, npages); 1876 pagedaemon_wakeup(); 1877 return (NULL); 1878 } 1879 if (m_ret != NULL) 1880 for (m = m_ret; m < &m_ret[npages]; m++) { 1881 drop = vm_page_alloc_init(m); 1882 if (drop != NULL) { 1883 /* 1884 * Enqueue the vnode for deferred vdrop(). 1885 */ 1886 m->plinks.s.pv = drop; 1887 SLIST_INSERT_HEAD(&deferred_vdrop_list, m, 1888 plinks.s.ss); 1889 } 1890 } 1891 else { 1892 #if VM_NRESERVLEVEL > 0 1893 if (vm_reserv_reclaim_contig(npages, low, high, alignment, 1894 boundary)) 1895 goto retry; 1896 #endif 1897 } 1898 mtx_unlock(&vm_page_queue_free_mtx); 1899 if (m_ret == NULL) 1900 return (NULL); 1901 1902 /* 1903 * Initialize the pages. Only the PG_ZERO flag is inherited. 1904 */ 1905 flags = 0; 1906 if ((req & VM_ALLOC_ZERO) != 0) 1907 flags = PG_ZERO; 1908 if ((req & VM_ALLOC_NODUMP) != 0) 1909 flags |= PG_NODUMP; 1910 if ((req & VM_ALLOC_WIRED) != 0) 1911 atomic_add_int(&vm_cnt.v_wire_count, npages); 1912 if (object != NULL) { 1913 if (object->memattr != VM_MEMATTR_DEFAULT && 1914 memattr == VM_MEMATTR_DEFAULT) 1915 memattr = object->memattr; 1916 } 1917 for (m = m_ret; m < &m_ret[npages]; m++) { 1918 m->aflags = 0; 1919 m->flags = (m->flags | PG_NODUMP) & flags; 1920 m->busy_lock = VPB_UNBUSIED; 1921 if (object != NULL) { 1922 if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) 1923 m->busy_lock = VPB_SINGLE_EXCLUSIVER; 1924 if ((req & VM_ALLOC_SBUSY) != 0) 1925 m->busy_lock = VPB_SHARERS_WORD(1); 1926 } 1927 if ((req & VM_ALLOC_WIRED) != 0) 1928 m->wire_count = 1; 1929 /* Unmanaged pages don't use "act_count". */ 1930 m->oflags = VPO_UNMANAGED; 1931 if (object != NULL) { 1932 if (vm_page_insert(m, object, pindex)) { 1933 vm_page_alloc_contig_vdrop( 1934 &deferred_vdrop_list); 1935 if (vm_paging_needed()) 1936 pagedaemon_wakeup(); 1937 if ((req & VM_ALLOC_WIRED) != 0) 1938 atomic_subtract_int(&vm_cnt.v_wire_count, 1939 npages); 1940 for (m_tmp = m, m = m_ret; 1941 m < &m_ret[npages]; m++) { 1942 if ((req & VM_ALLOC_WIRED) != 0) 1943 m->wire_count = 0; 1944 if (m >= m_tmp) 1945 m->object = NULL; 1946 vm_page_free(m); 1947 } 1948 return (NULL); 1949 } 1950 } else 1951 m->pindex = pindex; 1952 if (memattr != VM_MEMATTR_DEFAULT) 1953 pmap_page_set_memattr(m, memattr); 1954 pindex++; 1955 } 1956 vm_page_alloc_contig_vdrop(&deferred_vdrop_list); 1957 if (vm_paging_needed()) 1958 pagedaemon_wakeup(); 1959 return (m_ret); 1960 } 1961 1962 /* 1963 * Initialize a page that has been freshly dequeued from a freelist. 1964 * The caller has to drop the vnode returned, if it is not NULL. 1965 * 1966 * This function may only be used to initialize unmanaged pages. 1967 * 1968 * To be called with vm_page_queue_free_mtx held. 1969 */ 1970 static struct vnode * 1971 vm_page_alloc_init(vm_page_t m) 1972 { 1973 struct vnode *drop; 1974 vm_object_t m_object; 1975 1976 KASSERT(m->queue == PQ_NONE, 1977 ("vm_page_alloc_init: page %p has unexpected queue %d", 1978 m, m->queue)); 1979 KASSERT(m->wire_count == 0, 1980 ("vm_page_alloc_init: page %p is wired", m)); 1981 KASSERT(m->hold_count == 0, 1982 ("vm_page_alloc_init: page %p is held", m)); 1983 KASSERT(!vm_page_sbusied(m), 1984 ("vm_page_alloc_init: page %p is busy", m)); 1985 KASSERT(m->dirty == 0, 1986 ("vm_page_alloc_init: page %p is dirty", m)); 1987 KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, 1988 ("vm_page_alloc_init: page %p has unexpected memattr %d", 1989 m, pmap_page_get_memattr(m))); 1990 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1991 drop = NULL; 1992 if ((m->flags & PG_CACHED) != 0) { 1993 KASSERT((m->flags & PG_ZERO) == 0, 1994 ("vm_page_alloc_init: cached page %p is PG_ZERO", m)); 1995 m->valid = 0; 1996 m_object = m->object; 1997 vm_page_cache_remove(m); 1998 if (m_object->type == OBJT_VNODE && 1999 vm_object_cache_is_empty(m_object)) 2000 drop = m_object->handle; 2001 } else { 2002 KASSERT(m->valid == 0, 2003 ("vm_page_alloc_init: free page %p is valid", m)); 2004 vm_phys_freecnt_adj(m, -1); 2005 if ((m->flags & PG_ZERO) != 0) 2006 vm_page_zero_count--; 2007 } 2008 return (drop); 2009 } 2010 2011 /* 2012 * vm_page_alloc_freelist: 2013 * 2014 * Allocate a physical page from the specified free page list. 2015 * 2016 * The caller must always specify an allocation class. 2017 * 2018 * allocation classes: 2019 * VM_ALLOC_NORMAL normal process request 2020 * VM_ALLOC_SYSTEM system *really* needs a page 2021 * VM_ALLOC_INTERRUPT interrupt time request 2022 * 2023 * optional allocation flags: 2024 * VM_ALLOC_COUNT(number) the number of additional pages that the caller 2025 * intends to allocate 2026 * VM_ALLOC_WIRED wire the allocated page 2027 * VM_ALLOC_ZERO prefer a zeroed page 2028 * 2029 * This routine may not sleep. 2030 */ 2031 vm_page_t 2032 vm_page_alloc_freelist(int flind, int req) 2033 { 2034 struct vnode *drop; 2035 vm_page_t m; 2036 u_int flags; 2037 int req_class; 2038 2039 req_class = req & VM_ALLOC_CLASS_MASK; 2040 2041 /* 2042 * The page daemon is allowed to dig deeper into the free page list. 2043 */ 2044 if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) 2045 req_class = VM_ALLOC_SYSTEM; 2046 2047 /* 2048 * Do not allocate reserved pages unless the req has asked for it. 2049 */ 2050 mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE); 2051 if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved || 2052 (req_class == VM_ALLOC_SYSTEM && 2053 vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) || 2054 (req_class == VM_ALLOC_INTERRUPT && 2055 vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) 2056 m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0); 2057 else { 2058 mtx_unlock(&vm_page_queue_free_mtx); 2059 atomic_add_int(&vm_pageout_deficit, 2060 max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); 2061 pagedaemon_wakeup(); 2062 return (NULL); 2063 } 2064 if (m == NULL) { 2065 mtx_unlock(&vm_page_queue_free_mtx); 2066 return (NULL); 2067 } 2068 drop = vm_page_alloc_init(m); 2069 mtx_unlock(&vm_page_queue_free_mtx); 2070 2071 /* 2072 * Initialize the page. Only the PG_ZERO flag is inherited. 2073 */ 2074 m->aflags = 0; 2075 flags = 0; 2076 if ((req & VM_ALLOC_ZERO) != 0) 2077 flags = PG_ZERO; 2078 m->flags &= flags; 2079 if ((req & VM_ALLOC_WIRED) != 0) { 2080 /* 2081 * The page lock is not required for wiring a page that does 2082 * not belong to an object. 2083 */ 2084 atomic_add_int(&vm_cnt.v_wire_count, 1); 2085 m->wire_count = 1; 2086 } 2087 /* Unmanaged pages don't use "act_count". */ 2088 m->oflags = VPO_UNMANAGED; 2089 if (drop != NULL) 2090 vdrop(drop); 2091 if (vm_paging_needed()) 2092 pagedaemon_wakeup(); 2093 return (m); 2094 } 2095 2096 /* 2097 * vm_wait: (also see VM_WAIT macro) 2098 * 2099 * Sleep until free pages are available for allocation. 2100 * - Called in various places before memory allocations. 2101 */ 2102 void 2103 vm_wait(void) 2104 { 2105 2106 mtx_lock(&vm_page_queue_free_mtx); 2107 if (curproc == pageproc) { 2108 vm_pageout_pages_needed = 1; 2109 msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx, 2110 PDROP | PSWP, "VMWait", 0); 2111 } else { 2112 if (!vm_pages_needed) { 2113 vm_pages_needed = 1; 2114 wakeup(&vm_pages_needed); 2115 } 2116 msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM, 2117 "vmwait", 0); 2118 } 2119 } 2120 2121 /* 2122 * vm_waitpfault: (also see VM_WAITPFAULT macro) 2123 * 2124 * Sleep until free pages are available for allocation. 2125 * - Called only in vm_fault so that processes page faulting 2126 * can be easily tracked. 2127 * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing 2128 * processes will be able to grab memory first. Do not change 2129 * this balance without careful testing first. 2130 */ 2131 void 2132 vm_waitpfault(void) 2133 { 2134 2135 mtx_lock(&vm_page_queue_free_mtx); 2136 if (!vm_pages_needed) { 2137 vm_pages_needed = 1; 2138 wakeup(&vm_pages_needed); 2139 } 2140 msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER, 2141 "pfault", 0); 2142 } 2143 2144 struct vm_pagequeue * 2145 vm_page_pagequeue(vm_page_t m) 2146 { 2147 2148 return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]); 2149 } 2150 2151 /* 2152 * vm_page_dequeue: 2153 * 2154 * Remove the given page from its current page queue. 2155 * 2156 * The page must be locked. 2157 */ 2158 void 2159 vm_page_dequeue(vm_page_t m) 2160 { 2161 struct vm_pagequeue *pq; 2162 2163 vm_page_assert_locked(m); 2164 KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued", 2165 m)); 2166 pq = vm_page_pagequeue(m); 2167 vm_pagequeue_lock(pq); 2168 m->queue = PQ_NONE; 2169 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2170 vm_pagequeue_cnt_dec(pq); 2171 vm_pagequeue_unlock(pq); 2172 } 2173 2174 /* 2175 * vm_page_dequeue_locked: 2176 * 2177 * Remove the given page from its current page queue. 2178 * 2179 * The page and page queue must be locked. 2180 */ 2181 void 2182 vm_page_dequeue_locked(vm_page_t m) 2183 { 2184 struct vm_pagequeue *pq; 2185 2186 vm_page_lock_assert(m, MA_OWNED); 2187 pq = vm_page_pagequeue(m); 2188 vm_pagequeue_assert_locked(pq); 2189 m->queue = PQ_NONE; 2190 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2191 vm_pagequeue_cnt_dec(pq); 2192 } 2193 2194 /* 2195 * vm_page_enqueue: 2196 * 2197 * Add the given page to the specified page queue. 2198 * 2199 * The page must be locked. 2200 */ 2201 static void 2202 vm_page_enqueue(uint8_t queue, vm_page_t m) 2203 { 2204 struct vm_pagequeue *pq; 2205 2206 vm_page_lock_assert(m, MA_OWNED); 2207 KASSERT(queue < PQ_COUNT, 2208 ("vm_page_enqueue: invalid queue %u request for page %p", 2209 queue, m)); 2210 pq = &vm_phys_domain(m)->vmd_pagequeues[queue]; 2211 vm_pagequeue_lock(pq); 2212 m->queue = queue; 2213 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 2214 vm_pagequeue_cnt_inc(pq); 2215 vm_pagequeue_unlock(pq); 2216 } 2217 2218 /* 2219 * vm_page_requeue: 2220 * 2221 * Move the given page to the tail of its current page queue. 2222 * 2223 * The page must be locked. 2224 */ 2225 void 2226 vm_page_requeue(vm_page_t m) 2227 { 2228 struct vm_pagequeue *pq; 2229 2230 vm_page_lock_assert(m, MA_OWNED); 2231 KASSERT(m->queue != PQ_NONE, 2232 ("vm_page_requeue: page %p is not queued", m)); 2233 pq = vm_page_pagequeue(m); 2234 vm_pagequeue_lock(pq); 2235 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2236 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 2237 vm_pagequeue_unlock(pq); 2238 } 2239 2240 /* 2241 * vm_page_requeue_locked: 2242 * 2243 * Move the given page to the tail of its current page queue. 2244 * 2245 * The page queue must be locked. 2246 */ 2247 void 2248 vm_page_requeue_locked(vm_page_t m) 2249 { 2250 struct vm_pagequeue *pq; 2251 2252 KASSERT(m->queue != PQ_NONE, 2253 ("vm_page_requeue_locked: page %p is not queued", m)); 2254 pq = vm_page_pagequeue(m); 2255 vm_pagequeue_assert_locked(pq); 2256 TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); 2257 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 2258 } 2259 2260 /* 2261 * vm_page_activate: 2262 * 2263 * Put the specified page on the active list (if appropriate). 2264 * Ensure that act_count is at least ACT_INIT but do not otherwise 2265 * mess with it. 2266 * 2267 * The page must be locked. 2268 */ 2269 void 2270 vm_page_activate(vm_page_t m) 2271 { 2272 int queue; 2273 2274 vm_page_lock_assert(m, MA_OWNED); 2275 if ((queue = m->queue) != PQ_ACTIVE) { 2276 if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { 2277 if (m->act_count < ACT_INIT) 2278 m->act_count = ACT_INIT; 2279 if (queue != PQ_NONE) 2280 vm_page_dequeue(m); 2281 vm_page_enqueue(PQ_ACTIVE, m); 2282 } else 2283 KASSERT(queue == PQ_NONE, 2284 ("vm_page_activate: wired page %p is queued", m)); 2285 } else { 2286 if (m->act_count < ACT_INIT) 2287 m->act_count = ACT_INIT; 2288 } 2289 } 2290 2291 /* 2292 * vm_page_free_wakeup: 2293 * 2294 * Helper routine for vm_page_free_toq() and vm_page_cache(). This 2295 * routine is called when a page has been added to the cache or free 2296 * queues. 2297 * 2298 * The page queues must be locked. 2299 */ 2300 static inline void 2301 vm_page_free_wakeup(void) 2302 { 2303 2304 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 2305 /* 2306 * if pageout daemon needs pages, then tell it that there are 2307 * some free. 2308 */ 2309 if (vm_pageout_pages_needed && 2310 vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) { 2311 wakeup(&vm_pageout_pages_needed); 2312 vm_pageout_pages_needed = 0; 2313 } 2314 /* 2315 * wakeup processes that are waiting on memory if we hit a 2316 * high water mark. And wakeup scheduler process if we have 2317 * lots of memory. this process will swapin processes. 2318 */ 2319 if (vm_pages_needed && !vm_page_count_min()) { 2320 vm_pages_needed = 0; 2321 wakeup(&vm_cnt.v_free_count); 2322 } 2323 } 2324 2325 /* 2326 * Turn a cached page into a free page, by changing its attributes. 2327 * Keep the statistics up-to-date. 2328 * 2329 * The free page queue must be locked. 2330 */ 2331 static void 2332 vm_page_cache_turn_free(vm_page_t m) 2333 { 2334 2335 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 2336 2337 m->object = NULL; 2338 m->valid = 0; 2339 KASSERT((m->flags & PG_CACHED) != 0, 2340 ("vm_page_cache_turn_free: page %p is not cached", m)); 2341 m->flags &= ~PG_CACHED; 2342 vm_cnt.v_cache_count--; 2343 vm_phys_freecnt_adj(m, 1); 2344 } 2345 2346 /* 2347 * vm_page_free_toq: 2348 * 2349 * Returns the given page to the free list, 2350 * disassociating it with any VM object. 2351 * 2352 * The object must be locked. The page must be locked if it is managed. 2353 */ 2354 void 2355 vm_page_free_toq(vm_page_t m) 2356 { 2357 2358 if ((m->oflags & VPO_UNMANAGED) == 0) { 2359 vm_page_lock_assert(m, MA_OWNED); 2360 KASSERT(!pmap_page_is_mapped(m), 2361 ("vm_page_free_toq: freeing mapped page %p", m)); 2362 } else 2363 KASSERT(m->queue == PQ_NONE, 2364 ("vm_page_free_toq: unmanaged page %p is queued", m)); 2365 PCPU_INC(cnt.v_tfree); 2366 2367 if (vm_page_sbusied(m)) 2368 panic("vm_page_free: freeing busy page %p", m); 2369 2370 /* 2371 * Unqueue, then remove page. Note that we cannot destroy 2372 * the page here because we do not want to call the pager's 2373 * callback routine until after we've put the page on the 2374 * appropriate free queue. 2375 */ 2376 vm_page_remque(m); 2377 vm_page_remove(m); 2378 2379 /* 2380 * If fictitious remove object association and 2381 * return, otherwise delay object association removal. 2382 */ 2383 if ((m->flags & PG_FICTITIOUS) != 0) { 2384 return; 2385 } 2386 2387 m->valid = 0; 2388 vm_page_undirty(m); 2389 2390 if (m->wire_count != 0) 2391 panic("vm_page_free: freeing wired page %p", m); 2392 if (m->hold_count != 0) { 2393 m->flags &= ~PG_ZERO; 2394 KASSERT((m->flags & PG_UNHOLDFREE) == 0, 2395 ("vm_page_free: freeing PG_UNHOLDFREE page %p", m)); 2396 m->flags |= PG_UNHOLDFREE; 2397 } else { 2398 /* 2399 * Restore the default memory attribute to the page. 2400 */ 2401 if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) 2402 pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); 2403 2404 /* 2405 * Insert the page into the physical memory allocator's 2406 * cache/free page queues. 2407 */ 2408 mtx_lock(&vm_page_queue_free_mtx); 2409 vm_phys_freecnt_adj(m, 1); 2410 #if VM_NRESERVLEVEL > 0 2411 if (!vm_reserv_free_page(m)) 2412 #else 2413 if (TRUE) 2414 #endif 2415 vm_phys_free_pages(m, 0); 2416 if ((m->flags & PG_ZERO) != 0) 2417 ++vm_page_zero_count; 2418 else 2419 vm_page_zero_idle_wakeup(); 2420 vm_page_free_wakeup(); 2421 mtx_unlock(&vm_page_queue_free_mtx); 2422 } 2423 } 2424 2425 /* 2426 * vm_page_wire: 2427 * 2428 * Mark this page as wired down by yet 2429 * another map, removing it from paging queues 2430 * as necessary. 2431 * 2432 * If the page is fictitious, then its wire count must remain one. 2433 * 2434 * The page must be locked. 2435 */ 2436 void 2437 vm_page_wire(vm_page_t m) 2438 { 2439 2440 /* 2441 * Only bump the wire statistics if the page is not already wired, 2442 * and only unqueue the page if it is on some queue (if it is unmanaged 2443 * it is already off the queues). 2444 */ 2445 vm_page_lock_assert(m, MA_OWNED); 2446 if ((m->flags & PG_FICTITIOUS) != 0) { 2447 KASSERT(m->wire_count == 1, 2448 ("vm_page_wire: fictitious page %p's wire count isn't one", 2449 m)); 2450 return; 2451 } 2452 if (m->wire_count == 0) { 2453 KASSERT((m->oflags & VPO_UNMANAGED) == 0 || 2454 m->queue == PQ_NONE, 2455 ("vm_page_wire: unmanaged page %p is queued", m)); 2456 vm_page_remque(m); 2457 atomic_add_int(&vm_cnt.v_wire_count, 1); 2458 } 2459 m->wire_count++; 2460 KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m)); 2461 } 2462 2463 /* 2464 * vm_page_unwire: 2465 * 2466 * Release one wiring of the specified page, potentially allowing it to be 2467 * paged out. Returns TRUE if the number of wirings transitions to zero and 2468 * FALSE otherwise. 2469 * 2470 * Only managed pages belonging to an object can be paged out. If the number 2471 * of wirings transitions to zero and the page is eligible for page out, then 2472 * the page is added to the specified paging queue (unless PQ_NONE is 2473 * specified). 2474 * 2475 * If a page is fictitious, then its wire count must always be one. 2476 * 2477 * A managed page must be locked. 2478 */ 2479 boolean_t 2480 vm_page_unwire(vm_page_t m, uint8_t queue) 2481 { 2482 2483 KASSERT(queue < PQ_COUNT || queue == PQ_NONE, 2484 ("vm_page_unwire: invalid queue %u request for page %p", 2485 queue, m)); 2486 if ((m->oflags & VPO_UNMANAGED) == 0) 2487 vm_page_assert_locked(m); 2488 if ((m->flags & PG_FICTITIOUS) != 0) { 2489 KASSERT(m->wire_count == 1, 2490 ("vm_page_unwire: fictitious page %p's wire count isn't one", m)); 2491 return (FALSE); 2492 } 2493 if (m->wire_count > 0) { 2494 m->wire_count--; 2495 if (m->wire_count == 0) { 2496 atomic_subtract_int(&vm_cnt.v_wire_count, 1); 2497 if ((m->oflags & VPO_UNMANAGED) == 0 && 2498 m->object != NULL && queue != PQ_NONE) { 2499 if (queue == PQ_INACTIVE) 2500 m->flags &= ~PG_WINATCFLS; 2501 vm_page_enqueue(queue, m); 2502 } 2503 return (TRUE); 2504 } else 2505 return (FALSE); 2506 } else 2507 panic("vm_page_unwire: page %p's wire count is zero", m); 2508 } 2509 2510 /* 2511 * Move the specified page to the inactive queue. 2512 * 2513 * Many pages placed on the inactive queue should actually go 2514 * into the cache, but it is difficult to figure out which. What 2515 * we do instead, if the inactive target is well met, is to put 2516 * clean pages at the head of the inactive queue instead of the tail. 2517 * This will cause them to be moved to the cache more quickly and 2518 * if not actively re-referenced, reclaimed more quickly. If we just 2519 * stick these pages at the end of the inactive queue, heavy filesystem 2520 * meta-data accesses can cause an unnecessary paging load on memory bound 2521 * processes. This optimization causes one-time-use metadata to be 2522 * reused more quickly. 2523 * 2524 * Normally noreuse is FALSE, resulting in LRU operation. noreuse is set 2525 * to TRUE if we want this page to be 'as if it were placed in the cache', 2526 * except without unmapping it from the process address space. In 2527 * practice this is implemented by inserting the page at the head of the 2528 * queue, using a marker page to guide FIFO insertion ordering. 2529 * 2530 * The page must be locked. 2531 */ 2532 static inline void 2533 _vm_page_deactivate(vm_page_t m, boolean_t noreuse) 2534 { 2535 struct vm_pagequeue *pq; 2536 int queue; 2537 2538 vm_page_assert_locked(m); 2539 2540 /* 2541 * Ignore if the page is already inactive, unless it is unlikely to be 2542 * reactivated. 2543 */ 2544 if ((queue = m->queue) == PQ_INACTIVE && !noreuse) 2545 return; 2546 if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { 2547 pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE]; 2548 /* Avoid multiple acquisitions of the inactive queue lock. */ 2549 if (queue == PQ_INACTIVE) { 2550 vm_pagequeue_lock(pq); 2551 vm_page_dequeue_locked(m); 2552 } else { 2553 if (queue != PQ_NONE) 2554 vm_page_dequeue(m); 2555 m->flags &= ~PG_WINATCFLS; 2556 vm_pagequeue_lock(pq); 2557 } 2558 m->queue = PQ_INACTIVE; 2559 if (noreuse) 2560 TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead, 2561 m, plinks.q); 2562 else 2563 TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); 2564 vm_pagequeue_cnt_inc(pq); 2565 vm_pagequeue_unlock(pq); 2566 } 2567 } 2568 2569 /* 2570 * Move the specified page to the inactive queue. 2571 * 2572 * The page must be locked. 2573 */ 2574 void 2575 vm_page_deactivate(vm_page_t m) 2576 { 2577 2578 _vm_page_deactivate(m, FALSE); 2579 } 2580 2581 /* 2582 * Move the specified page to the inactive queue with the expectation 2583 * that it is unlikely to be reused. 2584 * 2585 * The page must be locked. 2586 */ 2587 void 2588 vm_page_deactivate_noreuse(vm_page_t m) 2589 { 2590 2591 _vm_page_deactivate(m, TRUE); 2592 } 2593 2594 /* 2595 * vm_page_try_to_cache: 2596 * 2597 * Returns 0 on failure, 1 on success 2598 */ 2599 int 2600 vm_page_try_to_cache(vm_page_t m) 2601 { 2602 2603 vm_page_lock_assert(m, MA_OWNED); 2604 VM_OBJECT_ASSERT_WLOCKED(m->object); 2605 if (m->dirty || m->hold_count || m->wire_count || 2606 (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) 2607 return (0); 2608 pmap_remove_all(m); 2609 if (m->dirty) 2610 return (0); 2611 vm_page_cache(m); 2612 return (1); 2613 } 2614 2615 /* 2616 * vm_page_try_to_free() 2617 * 2618 * Attempt to free the page. If we cannot free it, we do nothing. 2619 * 1 is returned on success, 0 on failure. 2620 */ 2621 int 2622 vm_page_try_to_free(vm_page_t m) 2623 { 2624 2625 vm_page_lock_assert(m, MA_OWNED); 2626 if (m->object != NULL) 2627 VM_OBJECT_ASSERT_WLOCKED(m->object); 2628 if (m->dirty || m->hold_count || m->wire_count || 2629 (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) 2630 return (0); 2631 pmap_remove_all(m); 2632 if (m->dirty) 2633 return (0); 2634 vm_page_free(m); 2635 return (1); 2636 } 2637 2638 /* 2639 * vm_page_cache 2640 * 2641 * Put the specified page onto the page cache queue (if appropriate). 2642 * 2643 * The object and page must be locked. 2644 */ 2645 void 2646 vm_page_cache(vm_page_t m) 2647 { 2648 vm_object_t object; 2649 boolean_t cache_was_empty; 2650 2651 vm_page_lock_assert(m, MA_OWNED); 2652 object = m->object; 2653 VM_OBJECT_ASSERT_WLOCKED(object); 2654 if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) || 2655 m->hold_count || m->wire_count) 2656 panic("vm_page_cache: attempting to cache busy page"); 2657 KASSERT(!pmap_page_is_mapped(m), 2658 ("vm_page_cache: page %p is mapped", m)); 2659 KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m)); 2660 if (m->valid == 0 || object->type == OBJT_DEFAULT || 2661 (object->type == OBJT_SWAP && 2662 !vm_pager_has_page(object, m->pindex, NULL, NULL))) { 2663 /* 2664 * Hypothesis: A cache-eligible page belonging to a 2665 * default object or swap object but without a backing 2666 * store must be zero filled. 2667 */ 2668 vm_page_free(m); 2669 return; 2670 } 2671 KASSERT((m->flags & PG_CACHED) == 0, 2672 ("vm_page_cache: page %p is already cached", m)); 2673 2674 /* 2675 * Remove the page from the paging queues. 2676 */ 2677 vm_page_remque(m); 2678 2679 /* 2680 * Remove the page from the object's collection of resident 2681 * pages. 2682 */ 2683 vm_radix_remove(&object->rtree, m->pindex); 2684 TAILQ_REMOVE(&object->memq, m, listq); 2685 object->resident_page_count--; 2686 2687 /* 2688 * Restore the default memory attribute to the page. 2689 */ 2690 if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) 2691 pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); 2692 2693 /* 2694 * Insert the page into the object's collection of cached pages 2695 * and the physical memory allocator's cache/free page queues. 2696 */ 2697 m->flags &= ~PG_ZERO; 2698 mtx_lock(&vm_page_queue_free_mtx); 2699 cache_was_empty = vm_radix_is_empty(&object->cache); 2700 if (vm_radix_insert(&object->cache, m)) { 2701 mtx_unlock(&vm_page_queue_free_mtx); 2702 if (object->resident_page_count == 0) 2703 vdrop(object->handle); 2704 m->object = NULL; 2705 vm_page_free(m); 2706 return; 2707 } 2708 2709 /* 2710 * The above call to vm_radix_insert() could reclaim the one pre- 2711 * existing cached page from this object, resulting in a call to 2712 * vdrop(). 2713 */ 2714 if (!cache_was_empty) 2715 cache_was_empty = vm_radix_is_singleton(&object->cache); 2716 2717 m->flags |= PG_CACHED; 2718 vm_cnt.v_cache_count++; 2719 PCPU_INC(cnt.v_tcached); 2720 #if VM_NRESERVLEVEL > 0 2721 if (!vm_reserv_free_page(m)) { 2722 #else 2723 if (TRUE) { 2724 #endif 2725 vm_phys_free_pages(m, 0); 2726 } 2727 vm_page_free_wakeup(); 2728 mtx_unlock(&vm_page_queue_free_mtx); 2729 2730 /* 2731 * Increment the vnode's hold count if this is the object's only 2732 * cached page. Decrement the vnode's hold count if this was 2733 * the object's only resident page. 2734 */ 2735 if (object->type == OBJT_VNODE) { 2736 if (cache_was_empty && object->resident_page_count != 0) 2737 vhold(object->handle); 2738 else if (!cache_was_empty && object->resident_page_count == 0) 2739 vdrop(object->handle); 2740 } 2741 } 2742 2743 /* 2744 * vm_page_advise 2745 * 2746 * Deactivate or do nothing, as appropriate. 2747 * 2748 * The object and page must be locked. 2749 */ 2750 void 2751 vm_page_advise(vm_page_t m, int advice) 2752 { 2753 2754 vm_page_assert_locked(m); 2755 VM_OBJECT_ASSERT_WLOCKED(m->object); 2756 if (advice == MADV_FREE) 2757 /* 2758 * Mark the page clean. This will allow the page to be freed 2759 * up by the system. However, such pages are often reused 2760 * quickly by malloc() so we do not do anything that would 2761 * cause a page fault if we can help it. 2762 * 2763 * Specifically, we do not try to actually free the page now 2764 * nor do we try to put it in the cache (which would cause a 2765 * page fault on reuse). 2766 * 2767 * But we do make the page as freeable as we can without 2768 * actually taking the step of unmapping it. 2769 */ 2770 m->dirty = 0; 2771 else if (advice != MADV_DONTNEED) 2772 return; 2773 2774 /* 2775 * Clear any references to the page. Otherwise, the page daemon will 2776 * immediately reactivate the page. 2777 */ 2778 vm_page_aflag_clear(m, PGA_REFERENCED); 2779 2780 if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m)) 2781 vm_page_dirty(m); 2782 2783 /* 2784 * Place clean pages at the head of the inactive queue rather than the 2785 * tail, thus defeating the queue's LRU operation and ensuring that the 2786 * page will be reused quickly. 2787 */ 2788 _vm_page_deactivate(m, m->dirty == 0); 2789 } 2790 2791 /* 2792 * Grab a page, waiting until we are waken up due to the page 2793 * changing state. We keep on waiting, if the page continues 2794 * to be in the object. If the page doesn't exist, first allocate it 2795 * and then conditionally zero it. 2796 * 2797 * This routine may sleep. 2798 * 2799 * The object must be locked on entry. The lock will, however, be released 2800 * and reacquired if the routine sleeps. 2801 */ 2802 vm_page_t 2803 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags) 2804 { 2805 vm_page_t m; 2806 int sleep; 2807 2808 VM_OBJECT_ASSERT_WLOCKED(object); 2809 KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || 2810 (allocflags & VM_ALLOC_IGN_SBUSY) != 0, 2811 ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch")); 2812 retrylookup: 2813 if ((m = vm_page_lookup(object, pindex)) != NULL) { 2814 sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ? 2815 vm_page_xbusied(m) : vm_page_busied(m); 2816 if (sleep) { 2817 if ((allocflags & VM_ALLOC_NOWAIT) != 0) 2818 return (NULL); 2819 /* 2820 * Reference the page before unlocking and 2821 * sleeping so that the page daemon is less 2822 * likely to reclaim it. 2823 */ 2824 vm_page_aflag_set(m, PGA_REFERENCED); 2825 vm_page_lock(m); 2826 VM_OBJECT_WUNLOCK(object); 2827 vm_page_busy_sleep(m, "pgrbwt"); 2828 VM_OBJECT_WLOCK(object); 2829 goto retrylookup; 2830 } else { 2831 if ((allocflags & VM_ALLOC_WIRED) != 0) { 2832 vm_page_lock(m); 2833 vm_page_wire(m); 2834 vm_page_unlock(m); 2835 } 2836 if ((allocflags & 2837 (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) 2838 vm_page_xbusy(m); 2839 if ((allocflags & VM_ALLOC_SBUSY) != 0) 2840 vm_page_sbusy(m); 2841 return (m); 2842 } 2843 } 2844 m = vm_page_alloc(object, pindex, allocflags); 2845 if (m == NULL) { 2846 if ((allocflags & VM_ALLOC_NOWAIT) != 0) 2847 return (NULL); 2848 VM_OBJECT_WUNLOCK(object); 2849 VM_WAIT; 2850 VM_OBJECT_WLOCK(object); 2851 goto retrylookup; 2852 } else if (m->valid != 0) 2853 return (m); 2854 if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0) 2855 pmap_zero_page(m); 2856 return (m); 2857 } 2858 2859 /* 2860 * Mapping function for valid or dirty bits in a page. 2861 * 2862 * Inputs are required to range within a page. 2863 */ 2864 vm_page_bits_t 2865 vm_page_bits(int base, int size) 2866 { 2867 int first_bit; 2868 int last_bit; 2869 2870 KASSERT( 2871 base + size <= PAGE_SIZE, 2872 ("vm_page_bits: illegal base/size %d/%d", base, size) 2873 ); 2874 2875 if (size == 0) /* handle degenerate case */ 2876 return (0); 2877 2878 first_bit = base >> DEV_BSHIFT; 2879 last_bit = (base + size - 1) >> DEV_BSHIFT; 2880 2881 return (((vm_page_bits_t)2 << last_bit) - 2882 ((vm_page_bits_t)1 << first_bit)); 2883 } 2884 2885 /* 2886 * vm_page_set_valid_range: 2887 * 2888 * Sets portions of a page valid. The arguments are expected 2889 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive 2890 * of any partial chunks touched by the range. The invalid portion of 2891 * such chunks will be zeroed. 2892 * 2893 * (base + size) must be less then or equal to PAGE_SIZE. 2894 */ 2895 void 2896 vm_page_set_valid_range(vm_page_t m, int base, int size) 2897 { 2898 int endoff, frag; 2899 2900 VM_OBJECT_ASSERT_WLOCKED(m->object); 2901 if (size == 0) /* handle degenerate case */ 2902 return; 2903 2904 /* 2905 * If the base is not DEV_BSIZE aligned and the valid 2906 * bit is clear, we have to zero out a portion of the 2907 * first block. 2908 */ 2909 if ((frag = base & ~(DEV_BSIZE - 1)) != base && 2910 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0) 2911 pmap_zero_page_area(m, frag, base - frag); 2912 2913 /* 2914 * If the ending offset is not DEV_BSIZE aligned and the 2915 * valid bit is clear, we have to zero out a portion of 2916 * the last block. 2917 */ 2918 endoff = base + size; 2919 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff && 2920 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0) 2921 pmap_zero_page_area(m, endoff, 2922 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); 2923 2924 /* 2925 * Assert that no previously invalid block that is now being validated 2926 * is already dirty. 2927 */ 2928 KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0, 2929 ("vm_page_set_valid_range: page %p is dirty", m)); 2930 2931 /* 2932 * Set valid bits inclusive of any overlap. 2933 */ 2934 m->valid |= vm_page_bits(base, size); 2935 } 2936 2937 /* 2938 * Clear the given bits from the specified page's dirty field. 2939 */ 2940 static __inline void 2941 vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits) 2942 { 2943 uintptr_t addr; 2944 #if PAGE_SIZE < 16384 2945 int shift; 2946 #endif 2947 2948 /* 2949 * If the object is locked and the page is neither exclusive busy nor 2950 * write mapped, then the page's dirty field cannot possibly be 2951 * set by a concurrent pmap operation. 2952 */ 2953 VM_OBJECT_ASSERT_WLOCKED(m->object); 2954 if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m)) 2955 m->dirty &= ~pagebits; 2956 else { 2957 /* 2958 * The pmap layer can call vm_page_dirty() without 2959 * holding a distinguished lock. The combination of 2960 * the object's lock and an atomic operation suffice 2961 * to guarantee consistency of the page dirty field. 2962 * 2963 * For PAGE_SIZE == 32768 case, compiler already 2964 * properly aligns the dirty field, so no forcible 2965 * alignment is needed. Only require existence of 2966 * atomic_clear_64 when page size is 32768. 2967 */ 2968 addr = (uintptr_t)&m->dirty; 2969 #if PAGE_SIZE == 32768 2970 atomic_clear_64((uint64_t *)addr, pagebits); 2971 #elif PAGE_SIZE == 16384 2972 atomic_clear_32((uint32_t *)addr, pagebits); 2973 #else /* PAGE_SIZE <= 8192 */ 2974 /* 2975 * Use a trick to perform a 32-bit atomic on the 2976 * containing aligned word, to not depend on the existence 2977 * of atomic_clear_{8, 16}. 2978 */ 2979 shift = addr & (sizeof(uint32_t) - 1); 2980 #if BYTE_ORDER == BIG_ENDIAN 2981 shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY; 2982 #else 2983 shift *= NBBY; 2984 #endif 2985 addr &= ~(sizeof(uint32_t) - 1); 2986 atomic_clear_32((uint32_t *)addr, pagebits << shift); 2987 #endif /* PAGE_SIZE */ 2988 } 2989 } 2990 2991 /* 2992 * vm_page_set_validclean: 2993 * 2994 * Sets portions of a page valid and clean. The arguments are expected 2995 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive 2996 * of any partial chunks touched by the range. The invalid portion of 2997 * such chunks will be zero'd. 2998 * 2999 * (base + size) must be less then or equal to PAGE_SIZE. 3000 */ 3001 void 3002 vm_page_set_validclean(vm_page_t m, int base, int size) 3003 { 3004 vm_page_bits_t oldvalid, pagebits; 3005 int endoff, frag; 3006 3007 VM_OBJECT_ASSERT_WLOCKED(m->object); 3008 if (size == 0) /* handle degenerate case */ 3009 return; 3010 3011 /* 3012 * If the base is not DEV_BSIZE aligned and the valid 3013 * bit is clear, we have to zero out a portion of the 3014 * first block. 3015 */ 3016 if ((frag = base & ~(DEV_BSIZE - 1)) != base && 3017 (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0) 3018 pmap_zero_page_area(m, frag, base - frag); 3019 3020 /* 3021 * If the ending offset is not DEV_BSIZE aligned and the 3022 * valid bit is clear, we have to zero out a portion of 3023 * the last block. 3024 */ 3025 endoff = base + size; 3026 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff && 3027 (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0) 3028 pmap_zero_page_area(m, endoff, 3029 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); 3030 3031 /* 3032 * Set valid, clear dirty bits. If validating the entire 3033 * page we can safely clear the pmap modify bit. We also 3034 * use this opportunity to clear the VPO_NOSYNC flag. If a process 3035 * takes a write fault on a MAP_NOSYNC memory area the flag will 3036 * be set again. 3037 * 3038 * We set valid bits inclusive of any overlap, but we can only 3039 * clear dirty bits for DEV_BSIZE chunks that are fully within 3040 * the range. 3041 */ 3042 oldvalid = m->valid; 3043 pagebits = vm_page_bits(base, size); 3044 m->valid |= pagebits; 3045 #if 0 /* NOT YET */ 3046 if ((frag = base & (DEV_BSIZE - 1)) != 0) { 3047 frag = DEV_BSIZE - frag; 3048 base += frag; 3049 size -= frag; 3050 if (size < 0) 3051 size = 0; 3052 } 3053 pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1)); 3054 #endif 3055 if (base == 0 && size == PAGE_SIZE) { 3056 /* 3057 * The page can only be modified within the pmap if it is 3058 * mapped, and it can only be mapped if it was previously 3059 * fully valid. 3060 */ 3061 if (oldvalid == VM_PAGE_BITS_ALL) 3062 /* 3063 * Perform the pmap_clear_modify() first. Otherwise, 3064 * a concurrent pmap operation, such as 3065 * pmap_protect(), could clear a modification in the 3066 * pmap and set the dirty field on the page before 3067 * pmap_clear_modify() had begun and after the dirty 3068 * field was cleared here. 3069 */ 3070 pmap_clear_modify(m); 3071 m->dirty = 0; 3072 m->oflags &= ~VPO_NOSYNC; 3073 } else if (oldvalid != VM_PAGE_BITS_ALL) 3074 m->dirty &= ~pagebits; 3075 else 3076 vm_page_clear_dirty_mask(m, pagebits); 3077 } 3078 3079 void 3080 vm_page_clear_dirty(vm_page_t m, int base, int size) 3081 { 3082 3083 vm_page_clear_dirty_mask(m, vm_page_bits(base, size)); 3084 } 3085 3086 /* 3087 * vm_page_set_invalid: 3088 * 3089 * Invalidates DEV_BSIZE'd chunks within a page. Both the 3090 * valid and dirty bits for the effected areas are cleared. 3091 */ 3092 void 3093 vm_page_set_invalid(vm_page_t m, int base, int size) 3094 { 3095 vm_page_bits_t bits; 3096 vm_object_t object; 3097 3098 object = m->object; 3099 VM_OBJECT_ASSERT_WLOCKED(object); 3100 if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) + 3101 size >= object->un_pager.vnp.vnp_size) 3102 bits = VM_PAGE_BITS_ALL; 3103 else 3104 bits = vm_page_bits(base, size); 3105 if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL && 3106 bits != 0) 3107 pmap_remove_all(m); 3108 KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) || 3109 !pmap_page_is_mapped(m), 3110 ("vm_page_set_invalid: page %p is mapped", m)); 3111 m->valid &= ~bits; 3112 m->dirty &= ~bits; 3113 } 3114 3115 /* 3116 * vm_page_zero_invalid() 3117 * 3118 * The kernel assumes that the invalid portions of a page contain 3119 * garbage, but such pages can be mapped into memory by user code. 3120 * When this occurs, we must zero out the non-valid portions of the 3121 * page so user code sees what it expects. 3122 * 3123 * Pages are most often semi-valid when the end of a file is mapped 3124 * into memory and the file's size is not page aligned. 3125 */ 3126 void 3127 vm_page_zero_invalid(vm_page_t m, boolean_t setvalid) 3128 { 3129 int b; 3130 int i; 3131 3132 VM_OBJECT_ASSERT_WLOCKED(m->object); 3133 /* 3134 * Scan the valid bits looking for invalid sections that 3135 * must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the 3136 * valid bit may be set ) have already been zeroed by 3137 * vm_page_set_validclean(). 3138 */ 3139 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) { 3140 if (i == (PAGE_SIZE / DEV_BSIZE) || 3141 (m->valid & ((vm_page_bits_t)1 << i))) { 3142 if (i > b) { 3143 pmap_zero_page_area(m, 3144 b << DEV_BSHIFT, (i - b) << DEV_BSHIFT); 3145 } 3146 b = i + 1; 3147 } 3148 } 3149 3150 /* 3151 * setvalid is TRUE when we can safely set the zero'd areas 3152 * as being valid. We can do this if there are no cache consistancy 3153 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS. 3154 */ 3155 if (setvalid) 3156 m->valid = VM_PAGE_BITS_ALL; 3157 } 3158 3159 /* 3160 * vm_page_is_valid: 3161 * 3162 * Is (partial) page valid? Note that the case where size == 0 3163 * will return FALSE in the degenerate case where the page is 3164 * entirely invalid, and TRUE otherwise. 3165 */ 3166 int 3167 vm_page_is_valid(vm_page_t m, int base, int size) 3168 { 3169 vm_page_bits_t bits; 3170 3171 VM_OBJECT_ASSERT_LOCKED(m->object); 3172 bits = vm_page_bits(base, size); 3173 return (m->valid != 0 && (m->valid & bits) == bits); 3174 } 3175 3176 /* 3177 * vm_page_ps_is_valid: 3178 * 3179 * Returns TRUE if the entire (super)page is valid and FALSE otherwise. 3180 */ 3181 boolean_t 3182 vm_page_ps_is_valid(vm_page_t m) 3183 { 3184 int i, npages; 3185 3186 VM_OBJECT_ASSERT_LOCKED(m->object); 3187 npages = atop(pagesizes[m->psind]); 3188 3189 /* 3190 * The physically contiguous pages that make up a superpage, i.e., a 3191 * page with a page size index ("psind") greater than zero, will 3192 * occupy adjacent entries in vm_page_array[]. 3193 */ 3194 for (i = 0; i < npages; i++) { 3195 if (m[i].valid != VM_PAGE_BITS_ALL) 3196 return (FALSE); 3197 } 3198 return (TRUE); 3199 } 3200 3201 /* 3202 * Set the page's dirty bits if the page is modified. 3203 */ 3204 void 3205 vm_page_test_dirty(vm_page_t m) 3206 { 3207 3208 VM_OBJECT_ASSERT_WLOCKED(m->object); 3209 if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m)) 3210 vm_page_dirty(m); 3211 } 3212 3213 void 3214 vm_page_lock_KBI(vm_page_t m, const char *file, int line) 3215 { 3216 3217 mtx_lock_flags_(vm_page_lockptr(m), 0, file, line); 3218 } 3219 3220 void 3221 vm_page_unlock_KBI(vm_page_t m, const char *file, int line) 3222 { 3223 3224 mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line); 3225 } 3226 3227 int 3228 vm_page_trylock_KBI(vm_page_t m, const char *file, int line) 3229 { 3230 3231 return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line)); 3232 } 3233 3234 #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 3235 void 3236 vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line) 3237 { 3238 3239 vm_page_lock_assert_KBI(m, MA_OWNED, file, line); 3240 } 3241 3242 void 3243 vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line) 3244 { 3245 3246 mtx_assert_(vm_page_lockptr(m), a, file, line); 3247 } 3248 #endif 3249 3250 #ifdef INVARIANTS 3251 void 3252 vm_page_object_lock_assert(vm_page_t m) 3253 { 3254 3255 /* 3256 * Certain of the page's fields may only be modified by the 3257 * holder of the containing object's lock or the exclusive busy. 3258 * holder. Unfortunately, the holder of the write busy is 3259 * not recorded, and thus cannot be checked here. 3260 */ 3261 if (m->object != NULL && !vm_page_xbusied(m)) 3262 VM_OBJECT_ASSERT_WLOCKED(m->object); 3263 } 3264 3265 void 3266 vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits) 3267 { 3268 3269 if ((bits & PGA_WRITEABLE) == 0) 3270 return; 3271 3272 /* 3273 * The PGA_WRITEABLE flag can only be set if the page is 3274 * managed, is exclusively busied or the object is locked. 3275 * Currently, this flag is only set by pmap_enter(). 3276 */ 3277 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 3278 ("PGA_WRITEABLE on unmanaged page")); 3279 if (!vm_page_xbusied(m)) 3280 VM_OBJECT_ASSERT_LOCKED(m->object); 3281 } 3282 #endif 3283 3284 #include "opt_ddb.h" 3285 #ifdef DDB 3286 #include <sys/kernel.h> 3287 3288 #include <ddb/ddb.h> 3289 3290 DB_SHOW_COMMAND(page, vm_page_print_page_info) 3291 { 3292 db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count); 3293 db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count); 3294 db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count); 3295 db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count); 3296 db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count); 3297 db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved); 3298 db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min); 3299 db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target); 3300 db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target); 3301 } 3302 3303 DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) 3304 { 3305 int dom; 3306 3307 db_printf("pq_free %d pq_cache %d\n", 3308 vm_cnt.v_free_count, vm_cnt.v_cache_count); 3309 for (dom = 0; dom < vm_ndomains; dom++) { 3310 db_printf( 3311 "dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n", 3312 dom, 3313 vm_dom[dom].vmd_page_count, 3314 vm_dom[dom].vmd_free_count, 3315 vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt, 3316 vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt, 3317 vm_dom[dom].vmd_pass); 3318 } 3319 } 3320 3321 DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo) 3322 { 3323 vm_page_t m; 3324 boolean_t phys; 3325 3326 if (!have_addr) { 3327 db_printf("show pginfo addr\n"); 3328 return; 3329 } 3330 3331 phys = strchr(modif, 'p') != NULL; 3332 if (phys) 3333 m = PHYS_TO_VM_PAGE(addr); 3334 else 3335 m = (vm_page_t)addr; 3336 db_printf( 3337 "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n" 3338 " af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n", 3339 m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr, 3340 m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags, 3341 m->flags, m->act_count, m->busy_lock, m->valid, m->dirty); 3342 } 3343 #endif /* DDB */ 3344