vm_fault.c revision 270630
1238104Sdes/*- 2238104Sdes * Copyright (c) 1991, 1993 3238104Sdes * The Regents of the University of California. All rights reserved. 4238104Sdes * Copyright (c) 1994 John S. Dyson 5269257Sdes * All rights reserved. 6269257Sdes * Copyright (c) 1994 David Greenman 7269257Sdes * All rights reserved. 8269257Sdes * 9269257Sdes * 10269257Sdes * This code is derived from software contributed to Berkeley by 11238104Sdes * The Mach Operating System project at Carnegie-Mellon University. 12238104Sdes * 13238104Sdes * Redistribution and use in source and binary forms, with or without 14238104Sdes * modification, are permitted provided that the following conditions 15238104Sdes * are met: 16238104Sdes * 1. Redistributions of source code must retain the above copyright 17238104Sdes * notice, this list of conditions and the following disclaimer. 18238104Sdes * 2. Redistributions in binary form must reproduce the above copyright 19238104Sdes * notice, this list of conditions and the following disclaimer in the 20238104Sdes * documentation and/or other materials provided with the distribution. 21238104Sdes * 3. All advertising materials mentioning features or use of this software 22238104Sdes * must display the following acknowledgement: 23238104Sdes * This product includes software developed by the University of 24238104Sdes * California, Berkeley and its contributors. 25269257Sdes * 4. Neither the name of the University nor the names of its contributors 26238104Sdes * may be used to endorse or promote products derived from this software 27238104Sdes * without specific prior written permission. 28238104Sdes * 29238104Sdes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30238104Sdes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31238104Sdes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32238104Sdes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33238104Sdes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34238104Sdes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35238104Sdes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36238104Sdes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37238104Sdes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38238104Sdes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39238104Sdes * SUCH DAMAGE. 40238104Sdes * 41238104Sdes * from: @(#)vm_fault.c 8.4 (Berkeley) 1/12/94 42238104Sdes * 43238104Sdes * 44238104Sdes * Copyright (c) 1987, 1990 Carnegie-Mellon University. 45238104Sdes * All rights reserved. 46238104Sdes * 47238104Sdes * Authors: Avadis Tevanian, Jr., Michael Wayne Young 48238104Sdes * 49238104Sdes * Permission to use, copy, modify and distribute this software and 50238104Sdes * its documentation is hereby granted, provided that both the copyright 51238104Sdes * notice and this permission notice appear in all copies of the 52238104Sdes * software, derivative works or modified versions, and any portions 53238104Sdes * thereof, and that both notices appear in supporting documentation. 54238104Sdes * 55238104Sdes * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 56238104Sdes * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 57238104Sdes * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 58238104Sdes * 59238104Sdes * Carnegie Mellon requests users of this software to return to 60238104Sdes * 61238104Sdes * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 62238104Sdes * School of Computer Science 63238104Sdes * Carnegie Mellon University 64238104Sdes * Pittsburgh PA 15213-3890 65238104Sdes * 66238104Sdes * any improvements or extensions that they make and grant Carnegie the 67238104Sdes * rights to redistribute these changes. 68238104Sdes */ 69238104Sdes 70238104Sdes/* 71238104Sdes * Page fault handling module. 72238104Sdes */ 73238104Sdes 74238104Sdes#include <sys/cdefs.h> 75238104Sdes__FBSDID("$FreeBSD: stable/10/sys/vm/vm_fault.c 270630 2014-08-25 21:21:29Z kib $"); 76238104Sdes 77238104Sdes#include "opt_ktrace.h" 78238104Sdes#include "opt_vm.h" 79238104Sdes 80238104Sdes#include <sys/param.h> 81238104Sdes#include <sys/systm.h> 82238104Sdes#include <sys/kernel.h> 83238104Sdes#include <sys/lock.h> 84238104Sdes#include <sys/proc.h> 85238104Sdes#include <sys/resourcevar.h> 86238104Sdes#include <sys/rwlock.h> 87238104Sdes#include <sys/sysctl.h> 88238104Sdes#include <sys/vmmeter.h> 89238104Sdes#include <sys/vnode.h> 90238104Sdes#ifdef KTRACE 91238104Sdes#include <sys/ktrace.h> 92238104Sdes#endif 93238104Sdes 94238104Sdes#include <vm/vm.h> 95238104Sdes#include <vm/vm_param.h> 96238104Sdes#include <vm/pmap.h> 97238104Sdes#include <vm/vm_map.h> 98238104Sdes#include <vm/vm_object.h> 99238104Sdes#include <vm/vm_page.h> 100238104Sdes#include <vm/vm_pageout.h> 101238104Sdes#include <vm/vm_kern.h> 102238104Sdes#include <vm/vm_pager.h> 103238104Sdes#include <vm/vm_extern.h> 104238104Sdes 105238104Sdes#define PFBAK 4 106238104Sdes#define PFFOR 4 107238104Sdes 108238104Sdesstatic int vm_fault_additional_pages(vm_page_t, int, int, vm_page_t *, int *); 109269257Sdes 110238104Sdes#define VM_FAULT_READ_BEHIND 8 111238104Sdes#define VM_FAULT_READ_MAX (1 + VM_FAULT_READ_AHEAD_MAX) 112238104Sdes#define VM_FAULT_NINCR (VM_FAULT_READ_MAX / VM_FAULT_READ_BEHIND) 113238104Sdes#define VM_FAULT_SUM (VM_FAULT_NINCR * (VM_FAULT_NINCR + 1) / 2) 114238104Sdes#define VM_FAULT_CACHE_BEHIND (VM_FAULT_READ_BEHIND * VM_FAULT_SUM) 115238104Sdes 116238104Sdesstruct faultstate { 117238104Sdes vm_page_t m; 118238104Sdes vm_object_t object; 119238104Sdes vm_pindex_t pindex; 120238104Sdes vm_page_t first_m; 121238104Sdes vm_object_t first_object; 122238104Sdes vm_pindex_t first_pindex; 123238104Sdes vm_map_t map; 124269257Sdes vm_map_entry_t entry; 125238104Sdes int lookup_still_valid; 126238104Sdes struct vnode *vp; 127238104Sdes}; 128238104Sdes 129238104Sdesstatic void vm_fault_cache_behind(const struct faultstate *fs, int distance); 130238104Sdesstatic void vm_fault_prefault(const struct faultstate *fs, vm_offset_t addra, 131238104Sdes int faultcount, int reqpage); 132238104Sdes 133238104Sdesstatic inline void 134238104Sdesrelease_page(struct faultstate *fs) 135238104Sdes{ 136238104Sdes 137238104Sdes vm_page_xunbusy(fs->m); 138238104Sdes vm_page_lock(fs->m); 139238104Sdes vm_page_deactivate(fs->m); 140238104Sdes vm_page_unlock(fs->m); 141238104Sdes fs->m = NULL; 142238104Sdes} 143238104Sdes 144238104Sdesstatic inline void 145238104Sdesunlock_map(struct faultstate *fs) 146238104Sdes{ 147238104Sdes 148238104Sdes if (fs->lookup_still_valid) { 149238104Sdes vm_map_lookup_done(fs->map, fs->entry); 150238104Sdes fs->lookup_still_valid = FALSE; 151238104Sdes } 152238104Sdes} 153238104Sdes 154238104Sdesstatic void 155238104Sdesunlock_and_deallocate(struct faultstate *fs) 156238104Sdes{ 157238104Sdes 158238104Sdes vm_object_pip_wakeup(fs->object); 159238104Sdes VM_OBJECT_WUNLOCK(fs->object); 160238104Sdes if (fs->object != fs->first_object) { 161238104Sdes VM_OBJECT_WLOCK(fs->first_object); 162238104Sdes vm_page_lock(fs->first_m); 163238104Sdes vm_page_free(fs->first_m); 164238104Sdes vm_page_unlock(fs->first_m); 165238104Sdes vm_object_pip_wakeup(fs->first_object); 166238104Sdes VM_OBJECT_WUNLOCK(fs->first_object); 167238104Sdes fs->first_m = NULL; 168238104Sdes } 169238104Sdes vm_object_deallocate(fs->first_object); 170238104Sdes unlock_map(fs); 171238104Sdes if (fs->vp != NULL) { 172238104Sdes vput(fs->vp); 173238104Sdes fs->vp = NULL; 174238104Sdes } 175238104Sdes} 176238104Sdes 177238104Sdes/* 178238104Sdes * TRYPAGER - used by vm_fault to calculate whether the pager for the 179238104Sdes * current object *might* contain the page. 180238104Sdes * 181238104Sdes * default objects are zero-fill, there is no real pager. 182238104Sdes */ 183238104Sdes#define TRYPAGER (fs.object->type != OBJT_DEFAULT && \ 184238104Sdes ((fault_flags & VM_FAULT_CHANGE_WIRING) == 0 || wired)) 185238104Sdes 186238104Sdes/* 187238104Sdes * vm_fault: 188238104Sdes * 189238104Sdes * Handle a page fault occurring at the given address, 190238104Sdes * requiring the given permissions, in the map specified. 191238104Sdes * If successful, the page is inserted into the 192238104Sdes * associated physical map. 193238104Sdes * 194238104Sdes * NOTE: the given address should be truncated to the 195238104Sdes * proper page address. 196238104Sdes * 197238104Sdes * KERN_SUCCESS is returned if the page fault is handled; otherwise, 198238104Sdes * a standard error specifying why the fault is fatal is returned. 199238104Sdes * 200238104Sdes * The map in question must be referenced, and remains so. 201238104Sdes * Caller may hold no locks. 202238104Sdes */ 203238104Sdesint 204238104Sdesvm_fault(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type, 205238104Sdes int fault_flags) 206238104Sdes{ 207238104Sdes struct thread *td; 208238104Sdes int result; 209238104Sdes 210238104Sdes td = curthread; 211238104Sdes if ((td->td_pflags & TDP_NOFAULTING) != 0) 212238104Sdes return (KERN_PROTECTION_FAILURE); 213238104Sdes#ifdef KTRACE 214238104Sdes if (map != kernel_map && KTRPOINT(td, KTR_FAULT)) 215238104Sdes ktrfault(vaddr, fault_type); 216238104Sdes#endif 217238104Sdes result = vm_fault_hold(map, trunc_page(vaddr), fault_type, fault_flags, 218238104Sdes NULL); 219238104Sdes#ifdef KTRACE 220238104Sdes if (map != kernel_map && KTRPOINT(td, KTR_FAULTEND)) 221238104Sdes ktrfaultend(result); 222238104Sdes#endif 223238104Sdes return (result); 224238104Sdes} 225238104Sdes 226238104Sdesint 227238104Sdesvm_fault_hold(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type, 228238104Sdes int fault_flags, vm_page_t *m_hold) 229238104Sdes{ 230238104Sdes vm_prot_t prot; 231238104Sdes long ahead, behind; 232238104Sdes int alloc_req, era, faultcount, nera, reqpage, result; 233238104Sdes boolean_t growstack, is_first_object_locked, wired; 234238104Sdes int map_generation; 235238104Sdes vm_object_t next_object; 236238104Sdes vm_page_t marray[VM_FAULT_READ_MAX]; 237238104Sdes int hardfault; 238238104Sdes struct faultstate fs; 239238104Sdes struct vnode *vp; 240238104Sdes vm_page_t m; 241238104Sdes int locked, error; 242238104Sdes 243238104Sdes hardfault = 0; 244238104Sdes growstack = TRUE; 245238104Sdes PCPU_INC(cnt.v_vm_faults); 246238104Sdes fs.vp = NULL; 247238104Sdes faultcount = reqpage = 0; 248238104Sdes 249238104SdesRetryFault:; 250238104Sdes 251238104Sdes /* 252238104Sdes * Find the backing store object and offset into it to begin the 253238104Sdes * search. 254238104Sdes */ 255238104Sdes fs.map = map; 256238104Sdes result = vm_map_lookup(&fs.map, vaddr, fault_type, &fs.entry, 257238104Sdes &fs.first_object, &fs.first_pindex, &prot, &wired); 258238104Sdes if (result != KERN_SUCCESS) { 259238104Sdes if (growstack && result == KERN_INVALID_ADDRESS && 260238104Sdes map != kernel_map) { 261238104Sdes result = vm_map_growstack(curproc, vaddr); 262238104Sdes if (result != KERN_SUCCESS) 263238104Sdes return (KERN_FAILURE); 264238104Sdes growstack = FALSE; 265238104Sdes goto RetryFault; 266238104Sdes } 267238104Sdes return (result); 268238104Sdes } 269238104Sdes 270238104Sdes map_generation = fs.map->timestamp; 271238104Sdes 272238104Sdes if (fs.entry->eflags & MAP_ENTRY_NOFAULT) { 273238104Sdes if ((curthread->td_pflags & TDP_DEVMEMIO) != 0) { 274238104Sdes vm_map_unlock_read(fs.map); 275238104Sdes return (KERN_FAILURE); 276238104Sdes } 277238104Sdes panic("vm_fault: fault on nofault entry, addr: %lx", 278238104Sdes (u_long)vaddr); 279238104Sdes } 280238104Sdes 281238104Sdes if (fs.entry->eflags & MAP_ENTRY_IN_TRANSITION && 282238104Sdes fs.entry->wiring_thread != curthread) { 283238104Sdes vm_map_unlock_read(fs.map); 284238104Sdes vm_map_lock(fs.map); 285238104Sdes if (vm_map_lookup_entry(fs.map, vaddr, &fs.entry) && 286238104Sdes (fs.entry->eflags & MAP_ENTRY_IN_TRANSITION)) { 287238104Sdes fs.entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 288238104Sdes vm_map_unlock_and_wait(fs.map, 0); 289238104Sdes } else 290238104Sdes vm_map_unlock(fs.map); 291238104Sdes goto RetryFault; 292238104Sdes } 293238104Sdes 294238104Sdes if (wired) 295238104Sdes fault_type = prot | (fault_type & VM_PROT_COPY); 296238104Sdes 297238104Sdes if (fs.vp == NULL /* avoid locked vnode leak */ && 298238104Sdes (fault_flags & (VM_FAULT_CHANGE_WIRING | VM_FAULT_DIRTY)) == 0 && 299238104Sdes /* avoid calling vm_object_set_writeable_dirty() */ 300238104Sdes ((prot & VM_PROT_WRITE) == 0 || 301238104Sdes fs.first_object->type != OBJT_VNODE || 302238104Sdes (fs.first_object->flags & OBJ_MIGHTBEDIRTY) != 0)) { 303238104Sdes VM_OBJECT_RLOCK(fs.first_object); 304238104Sdes if ((prot & VM_PROT_WRITE) != 0 && 305238104Sdes fs.first_object->type == OBJT_VNODE && 306238104Sdes (fs.first_object->flags & OBJ_MIGHTBEDIRTY) == 0) 307238104Sdes goto fast_failed; 308238104Sdes m = vm_page_lookup(fs.first_object, fs.first_pindex); 309238104Sdes /* A busy page can be mapped for read|execute access. */ 310238104Sdes if (m == NULL || ((prot & VM_PROT_WRITE) != 0 && 311238104Sdes vm_page_busied(m)) || m->valid != VM_PAGE_BITS_ALL) 312238104Sdes goto fast_failed; 313238104Sdes result = pmap_enter(fs.map->pmap, vaddr, m, prot, 314238104Sdes fault_type | PMAP_ENTER_NOSLEEP | (wired ? PMAP_ENTER_WIRED : 315238104Sdes 0), 0); 316238104Sdes if (result != KERN_SUCCESS) 317238104Sdes goto fast_failed; 318238104Sdes if (m_hold != NULL) { 319238104Sdes *m_hold = m; 320238104Sdes vm_page_lock(m); 321238104Sdes vm_page_hold(m); 322238104Sdes vm_page_unlock(m); 323238104Sdes } 324238104Sdes if ((fault_type & VM_PROT_WRITE) != 0 && 325238104Sdes (m->oflags & VPO_UNMANAGED) == 0) { 326238104Sdes vm_page_dirty(m); 327238104Sdes vm_pager_page_unswapped(m); 328238104Sdes } 329238104Sdes VM_OBJECT_RUNLOCK(fs.first_object); 330238104Sdes if (!wired) 331238104Sdes vm_fault_prefault(&fs, vaddr, 0, 0); 332238104Sdes vm_map_lookup_done(fs.map, fs.entry); 333238104Sdes curthread->td_ru.ru_minflt++; 334238104Sdes return (KERN_SUCCESS); 335238104Sdesfast_failed: 336238104Sdes if (!VM_OBJECT_TRYUPGRADE(fs.first_object)) { 337238104Sdes VM_OBJECT_RUNLOCK(fs.first_object); 338238104Sdes VM_OBJECT_WLOCK(fs.first_object); 339238104Sdes } 340238104Sdes } else { 341238104Sdes VM_OBJECT_WLOCK(fs.first_object); 342238104Sdes } 343238104Sdes 344238104Sdes /* 345238104Sdes * Make a reference to this object to prevent its disposal while we 346238104Sdes * are messing with it. Once we have the reference, the map is free 347238104Sdes * to be diddled. Since objects reference their shadows (and copies), 348238104Sdes * they will stay around as well. 349238104Sdes * 350238104Sdes * Bump the paging-in-progress count to prevent size changes (e.g. 351238104Sdes * truncation operations) during I/O. This must be done after 352238104Sdes * obtaining the vnode lock in order to avoid possible deadlocks. 353238104Sdes */ 354238104Sdes vm_object_reference_locked(fs.first_object); 355238104Sdes vm_object_pip_add(fs.first_object, 1); 356238104Sdes 357238104Sdes fs.lookup_still_valid = TRUE; 358238104Sdes 359238104Sdes fs.first_m = NULL; 360238104Sdes 361238104Sdes /* 362238104Sdes * Search for the page at object/offset. 363238104Sdes */ 364238104Sdes fs.object = fs.first_object; 365238104Sdes fs.pindex = fs.first_pindex; 366238104Sdes while (TRUE) { 367238104Sdes /* 368238104Sdes * If the object is dead, we stop here 369238104Sdes */ 370238104Sdes if (fs.object->flags & OBJ_DEAD) { 371238104Sdes unlock_and_deallocate(&fs); 372238104Sdes return (KERN_PROTECTION_FAILURE); 373238104Sdes } 374238104Sdes 375238104Sdes /* 376238104Sdes * See if page is resident 377238104Sdes */ 378238104Sdes fs.m = vm_page_lookup(fs.object, fs.pindex); 379238104Sdes if (fs.m != NULL) { 380238104Sdes /* 381238104Sdes * Wait/Retry if the page is busy. We have to do this 382238104Sdes * if the page is either exclusive or shared busy 383238104Sdes * because the vm_pager may be using read busy for 384238104Sdes * pageouts (and even pageins if it is the vnode 385238104Sdes * pager), and we could end up trying to pagein and 386238104Sdes * pageout the same page simultaneously. 387238104Sdes * 388238104Sdes * We can theoretically allow the busy case on a read 389238104Sdes * fault if the page is marked valid, but since such 390238104Sdes * pages are typically already pmap'd, putting that 391238104Sdes * special case in might be more effort then it is 392238104Sdes * worth. We cannot under any circumstances mess 393238104Sdes * around with a shared busied page except, perhaps, 394238104Sdes * to pmap it. 395238104Sdes */ 396238104Sdes if (vm_page_busied(fs.m)) { 397238104Sdes /* 398238104Sdes * Reference the page before unlocking and 399238104Sdes * sleeping so that the page daemon is less 400238104Sdes * likely to reclaim it. 401238104Sdes */ 402238104Sdes vm_page_aflag_set(fs.m, PGA_REFERENCED); 403238104Sdes if (fs.object != fs.first_object) { 404238104Sdes if (!VM_OBJECT_TRYWLOCK( 405238104Sdes fs.first_object)) { 406238104Sdes VM_OBJECT_WUNLOCK(fs.object); 407238104Sdes VM_OBJECT_WLOCK(fs.first_object); 408238104Sdes VM_OBJECT_WLOCK(fs.object); 409238104Sdes } 410238104Sdes vm_page_lock(fs.first_m); 411238104Sdes vm_page_free(fs.first_m); 412238104Sdes vm_page_unlock(fs.first_m); 413269257Sdes vm_object_pip_wakeup(fs.first_object); 414269257Sdes VM_OBJECT_WUNLOCK(fs.first_object); 415269257Sdes fs.first_m = NULL; 416269257Sdes } 417269257Sdes unlock_map(&fs); 418269257Sdes if (fs.m == vm_page_lookup(fs.object, 419269257Sdes fs.pindex)) { 420269257Sdes vm_page_sleep_if_busy(fs.m, "vmpfw"); 421269257Sdes } 422269257Sdes vm_object_pip_wakeup(fs.object); 423269257Sdes VM_OBJECT_WUNLOCK(fs.object); 424269257Sdes PCPU_INC(cnt.v_intrans); 425269257Sdes vm_object_deallocate(fs.first_object); 426269257Sdes goto RetryFault; 427269257Sdes } 428269257Sdes vm_page_lock(fs.m); 429238104Sdes vm_page_remque(fs.m); 430238104Sdes vm_page_unlock(fs.m); 431238104Sdes 432238104Sdes /* 433238104Sdes * Mark page busy for other processes, and the 434238104Sdes * pagedaemon. If it still isn't completely valid 435238104Sdes * (readable), jump to readrest, else break-out ( we 436238104Sdes * found the page ). 437238104Sdes */ 438238104Sdes vm_page_xbusy(fs.m); 439238104Sdes if (fs.m->valid != VM_PAGE_BITS_ALL) 440238104Sdes goto readrest; 441238104Sdes break; 442238104Sdes } 443238104Sdes 444238104Sdes /* 445238104Sdes * Page is not resident, If this is the search termination 446238104Sdes * or the pager might contain the page, allocate a new page. 447238104Sdes */ 448238104Sdes if (TRYPAGER || fs.object == fs.first_object) { 449238104Sdes if (fs.pindex >= fs.object->size) { 450238104Sdes unlock_and_deallocate(&fs); 451238104Sdes return (KERN_PROTECTION_FAILURE); 452238104Sdes } 453238104Sdes 454238104Sdes /* 455238104Sdes * Allocate a new page for this object/offset pair. 456238104Sdes * 457238104Sdes * Unlocked read of the p_flag is harmless. At 458238104Sdes * worst, the P_KILLED might be not observed 459238104Sdes * there, and allocation can fail, causing 460238104Sdes * restart and new reading of the p_flag. 461238104Sdes */ 462238104Sdes fs.m = NULL; 463238104Sdes if (!vm_page_count_severe() || P_KILLED(curproc)) { 464238104Sdes#if VM_NRESERVLEVEL > 0 465238104Sdes if ((fs.object->flags & OBJ_COLORED) == 0) { 466238104Sdes fs.object->flags |= OBJ_COLORED; 467238104Sdes fs.object->pg_color = atop(vaddr) - 468238104Sdes fs.pindex; 469238104Sdes } 470238104Sdes#endif 471238104Sdes alloc_req = P_KILLED(curproc) ? 472238104Sdes VM_ALLOC_SYSTEM : VM_ALLOC_NORMAL; 473238104Sdes if (fs.object->type != OBJT_VNODE && 474238104Sdes fs.object->backing_object == NULL) 475238104Sdes alloc_req |= VM_ALLOC_ZERO; 476238104Sdes fs.m = vm_page_alloc(fs.object, fs.pindex, 477238104Sdes alloc_req); 478238104Sdes } 479238104Sdes if (fs.m == NULL) { 480238104Sdes unlock_and_deallocate(&fs); 481238104Sdes VM_WAITPFAULT; 482238104Sdes goto RetryFault; 483238104Sdes } else if (fs.m->valid == VM_PAGE_BITS_ALL) 484238104Sdes break; 485238104Sdes } 486238104Sdes 487238104Sdesreadrest: 488238104Sdes /* 489238104Sdes * We have found a valid page or we have allocated a new page. 490238104Sdes * The page thus may not be valid or may not be entirely 491238104Sdes * valid. 492238104Sdes * 493238104Sdes * Attempt to fault-in the page if there is a chance that the 494238104Sdes * pager has it, and potentially fault in additional pages 495238104Sdes * at the same time. 496238104Sdes */ 497238104Sdes if (TRYPAGER) { 498238104Sdes int rv; 499238104Sdes u_char behavior = vm_map_entry_behavior(fs.entry); 500238104Sdes 501238104Sdes if (behavior == MAP_ENTRY_BEHAV_RANDOM || 502238104Sdes P_KILLED(curproc)) { 503238104Sdes behind = 0; 504238104Sdes ahead = 0; 505238104Sdes } else if (behavior == MAP_ENTRY_BEHAV_SEQUENTIAL) { 506238104Sdes behind = 0; 507238104Sdes ahead = atop(fs.entry->end - vaddr) - 1; 508238104Sdes if (ahead > VM_FAULT_READ_AHEAD_MAX) 509238104Sdes ahead = VM_FAULT_READ_AHEAD_MAX; 510238104Sdes if (fs.pindex == fs.entry->next_read) 511238104Sdes vm_fault_cache_behind(&fs, 512238104Sdes VM_FAULT_READ_MAX); 513238104Sdes } else { 514238104Sdes /* 515238104Sdes * If this is a sequential page fault, then 516238104Sdes * arithmetically increase the number of pages 517238104Sdes * in the read-ahead window. Otherwise, reset 518238104Sdes * the read-ahead window to its smallest size. 519238104Sdes */ 520238104Sdes behind = atop(vaddr - fs.entry->start); 521238104Sdes if (behind > VM_FAULT_READ_BEHIND) 522238104Sdes behind = VM_FAULT_READ_BEHIND; 523238104Sdes ahead = atop(fs.entry->end - vaddr) - 1; 524238104Sdes era = fs.entry->read_ahead; 525238104Sdes if (fs.pindex == fs.entry->next_read) { 526238104Sdes nera = era + behind; 527238104Sdes if (nera > VM_FAULT_READ_AHEAD_MAX) 528238104Sdes nera = VM_FAULT_READ_AHEAD_MAX; 529238104Sdes behind = 0; 530238104Sdes if (ahead > nera) 531238104Sdes ahead = nera; 532238104Sdes if (era == VM_FAULT_READ_AHEAD_MAX) 533238104Sdes vm_fault_cache_behind(&fs, 534238104Sdes VM_FAULT_CACHE_BEHIND); 535238104Sdes } else if (ahead > VM_FAULT_READ_AHEAD_MIN) 536238104Sdes ahead = VM_FAULT_READ_AHEAD_MIN; 537238104Sdes if (era != ahead) 538238104Sdes fs.entry->read_ahead = ahead; 539238104Sdes } 540238104Sdes 541238104Sdes /* 542238104Sdes * Call the pager to retrieve the data, if any, after 543238104Sdes * releasing the lock on the map. We hold a ref on 544238104Sdes * fs.object and the pages are exclusive busied. 545238104Sdes */ 546238104Sdes unlock_map(&fs); 547238104Sdes 548238104Sdes if (fs.object->type == OBJT_VNODE) { 549238104Sdes vp = fs.object->handle; 550238104Sdes if (vp == fs.vp) 551238104Sdes goto vnode_locked; 552238104Sdes else if (fs.vp != NULL) { 553238104Sdes vput(fs.vp); 554238104Sdes fs.vp = NULL; 555238104Sdes } 556238104Sdes locked = VOP_ISLOCKED(vp); 557238104Sdes 558238104Sdes if (locked != LK_EXCLUSIVE) 559238104Sdes locked = LK_SHARED; 560238104Sdes /* Do not sleep for vnode lock while fs.m is busy */ 561238104Sdes error = vget(vp, locked | LK_CANRECURSE | 562238104Sdes LK_NOWAIT, curthread); 563238104Sdes if (error != 0) { 564238104Sdes vhold(vp); 565238104Sdes release_page(&fs); 566238104Sdes unlock_and_deallocate(&fs); 567238104Sdes error = vget(vp, locked | LK_RETRY | 568238104Sdes LK_CANRECURSE, curthread); 569238104Sdes vdrop(vp); 570238104Sdes fs.vp = vp; 571238104Sdes KASSERT(error == 0, 572238104Sdes ("vm_fault: vget failed")); 573238104Sdes goto RetryFault; 574238104Sdes } 575238104Sdes fs.vp = vp; 576238104Sdes } 577238104Sdesvnode_locked: 578238104Sdes KASSERT(fs.vp == NULL || !fs.map->system_map, 579238104Sdes ("vm_fault: vnode-backed object mapped by system map")); 580238104Sdes 581238104Sdes /* 582238104Sdes * now we find out if any other pages should be paged 583238104Sdes * in at this time this routine checks to see if the 584238104Sdes * pages surrounding this fault reside in the same 585238104Sdes * object as the page for this fault. If they do, 586238104Sdes * then they are faulted in also into the object. The 587238104Sdes * array "marray" returned contains an array of 588238104Sdes * vm_page_t structs where one of them is the 589238104Sdes * vm_page_t passed to the routine. The reqpage 590238104Sdes * return value is the index into the marray for the 591238104Sdes * vm_page_t passed to the routine. 592238104Sdes * 593238104Sdes * fs.m plus the additional pages are exclusive busied. 594238104Sdes */ 595238104Sdes faultcount = vm_fault_additional_pages( 596238104Sdes fs.m, behind, ahead, marray, &reqpage); 597238104Sdes 598238104Sdes rv = faultcount ? 599238104Sdes vm_pager_get_pages(fs.object, marray, faultcount, 600238104Sdes reqpage) : VM_PAGER_FAIL; 601238104Sdes 602238104Sdes if (rv == VM_PAGER_OK) { 603238104Sdes /* 604238104Sdes * Found the page. Leave it busy while we play 605238104Sdes * with it. 606238104Sdes */ 607238104Sdes 608238104Sdes /* 609238104Sdes * Relookup in case pager changed page. Pager 610238104Sdes * is responsible for disposition of old page 611238104Sdes * if moved. 612238104Sdes */ 613238104Sdes fs.m = vm_page_lookup(fs.object, fs.pindex); 614238104Sdes if (!fs.m) { 615238104Sdes unlock_and_deallocate(&fs); 616238104Sdes goto RetryFault; 617238104Sdes } 618238104Sdes 619238104Sdes hardfault++; 620238104Sdes break; /* break to PAGE HAS BEEN FOUND */ 621238104Sdes } 622238104Sdes /* 623238104Sdes * Remove the bogus page (which does not exist at this 624238104Sdes * object/offset); before doing so, we must get back 625238104Sdes * our object lock to preserve our invariant. 626238104Sdes * 627238104Sdes * Also wake up any other process that may want to bring 628238104Sdes * in this page. 629238104Sdes * 630238104Sdes * If this is the top-level object, we must leave the 631238104Sdes * busy page to prevent another process from rushing 632238104Sdes * past us, and inserting the page in that object at 633238104Sdes * the same time that we are. 634238104Sdes */ 635238104Sdes if (rv == VM_PAGER_ERROR) 636238104Sdes printf("vm_fault: pager read error, pid %d (%s)\n", 637238104Sdes curproc->p_pid, curproc->p_comm); 638238104Sdes /* 639238104Sdes * Data outside the range of the pager or an I/O error 640238104Sdes */ 641238104Sdes /* 642238104Sdes * XXX - the check for kernel_map is a kludge to work 643238104Sdes * around having the machine panic on a kernel space 644238104Sdes * fault w/ I/O error. 645238104Sdes */ 646238104Sdes if (((fs.map != kernel_map) && (rv == VM_PAGER_ERROR)) || 647238104Sdes (rv == VM_PAGER_BAD)) { 648238104Sdes vm_page_lock(fs.m); 649238104Sdes vm_page_free(fs.m); 650238104Sdes vm_page_unlock(fs.m); 651238104Sdes fs.m = NULL; 652238104Sdes unlock_and_deallocate(&fs); 653238104Sdes return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE); 654238104Sdes } 655238104Sdes if (fs.object != fs.first_object) { 656238104Sdes vm_page_lock(fs.m); 657238104Sdes vm_page_free(fs.m); 658238104Sdes vm_page_unlock(fs.m); 659238104Sdes fs.m = NULL; 660238104Sdes /* 661238104Sdes * XXX - we cannot just fall out at this 662238104Sdes * point, m has been freed and is invalid! 663238104Sdes */ 664238104Sdes } 665238104Sdes } 666238104Sdes 667238104Sdes /* 668238104Sdes * We get here if the object has default pager (or unwiring) 669238104Sdes * or the pager doesn't have the page. 670238104Sdes */ 671238104Sdes if (fs.object == fs.first_object) 672238104Sdes fs.first_m = fs.m; 673238104Sdes 674238104Sdes /* 675238104Sdes * Move on to the next object. Lock the next object before 676238104Sdes * unlocking the current one. 677238104Sdes */ 678238104Sdes fs.pindex += OFF_TO_IDX(fs.object->backing_object_offset); 679238104Sdes next_object = fs.object->backing_object; 680238104Sdes if (next_object == NULL) { 681238104Sdes /* 682238104Sdes * If there's no object left, fill the page in the top 683238104Sdes * object with zeros. 684238104Sdes */ 685238104Sdes if (fs.object != fs.first_object) { 686238104Sdes vm_object_pip_wakeup(fs.object); 687238104Sdes VM_OBJECT_WUNLOCK(fs.object); 688238104Sdes 689238104Sdes fs.object = fs.first_object; 690238104Sdes fs.pindex = fs.first_pindex; 691238104Sdes fs.m = fs.first_m; 692238104Sdes VM_OBJECT_WLOCK(fs.object); 693238104Sdes } 694238104Sdes fs.first_m = NULL; 695238104Sdes 696238104Sdes /* 697238104Sdes * Zero the page if necessary and mark it valid. 698238104Sdes */ 699238104Sdes if ((fs.m->flags & PG_ZERO) == 0) { 700238104Sdes pmap_zero_page(fs.m); 701238104Sdes } else { 702238104Sdes PCPU_INC(cnt.v_ozfod); 703238104Sdes } 704238104Sdes PCPU_INC(cnt.v_zfod); 705238104Sdes fs.m->valid = VM_PAGE_BITS_ALL; 706238104Sdes /* Don't try to prefault neighboring pages. */ 707238104Sdes faultcount = 1; 708238104Sdes break; /* break to PAGE HAS BEEN FOUND */ 709238104Sdes } else { 710238104Sdes KASSERT(fs.object != next_object, 711238104Sdes ("object loop %p", next_object)); 712238104Sdes VM_OBJECT_WLOCK(next_object); 713238104Sdes vm_object_pip_add(next_object, 1); 714238104Sdes if (fs.object != fs.first_object) 715238104Sdes vm_object_pip_wakeup(fs.object); 716238104Sdes VM_OBJECT_WUNLOCK(fs.object); 717238104Sdes fs.object = next_object; 718238104Sdes } 719238104Sdes } 720238104Sdes 721238104Sdes vm_page_assert_xbusied(fs.m); 722238104Sdes 723238104Sdes /* 724238104Sdes * PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock 725238104Sdes * is held.] 726238104Sdes */ 727238104Sdes 728238104Sdes /* 729238104Sdes * If the page is being written, but isn't already owned by the 730238104Sdes * top-level object, we have to copy it into a new page owned by the 731238104Sdes * top-level object. 732238104Sdes */ 733238104Sdes if (fs.object != fs.first_object) { 734238104Sdes /* 735238104Sdes * We only really need to copy if we want to write it. 736238104Sdes */ 737238104Sdes if ((fault_type & (VM_PROT_COPY | VM_PROT_WRITE)) != 0) { 738238104Sdes /* 739238104Sdes * This allows pages to be virtually copied from a 740238104Sdes * backing_object into the first_object, where the 741238104Sdes * backing object has no other refs to it, and cannot 742238104Sdes * gain any more refs. Instead of a bcopy, we just 743238104Sdes * move the page from the backing object to the 744238104Sdes * first object. Note that we must mark the page 745238104Sdes * dirty in the first object so that it will go out 746238104Sdes * to swap when needed. 747238104Sdes */ 748238104Sdes is_first_object_locked = FALSE; 749238104Sdes if ( 750238104Sdes /* 751238104Sdes * Only one shadow object 752238104Sdes */ 753238104Sdes (fs.object->shadow_count == 1) && 754238104Sdes /* 755238104Sdes * No COW refs, except us 756238104Sdes */ 757238104Sdes (fs.object->ref_count == 1) && 758238104Sdes /* 759238104Sdes * No one else can look this object up 760238104Sdes */ 761238104Sdes (fs.object->handle == NULL) && 762238104Sdes /* 763238104Sdes * No other ways to look the object up 764238104Sdes */ 765238104Sdes ((fs.object->type == OBJT_DEFAULT) || 766238104Sdes (fs.object->type == OBJT_SWAP)) && 767238104Sdes (is_first_object_locked = VM_OBJECT_TRYWLOCK(fs.first_object)) && 768238104Sdes /* 769238104Sdes * We don't chase down the shadow chain 770238104Sdes */ 771238104Sdes fs.object == fs.first_object->backing_object) { 772238104Sdes /* 773238104Sdes * get rid of the unnecessary page 774238104Sdes */ 775238104Sdes vm_page_lock(fs.first_m); 776238104Sdes vm_page_free(fs.first_m); 777238104Sdes vm_page_unlock(fs.first_m); 778238104Sdes /* 779238104Sdes * grab the page and put it into the 780238104Sdes * process'es object. The page is 781238104Sdes * automatically made dirty. 782238104Sdes */ 783238104Sdes if (vm_page_rename(fs.m, fs.first_object, 784238104Sdes fs.first_pindex)) { 785238104Sdes unlock_and_deallocate(&fs); 786238104Sdes goto RetryFault; 787238104Sdes } 788238104Sdes vm_page_xbusy(fs.m); 789238104Sdes fs.first_m = fs.m; 790238104Sdes fs.m = NULL; 791238104Sdes PCPU_INC(cnt.v_cow_optim); 792238104Sdes } else { 793238104Sdes /* 794238104Sdes * Oh, well, lets copy it. 795238104Sdes */ 796238104Sdes pmap_copy_page(fs.m, fs.first_m); 797238104Sdes fs.first_m->valid = VM_PAGE_BITS_ALL; 798238104Sdes if (wired && (fault_flags & 799238104Sdes VM_FAULT_CHANGE_WIRING) == 0) { 800238104Sdes vm_page_lock(fs.first_m); 801238104Sdes vm_page_wire(fs.first_m); 802238104Sdes vm_page_unlock(fs.first_m); 803238104Sdes 804238104Sdes vm_page_lock(fs.m); 805238104Sdes vm_page_unwire(fs.m, FALSE); 806238104Sdes vm_page_unlock(fs.m); 807238104Sdes } 808238104Sdes /* 809238104Sdes * We no longer need the old page or object. 810238104Sdes */ 811238104Sdes release_page(&fs); 812238104Sdes } 813238104Sdes /* 814238104Sdes * fs.object != fs.first_object due to above 815238104Sdes * conditional 816238104Sdes */ 817238104Sdes vm_object_pip_wakeup(fs.object); 818238104Sdes VM_OBJECT_WUNLOCK(fs.object); 819238104Sdes /* 820238104Sdes * Only use the new page below... 821238104Sdes */ 822238104Sdes fs.object = fs.first_object; 823238104Sdes fs.pindex = fs.first_pindex; 824238104Sdes fs.m = fs.first_m; 825238104Sdes if (!is_first_object_locked) 826238104Sdes VM_OBJECT_WLOCK(fs.object); 827238104Sdes PCPU_INC(cnt.v_cow_faults); 828238104Sdes curthread->td_cow++; 829238104Sdes } else { 830238104Sdes prot &= ~VM_PROT_WRITE; 831238104Sdes } 832238104Sdes } 833238104Sdes 834238104Sdes /* 835238104Sdes * We must verify that the maps have not changed since our last 836238104Sdes * lookup. 837238104Sdes */ 838238104Sdes if (!fs.lookup_still_valid) { 839238104Sdes vm_object_t retry_object; 840238104Sdes vm_pindex_t retry_pindex; 841238104Sdes vm_prot_t retry_prot; 842238104Sdes 843238104Sdes if (!vm_map_trylock_read(fs.map)) { 844238104Sdes release_page(&fs); 845238104Sdes unlock_and_deallocate(&fs); 846238104Sdes goto RetryFault; 847238104Sdes } 848238104Sdes fs.lookup_still_valid = TRUE; 849238104Sdes if (fs.map->timestamp != map_generation) { 850238104Sdes result = vm_map_lookup_locked(&fs.map, vaddr, fault_type, 851238104Sdes &fs.entry, &retry_object, &retry_pindex, &retry_prot, &wired); 852238104Sdes 853238104Sdes /* 854238104Sdes * If we don't need the page any longer, put it on the inactive 855238104Sdes * list (the easiest thing to do here). If no one needs it, 856238104Sdes * pageout will grab it eventually. 857238104Sdes */ 858238104Sdes if (result != KERN_SUCCESS) { 859238104Sdes release_page(&fs); 860238104Sdes unlock_and_deallocate(&fs); 861238104Sdes 862238104Sdes /* 863238104Sdes * If retry of map lookup would have blocked then 864238104Sdes * retry fault from start. 865238104Sdes */ 866238104Sdes if (result == KERN_FAILURE) 867238104Sdes goto RetryFault; 868238104Sdes return (result); 869238104Sdes } 870238104Sdes if ((retry_object != fs.first_object) || 871238104Sdes (retry_pindex != fs.first_pindex)) { 872238104Sdes release_page(&fs); 873238104Sdes unlock_and_deallocate(&fs); 874238104Sdes goto RetryFault; 875238104Sdes } 876238104Sdes 877238104Sdes /* 878238104Sdes * Check whether the protection has changed or the object has 879238104Sdes * been copied while we left the map unlocked. Changing from 880238104Sdes * read to write permission is OK - we leave the page 881238104Sdes * write-protected, and catch the write fault. Changing from 882238104Sdes * write to read permission means that we can't mark the page 883238104Sdes * write-enabled after all. 884238104Sdes */ 885238104Sdes prot &= retry_prot; 886238104Sdes } 887238104Sdes } 888238104Sdes /* 889238104Sdes * If the page was filled by a pager, update the map entry's 890238104Sdes * last read offset. Since the pager does not return the 891238104Sdes * actual set of pages that it read, this update is based on 892238104Sdes * the requested set. Typically, the requested and actual 893238104Sdes * sets are the same. 894238104Sdes * 895238104Sdes * XXX The following assignment modifies the map 896238104Sdes * without holding a write lock on it. 897238104Sdes */ 898238104Sdes if (hardfault) 899238104Sdes fs.entry->next_read = fs.pindex + faultcount - reqpage; 900238104Sdes 901238104Sdes if (((prot & VM_PROT_WRITE) != 0 || 902238104Sdes (fault_flags & VM_FAULT_DIRTY) != 0) && 903238104Sdes (fs.m->oflags & VPO_UNMANAGED) == 0) { 904238104Sdes vm_object_set_writeable_dirty(fs.object); 905238104Sdes 906238104Sdes /* 907238104Sdes * If this is a NOSYNC mmap we do not want to set VPO_NOSYNC 908238104Sdes * if the page is already dirty to prevent data written with 909238104Sdes * the expectation of being synced from not being synced. 910238104Sdes * Likewise if this entry does not request NOSYNC then make 911238104Sdes * sure the page isn't marked NOSYNC. Applications sharing 912238104Sdes * data should use the same flags to avoid ping ponging. 913238104Sdes */ 914238104Sdes if (fs.entry->eflags & MAP_ENTRY_NOSYNC) { 915238104Sdes if (fs.m->dirty == 0) 916238104Sdes fs.m->oflags |= VPO_NOSYNC; 917238104Sdes } else { 918238104Sdes fs.m->oflags &= ~VPO_NOSYNC; 919238104Sdes } 920238104Sdes 921238104Sdes /* 922238104Sdes * If the fault is a write, we know that this page is being 923238104Sdes * written NOW so dirty it explicitly to save on 924238104Sdes * pmap_is_modified() calls later. 925238104Sdes * 926238104Sdes * Also tell the backing pager, if any, that it should remove 927238104Sdes * any swap backing since the page is now dirty. 928238104Sdes */ 929238104Sdes if (((fault_type & VM_PROT_WRITE) != 0 && 930238104Sdes (fault_flags & VM_FAULT_CHANGE_WIRING) == 0) || 931238104Sdes (fault_flags & VM_FAULT_DIRTY) != 0) { 932238104Sdes vm_page_dirty(fs.m); 933238104Sdes vm_pager_page_unswapped(fs.m); 934238104Sdes } 935238104Sdes } 936238104Sdes 937238104Sdes vm_page_assert_xbusied(fs.m); 938238104Sdes 939238104Sdes /* 940238104Sdes * Page must be completely valid or it is not fit to 941238104Sdes * map into user space. vm_pager_get_pages() ensures this. 942238104Sdes */ 943238104Sdes KASSERT(fs.m->valid == VM_PAGE_BITS_ALL, 944238104Sdes ("vm_fault: page %p partially invalid", fs.m)); 945238104Sdes VM_OBJECT_WUNLOCK(fs.object); 946238104Sdes 947238104Sdes /* 948238104Sdes * Put this page into the physical map. We had to do the unlock above 949238104Sdes * because pmap_enter() may sleep. We don't put the page 950238104Sdes * back on the active queue until later so that the pageout daemon 951238104Sdes * won't find it (yet). 952238104Sdes */ 953238104Sdes pmap_enter(fs.map->pmap, vaddr, fs.m, prot, 954238104Sdes fault_type | (wired ? PMAP_ENTER_WIRED : 0), 0); 955238104Sdes if (faultcount != 1 && (fault_flags & VM_FAULT_CHANGE_WIRING) == 0 && 956238104Sdes wired == 0) 957238104Sdes vm_fault_prefault(&fs, vaddr, faultcount, reqpage); 958238104Sdes VM_OBJECT_WLOCK(fs.object); 959238104Sdes vm_page_lock(fs.m); 960238104Sdes 961238104Sdes /* 962238104Sdes * If the page is not wired down, then put it where the pageout daemon 963238104Sdes * can find it. 964238104Sdes */ 965238104Sdes if (fault_flags & VM_FAULT_CHANGE_WIRING) { 966238104Sdes if (wired) 967238104Sdes vm_page_wire(fs.m); 968238104Sdes else 969238104Sdes vm_page_unwire(fs.m, 1); 970238104Sdes } else 971238104Sdes vm_page_activate(fs.m); 972238104Sdes if (m_hold != NULL) { 973238104Sdes *m_hold = fs.m; 974238104Sdes vm_page_hold(fs.m); 975238104Sdes } 976238104Sdes vm_page_unlock(fs.m); 977238104Sdes vm_page_xunbusy(fs.m); 978238104Sdes 979238104Sdes /* 980238104Sdes * Unlock everything, and return 981238104Sdes */ 982238104Sdes unlock_and_deallocate(&fs); 983238104Sdes if (hardfault) { 984238104Sdes PCPU_INC(cnt.v_io_faults); 985238104Sdes curthread->td_ru.ru_majflt++; 986238104Sdes } else 987238104Sdes curthread->td_ru.ru_minflt++; 988238104Sdes 989238104Sdes return (KERN_SUCCESS); 990238104Sdes} 991238104Sdes 992238104Sdes/* 993238104Sdes * Speed up the reclamation of up to "distance" pages that precede the 994238104Sdes * faulting pindex within the first object of the shadow chain. 995238104Sdes */ 996238104Sdesstatic void 997238104Sdesvm_fault_cache_behind(const struct faultstate *fs, int distance) 998238104Sdes{ 999238104Sdes vm_object_t first_object, object; 1000238104Sdes vm_page_t m, m_prev; 1001238104Sdes vm_pindex_t pindex; 1002238104Sdes 1003238104Sdes object = fs->object; 1004238104Sdes VM_OBJECT_ASSERT_WLOCKED(object); 1005238104Sdes first_object = fs->first_object; 1006238104Sdes if (first_object != object) { 1007238104Sdes if (!VM_OBJECT_TRYWLOCK(first_object)) { 1008238104Sdes VM_OBJECT_WUNLOCK(object); 1009238104Sdes VM_OBJECT_WLOCK(first_object); 1010238104Sdes VM_OBJECT_WLOCK(object); 1011238104Sdes } 1012238104Sdes } 1013238104Sdes /* Neither fictitious nor unmanaged pages can be cached. */ 1014238104Sdes if ((first_object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0) { 1015238104Sdes if (fs->first_pindex < distance) 1016238104Sdes pindex = 0; 1017238104Sdes else 1018238104Sdes pindex = fs->first_pindex - distance; 1019238104Sdes if (pindex < OFF_TO_IDX(fs->entry->offset)) 1020238104Sdes pindex = OFF_TO_IDX(fs->entry->offset); 1021238104Sdes m = first_object != object ? fs->first_m : fs->m; 1022238104Sdes vm_page_assert_xbusied(m); 1023238104Sdes m_prev = vm_page_prev(m); 1024238104Sdes while ((m = m_prev) != NULL && m->pindex >= pindex && 1025238104Sdes m->valid == VM_PAGE_BITS_ALL) { 1026238104Sdes m_prev = vm_page_prev(m); 1027238104Sdes if (vm_page_busied(m)) 1028238104Sdes continue; 1029238104Sdes vm_page_lock(m); 1030238104Sdes if (m->hold_count == 0 && m->wire_count == 0) { 1031238104Sdes pmap_remove_all(m); 1032238104Sdes vm_page_aflag_clear(m, PGA_REFERENCED); 1033238104Sdes if (m->dirty != 0) 1034238104Sdes vm_page_deactivate(m); 1035238104Sdes else 1036238104Sdes vm_page_cache(m); 1037238104Sdes } 1038238104Sdes vm_page_unlock(m); 1039238104Sdes } 1040238104Sdes } 1041238104Sdes if (first_object != object) 1042238104Sdes VM_OBJECT_WUNLOCK(first_object); 1043238104Sdes} 1044238104Sdes 1045238104Sdes/* 1046238104Sdes * vm_fault_prefault provides a quick way of clustering 1047238104Sdes * pagefaults into a processes address space. It is a "cousin" 1048238104Sdes * of vm_map_pmap_enter, except it runs at page fault time instead 1049238104Sdes * of mmap time. 1050238104Sdes */ 1051238104Sdesstatic void 1052238104Sdesvm_fault_prefault(const struct faultstate *fs, vm_offset_t addra, 1053238104Sdes int faultcount, int reqpage) 1054238104Sdes{ 1055238104Sdes pmap_t pmap; 1056238104Sdes vm_map_entry_t entry; 1057238104Sdes vm_object_t backing_object, lobject; 1058238104Sdes vm_offset_t addr, starta; 1059238104Sdes vm_pindex_t pindex; 1060238104Sdes vm_page_t m; 1061238104Sdes int backward, forward, i; 1062238104Sdes 1063238104Sdes pmap = fs->map->pmap; 1064238104Sdes if (pmap != vmspace_pmap(curthread->td_proc->p_vmspace)) 1065238104Sdes return; 1066238104Sdes 1067238104Sdes if (faultcount > 0) { 1068238104Sdes backward = reqpage; 1069238104Sdes forward = faultcount - reqpage - 1; 1070238104Sdes } else { 1071238104Sdes backward = PFBAK; 1072238104Sdes forward = PFFOR; 1073238104Sdes } 1074238104Sdes entry = fs->entry; 1075238104Sdes 1076238104Sdes starta = addra - backward * PAGE_SIZE; 1077238104Sdes if (starta < entry->start) { 1078238104Sdes starta = entry->start; 1079238104Sdes } else if (starta > addra) { 1080238104Sdes starta = 0; 1081238104Sdes } 1082238104Sdes 1083238104Sdes /* 1084238104Sdes * Generate the sequence of virtual addresses that are candidates for 1085238104Sdes * prefaulting in an outward spiral from the faulting virtual address, 1086238104Sdes * "addra". Specifically, the sequence is "addra - PAGE_SIZE", "addra 1087238104Sdes * + PAGE_SIZE", "addra - 2 * PAGE_SIZE", "addra + 2 * PAGE_SIZE", ... 1088238104Sdes * If the candidate address doesn't have a backing physical page, then 1089238104Sdes * the loop immediately terminates. 1090238104Sdes */ 1091238104Sdes for (i = 0; i < 2 * imax(backward, forward); i++) { 1092238104Sdes addr = addra + ((i >> 1) + 1) * ((i & 1) == 0 ? -PAGE_SIZE : 1093238104Sdes PAGE_SIZE); 1094238104Sdes if (addr > addra + forward * PAGE_SIZE) 1095238104Sdes addr = 0; 1096238104Sdes 1097238104Sdes if (addr < starta || addr >= entry->end) 1098238104Sdes continue; 1099238104Sdes 1100238104Sdes if (!pmap_is_prefaultable(pmap, addr)) 1101238104Sdes continue; 1102238104Sdes 1103238104Sdes pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 1104238104Sdes lobject = entry->object.vm_object; 1105238104Sdes VM_OBJECT_RLOCK(lobject); 1106238104Sdes while ((m = vm_page_lookup(lobject, pindex)) == NULL && 1107238104Sdes lobject->type == OBJT_DEFAULT && 1108238104Sdes (backing_object = lobject->backing_object) != NULL) { 1109238104Sdes KASSERT((lobject->backing_object_offset & PAGE_MASK) == 1110238104Sdes 0, ("vm_fault_prefault: unaligned object offset")); 1111238104Sdes pindex += lobject->backing_object_offset >> PAGE_SHIFT; 1112238104Sdes VM_OBJECT_RLOCK(backing_object); 1113238104Sdes VM_OBJECT_RUNLOCK(lobject); 1114238104Sdes lobject = backing_object; 1115238104Sdes } 1116238104Sdes if (m == NULL) { 1117238104Sdes VM_OBJECT_RUNLOCK(lobject); 1118238104Sdes break; 1119238104Sdes } 1120238104Sdes if (m->valid == VM_PAGE_BITS_ALL && 1121238104Sdes (m->flags & PG_FICTITIOUS) == 0) 1122238104Sdes pmap_enter_quick(pmap, addr, m, entry->protection); 1123238104Sdes VM_OBJECT_RUNLOCK(lobject); 1124238104Sdes } 1125238104Sdes} 1126238104Sdes 1127238104Sdes/* 1128238104Sdes * Hold each of the physical pages that are mapped by the specified range of 1129238104Sdes * virtual addresses, ["addr", "addr" + "len"), if those mappings are valid 1130238104Sdes * and allow the specified types of access, "prot". If all of the implied 1131238104Sdes * pages are successfully held, then the number of held pages is returned 1132238104Sdes * together with pointers to those pages in the array "ma". However, if any 1133238104Sdes * of the pages cannot be held, -1 is returned. 1134238104Sdes */ 1135238104Sdesint 1136238104Sdesvm_fault_quick_hold_pages(vm_map_t map, vm_offset_t addr, vm_size_t len, 1137238104Sdes vm_prot_t prot, vm_page_t *ma, int max_count) 1138238104Sdes{ 1139238104Sdes vm_offset_t end, va; 1140238104Sdes vm_page_t *mp; 1141238104Sdes int count; 1142238104Sdes boolean_t pmap_failed; 1143238104Sdes 1144238104Sdes if (len == 0) 1145238104Sdes return (0); 1146238104Sdes end = round_page(addr + len); 1147238104Sdes addr = trunc_page(addr); 1148238104Sdes 1149238104Sdes /* 1150238104Sdes * Check for illegal addresses. 1151238104Sdes */ 1152238104Sdes if (addr < vm_map_min(map) || addr > end || end > vm_map_max(map)) 1153238104Sdes return (-1); 1154238104Sdes 1155238104Sdes if (atop(end - addr) > max_count) 1156238104Sdes panic("vm_fault_quick_hold_pages: count > max_count"); 1157238104Sdes count = atop(end - addr); 1158238104Sdes 1159238104Sdes /* 1160238104Sdes * Most likely, the physical pages are resident in the pmap, so it is 1161238104Sdes * faster to try pmap_extract_and_hold() first. 1162238104Sdes */ 1163238104Sdes pmap_failed = FALSE; 1164238104Sdes for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE) { 1165238104Sdes *mp = pmap_extract_and_hold(map->pmap, va, prot); 1166238104Sdes if (*mp == NULL) 1167238104Sdes pmap_failed = TRUE; 1168238104Sdes else if ((prot & VM_PROT_WRITE) != 0 && 1169238104Sdes (*mp)->dirty != VM_PAGE_BITS_ALL) { 1170238104Sdes /* 1171238104Sdes * Explicitly dirty the physical page. Otherwise, the 1172238104Sdes * caller's changes may go unnoticed because they are 1173238104Sdes * performed through an unmanaged mapping or by a DMA 1174238104Sdes * operation. 1175238104Sdes * 1176238104Sdes * The object lock is not held here. 1177238104Sdes * See vm_page_clear_dirty_mask(). 1178238104Sdes */ 1179238104Sdes vm_page_dirty(*mp); 1180238104Sdes } 1181238104Sdes } 1182238104Sdes if (pmap_failed) { 1183238104Sdes /* 1184238104Sdes * One or more pages could not be held by the pmap. Either no 1185238104Sdes * page was mapped at the specified virtual address or that 1186238104Sdes * mapping had insufficient permissions. Attempt to fault in 1187238104Sdes * and hold these pages. 1188238104Sdes */ 1189238104Sdes for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE) 1190238104Sdes if (*mp == NULL && vm_fault_hold(map, va, prot, 1191238104Sdes VM_FAULT_NORMAL, mp) != KERN_SUCCESS) 1192238104Sdes goto error; 1193238104Sdes } 1194238104Sdes return (count); 1195238104Sdeserror: 1196238104Sdes for (mp = ma; mp < ma + count; mp++) 1197238104Sdes if (*mp != NULL) { 1198238104Sdes vm_page_lock(*mp); 1199238104Sdes vm_page_unhold(*mp); 1200238104Sdes vm_page_unlock(*mp); 1201238104Sdes } 1202238104Sdes return (-1); 1203238104Sdes} 1204238104Sdes 1205238104Sdes/* 1206238104Sdes * vm_fault_wire: 1207238104Sdes * 1208238104Sdes * Wire down a range of virtual addresses in a map. 1209238104Sdes */ 1210238104Sdesint 1211238104Sdesvm_fault_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1212238104Sdes boolean_t fictitious) 1213238104Sdes{ 1214238104Sdes vm_offset_t va; 1215238104Sdes int rv; 1216238104Sdes 1217238104Sdes /* 1218238104Sdes * We simulate a fault to get the page and enter it in the physical 1219269257Sdes * map. For user wiring, we only ask for read access on currently 1220238104Sdes * read-only sections. 1221238104Sdes */ 1222238104Sdes for (va = start; va < end; va += PAGE_SIZE) { 1223238104Sdes rv = vm_fault(map, va, VM_PROT_NONE, VM_FAULT_CHANGE_WIRING); 1224238104Sdes if (rv) { 1225238104Sdes if (va != start) 1226238104Sdes vm_fault_unwire(map, start, va, fictitious); 1227238104Sdes return (rv); 1228238104Sdes } 1229238104Sdes } 1230238104Sdes return (KERN_SUCCESS); 1231238104Sdes} 1232238104Sdes 1233238104Sdes/* 1234238104Sdes * vm_fault_unwire: 1235238104Sdes * 1236238104Sdes * Unwire a range of virtual addresses in a map. 1237238104Sdes */ 1238238104Sdesvoid 1239238104Sdesvm_fault_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1240238104Sdes boolean_t fictitious) 1241238104Sdes{ 1242238104Sdes vm_paddr_t pa; 1243238104Sdes vm_offset_t va; 1244238104Sdes vm_page_t m; 1245238104Sdes pmap_t pmap; 1246238104Sdes 1247238104Sdes pmap = vm_map_pmap(map); 1248238104Sdes 1249238104Sdes /* 1250238104Sdes * Since the pages are wired down, we must be able to get their 1251238104Sdes * mappings from the physical map system. 1252238104Sdes */ 1253238104Sdes for (va = start; va < end; va += PAGE_SIZE) { 1254238104Sdes pa = pmap_extract(pmap, va); 1255238104Sdes if (pa != 0) { 1256238104Sdes pmap_change_wiring(pmap, va, FALSE); 1257238104Sdes if (!fictitious) { 1258238104Sdes m = PHYS_TO_VM_PAGE(pa); 1259238104Sdes vm_page_lock(m); 1260238104Sdes vm_page_unwire(m, TRUE); 1261238104Sdes vm_page_unlock(m); 1262238104Sdes } 1263238104Sdes } 1264238104Sdes } 1265238104Sdes} 1266238104Sdes 1267238104Sdes/* 1268238104Sdes * Routine: 1269238104Sdes * vm_fault_copy_entry 1270238104Sdes * Function: 1271238104Sdes * Create new shadow object backing dst_entry with private copy of 1272238104Sdes * all underlying pages. When src_entry is equal to dst_entry, 1273238104Sdes * function implements COW for wired-down map entry. Otherwise, 1274238104Sdes * it forks wired entry into dst_map. 1275238104Sdes * 1276238104Sdes * In/out conditions: 1277238104Sdes * The source and destination maps must be locked for write. 1278238104Sdes * The source map entry must be wired down (or be a sharing map 1279238104Sdes * entry corresponding to a main map entry that is wired down). 1280238104Sdes */ 1281238104Sdesvoid 1282238104Sdesvm_fault_copy_entry(vm_map_t dst_map, vm_map_t src_map, 1283238104Sdes vm_map_entry_t dst_entry, vm_map_entry_t src_entry, 1284238104Sdes vm_ooffset_t *fork_charge) 1285238104Sdes{ 1286238104Sdes vm_object_t backing_object, dst_object, object, src_object; 1287238104Sdes vm_pindex_t dst_pindex, pindex, src_pindex; 1288238104Sdes vm_prot_t access, prot; 1289238104Sdes vm_offset_t vaddr; 1290238104Sdes vm_page_t dst_m; 1291238104Sdes vm_page_t src_m; 1292238104Sdes boolean_t upgrade; 1293238104Sdes 1294238104Sdes#ifdef lint 1295238104Sdes src_map++; 1296238104Sdes#endif /* lint */ 1297238104Sdes 1298238104Sdes upgrade = src_entry == dst_entry; 1299238104Sdes access = prot = dst_entry->protection; 1300238104Sdes 1301238104Sdes src_object = src_entry->object.vm_object; 1302238104Sdes src_pindex = OFF_TO_IDX(src_entry->offset); 1303238104Sdes 1304238104Sdes if (upgrade && (dst_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 1305238104Sdes dst_object = src_object; 1306238104Sdes vm_object_reference(dst_object); 1307238104Sdes } else { 1308238104Sdes /* 1309238104Sdes * Create the top-level object for the destination entry. (Doesn't 1310238104Sdes * actually shadow anything - we copy the pages directly.) 1311238104Sdes */ 1312238104Sdes dst_object = vm_object_allocate(OBJT_DEFAULT, 1313238104Sdes OFF_TO_IDX(dst_entry->end - dst_entry->start)); 1314238104Sdes#if VM_NRESERVLEVEL > 0 1315238104Sdes dst_object->flags |= OBJ_COLORED; 1316238104Sdes dst_object->pg_color = atop(dst_entry->start); 1317238104Sdes#endif 1318238104Sdes } 1319238104Sdes 1320238104Sdes VM_OBJECT_WLOCK(dst_object); 1321238104Sdes KASSERT(upgrade || dst_entry->object.vm_object == NULL, 1322238104Sdes ("vm_fault_copy_entry: vm_object not NULL")); 1323238104Sdes if (src_object != dst_object) { 1324238104Sdes dst_entry->object.vm_object = dst_object; 1325238104Sdes dst_entry->offset = 0; 1326238104Sdes dst_object->charge = dst_entry->end - dst_entry->start; 1327238104Sdes } 1328238104Sdes if (fork_charge != NULL) { 1329238104Sdes KASSERT(dst_entry->cred == NULL, 1330238104Sdes ("vm_fault_copy_entry: leaked swp charge")); 1331238104Sdes dst_object->cred = curthread->td_ucred; 1332238104Sdes crhold(dst_object->cred); 1333238104Sdes *fork_charge += dst_object->charge; 1334238104Sdes } else if (dst_object->cred == NULL) { 1335238104Sdes KASSERT(dst_entry->cred != NULL, ("no cred for entry %p", 1336238104Sdes dst_entry)); 1337238104Sdes dst_object->cred = dst_entry->cred; 1338238104Sdes dst_entry->cred = NULL; 1339238104Sdes } 1340238104Sdes 1341238104Sdes /* 1342238104Sdes * If not an upgrade, then enter the mappings in the pmap as 1343238104Sdes * read and/or execute accesses. Otherwise, enter them as 1344238104Sdes * write accesses. 1345238104Sdes * 1346238104Sdes * A writeable large page mapping is only created if all of 1347238104Sdes * the constituent small page mappings are modified. Marking 1348238104Sdes * PTEs as modified on inception allows promotion to happen 1349238104Sdes * without taking potentially large number of soft faults. 1350238104Sdes */ 1351238104Sdes if (!upgrade) 1352238104Sdes access &= ~VM_PROT_WRITE; 1353238104Sdes 1354238104Sdes /* 1355238104Sdes * Loop through all of the virtual pages within the entry's 1356238104Sdes * range, copying each page from the source object to the 1357238104Sdes * destination object. Since the source is wired, those pages 1358238104Sdes * must exist. In contrast, the destination is pageable. 1359238104Sdes * Since the destination object does share any backing storage 1360238104Sdes * with the source object, all of its pages must be dirtied, 1361238104Sdes * regardless of whether they can be written. 1362238104Sdes */ 1363238104Sdes for (vaddr = dst_entry->start, dst_pindex = 0; 1364238104Sdes vaddr < dst_entry->end; 1365238104Sdes vaddr += PAGE_SIZE, dst_pindex++) { 1366238104Sdesagain: 1367238104Sdes /* 1368238104Sdes * Find the page in the source object, and copy it in. 1369238104Sdes * Because the source is wired down, the page will be 1370238104Sdes * in memory. 1371238104Sdes */ 1372238104Sdes if (src_object != dst_object) 1373238104Sdes VM_OBJECT_RLOCK(src_object); 1374238104Sdes object = src_object; 1375238104Sdes pindex = src_pindex + dst_pindex; 1376238104Sdes while ((src_m = vm_page_lookup(object, pindex)) == NULL && 1377238104Sdes (backing_object = object->backing_object) != NULL) { 1378238104Sdes /* 1379 * Unless the source mapping is read-only or 1380 * it is presently being upgraded from 1381 * read-only, the first object in the shadow 1382 * chain should provide all of the pages. In 1383 * other words, this loop body should never be 1384 * executed when the source mapping is already 1385 * read/write. 1386 */ 1387 KASSERT((src_entry->protection & VM_PROT_WRITE) == 0 || 1388 upgrade, 1389 ("vm_fault_copy_entry: main object missing page")); 1390 1391 VM_OBJECT_RLOCK(backing_object); 1392 pindex += OFF_TO_IDX(object->backing_object_offset); 1393 if (object != dst_object) 1394 VM_OBJECT_RUNLOCK(object); 1395 object = backing_object; 1396 } 1397 KASSERT(src_m != NULL, ("vm_fault_copy_entry: page missing")); 1398 1399 if (object != dst_object) { 1400 /* 1401 * Allocate a page in the destination object. 1402 */ 1403 dst_m = vm_page_alloc(dst_object, (src_object == 1404 dst_object ? src_pindex : 0) + dst_pindex, 1405 VM_ALLOC_NORMAL); 1406 if (dst_m == NULL) { 1407 VM_OBJECT_WUNLOCK(dst_object); 1408 VM_OBJECT_RUNLOCK(object); 1409 VM_WAIT; 1410 VM_OBJECT_WLOCK(dst_object); 1411 goto again; 1412 } 1413 pmap_copy_page(src_m, dst_m); 1414 VM_OBJECT_RUNLOCK(object); 1415 dst_m->valid = VM_PAGE_BITS_ALL; 1416 dst_m->dirty = VM_PAGE_BITS_ALL; 1417 } else { 1418 dst_m = src_m; 1419 if (vm_page_sleep_if_busy(dst_m, "fltupg")) 1420 goto again; 1421 vm_page_xbusy(dst_m); 1422 KASSERT(dst_m->valid == VM_PAGE_BITS_ALL, 1423 ("invalid dst page %p", dst_m)); 1424 } 1425 VM_OBJECT_WUNLOCK(dst_object); 1426 1427 /* 1428 * Enter it in the pmap. If a wired, copy-on-write 1429 * mapping is being replaced by a write-enabled 1430 * mapping, then wire that new mapping. 1431 */ 1432 pmap_enter(dst_map->pmap, vaddr, dst_m, prot, 1433 access | (upgrade ? PMAP_ENTER_WIRED : 0), 0); 1434 1435 /* 1436 * Mark it no longer busy, and put it on the active list. 1437 */ 1438 VM_OBJECT_WLOCK(dst_object); 1439 1440 if (upgrade) { 1441 if (src_m != dst_m) { 1442 vm_page_lock(src_m); 1443 vm_page_unwire(src_m, 0); 1444 vm_page_unlock(src_m); 1445 vm_page_lock(dst_m); 1446 vm_page_wire(dst_m); 1447 vm_page_unlock(dst_m); 1448 } else { 1449 KASSERT(dst_m->wire_count > 0, 1450 ("dst_m %p is not wired", dst_m)); 1451 } 1452 } else { 1453 vm_page_lock(dst_m); 1454 vm_page_activate(dst_m); 1455 vm_page_unlock(dst_m); 1456 } 1457 vm_page_xunbusy(dst_m); 1458 } 1459 VM_OBJECT_WUNLOCK(dst_object); 1460 if (upgrade) { 1461 dst_entry->eflags &= ~(MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY); 1462 vm_object_deallocate(src_object); 1463 } 1464} 1465 1466 1467/* 1468 * This routine checks around the requested page for other pages that 1469 * might be able to be faulted in. This routine brackets the viable 1470 * pages for the pages to be paged in. 1471 * 1472 * Inputs: 1473 * m, rbehind, rahead 1474 * 1475 * Outputs: 1476 * marray (array of vm_page_t), reqpage (index of requested page) 1477 * 1478 * Return value: 1479 * number of pages in marray 1480 */ 1481static int 1482vm_fault_additional_pages(m, rbehind, rahead, marray, reqpage) 1483 vm_page_t m; 1484 int rbehind; 1485 int rahead; 1486 vm_page_t *marray; 1487 int *reqpage; 1488{ 1489 int i,j; 1490 vm_object_t object; 1491 vm_pindex_t pindex, startpindex, endpindex, tpindex; 1492 vm_page_t rtm; 1493 int cbehind, cahead; 1494 1495 VM_OBJECT_ASSERT_WLOCKED(m->object); 1496 1497 object = m->object; 1498 pindex = m->pindex; 1499 cbehind = cahead = 0; 1500 1501 /* 1502 * if the requested page is not available, then give up now 1503 */ 1504 if (!vm_pager_has_page(object, pindex, &cbehind, &cahead)) { 1505 return 0; 1506 } 1507 1508 if ((cbehind == 0) && (cahead == 0)) { 1509 *reqpage = 0; 1510 marray[0] = m; 1511 return 1; 1512 } 1513 1514 if (rahead > cahead) { 1515 rahead = cahead; 1516 } 1517 1518 if (rbehind > cbehind) { 1519 rbehind = cbehind; 1520 } 1521 1522 /* 1523 * scan backward for the read behind pages -- in memory 1524 */ 1525 if (pindex > 0) { 1526 if (rbehind > pindex) { 1527 rbehind = pindex; 1528 startpindex = 0; 1529 } else { 1530 startpindex = pindex - rbehind; 1531 } 1532 1533 if ((rtm = TAILQ_PREV(m, pglist, listq)) != NULL && 1534 rtm->pindex >= startpindex) 1535 startpindex = rtm->pindex + 1; 1536 1537 /* tpindex is unsigned; beware of numeric underflow. */ 1538 for (i = 0, tpindex = pindex - 1; tpindex >= startpindex && 1539 tpindex < pindex; i++, tpindex--) { 1540 1541 rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL | 1542 VM_ALLOC_IFNOTCACHED); 1543 if (rtm == NULL) { 1544 /* 1545 * Shift the allocated pages to the 1546 * beginning of the array. 1547 */ 1548 for (j = 0; j < i; j++) { 1549 marray[j] = marray[j + tpindex + 1 - 1550 startpindex]; 1551 } 1552 break; 1553 } 1554 1555 marray[tpindex - startpindex] = rtm; 1556 } 1557 } else { 1558 startpindex = 0; 1559 i = 0; 1560 } 1561 1562 marray[i] = m; 1563 /* page offset of the required page */ 1564 *reqpage = i; 1565 1566 tpindex = pindex + 1; 1567 i++; 1568 1569 /* 1570 * scan forward for the read ahead pages 1571 */ 1572 endpindex = tpindex + rahead; 1573 if ((rtm = TAILQ_NEXT(m, listq)) != NULL && rtm->pindex < endpindex) 1574 endpindex = rtm->pindex; 1575 if (endpindex > object->size) 1576 endpindex = object->size; 1577 1578 for (; tpindex < endpindex; i++, tpindex++) { 1579 1580 rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL | 1581 VM_ALLOC_IFNOTCACHED); 1582 if (rtm == NULL) { 1583 break; 1584 } 1585 1586 marray[i] = rtm; 1587 } 1588 1589 /* return number of pages */ 1590 return i; 1591} 1592 1593/* 1594 * Block entry into the machine-independent layer's page fault handler by 1595 * the calling thread. Subsequent calls to vm_fault() by that thread will 1596 * return KERN_PROTECTION_FAILURE. Enable machine-dependent handling of 1597 * spurious page faults. 1598 */ 1599int 1600vm_fault_disable_pagefaults(void) 1601{ 1602 1603 return (curthread_pflags_set(TDP_NOFAULTING | TDP_RESETSPUR)); 1604} 1605 1606void 1607vm_fault_enable_pagefaults(int save) 1608{ 1609 1610 curthread_pflags_restore(save); 1611} 1612