2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 1994 John S. Dyson
6 * Copyright (c) 1994 David Greenman
10 * This code is derived from software contributed to Berkeley by
11 * The Mach Operating System project at Carnegie-Mellon University.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41 * from: @(#)vm_fault.c 8.4 (Berkeley) 1/12/94
44 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
45 * All rights reserved.
47 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
49 * Permission to use, copy, modify and distribute this software and
50 * its documentation is hereby granted, provided that both the copyright
51 * notice and this permission notice appear in all copies of the
52 * software, derivative works or modified versions, and any portions
53 * thereof, and that both notices appear in supporting documentation.
55 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
59 * Carnegie Mellon requests users of this software to return to
61 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
62 * School of Computer Science
63 * Carnegie Mellon University
64 * Pittsburgh PA 15213-3890
66 * any improvements or extensions that they make and grant Carnegie the
67 * rights to redistribute these changes.
73 * Page fault handling module.
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
80 #include <sys/mutex.h>
82 #include <sys/resourcevar.h>
83 #include <sys/sysctl.h>
84 #include <sys/vmmeter.h>
85 #include <sys/vnode.h>
88 #include <vm/vm_param.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_pageout.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vnode_pager.h>
97 #include <vm/vm_extern.h>
99 static int vm_fault_additional_pages __P((vm_page_t, int,
100 int, vm_page_t *, int *));
102 #define VM_FAULT_READ_AHEAD 8
103 #define VM_FAULT_READ_BEHIND 7
104 #define VM_FAULT_READ (VM_FAULT_READ_AHEAD+VM_FAULT_READ_BEHIND+1)
111 vm_object_t first_object;
112 vm_pindex_t first_pindex;
114 vm_map_entry_t entry;
115 int lookup_still_valid;
120 release_page(struct faultstate *fs)
122 vm_page_wakeup(fs->m);
123 vm_page_deactivate(fs->m);
128 unlock_map(struct faultstate *fs)
130 if (fs->lookup_still_valid) {
131 vm_map_lookup_done(fs->map, fs->entry);
132 fs->lookup_still_valid = FALSE;
137 _unlock_things(struct faultstate *fs, int dealloc)
140 vm_object_pip_wakeup(fs->object);
141 if (fs->object != fs->first_object) {
142 vm_page_free(fs->first_m);
143 vm_object_pip_wakeup(fs->first_object);
147 vm_object_deallocate(fs->first_object);
150 if (fs->vp != NULL) {
156 #define unlock_things(fs) _unlock_things(fs, 0)
157 #define unlock_and_deallocate(fs) _unlock_things(fs, 1)
160 * TRYPAGER - used by vm_fault to calculate whether the pager for the
161 * current object *might* contain the page.
163 * default objects are zero-fill, there is no real pager.
166 #define TRYPAGER (fs.object->type != OBJT_DEFAULT && \
167 (((fault_flags & VM_FAULT_WIRE_MASK) == 0) || wired))
172 * Handle a page fault occurring at the given address,
173 * requiring the given permissions, in the map specified.
174 * If successful, the page is inserted into the
175 * associated physical map.
177 * NOTE: the given address should be truncated to the
178 * proper page address.
180 * KERN_SUCCESS is returned if the page fault is handled; otherwise,
181 * a standard error specifying why the fault is fatal is returned.
184 * The map in question must be referenced, and remains so.
185 * Caller may hold no locks.
187 static int vm_fault1 __P((vm_map_t, vm_offset_t, vm_prot_t, int));
190 vm_fault(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
198 ret = vm_fault1(map, vaddr, fault_type, fault_flags);
204 vm_fault1(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
211 vm_object_t next_object;
212 vm_page_t marray[VM_FAULT_READ];
215 struct faultstate fs;
225 * Find the backing store object and offset into it to begin the
229 if ((result = vm_map_lookup(&fs.map, vaddr,
230 fault_type, &fs.entry, &fs.first_object,
231 &fs.first_pindex, &prot, &wired)) != KERN_SUCCESS) {
232 if ((result != KERN_PROTECTION_FAILURE) ||
233 ((fault_flags & VM_FAULT_WIRE_MASK) != VM_FAULT_USER_WIRE)) {
238 * If we are user-wiring a r/w segment, and it is COW, then
239 * we need to do the COW operation. Note that we don't COW
240 * currently RO sections now, because it is NOT desirable
241 * to COW .text. We simply keep .text from ever being COW'ed
242 * and take the heat that one cannot debug wired .text sections.
244 result = vm_map_lookup(&fs.map, vaddr,
245 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_OVERRIDE_WRITE,
246 &fs.entry, &fs.first_object, &fs.first_pindex, &prot, &wired);
247 if (result != KERN_SUCCESS) {
252 * If we don't COW now, on a user wire, the user will never
253 * be able to write to the mapping. If we don't make this
254 * restriction, the bookkeeping would be nearly impossible.
256 if ((fs.entry->protection & VM_PROT_WRITE) == 0)
257 fs.entry->max_protection &= ~VM_PROT_WRITE;
260 map_generation = fs.map->timestamp;
262 if (fs.entry->eflags & MAP_ENTRY_NOFAULT) {
263 panic("vm_fault: fault on nofault entry, addr: %lx",
268 * Make a reference to this object to prevent its disposal while we
269 * are messing with it. Once we have the reference, the map is free
270 * to be diddled. Since objects reference their shadows (and copies),
271 * they will stay around as well.
273 vm_object_reference(fs.first_object);
274 vm_object_pip_add(fs.first_object, 1);
276 fs.vp = vnode_pager_lock(fs.first_object);
277 if ((fault_type & VM_PROT_WRITE) &&
278 (fs.first_object->type == OBJT_VNODE)) {
279 vm_freeze_copyopts(fs.first_object,
280 fs.first_pindex, fs.first_pindex + 1);
283 fs.lookup_still_valid = TRUE;
291 * Search for the page at object/offset.
294 fs.object = fs.first_object;
295 fs.pindex = fs.first_pindex;
299 * If the object is dead, we stop here
302 if (fs.object->flags & OBJ_DEAD) {
303 unlock_and_deallocate(&fs);
304 return (KERN_PROTECTION_FAILURE);
308 * See if page is resident
311 fs.m = vm_page_lookup(fs.object, fs.pindex);
315 * Wait/Retry if the page is busy. We have to do this
316 * if the page is busy via either PG_BUSY or
317 * vm_page_t->busy because the vm_pager may be using
318 * vm_page_t->busy for pageouts ( and even pageins if
319 * it is the vnode pager ), and we could end up trying
320 * to pagein and pageout the same page simultaneously.
322 * We can theoretically allow the busy case on a read
323 * fault if the page is marked valid, but since such
324 * pages are typically already pmap'd, putting that
325 * special case in might be more effort then it is
326 * worth. We cannot under any circumstances mess
327 * around with a vm_page_t->busy page except, perhaps,
330 if ((fs.m->flags & PG_BUSY) || fs.m->busy) {
332 (void)vm_page_sleep_busy(fs.m, TRUE, "vmpfw");
334 vm_object_deallocate(fs.first_object);
340 vm_pageq_remove_nowakeup(fs.m);
343 if ((queue - fs.m->pc) == PQ_CACHE && vm_page_count_severe()) {
344 vm_page_activate(fs.m);
345 unlock_and_deallocate(&fs);
351 * Mark page busy for other processes, and the
352 * pagedaemon. If it still isn't completely valid
353 * (readable), jump to readrest, else break-out ( we
358 if (((fs.m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) &&
359 fs.m->object != kernel_object && fs.m->object != kmem_object) {
367 * Page is not resident, If this is the search termination
368 * or the pager might contain the page, allocate a new page.
371 if (TRYPAGER || fs.object == fs.first_object) {
372 if (fs.pindex >= fs.object->size) {
373 unlock_and_deallocate(&fs);
374 return (KERN_PROTECTION_FAILURE);
378 * Allocate a new page for this object/offset pair.
381 if (!vm_page_count_severe()) {
382 fs.m = vm_page_alloc(fs.object, fs.pindex,
383 (fs.vp || fs.object->backing_object)? VM_ALLOC_NORMAL: VM_ALLOC_ZERO);
386 unlock_and_deallocate(&fs);
394 * We have found a valid page or we have allocated a new page.
395 * The page thus may not be valid or may not be entirely
398 * Attempt to fault-in the page if there is a chance that the
399 * pager has it, and potentially fault in additional pages
407 u_char behavior = vm_map_entry_behavior(fs.entry);
409 if (behavior == MAP_ENTRY_BEHAV_RANDOM) {
413 behind = (vaddr - fs.entry->start) >> PAGE_SHIFT;
414 if (behind > VM_FAULT_READ_BEHIND)
415 behind = VM_FAULT_READ_BEHIND;
417 ahead = ((fs.entry->end - vaddr) >> PAGE_SHIFT) - 1;
418 if (ahead > VM_FAULT_READ_AHEAD)
419 ahead = VM_FAULT_READ_AHEAD;
422 if ((fs.first_object->type != OBJT_DEVICE) &&
423 (behavior == MAP_ENTRY_BEHAV_SEQUENTIAL ||
424 (behavior != MAP_ENTRY_BEHAV_RANDOM &&
425 fs.pindex >= fs.entry->lastr &&
426 fs.pindex < fs.entry->lastr + VM_FAULT_READ))
428 vm_pindex_t firstpindex, tmppindex;
430 if (fs.first_pindex < 2 * VM_FAULT_READ)
433 firstpindex = fs.first_pindex - 2 * VM_FAULT_READ;
436 * note: partially valid pages cannot be
437 * included in the lookahead - NFS piecemeal
438 * writes will barf on it badly.
441 for(tmppindex = fs.first_pindex - 1;
442 tmppindex >= firstpindex;
445 mt = vm_page_lookup( fs.first_object, tmppindex);
446 if (mt == NULL || (mt->valid != VM_PAGE_BITS_ALL))
449 (mt->flags & (PG_BUSY | PG_FICTITIOUS | PG_UNMANAGED)) ||
454 vm_page_test_dirty(mt);
456 vm_page_protect(mt, VM_PROT_NONE);
457 vm_page_deactivate(mt);
468 * now we find out if any other pages should be paged
469 * in at this time this routine checks to see if the
470 * pages surrounding this fault reside in the same
471 * object as the page for this fault. If they do,
472 * then they are faulted in also into the object. The
473 * array "marray" returned contains an array of
474 * vm_page_t structs where one of them is the
475 * vm_page_t passed to the routine. The reqpage
476 * return value is the index into the marray for the
477 * vm_page_t passed to the routine.
479 * fs.m plus the additional pages are PG_BUSY'd.
481 faultcount = vm_fault_additional_pages(
482 fs.m, behind, ahead, marray, &reqpage);
485 * update lastr imperfectly (we do not know how much
486 * getpages will actually read), but good enough.
488 fs.entry->lastr = fs.pindex + faultcount - behind;
491 * Call the pager to retrieve the data, if any, after
492 * releasing the lock on the map. We hold a ref on
493 * fs.object and the pages are PG_BUSY'd.
498 vm_pager_get_pages(fs.object, marray, faultcount,
499 reqpage) : VM_PAGER_FAIL;
501 if (rv == VM_PAGER_OK) {
503 * Found the page. Leave it busy while we play
508 * Relookup in case pager changed page. Pager
509 * is responsible for disposition of old page
512 fs.m = vm_page_lookup(fs.object, fs.pindex);
514 unlock_and_deallocate(&fs);
519 break; /* break to PAGE HAS BEEN FOUND */
522 * Remove the bogus page (which does not exist at this
523 * object/offset); before doing so, we must get back
524 * our object lock to preserve our invariant.
526 * Also wake up any other process that may want to bring
529 * If this is the top-level object, we must leave the
530 * busy page to prevent another process from rushing
531 * past us, and inserting the page in that object at
532 * the same time that we are.
535 if (rv == VM_PAGER_ERROR)
536 printf("vm_fault: pager read error, pid %d (%s)\n",
537 curproc->p_pid, curproc->p_comm);
539 * Data outside the range of the pager or an I/O error
542 * XXX - the check for kernel_map is a kludge to work
543 * around having the machine panic on a kernel space
544 * fault w/ I/O error.
546 if (((fs.map != kernel_map) && (rv == VM_PAGER_ERROR)) ||
547 (rv == VM_PAGER_BAD)) {
550 unlock_and_deallocate(&fs);
551 return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE);
553 if (fs.object != fs.first_object) {
557 * XXX - we cannot just fall out at this
558 * point, m has been freed and is invalid!
564 * We get here if the object has default pager (or unwiring)
565 * or the pager doesn't have the page.
567 if (fs.object == fs.first_object)
571 * Move on to the next object. Lock the next object before
572 * unlocking the current one.
575 fs.pindex += OFF_TO_IDX(fs.object->backing_object_offset);
576 next_object = fs.object->backing_object;
577 if (next_object == NULL) {
579 * If there's no object left, fill the page in the top
582 if (fs.object != fs.first_object) {
583 vm_object_pip_wakeup(fs.object);
585 fs.object = fs.first_object;
586 fs.pindex = fs.first_pindex;
592 * Zero the page if necessary and mark it valid.
594 if ((fs.m->flags & PG_ZERO) == 0) {
595 vm_page_zero_fill(fs.m);
600 fs.m->valid = VM_PAGE_BITS_ALL;
601 break; /* break to PAGE HAS BEEN FOUND */
603 if (fs.object != fs.first_object) {
604 vm_object_pip_wakeup(fs.object);
606 KASSERT(fs.object != next_object, ("object loop %p", next_object));
607 fs.object = next_object;
608 vm_object_pip_add(fs.object, 1);
612 KASSERT((fs.m->flags & PG_BUSY) != 0,
613 ("vm_fault: not busy after main loop"));
616 * PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock
621 * If the page is being written, but isn't already owned by the
622 * top-level object, we have to copy it into a new page owned by the
626 if (fs.object != fs.first_object) {
628 * We only really need to copy if we want to write it.
631 if (fault_type & VM_PROT_WRITE) {
633 * This allows pages to be virtually copied from a
634 * backing_object into the first_object, where the
635 * backing object has no other refs to it, and cannot
636 * gain any more refs. Instead of a bcopy, we just
637 * move the page from the backing object to the
638 * first object. Note that we must mark the page
639 * dirty in the first object so that it will go out
640 * to swap when needed.
642 if (map_generation == fs.map->timestamp &&
644 * Only one shadow object
646 (fs.object->shadow_count == 1) &&
648 * No COW refs, except us
650 (fs.object->ref_count == 1) &&
652 * No one else can look this object up
654 (fs.object->handle == NULL) &&
656 * No other ways to look the object up
658 ((fs.object->type == OBJT_DEFAULT) ||
659 (fs.object->type == OBJT_SWAP)) &&
661 * We don't chase down the shadow chain
663 (fs.object == fs.first_object->backing_object) &&
666 * grab the lock if we need to
668 (fs.lookup_still_valid ||
669 lockmgr(&fs.map->lock, LK_EXCLUSIVE|LK_NOWAIT, (void *)0, curproc) == 0)
672 fs.lookup_still_valid = 1;
674 * get rid of the unnecessary page
676 vm_page_protect(fs.first_m, VM_PROT_NONE);
677 vm_page_free(fs.first_m);
681 * grab the page and put it into the
682 * process'es object. The page is
683 * automatically made dirty.
685 vm_page_rename(fs.m, fs.first_object, fs.first_pindex);
687 vm_page_busy(fs.first_m);
692 * Oh, well, lets copy it.
694 vm_page_copy(fs.m, fs.first_m);
699 * We no longer need the old page or object.
705 * fs.object != fs.first_object due to above
709 vm_object_pip_wakeup(fs.object);
712 * Only use the new page below...
717 fs.object = fs.first_object;
718 fs.pindex = fs.first_pindex;
721 prot &= ~VM_PROT_WRITE;
726 * We must verify that the maps have not changed since our last
730 if (!fs.lookup_still_valid &&
731 (fs.map->timestamp != map_generation)) {
732 vm_object_t retry_object;
733 vm_pindex_t retry_pindex;
734 vm_prot_t retry_prot;
737 * Since map entries may be pageable, make sure we can take a
738 * page fault on them.
742 * Unlock vnode before the lookup to avoid deadlock. E.G.
743 * avoid a deadlock between the inode and exec_map that can
744 * occur due to locks being obtained in different orders.
752 if (fs.map->infork) {
754 unlock_and_deallocate(&fs);
759 * To avoid trying to write_lock the map while another process
760 * has it read_locked (in vm_map_pageable), we do not try for
761 * write permission. If the page is still writable, we will
762 * get write permission. If it is not, or has been marked
763 * needs_copy, we enter the mapping without write permission,
764 * and will merely take another fault.
766 result = vm_map_lookup(&fs.map, vaddr, fault_type & ~VM_PROT_WRITE,
767 &fs.entry, &retry_object, &retry_pindex, &retry_prot, &wired);
768 map_generation = fs.map->timestamp;
771 * If we don't need the page any longer, put it on the active
772 * list (the easiest thing to do here). If no one needs it,
773 * pageout will grab it eventually.
776 if (result != KERN_SUCCESS) {
778 unlock_and_deallocate(&fs);
781 fs.lookup_still_valid = TRUE;
783 if ((retry_object != fs.first_object) ||
784 (retry_pindex != fs.first_pindex)) {
786 unlock_and_deallocate(&fs);
790 * Check whether the protection has changed or the object has
791 * been copied while we left the map unlocked. Changing from
792 * read to write permission is OK - we leave the page
793 * write-protected, and catch the write fault. Changing from
794 * write to read permission means that we can't mark the page
795 * write-enabled after all.
801 * Put this page into the physical map. We had to do the unlock above
802 * because pmap_enter may cause other faults. We don't put the page
803 * back on the active queue until later so that the page-out daemon
804 * won't find us (yet).
807 if (prot & VM_PROT_WRITE) {
808 vm_page_flag_set(fs.m, PG_WRITEABLE);
809 vm_object_set_flag(fs.m->object,
810 OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
813 * If the fault is a write, we know that this page is being
814 * written NOW so dirty it explicitly to save on
815 * pmap_is_modified() calls later.
817 * If this is a NOSYNC mmap we do not want to set PG_NOSYNC
818 * if the page is already dirty to prevent data written with
819 * the expectation of being synced from not being synced.
820 * Likewise if this entry does not request NOSYNC then make
821 * sure the page isn't marked NOSYNC. Applications sharing
822 * data should use the same flags to avoid ping ponging.
824 * Also tell the backing pager, if any, that it should remove
825 * any swap backing since the page is now dirty.
827 if (fs.entry->eflags & MAP_ENTRY_NOSYNC) {
828 if (fs.m->dirty == 0)
829 vm_page_flag_set(fs.m, PG_NOSYNC);
831 vm_page_flag_clear(fs.m, PG_NOSYNC);
833 if (fault_flags & VM_FAULT_DIRTY) {
837 vm_pager_page_unswapped(fs.m);
843 * Page had better still be busy
846 KASSERT(fs.m->flags & PG_BUSY,
847 ("vm_fault: page %p not busy!", fs.m));
852 * Sanity check: page must be completely valid or it is not fit to
853 * map into user space. vm_pager_get_pages() ensures this.
856 if (fs.m->valid != VM_PAGE_BITS_ALL) {
857 vm_page_zero_invalid(fs.m, TRUE);
858 printf("Warning: page %p partially invalid on fault\n", fs.m);
861 pmap_enter(fs.map->pmap, vaddr, fs.m, prot, wired);
863 if (((fault_flags & VM_FAULT_WIRE_MASK) == 0) && (wired == 0)) {
864 pmap_prefault(fs.map->pmap, vaddr, fs.entry);
867 vm_page_flag_clear(fs.m, PG_ZERO);
868 vm_page_flag_set(fs.m, PG_MAPPED|PG_REFERENCED);
869 if (fault_flags & VM_FAULT_HOLD)
873 * If the page is not wired down, then put it where the pageout daemon
877 if (fault_flags & VM_FAULT_WIRE_MASK) {
881 vm_page_unwire(fs.m, 1);
883 vm_page_activate(fs.m);
886 mtx_lock_spin(&sched_lock);
887 if (curproc && (curproc->p_sflag & PS_INMEM) && curproc->p_stats) {
889 curproc->p_stats->p_ru.ru_majflt++;
891 curproc->p_stats->p_ru.ru_minflt++;
894 mtx_unlock_spin(&sched_lock);
897 * Unlock everything, and return
900 vm_page_wakeup(fs.m);
901 vm_object_deallocate(fs.first_object);
903 return (KERN_SUCCESS);
910 * Wire down a range of virtual addresses in a map.
913 vm_fault_wire(map, start, end)
915 vm_offset_t start, end;
922 pmap = vm_map_pmap(map);
925 * Inform the physical mapping system that the range of addresses may
926 * not fault, so that page tables and such can be locked down as well.
929 pmap_pageable(pmap, start, end, FALSE);
932 * We simulate a fault to get the page and enter it in the physical
936 for (va = start; va < end; va += PAGE_SIZE) {
937 rv = vm_fault(map, va, VM_PROT_READ|VM_PROT_WRITE,
938 VM_FAULT_CHANGE_WIRING);
941 vm_fault_unwire(map, start, va);
945 return (KERN_SUCCESS);
949 * vm_fault_user_wire:
951 * Wire down a range of virtual addresses in a map. This
952 * is for user mode though, so we only ask for read access
953 * on currently read only sections.
956 vm_fault_user_wire(map, start, end)
958 vm_offset_t start, end;
967 pmap = vm_map_pmap(map);
970 * Inform the physical mapping system that the range of addresses may
971 * not fault, so that page tables and such can be locked down as well.
974 pmap_pageable(pmap, start, end, FALSE);
977 * We simulate a fault to get the page and enter it in the physical
980 for (va = start; va < end; va += PAGE_SIZE) {
981 rv = vm_fault(map, va, VM_PROT_READ, VM_FAULT_USER_WIRE);
984 vm_fault_unwire(map, start, va);
988 return (KERN_SUCCESS);
995 * Unwire a range of virtual addresses in a map.
998 vm_fault_unwire(map, start, end)
1000 vm_offset_t start, end;
1006 pmap = vm_map_pmap(map);
1009 * Since the pages are wired down, we must be able to get their
1010 * mappings from the physical map system.
1013 for (va = start; va < end; va += PAGE_SIZE) {
1014 pa = pmap_extract(pmap, va);
1015 if (pa != (vm_offset_t) 0) {
1016 pmap_change_wiring(pmap, va, FALSE);
1017 vm_page_unwire(PHYS_TO_VM_PAGE(pa), 1);
1022 * Inform the physical mapping system that the range of addresses may
1023 * fault, so that page tables and such may be unwired themselves.
1026 pmap_pageable(pmap, start, end, TRUE);
1032 * vm_fault_copy_entry
1034 * Copy all of the pages from a wired-down map entry to another.
1036 * In/out conditions:
1037 * The source and destination maps must be locked for write.
1038 * The source map entry must be wired down (or be a sharing map
1039 * entry corresponding to a main map entry that is wired down).
1043 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry)
1046 vm_map_entry_t dst_entry;
1047 vm_map_entry_t src_entry;
1049 vm_object_t dst_object;
1050 vm_object_t src_object;
1051 vm_ooffset_t dst_offset;
1052 vm_ooffset_t src_offset;
1062 src_object = src_entry->object.vm_object;
1063 src_offset = src_entry->offset;
1066 * Create the top-level object for the destination entry. (Doesn't
1067 * actually shadow anything - we copy the pages directly.)
1069 dst_object = vm_object_allocate(OBJT_DEFAULT,
1070 (vm_size_t) OFF_TO_IDX(dst_entry->end - dst_entry->start));
1072 dst_entry->object.vm_object = dst_object;
1073 dst_entry->offset = 0;
1075 prot = dst_entry->max_protection;
1078 * Loop through all of the pages in the entry's range, copying each
1079 * one from the source object (it should be there) to the destination
1082 for (vaddr = dst_entry->start, dst_offset = 0;
1083 vaddr < dst_entry->end;
1084 vaddr += PAGE_SIZE, dst_offset += PAGE_SIZE) {
1087 * Allocate a page in the destination object
1090 dst_m = vm_page_alloc(dst_object,
1091 OFF_TO_IDX(dst_offset), VM_ALLOC_NORMAL);
1092 if (dst_m == NULL) {
1095 } while (dst_m == NULL);
1098 * Find the page in the source object, and copy it in.
1099 * (Because the source is wired down, the page will be in
1102 src_m = vm_page_lookup(src_object,
1103 OFF_TO_IDX(dst_offset + src_offset));
1105 panic("vm_fault_copy_wired: page missing");
1107 vm_page_copy(src_m, dst_m);
1110 * Enter it in the pmap...
1113 vm_page_flag_clear(dst_m, PG_ZERO);
1114 pmap_enter(dst_map->pmap, vaddr, dst_m, prot, FALSE);
1115 vm_page_flag_set(dst_m, PG_WRITEABLE|PG_MAPPED);
1118 * Mark it no longer busy, and put it on the active list.
1120 vm_page_activate(dst_m);
1121 vm_page_wakeup(dst_m);
1127 * This routine checks around the requested page for other pages that
1128 * might be able to be faulted in. This routine brackets the viable
1129 * pages for the pages to be paged in.
1132 * m, rbehind, rahead
1135 * marray (array of vm_page_t), reqpage (index of requested page)
1138 * number of pages in marray
1140 * This routine can't block.
1143 vm_fault_additional_pages(m, rbehind, rahead, marray, reqpage)
1152 vm_pindex_t pindex, startpindex, endpindex, tpindex;
1154 int cbehind, cahead;
1162 * we don't fault-ahead for device pager
1164 if (object->type == OBJT_DEVICE) {
1171 * if the requested page is not available, then give up now
1174 if (!vm_pager_has_page(object, pindex, &cbehind, &cahead)) {
1178 if ((cbehind == 0) && (cahead == 0)) {
1184 if (rahead > cahead) {
1188 if (rbehind > cbehind) {
1193 * try to do any readahead that we might have free pages for.
1195 if ((rahead + rbehind) >
1196 ((cnt.v_free_count + cnt.v_cache_count) - cnt.v_free_reserved)) {
1197 pagedaemon_wakeup();
1204 * scan backward for the read behind pages -- in memory
1207 if (rbehind > pindex) {
1211 startpindex = pindex - rbehind;
1214 for ( tpindex = pindex - 1; tpindex >= startpindex; tpindex -= 1) {
1215 if (vm_page_lookup( object, tpindex)) {
1216 startpindex = tpindex + 1;
1223 for(i = 0, tpindex = startpindex; tpindex < pindex; i++, tpindex++) {
1225 rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL);
1227 for (j = 0; j < i; j++) {
1228 vm_page_free(marray[j]);
1243 /* page offset of the required page */
1246 tpindex = pindex + 1;
1250 * scan forward for the read ahead pages
1252 endpindex = tpindex + rahead;
1253 if (endpindex > object->size)
1254 endpindex = object->size;
1256 for( ; tpindex < endpindex; i++, tpindex++) {
1258 if (vm_page_lookup(object, tpindex)) {
1262 rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL);
1270 /* return number of bytes of pages */