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>
79 #include <sys/vnode.h>
80 #include <sys/resourcevar.h>
81 #include <sys/vmmeter.h>
84 #include <vm/vm_param.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pageout.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vnode_pager.h>
94 #include <vm/vm_extern.h>
96 static int vm_fault_additional_pages __P((vm_page_t, int,
97 int, vm_page_t *, int *));
99 #define VM_FAULT_READ_AHEAD 8
100 #define VM_FAULT_READ_BEHIND 7
101 #define VM_FAULT_READ (VM_FAULT_READ_AHEAD+VM_FAULT_READ_BEHIND+1)
108 vm_object_t first_object;
109 vm_pindex_t first_pindex;
111 vm_map_entry_t entry;
112 int lookup_still_valid;
117 release_page(struct faultstate *fs)
119 vm_page_wakeup(fs->m);
120 vm_page_deactivate(fs->m);
125 unlock_map(struct faultstate *fs)
127 if (fs->lookup_still_valid) {
128 vm_map_lookup_done(fs->map, fs->entry);
129 fs->lookup_still_valid = FALSE;
134 _unlock_things(struct faultstate *fs, int dealloc)
136 vm_object_pip_wakeup(fs->object);
137 if (fs->object != fs->first_object) {
138 vm_page_free(fs->first_m);
139 vm_object_pip_wakeup(fs->first_object);
143 vm_object_deallocate(fs->first_object);
146 if (fs->vp != NULL) {
152 #define unlock_things(fs) _unlock_things(fs, 0)
153 #define unlock_and_deallocate(fs) _unlock_things(fs, 1)
156 * TRYPAGER - used by vm_fault to calculate whether the pager for the
157 * current object *might* contain the page.
159 * default objects are zero-fill, there is no real pager.
162 #define TRYPAGER (fs.object->type != OBJT_DEFAULT && \
163 (((fault_flags & VM_FAULT_WIRE_MASK) == 0) || wired))
168 * Handle a page fault occurring at the given address,
169 * requiring the given permissions, in the map specified.
170 * If successful, the page is inserted into the
171 * associated physical map.
173 * NOTE: the given address should be truncated to the
174 * proper page address.
176 * KERN_SUCCESS is returned if the page fault is handled; otherwise,
177 * a standard error specifying why the fault is fatal is returned.
180 * The map in question must be referenced, and remains so.
181 * Caller may hold no locks.
184 vm_fault(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type, int fault_flags)
190 vm_object_t next_object;
191 vm_page_t marray[VM_FAULT_READ];
194 struct faultstate fs;
196 cnt.v_vm_faults++; /* needs lock XXX */
202 * Find the backing store object and offset into it to begin the
206 if ((result = vm_map_lookup(&fs.map, vaddr,
207 fault_type, &fs.entry, &fs.first_object,
208 &fs.first_pindex, &prot, &wired)) != KERN_SUCCESS) {
209 if ((result != KERN_PROTECTION_FAILURE) ||
210 ((fault_flags & VM_FAULT_WIRE_MASK) != VM_FAULT_USER_WIRE)) {
215 * If we are user-wiring a r/w segment, and it is COW, then
216 * we need to do the COW operation. Note that we don't COW
217 * currently RO sections now, because it is NOT desirable
218 * to COW .text. We simply keep .text from ever being COW'ed
219 * and take the heat that one cannot debug wired .text sections.
221 result = vm_map_lookup(&fs.map, vaddr,
222 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_OVERRIDE_WRITE,
223 &fs.entry, &fs.first_object, &fs.first_pindex, &prot, &wired);
224 if (result != KERN_SUCCESS) {
229 * If we don't COW now, on a user wire, the user will never
230 * be able to write to the mapping. If we don't make this
231 * restriction, the bookkeeping would be nearly impossible.
233 if ((fs.entry->protection & VM_PROT_WRITE) == 0)
234 fs.entry->max_protection &= ~VM_PROT_WRITE;
237 map_generation = fs.map->timestamp;
239 if (fs.entry->eflags & MAP_ENTRY_NOFAULT) {
240 panic("vm_fault: fault on nofault entry, addr: %lx",
245 * Make a reference to this object to prevent its disposal while we
246 * are messing with it. Once we have the reference, the map is free
247 * to be diddled. Since objects reference their shadows (and copies),
248 * they will stay around as well.
250 vm_object_reference(fs.first_object);
251 vm_object_pip_add(fs.first_object, 1);
253 fs.vp = vnode_pager_lock(fs.first_object);
254 if ((fault_type & VM_PROT_WRITE) &&
255 (fs.first_object->type == OBJT_VNODE)) {
256 vm_freeze_copyopts(fs.first_object,
257 fs.first_pindex, fs.first_pindex + 1);
260 fs.lookup_still_valid = TRUE;
268 * Search for the page at object/offset.
271 fs.object = fs.first_object;
272 fs.pindex = fs.first_pindex;
276 * If the object is dead, we stop here
279 if (fs.object->flags & OBJ_DEAD) {
280 unlock_and_deallocate(&fs);
281 return (KERN_PROTECTION_FAILURE);
285 * See if page is resident
288 fs.m = vm_page_lookup(fs.object, fs.pindex);
292 * Wait/Retry if the page is busy. We have to do this
293 * if the page is busy via either PG_BUSY or
294 * vm_page_t->busy because the vm_pager may be using
295 * vm_page_t->busy for pageouts ( and even pageins if
296 * it is the vnode pager ), and we could end up trying
297 * to pagein and pageout the same page simultaneously.
299 * We can theoretically allow the busy case on a read
300 * fault if the page is marked valid, but since such
301 * pages are typically already pmap'd, putting that
302 * special case in might be more effort then it is
303 * worth. We cannot under any circumstances mess
304 * around with a vm_page_t->busy page except, perhaps,
307 if ((fs.m->flags & PG_BUSY) || fs.m->busy) {
309 (void)vm_page_sleep_busy(fs.m, TRUE, "vmpfw");
311 vm_object_deallocate(fs.first_object);
317 vm_page_unqueue_nowakeup(fs.m);
320 if ((queue - fs.m->pc) == PQ_CACHE && vm_page_count_severe()) {
321 vm_page_activate(fs.m);
322 unlock_and_deallocate(&fs);
328 * Mark page busy for other processes, and the
329 * pagedaemon. If it still isn't completely valid
330 * (readable), jump to readrest, else break-out ( we
335 if (((fs.m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) &&
336 fs.m->object != kernel_object && fs.m->object != kmem_object) {
344 * Page is not resident, If this is the search termination
345 * or the pager might contain the page, allocate a new page.
348 if (TRYPAGER || fs.object == fs.first_object) {
349 if (fs.pindex >= fs.object->size) {
350 unlock_and_deallocate(&fs);
351 return (KERN_PROTECTION_FAILURE);
355 * Allocate a new page for this object/offset pair.
358 if (!vm_page_count_severe()) {
359 fs.m = vm_page_alloc(fs.object, fs.pindex,
360 (fs.vp || fs.object->backing_object)? VM_ALLOC_NORMAL: VM_ALLOC_ZERO);
363 unlock_and_deallocate(&fs);
371 * We have found a valid page or we have allocated a new page.
372 * The page thus may not be valid or may not be entirely
375 * Attempt to fault-in the page if there is a chance that the
376 * pager has it, and potentially fault in additional pages
384 u_char behavior = vm_map_entry_behavior(fs.entry);
386 if (behavior == MAP_ENTRY_BEHAV_RANDOM) {
390 behind = (vaddr - fs.entry->start) >> PAGE_SHIFT;
391 if (behind > VM_FAULT_READ_BEHIND)
392 behind = VM_FAULT_READ_BEHIND;
394 ahead = ((fs.entry->end - vaddr) >> PAGE_SHIFT) - 1;
395 if (ahead > VM_FAULT_READ_AHEAD)
396 ahead = VM_FAULT_READ_AHEAD;
399 if ((fs.first_object->type != OBJT_DEVICE) &&
400 (behavior == MAP_ENTRY_BEHAV_SEQUENTIAL ||
401 (behavior != MAP_ENTRY_BEHAV_RANDOM &&
402 fs.pindex >= fs.entry->lastr &&
403 fs.pindex < fs.entry->lastr + VM_FAULT_READ))
405 vm_pindex_t firstpindex, tmppindex;
407 if (fs.first_pindex < 2 * VM_FAULT_READ)
410 firstpindex = fs.first_pindex - 2 * VM_FAULT_READ;
413 * note: partially valid pages cannot be
414 * included in the lookahead - NFS piecemeal
415 * writes will barf on it badly.
418 for(tmppindex = fs.first_pindex - 1;
419 tmppindex >= firstpindex;
422 mt = vm_page_lookup( fs.first_object, tmppindex);
423 if (mt == NULL || (mt->valid != VM_PAGE_BITS_ALL))
426 (mt->flags & (PG_BUSY | PG_FICTITIOUS | PG_UNMANAGED)) ||
431 vm_page_test_dirty(mt);
433 vm_page_protect(mt, VM_PROT_NONE);
434 vm_page_deactivate(mt);
445 * now we find out if any other pages should be paged
446 * in at this time this routine checks to see if the
447 * pages surrounding this fault reside in the same
448 * object as the page for this fault. If they do,
449 * then they are faulted in also into the object. The
450 * array "marray" returned contains an array of
451 * vm_page_t structs where one of them is the
452 * vm_page_t passed to the routine. The reqpage
453 * return value is the index into the marray for the
454 * vm_page_t passed to the routine.
456 * fs.m plus the additional pages are PG_BUSY'd.
458 faultcount = vm_fault_additional_pages(
459 fs.m, behind, ahead, marray, &reqpage);
462 * update lastr imperfectly (we do not know how much
463 * getpages will actually read), but good enough.
465 fs.entry->lastr = fs.pindex + faultcount - behind;
468 * Call the pager to retrieve the data, if any, after
469 * releasing the lock on the map. We hold a ref on
470 * fs.object and the pages are PG_BUSY'd.
475 vm_pager_get_pages(fs.object, marray, faultcount,
476 reqpage) : VM_PAGER_FAIL;
478 if (rv == VM_PAGER_OK) {
480 * Found the page. Leave it busy while we play
485 * Relookup in case pager changed page. Pager
486 * is responsible for disposition of old page
489 fs.m = vm_page_lookup(fs.object, fs.pindex);
491 unlock_and_deallocate(&fs);
496 break; /* break to PAGE HAS BEEN FOUND */
499 * Remove the bogus page (which does not exist at this
500 * object/offset); before doing so, we must get back
501 * our object lock to preserve our invariant.
503 * Also wake up any other process that may want to bring
506 * If this is the top-level object, we must leave the
507 * busy page to prevent another process from rushing
508 * past us, and inserting the page in that object at
509 * the same time that we are.
512 if (rv == VM_PAGER_ERROR)
513 printf("vm_fault: pager read error, pid %d (%s)\n",
514 curproc->p_pid, curproc->p_comm);
516 * Data outside the range of the pager or an I/O error
519 * XXX - the check for kernel_map is a kludge to work
520 * around having the machine panic on a kernel space
521 * fault w/ I/O error.
523 if (((fs.map != kernel_map) && (rv == VM_PAGER_ERROR)) ||
524 (rv == VM_PAGER_BAD)) {
527 unlock_and_deallocate(&fs);
528 return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE);
530 if (fs.object != fs.first_object) {
534 * XXX - we cannot just fall out at this
535 * point, m has been freed and is invalid!
541 * We get here if the object has default pager (or unwiring)
542 * or the pager doesn't have the page.
544 if (fs.object == fs.first_object)
548 * Move on to the next object. Lock the next object before
549 * unlocking the current one.
552 fs.pindex += OFF_TO_IDX(fs.object->backing_object_offset);
553 next_object = fs.object->backing_object;
554 if (next_object == NULL) {
556 * If there's no object left, fill the page in the top
559 if (fs.object != fs.first_object) {
560 vm_object_pip_wakeup(fs.object);
562 fs.object = fs.first_object;
563 fs.pindex = fs.first_pindex;
569 * Zero the page if necessary and mark it valid.
571 if ((fs.m->flags & PG_ZERO) == 0) {
572 vm_page_zero_fill(fs.m);
577 fs.m->valid = VM_PAGE_BITS_ALL;
578 break; /* break to PAGE HAS BEEN FOUND */
580 if (fs.object != fs.first_object) {
581 vm_object_pip_wakeup(fs.object);
583 KASSERT(fs.object != next_object, ("object loop %p", next_object));
584 fs.object = next_object;
585 vm_object_pip_add(fs.object, 1);
589 KASSERT((fs.m->flags & PG_BUSY) != 0,
590 ("vm_fault: not busy after main loop"));
593 * PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock
598 * If the page is being written, but isn't already owned by the
599 * top-level object, we have to copy it into a new page owned by the
603 if (fs.object != fs.first_object) {
605 * We only really need to copy if we want to write it.
608 if (fault_type & VM_PROT_WRITE) {
610 * This allows pages to be virtually copied from a
611 * backing_object into the first_object, where the
612 * backing object has no other refs to it, and cannot
613 * gain any more refs. Instead of a bcopy, we just
614 * move the page from the backing object to the
615 * first object. Note that we must mark the page
616 * dirty in the first object so that it will go out
617 * to swap when needed.
619 if (map_generation == fs.map->timestamp &&
621 * Only one shadow object
623 (fs.object->shadow_count == 1) &&
625 * No COW refs, except us
627 (fs.object->ref_count == 1) &&
629 * No one else can look this object up
631 (fs.object->handle == NULL) &&
633 * No other ways to look the object up
635 ((fs.object->type == OBJT_DEFAULT) ||
636 (fs.object->type == OBJT_SWAP)) &&
638 * We don't chase down the shadow chain
640 (fs.object == fs.first_object->backing_object) &&
643 * grab the lock if we need to
645 (fs.lookup_still_valid ||
646 lockmgr(&fs.map->lock, LK_EXCLUSIVE|LK_NOWAIT, (void *)0, curproc) == 0)
649 fs.lookup_still_valid = 1;
651 * get rid of the unnecessary page
653 vm_page_protect(fs.first_m, VM_PROT_NONE);
654 vm_page_free(fs.first_m);
658 * grab the page and put it into the
659 * process'es object. The page is
660 * automatically made dirty.
662 vm_page_rename(fs.m, fs.first_object, fs.first_pindex);
664 vm_page_busy(fs.first_m);
669 * Oh, well, lets copy it.
671 vm_page_copy(fs.m, fs.first_m);
676 * We no longer need the old page or object.
682 * fs.object != fs.first_object due to above
686 vm_object_pip_wakeup(fs.object);
689 * Only use the new page below...
694 fs.object = fs.first_object;
695 fs.pindex = fs.first_pindex;
698 prot &= ~VM_PROT_WRITE;
703 * We must verify that the maps have not changed since our last
707 if (!fs.lookup_still_valid &&
708 (fs.map->timestamp != map_generation)) {
709 vm_object_t retry_object;
710 vm_pindex_t retry_pindex;
711 vm_prot_t retry_prot;
714 * Since map entries may be pageable, make sure we can take a
715 * page fault on them.
719 * Unlock vnode before the lookup to avoid deadlock. E.G.
720 * avoid a deadlock between the inode and exec_map that can
721 * occur due to locks being obtained in different orders.
729 if (fs.map->infork) {
731 unlock_and_deallocate(&fs);
736 * To avoid trying to write_lock the map while another process
737 * has it read_locked (in vm_map_pageable), we do not try for
738 * write permission. If the page is still writable, we will
739 * get write permission. If it is not, or has been marked
740 * needs_copy, we enter the mapping without write permission,
741 * and will merely take another fault.
743 result = vm_map_lookup(&fs.map, vaddr, fault_type & ~VM_PROT_WRITE,
744 &fs.entry, &retry_object, &retry_pindex, &retry_prot, &wired);
745 map_generation = fs.map->timestamp;
748 * If we don't need the page any longer, put it on the active
749 * list (the easiest thing to do here). If no one needs it,
750 * pageout will grab it eventually.
753 if (result != KERN_SUCCESS) {
755 unlock_and_deallocate(&fs);
758 fs.lookup_still_valid = TRUE;
760 if ((retry_object != fs.first_object) ||
761 (retry_pindex != fs.first_pindex)) {
763 unlock_and_deallocate(&fs);
767 * Check whether the protection has changed or the object has
768 * been copied while we left the map unlocked. Changing from
769 * read to write permission is OK - we leave the page
770 * write-protected, and catch the write fault. Changing from
771 * write to read permission means that we can't mark the page
772 * write-enabled after all.
778 * Put this page into the physical map. We had to do the unlock above
779 * because pmap_enter may cause other faults. We don't put the page
780 * back on the active queue until later so that the page-out daemon
781 * won't find us (yet).
784 if (prot & VM_PROT_WRITE) {
785 vm_page_flag_set(fs.m, PG_WRITEABLE);
786 vm_object_set_flag(fs.m->object,
787 OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
790 * If the fault is a write, we know that this page is being
791 * written NOW so dirty it explicitly to save on
792 * pmap_is_modified() calls later.
794 * If this is a NOSYNC mmap we do not want to set PG_NOSYNC
795 * if the page is already dirty to prevent data written with
796 * the expectation of being synced from not being synced.
797 * Likewise if this entry does not request NOSYNC then make
798 * sure the page isn't marked NOSYNC. Applications sharing
799 * data should use the same flags to avoid ping ponging.
801 * Also tell the backing pager, if any, that it should remove
802 * any swap backing since the page is now dirty.
804 if (fs.entry->eflags & MAP_ENTRY_NOSYNC) {
805 if (fs.m->dirty == 0)
806 vm_page_flag_set(fs.m, PG_NOSYNC);
808 vm_page_flag_clear(fs.m, PG_NOSYNC);
810 if (fault_flags & VM_FAULT_DIRTY) {
814 vm_pager_page_unswapped(fs.m);
820 * Page had better still be busy
823 KASSERT(fs.m->flags & PG_BUSY,
824 ("vm_fault: page %p not busy!", fs.m));
829 * Sanity check: page must be completely valid or it is not fit to
830 * map into user space. vm_pager_get_pages() ensures this.
833 if (fs.m->valid != VM_PAGE_BITS_ALL) {
834 vm_page_zero_invalid(fs.m, TRUE);
835 printf("Warning: page %p partially invalid on fault\n", fs.m);
838 pmap_enter(fs.map->pmap, vaddr, fs.m, prot, wired);
840 if (((fault_flags & VM_FAULT_WIRE_MASK) == 0) && (wired == 0)) {
841 pmap_prefault(fs.map->pmap, vaddr, fs.entry);
844 vm_page_flag_clear(fs.m, PG_ZERO);
845 vm_page_flag_set(fs.m, PG_MAPPED|PG_REFERENCED);
846 if (fault_flags & VM_FAULT_HOLD)
850 * If the page is not wired down, then put it where the pageout daemon
854 if (fault_flags & VM_FAULT_WIRE_MASK) {
858 vm_page_unwire(fs.m, 1);
860 vm_page_activate(fs.m);
863 mtx_lock_spin(&sched_lock);
864 if (curproc && (curproc->p_sflag & PS_INMEM) && curproc->p_stats) {
866 curproc->p_stats->p_ru.ru_majflt++;
868 curproc->p_stats->p_ru.ru_minflt++;
871 mtx_unlock_spin(&sched_lock);
874 * Unlock everything, and return
877 vm_page_wakeup(fs.m);
878 vm_object_deallocate(fs.first_object);
880 return (KERN_SUCCESS);
887 * Wire down a range of virtual addresses in a map.
890 vm_fault_wire(map, start, end)
892 vm_offset_t start, end;
895 register vm_offset_t va;
896 register pmap_t pmap;
899 pmap = vm_map_pmap(map);
902 * Inform the physical mapping system that the range of addresses may
903 * not fault, so that page tables and such can be locked down as well.
906 pmap_pageable(pmap, start, end, FALSE);
909 * We simulate a fault to get the page and enter it in the physical
913 for (va = start; va < end; va += PAGE_SIZE) {
914 rv = vm_fault(map, va, VM_PROT_READ|VM_PROT_WRITE,
915 VM_FAULT_CHANGE_WIRING);
918 vm_fault_unwire(map, start, va);
922 return (KERN_SUCCESS);
926 * vm_fault_user_wire:
928 * Wire down a range of virtual addresses in a map. This
929 * is for user mode though, so we only ask for read access
930 * on currently read only sections.
933 vm_fault_user_wire(map, start, end)
935 vm_offset_t start, end;
938 register vm_offset_t va;
939 register pmap_t pmap;
942 pmap = vm_map_pmap(map);
945 * Inform the physical mapping system that the range of addresses may
946 * not fault, so that page tables and such can be locked down as well.
949 pmap_pageable(pmap, start, end, FALSE);
952 * We simulate a fault to get the page and enter it in the physical
955 for (va = start; va < end; va += PAGE_SIZE) {
956 rv = vm_fault(map, va, VM_PROT_READ, VM_FAULT_USER_WIRE);
959 vm_fault_unwire(map, start, va);
963 return (KERN_SUCCESS);
970 * Unwire a range of virtual addresses in a map.
973 vm_fault_unwire(map, start, end)
975 vm_offset_t start, end;
978 register vm_offset_t va, pa;
979 register pmap_t pmap;
981 pmap = vm_map_pmap(map);
984 * Since the pages are wired down, we must be able to get their
985 * mappings from the physical map system.
988 for (va = start; va < end; va += PAGE_SIZE) {
989 pa = pmap_extract(pmap, va);
990 if (pa != (vm_offset_t) 0) {
991 pmap_change_wiring(pmap, va, FALSE);
992 vm_page_unwire(PHYS_TO_VM_PAGE(pa), 1);
997 * Inform the physical mapping system that the range of addresses may
998 * fault, so that page tables and such may be unwired themselves.
1001 pmap_pageable(pmap, start, end, TRUE);
1007 * vm_fault_copy_entry
1009 * Copy all of the pages from a wired-down map entry to another.
1011 * In/out conditions:
1012 * The source and destination maps must be locked for write.
1013 * The source map entry must be wired down (or be a sharing map
1014 * entry corresponding to a main map entry that is wired down).
1018 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry)
1021 vm_map_entry_t dst_entry;
1022 vm_map_entry_t src_entry;
1024 vm_object_t dst_object;
1025 vm_object_t src_object;
1026 vm_ooffset_t dst_offset;
1027 vm_ooffset_t src_offset;
1037 src_object = src_entry->object.vm_object;
1038 src_offset = src_entry->offset;
1041 * Create the top-level object for the destination entry. (Doesn't
1042 * actually shadow anything - we copy the pages directly.)
1044 dst_object = vm_object_allocate(OBJT_DEFAULT,
1045 (vm_size_t) OFF_TO_IDX(dst_entry->end - dst_entry->start));
1047 dst_entry->object.vm_object = dst_object;
1048 dst_entry->offset = 0;
1050 prot = dst_entry->max_protection;
1053 * Loop through all of the pages in the entry's range, copying each
1054 * one from the source object (it should be there) to the destination
1057 for (vaddr = dst_entry->start, dst_offset = 0;
1058 vaddr < dst_entry->end;
1059 vaddr += PAGE_SIZE, dst_offset += PAGE_SIZE) {
1062 * Allocate a page in the destination object
1065 dst_m = vm_page_alloc(dst_object,
1066 OFF_TO_IDX(dst_offset), VM_ALLOC_NORMAL);
1067 if (dst_m == NULL) {
1070 } while (dst_m == NULL);
1073 * Find the page in the source object, and copy it in.
1074 * (Because the source is wired down, the page will be in
1077 src_m = vm_page_lookup(src_object,
1078 OFF_TO_IDX(dst_offset + src_offset));
1080 panic("vm_fault_copy_wired: page missing");
1082 vm_page_copy(src_m, dst_m);
1085 * Enter it in the pmap...
1088 vm_page_flag_clear(dst_m, PG_ZERO);
1089 pmap_enter(dst_map->pmap, vaddr, dst_m, prot, FALSE);
1090 vm_page_flag_set(dst_m, PG_WRITEABLE|PG_MAPPED);
1093 * Mark it no longer busy, and put it on the active list.
1095 vm_page_activate(dst_m);
1096 vm_page_wakeup(dst_m);
1102 * This routine checks around the requested page for other pages that
1103 * might be able to be faulted in. This routine brackets the viable
1104 * pages for the pages to be paged in.
1107 * m, rbehind, rahead
1110 * marray (array of vm_page_t), reqpage (index of requested page)
1113 * number of pages in marray
1116 vm_fault_additional_pages(m, rbehind, rahead, marray, reqpage)
1125 vm_pindex_t pindex, startpindex, endpindex, tpindex;
1127 int cbehind, cahead;
1133 * we don't fault-ahead for device pager
1135 if (object->type == OBJT_DEVICE) {
1142 * if the requested page is not available, then give up now
1145 if (!vm_pager_has_page(object, pindex, &cbehind, &cahead)) {
1149 if ((cbehind == 0) && (cahead == 0)) {
1155 if (rahead > cahead) {
1159 if (rbehind > cbehind) {
1164 * try to do any readahead that we might have free pages for.
1166 if ((rahead + rbehind) >
1167 ((cnt.v_free_count + cnt.v_cache_count) - cnt.v_free_reserved)) {
1168 pagedaemon_wakeup();
1175 * scan backward for the read behind pages -- in memory
1178 if (rbehind > pindex) {
1182 startpindex = pindex - rbehind;
1185 for ( tpindex = pindex - 1; tpindex >= startpindex; tpindex -= 1) {
1186 if (vm_page_lookup( object, tpindex)) {
1187 startpindex = tpindex + 1;
1194 for(i = 0, tpindex = startpindex; tpindex < pindex; i++, tpindex++) {
1196 rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL);
1198 for (j = 0; j < i; j++) {
1199 vm_page_free(marray[j]);
1214 /* page offset of the required page */
1217 tpindex = pindex + 1;
1221 * scan forward for the read ahead pages
1223 endpindex = tpindex + rahead;
1224 if (endpindex > object->size)
1225 endpindex = object->size;
1227 for( ; tpindex < endpindex; i++, tpindex++) {
1229 if (vm_page_lookup(object, tpindex)) {
1233 rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL);
1241 /* return number of bytes of pages */