4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
41 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42 * All rights reserved.
44 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
46 * Permission to use, copy, modify and distribute this software and
47 * its documentation is hereby granted, provided that both the copyright
48 * notice and this permission notice appear in all copies of the
49 * software, derivative works or modified versions, and any portions
50 * thereof, and that both notices appear in supporting documentation.
52 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
53 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
54 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
56 * Carnegie Mellon requests users of this software to return to
58 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
59 * School of Computer Science
60 * Carnegie Mellon University
61 * Pittsburgh PA 15213-3890
63 * any improvements or extensions that they make and grant Carnegie the
64 * rights to redistribute these changes.
66 * $FreeBSD: src/sys/vm/vm_map.c,v 1.187.2.19 2003/05/27 00:47:02 alc Exp $
67 * $DragonFly: src/sys/vm/vm_map.c,v 1.56 2007/04/29 18:25:41 dillon Exp $
71 * Virtual memory mapping module.
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/kernel.h>
79 #include <sys/vmmeter.h>
81 #include <sys/vnode.h>
82 #include <sys/resourcevar.h>
85 #include <sys/malloc.h>
88 #include <vm/vm_param.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_pager.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_extern.h>
96 #include <vm/swap_pager.h>
97 #include <vm/vm_zone.h>
99 #include <sys/thread2.h>
100 #include <sys/sysref2.h>
103 * Virtual memory maps provide for the mapping, protection, and sharing
104 * of virtual memory objects. In addition, this module provides for an
105 * efficient virtual copy of memory from one map to another.
107 * Synchronization is required prior to most operations.
109 * Maps consist of an ordered doubly-linked list of simple entries.
110 * A hint and a RB tree is used to speed-up lookups.
112 * Callers looking to modify maps specify start/end addresses which cause
113 * the related map entry to be clipped if necessary, and then later
114 * recombined if the pieces remained compatible.
116 * Virtual copy operations are performed by copying VM object references
117 * from one map to another, and then marking both regions as copy-on-write.
119 static void vmspace_terminate(struct vmspace *vm);
120 static void vmspace_lock(struct vmspace *vm);
121 static void vmspace_unlock(struct vmspace *vm);
122 static void vmspace_dtor(void *obj, void *private);
124 MALLOC_DEFINE(M_VMSPACE, "vmspace", "vmspace objcache backingstore");
126 struct sysref_class vmspace_sysref_class = {
129 .proto = SYSREF_PROTO_VMSPACE,
130 .offset = offsetof(struct vmspace, vm_sysref),
131 .objsize = sizeof(struct vmspace),
133 .flags = SRC_MANAGEDINIT,
134 .dtor = vmspace_dtor,
136 .terminate = (sysref_terminate_func_t)vmspace_terminate,
137 .lock = (sysref_lock_func_t)vmspace_lock,
138 .unlock = (sysref_lock_func_t)vmspace_unlock
144 static struct vm_zone mapentzone_store, mapzone_store;
145 static vm_zone_t mapentzone, mapzone;
146 static struct vm_object mapentobj, mapobj;
148 static struct vm_map_entry map_entry_init[MAX_MAPENT];
149 static struct vm_map_entry cpu_map_entry_init[MAXCPU][VMEPERCPU];
150 static struct vm_map map_init[MAX_KMAP];
152 static void vm_map_entry_shadow(vm_map_entry_t entry);
153 static vm_map_entry_t vm_map_entry_create(vm_map_t map, int *);
154 static void vm_map_entry_dispose (vm_map_t map, vm_map_entry_t entry, int *);
155 static void _vm_map_clip_end (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
156 static void _vm_map_clip_start (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
157 static void vm_map_entry_delete (vm_map_t, vm_map_entry_t, int *);
158 static void vm_map_entry_unwire (vm_map_t, vm_map_entry_t);
159 static void vm_map_copy_entry (vm_map_t, vm_map_t, vm_map_entry_t,
161 static void vm_map_split (vm_map_entry_t);
162 static void vm_map_unclip_range (vm_map_t map, vm_map_entry_t start_entry, vm_offset_t start, vm_offset_t end, int *count, int flags);
165 * Initialize the vm_map module. Must be called before any other vm_map
168 * Map and entry structures are allocated from the general purpose
169 * memory pool with some exceptions:
171 * - The kernel map is allocated statically.
172 * - Initial kernel map entries are allocated out of a static pool.
174 * These restrictions are necessary since malloc() uses the
175 * maps and requires map entries.
177 * Called from the low level boot code only.
182 mapzone = &mapzone_store;
183 zbootinit(mapzone, "MAP", sizeof (struct vm_map),
185 mapentzone = &mapentzone_store;
186 zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
187 map_entry_init, MAX_MAPENT);
191 * Called prior to any vmspace allocations.
193 * Called from the low level boot code only.
198 zinitna(mapentzone, &mapentobj, NULL, 0, 0,
199 ZONE_USE_RESERVE | ZONE_SPECIAL, 1);
200 zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1);
207 * Red black tree functions
209 * The caller must hold the related map lock.
211 static int rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b);
212 RB_GENERATE(vm_map_rb_tree, vm_map_entry, rb_entry, rb_vm_map_compare);
214 /* a->start is address, and the only field has to be initialized */
216 rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b)
218 if (a->start < b->start)
220 else if (a->start > b->start)
226 * Allocate a vmspace structure, including a vm_map and pmap.
227 * Initialize numerous fields. While the initial allocation is zerod,
228 * subsequence reuse from the objcache leaves elements of the structure
229 * intact (particularly the pmap), so portions must be zerod.
231 * The structure is not considered activated until we call sysref_activate().
236 vmspace_alloc(vm_offset_t min, vm_offset_t max)
240 lwkt_gettoken(&vmspace_token);
241 vm = sysref_alloc(&vmspace_sysref_class);
242 bzero(&vm->vm_startcopy,
243 (char *)&vm->vm_endcopy - (char *)&vm->vm_startcopy);
244 vm_map_init(&vm->vm_map, min, max, NULL);
245 pmap_pinit(vmspace_pmap(vm)); /* (some fields reused) */
246 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
248 vm->vm_exitingcnt = 0;
249 cpu_vmspace_alloc(vm);
250 sysref_activate(&vm->vm_sysref);
251 lwkt_reltoken(&vmspace_token);
257 * dtor function - Some elements of the pmap are retained in the
258 * free-cached vmspaces to improve performance. We have to clean them up
259 * here before returning the vmspace to the memory pool.
264 vmspace_dtor(void *obj, void *private)
266 struct vmspace *vm = obj;
268 pmap_puninit(vmspace_pmap(vm));
272 * Called in two cases:
274 * (1) When the last sysref is dropped, but exitingcnt might still be
277 * (2) When there are no sysrefs (i.e. refcnt is negative) left and the
278 * exitingcnt becomes zero
280 * sysref will not scrap the object until we call sysref_put() once more
281 * after the last ref has been dropped.
283 * Interlocked by the sysref API.
286 vmspace_terminate(struct vmspace *vm)
291 * If exitingcnt is non-zero we can't get rid of the entire vmspace
292 * yet, but we can scrap user memory.
294 lwkt_gettoken(&vmspace_token);
295 if (vm->vm_exitingcnt) {
297 pmap_remove_pages(vmspace_pmap(vm), VM_MIN_USER_ADDRESS,
298 VM_MAX_USER_ADDRESS);
299 vm_map_remove(&vm->vm_map, VM_MIN_USER_ADDRESS,
300 VM_MAX_USER_ADDRESS);
301 lwkt_reltoken(&vmspace_token);
304 cpu_vmspace_free(vm);
307 * Make sure any SysV shm is freed, it might not have in
312 KKASSERT(vm->vm_upcalls == NULL);
315 * Lock the map, to wait out all other references to it.
316 * Delete all of the mappings and pages they hold, then call
317 * the pmap module to reclaim anything left.
319 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
320 vm_map_lock(&vm->vm_map);
321 vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
322 vm->vm_map.max_offset, &count);
323 vm_map_unlock(&vm->vm_map);
324 vm_map_entry_release(count);
326 pmap_release(vmspace_pmap(vm));
327 sysref_put(&vm->vm_sysref);
328 lwkt_reltoken(&vmspace_token);
332 * vmspaces are not currently locked.
335 vmspace_lock(struct vmspace *vm __unused)
340 vmspace_unlock(struct vmspace *vm __unused)
345 * This is called during exit indicating that the vmspace is no
346 * longer in used by an exiting process, but the process has not yet
352 vmspace_exitbump(struct vmspace *vm)
354 lwkt_gettoken(&vmspace_token);
356 lwkt_reltoken(&vmspace_token);
360 * This is called in the wait*() handling code. The vmspace can be terminated
361 * after the last wait is finished using it.
366 vmspace_exitfree(struct proc *p)
370 lwkt_gettoken(&vmspace_token);
374 if (--vm->vm_exitingcnt == 0 && sysref_isinactive(&vm->vm_sysref))
375 vmspace_terminate(vm);
376 lwkt_reltoken(&vmspace_token);
380 * Swap useage is determined by taking the proportional swap used by
381 * VM objects backing the VM map. To make up for fractional losses,
382 * if the VM object has any swap use at all the associated map entries
383 * count for at least 1 swap page.
388 vmspace_swap_count(struct vmspace *vmspace)
390 vm_map_t map = &vmspace->vm_map;
396 lwkt_gettoken(&vmspace_token);
397 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
398 switch(cur->maptype) {
399 case VM_MAPTYPE_NORMAL:
400 case VM_MAPTYPE_VPAGETABLE:
401 if ((object = cur->object.vm_object) == NULL)
403 if (object->swblock_count) {
404 n = (cur->end - cur->start) / PAGE_SIZE;
405 count += object->swblock_count *
406 SWAP_META_PAGES * n / object->size + 1;
413 lwkt_reltoken(&vmspace_token);
418 * Calculate the approximate number of anonymous pages in use by
419 * this vmspace. To make up for fractional losses, we count each
420 * VM object as having at least 1 anonymous page.
425 vmspace_anonymous_count(struct vmspace *vmspace)
427 vm_map_t map = &vmspace->vm_map;
432 lwkt_gettoken(&vmspace_token);
433 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
434 switch(cur->maptype) {
435 case VM_MAPTYPE_NORMAL:
436 case VM_MAPTYPE_VPAGETABLE:
437 if ((object = cur->object.vm_object) == NULL)
439 if (object->type != OBJT_DEFAULT &&
440 object->type != OBJT_SWAP) {
443 count += object->resident_page_count;
449 lwkt_reltoken(&vmspace_token);
454 * Creates and returns a new empty VM map with the given physical map
455 * structure, and having the given lower and upper address bounds.
460 vm_map_create(vm_map_t result, pmap_t pmap, vm_offset_t min, vm_offset_t max)
463 result = zalloc(mapzone);
464 vm_map_init(result, min, max, pmap);
469 * Initialize an existing vm_map structure such as that in the vmspace
470 * structure. The pmap is initialized elsewhere.
475 vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max, pmap_t pmap)
477 map->header.next = map->header.prev = &map->header;
478 RB_INIT(&map->rb_root);
483 map->min_offset = min;
484 map->max_offset = max;
486 map->first_free = &map->header;
487 map->hint = &map->header;
489 lockinit(&map->lock, "thrd_sleep", 0, 0);
493 * Shadow the vm_map_entry's object. This typically needs to be done when
494 * a write fault is taken on an entry which had previously been cloned by
495 * fork(). The shared object (which might be NULL) must become private so
496 * we add a shadow layer above it.
498 * Object allocation for anonymous mappings is defered as long as possible.
499 * When creating a shadow, however, the underlying object must be instantiated
500 * so it can be shared.
502 * If the map segment is governed by a virtual page table then it is
503 * possible to address offsets beyond the mapped area. Just allocate
504 * a maximally sized object for this case.
506 * The vm_map must be exclusively locked.
507 * No other requirements.
511 vm_map_entry_shadow(vm_map_entry_t entry)
513 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
514 vm_object_shadow(&entry->object.vm_object, &entry->offset,
515 0x7FFFFFFF); /* XXX */
517 vm_object_shadow(&entry->object.vm_object, &entry->offset,
518 atop(entry->end - entry->start));
520 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
524 * Allocate an object for a vm_map_entry.
526 * Object allocation for anonymous mappings is defered as long as possible.
527 * This function is called when we can defer no longer, generally when a map
528 * entry might be split or forked or takes a page fault.
530 * If the map segment is governed by a virtual page table then it is
531 * possible to address offsets beyond the mapped area. Just allocate
532 * a maximally sized object for this case.
534 * The vm_map must be exclusively locked.
535 * No other requirements.
538 vm_map_entry_allocate_object(vm_map_entry_t entry)
542 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
543 obj = vm_object_allocate(OBJT_DEFAULT, 0x7FFFFFFF); /* XXX */
545 obj = vm_object_allocate(OBJT_DEFAULT,
546 atop(entry->end - entry->start));
548 entry->object.vm_object = obj;
553 * Set an initial negative count so the first attempt to reserve
554 * space preloads a bunch of vm_map_entry's for this cpu. Also
555 * pre-allocate 2 vm_map_entries which will be needed by zalloc() to
556 * map a new page for vm_map_entry structures. SMP systems are
557 * particularly sensitive.
559 * This routine is called in early boot so we cannot just call
560 * vm_map_entry_reserve().
562 * Called from the low level boot code only (for each cpu)
565 vm_map_entry_reserve_cpu_init(globaldata_t gd)
567 vm_map_entry_t entry;
570 gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2;
571 entry = &cpu_map_entry_init[gd->gd_cpuid][0];
572 for (i = 0; i < VMEPERCPU; ++i, ++entry) {
573 entry->next = gd->gd_vme_base;
574 gd->gd_vme_base = entry;
579 * Reserves vm_map_entry structures so code later on can manipulate
580 * map_entry structures within a locked map without blocking trying
581 * to allocate a new vm_map_entry.
586 vm_map_entry_reserve(int count)
588 struct globaldata *gd = mycpu;
589 vm_map_entry_t entry;
592 * Make sure we have enough structures in gd_vme_base to handle
593 * the reservation request.
596 while (gd->gd_vme_avail < count) {
597 entry = zalloc(mapentzone);
598 entry->next = gd->gd_vme_base;
599 gd->gd_vme_base = entry;
602 gd->gd_vme_avail -= count;
609 * Releases previously reserved vm_map_entry structures that were not
610 * used. If we have too much junk in our per-cpu cache clean some of
616 vm_map_entry_release(int count)
618 struct globaldata *gd = mycpu;
619 vm_map_entry_t entry;
622 gd->gd_vme_avail += count;
623 while (gd->gd_vme_avail > MAP_RESERVE_SLOP) {
624 entry = gd->gd_vme_base;
625 KKASSERT(entry != NULL);
626 gd->gd_vme_base = entry->next;
629 zfree(mapentzone, entry);
636 * Reserve map entry structures for use in kernel_map itself. These
637 * entries have *ALREADY* been reserved on a per-cpu basis when the map
638 * was inited. This function is used by zalloc() to avoid a recursion
639 * when zalloc() itself needs to allocate additional kernel memory.
641 * This function works like the normal reserve but does not load the
642 * vm_map_entry cache (because that would result in an infinite
643 * recursion). Note that gd_vme_avail may go negative. This is expected.
645 * Any caller of this function must be sure to renormalize after
646 * potentially eating entries to ensure that the reserve supply
652 vm_map_entry_kreserve(int count)
654 struct globaldata *gd = mycpu;
657 gd->gd_vme_avail -= count;
659 KASSERT(gd->gd_vme_base != NULL,
660 ("no reserved entries left, gd_vme_avail = %d\n",
666 * Release previously reserved map entries for kernel_map. We do not
667 * attempt to clean up like the normal release function as this would
668 * cause an unnecessary (but probably not fatal) deep procedure call.
673 vm_map_entry_krelease(int count)
675 struct globaldata *gd = mycpu;
678 gd->gd_vme_avail += count;
683 * Allocates a VM map entry for insertion. No entry fields are filled in.
685 * The entries should have previously been reserved. The reservation count
686 * is tracked in (*countp).
690 static vm_map_entry_t
691 vm_map_entry_create(vm_map_t map, int *countp)
693 struct globaldata *gd = mycpu;
694 vm_map_entry_t entry;
696 KKASSERT(*countp > 0);
699 entry = gd->gd_vme_base;
700 KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp));
701 gd->gd_vme_base = entry->next;
708 * Dispose of a vm_map_entry that is no longer being referenced.
713 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp)
715 struct globaldata *gd = mycpu;
717 KKASSERT(map->hint != entry);
718 KKASSERT(map->first_free != entry);
722 entry->next = gd->gd_vme_base;
723 gd->gd_vme_base = entry;
729 * Insert/remove entries from maps.
731 * The related map must be exclusively locked.
732 * No other requirements.
734 * NOTE! We currently acquire the vmspace_token only to avoid races
735 * against the pageout daemon's calls to vmspace_*_count(), which
736 * are unable to safely lock the vm_map without potentially
740 vm_map_entry_link(vm_map_t map,
741 vm_map_entry_t after_where,
742 vm_map_entry_t entry)
744 ASSERT_VM_MAP_LOCKED(map);
746 lwkt_gettoken(&vmspace_token);
748 entry->prev = after_where;
749 entry->next = after_where->next;
750 entry->next->prev = entry;
751 after_where->next = entry;
752 if (vm_map_rb_tree_RB_INSERT(&map->rb_root, entry))
753 panic("vm_map_entry_link: dup addr map %p ent %p", map, entry);
754 lwkt_reltoken(&vmspace_token);
758 vm_map_entry_unlink(vm_map_t map,
759 vm_map_entry_t entry)
764 ASSERT_VM_MAP_LOCKED(map);
766 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
767 panic("vm_map_entry_unlink: attempt to mess with "
768 "locked entry! %p", entry);
770 lwkt_gettoken(&vmspace_token);
775 vm_map_rb_tree_RB_REMOVE(&map->rb_root, entry);
777 lwkt_reltoken(&vmspace_token);
781 * Finds the map entry containing (or immediately preceding) the specified
782 * address in the given map. The entry is returned in (*entry).
784 * The boolean result indicates whether the address is actually contained
787 * The related map must be locked.
788 * No other requirements.
791 vm_map_lookup_entry(vm_map_t map, vm_offset_t address, vm_map_entry_t *entry)
796 ASSERT_VM_MAP_LOCKED(map);
799 * XXX TEMPORARILY DISABLED. For some reason our attempt to revive
800 * the hint code with the red-black lookup meets with system crashes
801 * and lockups. We do not yet know why.
803 * It is possible that the problem is related to the setting
804 * of the hint during map_entry deletion, in the code specified
805 * at the GGG comment later on in this file.
808 * Quickly check the cached hint, there's a good chance of a match.
810 if (map->hint != &map->header) {
812 if (address >= tmp->start && address < tmp->end) {
820 * Locate the record from the top of the tree. 'last' tracks the
821 * closest prior record and is returned if no match is found, which
822 * in binary tree terms means tracking the most recent right-branch
823 * taken. If there is no prior record, &map->header is returned.
826 tmp = RB_ROOT(&map->rb_root);
829 if (address >= tmp->start) {
830 if (address < tmp->end) {
836 tmp = RB_RIGHT(tmp, rb_entry);
838 tmp = RB_LEFT(tmp, rb_entry);
846 * Inserts the given whole VM object into the target map at the specified
847 * address range. The object's size should match that of the address range.
849 * The map must be exclusively locked.
850 * The caller must have reserved sufficient vm_map_entry structures.
852 * If object is non-NULL, ref count must be bumped by caller
853 * prior to making call to account for the new entry.
856 vm_map_insert(vm_map_t map, int *countp,
857 vm_object_t object, vm_ooffset_t offset,
858 vm_offset_t start, vm_offset_t end,
859 vm_maptype_t maptype,
860 vm_prot_t prot, vm_prot_t max,
863 vm_map_entry_t new_entry;
864 vm_map_entry_t prev_entry;
865 vm_map_entry_t temp_entry;
866 vm_eflags_t protoeflags;
868 ASSERT_VM_MAP_LOCKED(map);
871 * Check that the start and end points are not bogus.
873 if ((start < map->min_offset) || (end > map->max_offset) ||
875 return (KERN_INVALID_ADDRESS);
878 * Find the entry prior to the proposed starting address; if it's part
879 * of an existing entry, this range is bogus.
881 if (vm_map_lookup_entry(map, start, &temp_entry))
882 return (KERN_NO_SPACE);
884 prev_entry = temp_entry;
887 * Assert that the next entry doesn't overlap the end point.
890 if ((prev_entry->next != &map->header) &&
891 (prev_entry->next->start < end))
892 return (KERN_NO_SPACE);
896 if (cow & MAP_COPY_ON_WRITE)
897 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
899 if (cow & MAP_NOFAULT) {
900 protoeflags |= MAP_ENTRY_NOFAULT;
902 KASSERT(object == NULL,
903 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
905 if (cow & MAP_DISABLE_SYNCER)
906 protoeflags |= MAP_ENTRY_NOSYNC;
907 if (cow & MAP_DISABLE_COREDUMP)
908 protoeflags |= MAP_ENTRY_NOCOREDUMP;
909 if (cow & MAP_IS_STACK)
910 protoeflags |= MAP_ENTRY_STACK;
911 if (cow & MAP_IS_KSTACK)
912 protoeflags |= MAP_ENTRY_KSTACK;
914 lwkt_gettoken(&vm_token);
915 lwkt_gettoken(&vmobj_token);
919 * When object is non-NULL, it could be shared with another
920 * process. We have to set or clear OBJ_ONEMAPPING
923 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
924 vm_object_clear_flag(object, OBJ_ONEMAPPING);
927 else if ((prev_entry != &map->header) &&
928 (prev_entry->eflags == protoeflags) &&
929 (prev_entry->end == start) &&
930 (prev_entry->wired_count == 0) &&
931 prev_entry->maptype == maptype &&
932 ((prev_entry->object.vm_object == NULL) ||
933 vm_object_coalesce(prev_entry->object.vm_object,
934 OFF_TO_IDX(prev_entry->offset),
935 (vm_size_t)(prev_entry->end - prev_entry->start),
936 (vm_size_t)(end - prev_entry->end)))) {
938 * We were able to extend the object. Determine if we
939 * can extend the previous map entry to include the
942 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
943 (prev_entry->protection == prot) &&
944 (prev_entry->max_protection == max)) {
945 lwkt_reltoken(&vmobj_token);
946 lwkt_reltoken(&vm_token);
947 map->size += (end - prev_entry->end);
948 prev_entry->end = end;
949 vm_map_simplify_entry(map, prev_entry, countp);
950 return (KERN_SUCCESS);
954 * If we can extend the object but cannot extend the
955 * map entry, we have to create a new map entry. We
956 * must bump the ref count on the extended object to
957 * account for it. object may be NULL.
959 object = prev_entry->object.vm_object;
960 offset = prev_entry->offset +
961 (prev_entry->end - prev_entry->start);
962 vm_object_reference_locked(object);
965 lwkt_reltoken(&vmobj_token);
966 lwkt_reltoken(&vm_token);
969 * NOTE: if conditionals fail, object can be NULL here. This occurs
970 * in things like the buffer map where we manage kva but do not manage
978 new_entry = vm_map_entry_create(map, countp);
979 new_entry->start = start;
980 new_entry->end = end;
982 new_entry->maptype = maptype;
983 new_entry->eflags = protoeflags;
984 new_entry->object.vm_object = object;
985 new_entry->offset = offset;
986 new_entry->aux.master_pde = 0;
988 new_entry->inheritance = VM_INHERIT_DEFAULT;
989 new_entry->protection = prot;
990 new_entry->max_protection = max;
991 new_entry->wired_count = 0;
994 * Insert the new entry into the list
997 vm_map_entry_link(map, prev_entry, new_entry);
998 map->size += new_entry->end - new_entry->start;
1001 * Update the free space hint. Entries cannot overlap.
1002 * An exact comparison is needed to avoid matching
1003 * against the map->header.
1005 if ((map->first_free == prev_entry) &&
1006 (prev_entry->end == new_entry->start)) {
1007 map->first_free = new_entry;
1012 * Temporarily removed to avoid MAP_STACK panic, due to
1013 * MAP_STACK being a huge hack. Will be added back in
1014 * when MAP_STACK (and the user stack mapping) is fixed.
1017 * It may be possible to simplify the entry
1019 vm_map_simplify_entry(map, new_entry, countp);
1023 * Try to pre-populate the page table. Mappings governed by virtual
1024 * page tables cannot be prepopulated without a lot of work, so
1027 if ((cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) &&
1028 maptype != VM_MAPTYPE_VPAGETABLE) {
1029 pmap_object_init_pt(map->pmap, start, prot,
1030 object, OFF_TO_IDX(offset), end - start,
1031 cow & MAP_PREFAULT_PARTIAL);
1034 return (KERN_SUCCESS);
1038 * Find sufficient space for `length' bytes in the given map, starting at
1039 * `start'. Returns 0 on success, 1 on no space.
1041 * This function will returned an arbitrarily aligned pointer. If no
1042 * particular alignment is required you should pass align as 1. Note that
1043 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
1044 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
1047 * 'align' should be a power of 2 but is not required to be.
1049 * The map must be exclusively locked.
1050 * No other requirements.
1053 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1054 vm_size_t align, int flags, vm_offset_t *addr)
1056 vm_map_entry_t entry, next;
1058 vm_offset_t align_mask;
1060 if (start < map->min_offset)
1061 start = map->min_offset;
1062 if (start > map->max_offset)
1066 * If the alignment is not a power of 2 we will have to use
1067 * a mod/division, set align_mask to a special value.
1069 if ((align | (align - 1)) + 1 != (align << 1))
1070 align_mask = (vm_offset_t)-1;
1072 align_mask = align - 1;
1075 * Look for the first possible address; if there's already something
1076 * at this address, we have to start after it.
1078 if (start == map->min_offset) {
1079 if ((entry = map->first_free) != &map->header)
1084 if (vm_map_lookup_entry(map, start, &tmp))
1090 * Look through the rest of the map, trying to fit a new region in the
1091 * gap between existing regions, or after the very last region.
1093 for (;; start = (entry = next)->end) {
1095 * Adjust the proposed start by the requested alignment,
1096 * be sure that we didn't wrap the address.
1098 if (align_mask == (vm_offset_t)-1)
1099 end = ((start + align - 1) / align) * align;
1101 end = (start + align_mask) & ~align_mask;
1106 * Find the end of the proposed new region. Be sure we didn't
1107 * go beyond the end of the map, or wrap around the address.
1108 * Then check to see if this is the last entry or if the
1109 * proposed end fits in the gap between this and the next
1112 end = start + length;
1113 if (end > map->max_offset || end < start)
1118 * If the next entry's start address is beyond the desired
1119 * end address we may have found a good entry.
1121 * If the next entry is a stack mapping we do not map into
1122 * the stack's reserved space.
1124 * XXX continue to allow mapping into the stack's reserved
1125 * space if doing a MAP_STACK mapping inside a MAP_STACK
1126 * mapping, for backwards compatibility. But the caller
1127 * really should use MAP_STACK | MAP_TRYFIXED if they
1130 if (next == &map->header)
1132 if (next->start >= end) {
1133 if ((next->eflags & MAP_ENTRY_STACK) == 0)
1135 if (flags & MAP_STACK)
1137 if (next->start - next->aux.avail_ssize >= end)
1144 * Grow the kernel_map if necessary. pmap_growkernel() will panic
1145 * if it fails. The kernel_map is locked and nothing can steal
1146 * our address space if pmap_growkernel() blocks.
1148 * NOTE: This may be unconditionally called for kldload areas on
1149 * x86_64 because these do not bump kernel_vm_end (which would
1150 * fill 128G worth of page tables!). Therefore we must not
1153 if (map == &kernel_map) {
1156 kstop = round_page(start + length);
1157 if (kstop > kernel_vm_end)
1158 pmap_growkernel(start, kstop);
1165 * vm_map_find finds an unallocated region in the target address map with
1166 * the given length. The search is defined to be first-fit from the
1167 * specified address; the region found is returned in the same parameter.
1169 * If object is non-NULL, ref count must be bumped by caller
1170 * prior to making call to account for the new entry.
1172 * No requirements. This function will lock the map temporarily.
1175 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1176 vm_offset_t *addr, vm_size_t length, vm_size_t align,
1178 vm_maptype_t maptype,
1179 vm_prot_t prot, vm_prot_t max,
1188 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1191 if (vm_map_findspace(map, start, length, align, 0, addr)) {
1193 vm_map_entry_release(count);
1194 return (KERN_NO_SPACE);
1198 result = vm_map_insert(map, &count, object, offset,
1199 start, start + length,
1204 vm_map_entry_release(count);
1210 * Simplify the given map entry by merging with either neighbor. This
1211 * routine also has the ability to merge with both neighbors.
1213 * This routine guarentees that the passed entry remains valid (though
1214 * possibly extended). When merging, this routine may delete one or
1215 * both neighbors. No action is taken on entries which have their
1216 * in-transition flag set.
1218 * The map must be exclusively locked.
1221 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
1223 vm_map_entry_t next, prev;
1224 vm_size_t prevsize, esize;
1226 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1227 ++mycpu->gd_cnt.v_intrans_coll;
1231 if (entry->maptype == VM_MAPTYPE_SUBMAP)
1235 if (prev != &map->header) {
1236 prevsize = prev->end - prev->start;
1237 if ( (prev->end == entry->start) &&
1238 (prev->maptype == entry->maptype) &&
1239 (prev->object.vm_object == entry->object.vm_object) &&
1240 (!prev->object.vm_object ||
1241 (prev->offset + prevsize == entry->offset)) &&
1242 (prev->eflags == entry->eflags) &&
1243 (prev->protection == entry->protection) &&
1244 (prev->max_protection == entry->max_protection) &&
1245 (prev->inheritance == entry->inheritance) &&
1246 (prev->wired_count == entry->wired_count)) {
1247 if (map->first_free == prev)
1248 map->first_free = entry;
1249 if (map->hint == prev)
1251 vm_map_entry_unlink(map, prev);
1252 entry->start = prev->start;
1253 entry->offset = prev->offset;
1254 if (prev->object.vm_object)
1255 vm_object_deallocate(prev->object.vm_object);
1256 vm_map_entry_dispose(map, prev, countp);
1261 if (next != &map->header) {
1262 esize = entry->end - entry->start;
1263 if ((entry->end == next->start) &&
1264 (next->maptype == entry->maptype) &&
1265 (next->object.vm_object == entry->object.vm_object) &&
1266 (!entry->object.vm_object ||
1267 (entry->offset + esize == next->offset)) &&
1268 (next->eflags == entry->eflags) &&
1269 (next->protection == entry->protection) &&
1270 (next->max_protection == entry->max_protection) &&
1271 (next->inheritance == entry->inheritance) &&
1272 (next->wired_count == entry->wired_count)) {
1273 if (map->first_free == next)
1274 map->first_free = entry;
1275 if (map->hint == next)
1277 vm_map_entry_unlink(map, next);
1278 entry->end = next->end;
1279 if (next->object.vm_object)
1280 vm_object_deallocate(next->object.vm_object);
1281 vm_map_entry_dispose(map, next, countp);
1287 * Asserts that the given entry begins at or after the specified address.
1288 * If necessary, it splits the entry into two.
1290 #define vm_map_clip_start(map, entry, startaddr, countp) \
1292 if (startaddr > entry->start) \
1293 _vm_map_clip_start(map, entry, startaddr, countp); \
1297 * This routine is called only when it is known that the entry must be split.
1299 * The map must be exclusively locked.
1302 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start,
1305 vm_map_entry_t new_entry;
1308 * Split off the front portion -- note that we must insert the new
1309 * entry BEFORE this one, so that this entry has the specified
1313 vm_map_simplify_entry(map, entry, countp);
1316 * If there is no object backing this entry, we might as well create
1317 * one now. If we defer it, an object can get created after the map
1318 * is clipped, and individual objects will be created for the split-up
1319 * map. This is a bit of a hack, but is also about the best place to
1320 * put this improvement.
1322 if (entry->object.vm_object == NULL && !map->system_map) {
1323 vm_map_entry_allocate_object(entry);
1326 new_entry = vm_map_entry_create(map, countp);
1327 *new_entry = *entry;
1329 new_entry->end = start;
1330 entry->offset += (start - entry->start);
1331 entry->start = start;
1333 vm_map_entry_link(map, entry->prev, new_entry);
1335 switch(entry->maptype) {
1336 case VM_MAPTYPE_NORMAL:
1337 case VM_MAPTYPE_VPAGETABLE:
1338 vm_object_reference(new_entry->object.vm_object);
1346 * Asserts that the given entry ends at or before the specified address.
1347 * If necessary, it splits the entry into two.
1349 * The map must be exclusively locked.
1351 #define vm_map_clip_end(map, entry, endaddr, countp) \
1353 if (endaddr < entry->end) \
1354 _vm_map_clip_end(map, entry, endaddr, countp); \
1358 * This routine is called only when it is known that the entry must be split.
1360 * The map must be exclusively locked.
1363 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end,
1366 vm_map_entry_t new_entry;
1369 * If there is no object backing this entry, we might as well create
1370 * one now. If we defer it, an object can get created after the map
1371 * is clipped, and individual objects will be created for the split-up
1372 * map. This is a bit of a hack, but is also about the best place to
1373 * put this improvement.
1376 if (entry->object.vm_object == NULL && !map->system_map) {
1377 vm_map_entry_allocate_object(entry);
1381 * Create a new entry and insert it AFTER the specified entry
1384 new_entry = vm_map_entry_create(map, countp);
1385 *new_entry = *entry;
1387 new_entry->start = entry->end = end;
1388 new_entry->offset += (end - entry->start);
1390 vm_map_entry_link(map, entry, new_entry);
1392 switch(entry->maptype) {
1393 case VM_MAPTYPE_NORMAL:
1394 case VM_MAPTYPE_VPAGETABLE:
1395 vm_object_reference(new_entry->object.vm_object);
1403 * Asserts that the starting and ending region addresses fall within the
1404 * valid range for the map.
1406 #define VM_MAP_RANGE_CHECK(map, start, end) \
1408 if (start < vm_map_min(map)) \
1409 start = vm_map_min(map); \
1410 if (end > vm_map_max(map)) \
1411 end = vm_map_max(map); \
1417 * Used to block when an in-transition collison occurs. The map
1418 * is unlocked for the sleep and relocked before the return.
1422 vm_map_transition_wait(vm_map_t map)
1425 tsleep(map, 0, "vment", 0);
1430 * When we do blocking operations with the map lock held it is
1431 * possible that a clip might have occured on our in-transit entry,
1432 * requiring an adjustment to the entry in our loop. These macros
1433 * help the pageable and clip_range code deal with the case. The
1434 * conditional costs virtually nothing if no clipping has occured.
1437 #define CLIP_CHECK_BACK(entry, save_start) \
1439 while (entry->start != save_start) { \
1440 entry = entry->prev; \
1441 KASSERT(entry != &map->header, ("bad entry clip")); \
1445 #define CLIP_CHECK_FWD(entry, save_end) \
1447 while (entry->end != save_end) { \
1448 entry = entry->next; \
1449 KASSERT(entry != &map->header, ("bad entry clip")); \
1455 * Clip the specified range and return the base entry. The
1456 * range may cover several entries starting at the returned base
1457 * and the first and last entry in the covering sequence will be
1458 * properly clipped to the requested start and end address.
1460 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1463 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1464 * covered by the requested range.
1466 * The map must be exclusively locked on entry and will remain locked
1467 * on return. If no range exists or the range contains holes and you
1468 * specified that no holes were allowed, NULL will be returned. This
1469 * routine may temporarily unlock the map in order avoid a deadlock when
1474 vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
1475 int *countp, int flags)
1477 vm_map_entry_t start_entry;
1478 vm_map_entry_t entry;
1481 * Locate the entry and effect initial clipping. The in-transition
1482 * case does not occur very often so do not try to optimize it.
1485 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1487 entry = start_entry;
1488 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1489 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1490 ++mycpu->gd_cnt.v_intrans_coll;
1491 ++mycpu->gd_cnt.v_intrans_wait;
1492 vm_map_transition_wait(map);
1494 * entry and/or start_entry may have been clipped while
1495 * we slept, or may have gone away entirely. We have
1496 * to restart from the lookup.
1502 * Since we hold an exclusive map lock we do not have to restart
1503 * after clipping, even though clipping may block in zalloc.
1505 vm_map_clip_start(map, entry, start, countp);
1506 vm_map_clip_end(map, entry, end, countp);
1507 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1510 * Scan entries covered by the range. When working on the next
1511 * entry a restart need only re-loop on the current entry which
1512 * we have already locked, since 'next' may have changed. Also,
1513 * even though entry is safe, it may have been clipped so we
1514 * have to iterate forwards through the clip after sleeping.
1516 while (entry->next != &map->header && entry->next->start < end) {
1517 vm_map_entry_t next = entry->next;
1519 if (flags & MAP_CLIP_NO_HOLES) {
1520 if (next->start > entry->end) {
1521 vm_map_unclip_range(map, start_entry,
1522 start, entry->end, countp, flags);
1527 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1528 vm_offset_t save_end = entry->end;
1529 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1530 ++mycpu->gd_cnt.v_intrans_coll;
1531 ++mycpu->gd_cnt.v_intrans_wait;
1532 vm_map_transition_wait(map);
1535 * clips might have occured while we blocked.
1537 CLIP_CHECK_FWD(entry, save_end);
1538 CLIP_CHECK_BACK(start_entry, start);
1542 * No restart necessary even though clip_end may block, we
1543 * are holding the map lock.
1545 vm_map_clip_end(map, next, end, countp);
1546 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1549 if (flags & MAP_CLIP_NO_HOLES) {
1550 if (entry->end != end) {
1551 vm_map_unclip_range(map, start_entry,
1552 start, entry->end, countp, flags);
1556 return(start_entry);
1560 * Undo the effect of vm_map_clip_range(). You should pass the same
1561 * flags and the same range that you passed to vm_map_clip_range().
1562 * This code will clear the in-transition flag on the entries and
1563 * wake up anyone waiting. This code will also simplify the sequence
1564 * and attempt to merge it with entries before and after the sequence.
1566 * The map must be locked on entry and will remain locked on return.
1568 * Note that you should also pass the start_entry returned by
1569 * vm_map_clip_range(). However, if you block between the two calls
1570 * with the map unlocked please be aware that the start_entry may
1571 * have been clipped and you may need to scan it backwards to find
1572 * the entry corresponding with the original start address. You are
1573 * responsible for this, vm_map_unclip_range() expects the correct
1574 * start_entry to be passed to it and will KASSERT otherwise.
1578 vm_map_unclip_range(vm_map_t map, vm_map_entry_t start_entry,
1579 vm_offset_t start, vm_offset_t end,
1580 int *countp, int flags)
1582 vm_map_entry_t entry;
1584 entry = start_entry;
1586 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1587 while (entry != &map->header && entry->start < end) {
1588 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1589 ("in-transition flag not set during unclip on: %p",
1591 KASSERT(entry->end <= end,
1592 ("unclip_range: tail wasn't clipped"));
1593 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1594 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1595 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1598 entry = entry->next;
1602 * Simplification does not block so there is no restart case.
1604 entry = start_entry;
1605 while (entry != &map->header && entry->start < end) {
1606 vm_map_simplify_entry(map, entry, countp);
1607 entry = entry->next;
1612 * Mark the given range as handled by a subordinate map.
1614 * This range must have been created with vm_map_find(), and no other
1615 * operations may have been performed on this range prior to calling
1618 * Submappings cannot be removed.
1623 vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
1625 vm_map_entry_t entry;
1626 int result = KERN_INVALID_ARGUMENT;
1629 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1632 VM_MAP_RANGE_CHECK(map, start, end);
1634 if (vm_map_lookup_entry(map, start, &entry)) {
1635 vm_map_clip_start(map, entry, start, &count);
1637 entry = entry->next;
1640 vm_map_clip_end(map, entry, end, &count);
1642 if ((entry->start == start) && (entry->end == end) &&
1643 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1644 (entry->object.vm_object == NULL)) {
1645 entry->object.sub_map = submap;
1646 entry->maptype = VM_MAPTYPE_SUBMAP;
1647 result = KERN_SUCCESS;
1650 vm_map_entry_release(count);
1656 * Sets the protection of the specified address region in the target map.
1657 * If "set_max" is specified, the maximum protection is to be set;
1658 * otherwise, only the current protection is affected.
1660 * The protection is not applicable to submaps, but is applicable to normal
1661 * maps and maps governed by virtual page tables. For example, when operating
1662 * on a virtual page table our protection basically controls how COW occurs
1663 * on the backing object, whereas the virtual page table abstraction itself
1664 * is an abstraction for userland.
1669 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1670 vm_prot_t new_prot, boolean_t set_max)
1672 vm_map_entry_t current;
1673 vm_map_entry_t entry;
1676 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1679 VM_MAP_RANGE_CHECK(map, start, end);
1681 if (vm_map_lookup_entry(map, start, &entry)) {
1682 vm_map_clip_start(map, entry, start, &count);
1684 entry = entry->next;
1688 * Make a first pass to check for protection violations.
1691 while ((current != &map->header) && (current->start < end)) {
1692 if (current->maptype == VM_MAPTYPE_SUBMAP) {
1694 vm_map_entry_release(count);
1695 return (KERN_INVALID_ARGUMENT);
1697 if ((new_prot & current->max_protection) != new_prot) {
1699 vm_map_entry_release(count);
1700 return (KERN_PROTECTION_FAILURE);
1702 current = current->next;
1706 * Go back and fix up protections. [Note that clipping is not
1707 * necessary the second time.]
1711 while ((current != &map->header) && (current->start < end)) {
1714 vm_map_clip_end(map, current, end, &count);
1716 old_prot = current->protection;
1718 current->protection =
1719 (current->max_protection = new_prot) &
1722 current->protection = new_prot;
1726 * Update physical map if necessary. Worry about copy-on-write
1727 * here -- CHECK THIS XXX
1730 if (current->protection != old_prot) {
1731 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1734 pmap_protect(map->pmap, current->start,
1736 current->protection & MASK(current));
1740 vm_map_simplify_entry(map, current, &count);
1742 current = current->next;
1746 vm_map_entry_release(count);
1747 return (KERN_SUCCESS);
1751 * This routine traverses a processes map handling the madvise
1752 * system call. Advisories are classified as either those effecting
1753 * the vm_map_entry structure, or those effecting the underlying
1756 * The <value> argument is used for extended madvise calls.
1761 vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end,
1762 int behav, off_t value)
1764 vm_map_entry_t current, entry;
1770 * Some madvise calls directly modify the vm_map_entry, in which case
1771 * we need to use an exclusive lock on the map and we need to perform
1772 * various clipping operations. Otherwise we only need a read-lock
1776 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1780 case MADV_SEQUENTIAL:
1794 vm_map_lock_read(map);
1797 vm_map_entry_release(count);
1802 * Locate starting entry and clip if necessary.
1805 VM_MAP_RANGE_CHECK(map, start, end);
1807 if (vm_map_lookup_entry(map, start, &entry)) {
1809 vm_map_clip_start(map, entry, start, &count);
1811 entry = entry->next;
1816 * madvise behaviors that are implemented in the vm_map_entry.
1818 * We clip the vm_map_entry so that behavioral changes are
1819 * limited to the specified address range.
1821 for (current = entry;
1822 (current != &map->header) && (current->start < end);
1823 current = current->next
1825 if (current->maptype == VM_MAPTYPE_SUBMAP)
1828 vm_map_clip_end(map, current, end, &count);
1832 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1834 case MADV_SEQUENTIAL:
1835 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1838 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1841 current->eflags |= MAP_ENTRY_NOSYNC;
1844 current->eflags &= ~MAP_ENTRY_NOSYNC;
1847 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1850 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1854 * Invalidate the related pmap entries, used
1855 * to flush portions of the real kernel's
1856 * pmap when the caller has removed or
1857 * modified existing mappings in a virtual
1860 pmap_remove(map->pmap,
1861 current->start, current->end);
1865 * Set the page directory page for a map
1866 * governed by a virtual page table. Mark
1867 * the entry as being governed by a virtual
1868 * page table if it is not.
1870 * XXX the page directory page is stored
1871 * in the avail_ssize field if the map_entry.
1873 * XXX the map simplification code does not
1874 * compare this field so weird things may
1875 * happen if you do not apply this function
1876 * to the entire mapping governed by the
1877 * virtual page table.
1879 if (current->maptype != VM_MAPTYPE_VPAGETABLE) {
1883 current->aux.master_pde = value;
1884 pmap_remove(map->pmap,
1885 current->start, current->end);
1891 vm_map_simplify_entry(map, current, &count);
1899 * madvise behaviors that are implemented in the underlying
1902 * Since we don't clip the vm_map_entry, we have to clip
1903 * the vm_object pindex and count.
1905 * NOTE! We currently do not support these functions on
1906 * virtual page tables.
1908 for (current = entry;
1909 (current != &map->header) && (current->start < end);
1910 current = current->next
1912 vm_offset_t useStart;
1914 if (current->maptype != VM_MAPTYPE_NORMAL)
1917 pindex = OFF_TO_IDX(current->offset);
1918 count = atop(current->end - current->start);
1919 useStart = current->start;
1921 if (current->start < start) {
1922 pindex += atop(start - current->start);
1923 count -= atop(start - current->start);
1926 if (current->end > end)
1927 count -= atop(current->end - end);
1932 vm_object_madvise(current->object.vm_object,
1933 pindex, count, behav);
1936 * Try to populate the page table. Mappings governed
1937 * by virtual page tables cannot be pre-populated
1938 * without a lot of work so don't try.
1940 if (behav == MADV_WILLNEED &&
1941 current->maptype != VM_MAPTYPE_VPAGETABLE) {
1942 pmap_object_init_pt(
1945 current->protection,
1946 current->object.vm_object,
1948 (count << PAGE_SHIFT),
1949 MAP_PREFAULT_MADVISE
1953 vm_map_unlock_read(map);
1955 vm_map_entry_release(count);
1961 * Sets the inheritance of the specified address range in the target map.
1962 * Inheritance affects how the map will be shared with child maps at the
1963 * time of vm_map_fork.
1966 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1967 vm_inherit_t new_inheritance)
1969 vm_map_entry_t entry;
1970 vm_map_entry_t temp_entry;
1973 switch (new_inheritance) {
1974 case VM_INHERIT_NONE:
1975 case VM_INHERIT_COPY:
1976 case VM_INHERIT_SHARE:
1979 return (KERN_INVALID_ARGUMENT);
1982 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1985 VM_MAP_RANGE_CHECK(map, start, end);
1987 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1989 vm_map_clip_start(map, entry, start, &count);
1991 entry = temp_entry->next;
1993 while ((entry != &map->header) && (entry->start < end)) {
1994 vm_map_clip_end(map, entry, end, &count);
1996 entry->inheritance = new_inheritance;
1998 vm_map_simplify_entry(map, entry, &count);
2000 entry = entry->next;
2003 vm_map_entry_release(count);
2004 return (KERN_SUCCESS);
2008 * Implement the semantics of mlock
2011 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end,
2012 boolean_t new_pageable)
2014 vm_map_entry_t entry;
2015 vm_map_entry_t start_entry;
2017 int rv = KERN_SUCCESS;
2020 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2022 VM_MAP_RANGE_CHECK(map, start, real_end);
2025 start_entry = vm_map_clip_range(map, start, end, &count,
2027 if (start_entry == NULL) {
2029 vm_map_entry_release(count);
2030 return (KERN_INVALID_ADDRESS);
2033 if (new_pageable == 0) {
2034 entry = start_entry;
2035 while ((entry != &map->header) && (entry->start < end)) {
2036 vm_offset_t save_start;
2037 vm_offset_t save_end;
2040 * Already user wired or hard wired (trivial cases)
2042 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
2043 entry = entry->next;
2046 if (entry->wired_count != 0) {
2047 entry->wired_count++;
2048 entry->eflags |= MAP_ENTRY_USER_WIRED;
2049 entry = entry->next;
2054 * A new wiring requires instantiation of appropriate
2055 * management structures and the faulting in of the
2058 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
2059 int copyflag = entry->eflags &
2060 MAP_ENTRY_NEEDS_COPY;
2061 if (copyflag && ((entry->protection &
2062 VM_PROT_WRITE) != 0)) {
2063 vm_map_entry_shadow(entry);
2064 } else if (entry->object.vm_object == NULL &&
2066 vm_map_entry_allocate_object(entry);
2069 entry->wired_count++;
2070 entry->eflags |= MAP_ENTRY_USER_WIRED;
2073 * Now fault in the area. Note that vm_fault_wire()
2074 * may release the map lock temporarily, it will be
2075 * relocked on return. The in-transition
2076 * flag protects the entries.
2078 save_start = entry->start;
2079 save_end = entry->end;
2080 rv = vm_fault_wire(map, entry, TRUE);
2082 CLIP_CHECK_BACK(entry, save_start);
2084 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
2085 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2086 entry->wired_count = 0;
2087 if (entry->end == save_end)
2089 entry = entry->next;
2090 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2092 end = save_start; /* unwire the rest */
2096 * note that even though the entry might have been
2097 * clipped, the USER_WIRED flag we set prevents
2098 * duplication so we do not have to do a
2101 entry = entry->next;
2105 * If we failed fall through to the unwiring section to
2106 * unwire what we had wired so far. 'end' has already
2113 * start_entry might have been clipped if we unlocked the
2114 * map and blocked. No matter how clipped it has gotten
2115 * there should be a fragment that is on our start boundary.
2117 CLIP_CHECK_BACK(start_entry, start);
2121 * Deal with the unwiring case.
2125 * This is the unwiring case. We must first ensure that the
2126 * range to be unwired is really wired down. We know there
2129 entry = start_entry;
2130 while ((entry != &map->header) && (entry->start < end)) {
2131 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2132 rv = KERN_INVALID_ARGUMENT;
2135 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
2136 entry = entry->next;
2140 * Now decrement the wiring count for each region. If a region
2141 * becomes completely unwired, unwire its physical pages and
2145 * The map entries are processed in a loop, checking to
2146 * make sure the entry is wired and asserting it has a wired
2147 * count. However, another loop was inserted more-or-less in
2148 * the middle of the unwiring path. This loop picks up the
2149 * "entry" loop variable from the first loop without first
2150 * setting it to start_entry. Naturally, the secound loop
2151 * is never entered and the pages backing the entries are
2152 * never unwired. This can lead to a leak of wired pages.
2154 entry = start_entry;
2155 while ((entry != &map->header) && (entry->start < end)) {
2156 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED,
2157 ("expected USER_WIRED on entry %p", entry));
2158 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2159 entry->wired_count--;
2160 if (entry->wired_count == 0)
2161 vm_fault_unwire(map, entry);
2162 entry = entry->next;
2166 vm_map_unclip_range(map, start_entry, start, real_end, &count,
2170 vm_map_entry_release(count);
2175 * Sets the pageability of the specified address range in the target map.
2176 * Regions specified as not pageable require locked-down physical
2177 * memory and physical page maps.
2179 * The map must not be locked, but a reference must remain to the map
2180 * throughout the call.
2182 * This function may be called via the zalloc path and must properly
2183 * reserve map entries for kernel_map.
2188 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
2190 vm_map_entry_t entry;
2191 vm_map_entry_t start_entry;
2193 int rv = KERN_SUCCESS;
2196 if (kmflags & KM_KRESERVE)
2197 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
2199 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2201 VM_MAP_RANGE_CHECK(map, start, real_end);
2204 start_entry = vm_map_clip_range(map, start, end, &count,
2206 if (start_entry == NULL) {
2208 rv = KERN_INVALID_ADDRESS;
2211 if ((kmflags & KM_PAGEABLE) == 0) {
2215 * 1. Holding the write lock, we create any shadow or zero-fill
2216 * objects that need to be created. Then we clip each map
2217 * entry to the region to be wired and increment its wiring
2218 * count. We create objects before clipping the map entries
2219 * to avoid object proliferation.
2221 * 2. We downgrade to a read lock, and call vm_fault_wire to
2222 * fault in the pages for any newly wired area (wired_count is
2225 * Downgrading to a read lock for vm_fault_wire avoids a
2226 * possible deadlock with another process that may have faulted
2227 * on one of the pages to be wired (it would mark the page busy,
2228 * blocking us, then in turn block on the map lock that we
2229 * hold). Because of problems in the recursive lock package,
2230 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
2231 * any actions that require the write lock must be done
2232 * beforehand. Because we keep the read lock on the map, the
2233 * copy-on-write status of the entries we modify here cannot
2236 entry = start_entry;
2237 while ((entry != &map->header) && (entry->start < end)) {
2239 * Trivial case if the entry is already wired
2241 if (entry->wired_count) {
2242 entry->wired_count++;
2243 entry = entry->next;
2248 * The entry is being newly wired, we have to setup
2249 * appropriate management structures. A shadow
2250 * object is required for a copy-on-write region,
2251 * or a normal object for a zero-fill region. We
2252 * do not have to do this for entries that point to sub
2253 * maps because we won't hold the lock on the sub map.
2255 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
2256 int copyflag = entry->eflags &
2257 MAP_ENTRY_NEEDS_COPY;
2258 if (copyflag && ((entry->protection &
2259 VM_PROT_WRITE) != 0)) {
2260 vm_map_entry_shadow(entry);
2261 } else if (entry->object.vm_object == NULL &&
2263 vm_map_entry_allocate_object(entry);
2267 entry->wired_count++;
2268 entry = entry->next;
2276 * HACK HACK HACK HACK
2278 * vm_fault_wire() temporarily unlocks the map to avoid
2279 * deadlocks. The in-transition flag from vm_map_clip_range
2280 * call should protect us from changes while the map is
2283 * NOTE: Previously this comment stated that clipping might
2284 * still occur while the entry is unlocked, but from
2285 * what I can tell it actually cannot.
2287 * It is unclear whether the CLIP_CHECK_*() calls
2288 * are still needed but we keep them in anyway.
2290 * HACK HACK HACK HACK
2293 entry = start_entry;
2294 while (entry != &map->header && entry->start < end) {
2296 * If vm_fault_wire fails for any page we need to undo
2297 * what has been done. We decrement the wiring count
2298 * for those pages which have not yet been wired (now)
2299 * and unwire those that have (later).
2301 vm_offset_t save_start = entry->start;
2302 vm_offset_t save_end = entry->end;
2304 if (entry->wired_count == 1)
2305 rv = vm_fault_wire(map, entry, FALSE);
2307 CLIP_CHECK_BACK(entry, save_start);
2309 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
2310 entry->wired_count = 0;
2311 if (entry->end == save_end)
2313 entry = entry->next;
2314 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2319 CLIP_CHECK_FWD(entry, save_end);
2320 entry = entry->next;
2324 * If a failure occured undo everything by falling through
2325 * to the unwiring code. 'end' has already been adjusted
2329 kmflags |= KM_PAGEABLE;
2332 * start_entry is still IN_TRANSITION but may have been
2333 * clipped since vm_fault_wire() unlocks and relocks the
2334 * map. No matter how clipped it has gotten there should
2335 * be a fragment that is on our start boundary.
2337 CLIP_CHECK_BACK(start_entry, start);
2340 if (kmflags & KM_PAGEABLE) {
2342 * This is the unwiring case. We must first ensure that the
2343 * range to be unwired is really wired down. We know there
2346 entry = start_entry;
2347 while ((entry != &map->header) && (entry->start < end)) {
2348 if (entry->wired_count == 0) {
2349 rv = KERN_INVALID_ARGUMENT;
2352 entry = entry->next;
2356 * Now decrement the wiring count for each region. If a region
2357 * becomes completely unwired, unwire its physical pages and
2360 entry = start_entry;
2361 while ((entry != &map->header) && (entry->start < end)) {
2362 entry->wired_count--;
2363 if (entry->wired_count == 0)
2364 vm_fault_unwire(map, entry);
2365 entry = entry->next;
2369 vm_map_unclip_range(map, start_entry, start, real_end,
2370 &count, MAP_CLIP_NO_HOLES);
2374 if (kmflags & KM_KRESERVE)
2375 vm_map_entry_krelease(count);
2377 vm_map_entry_release(count);
2382 * Mark a newly allocated address range as wired but do not fault in
2383 * the pages. The caller is expected to load the pages into the object.
2385 * The map must be locked on entry and will remain locked on return.
2386 * No other requirements.
2389 vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size,
2392 vm_map_entry_t scan;
2393 vm_map_entry_t entry;
2395 entry = vm_map_clip_range(map, addr, addr + size,
2396 countp, MAP_CLIP_NO_HOLES);
2398 scan != &map->header && scan->start < addr + size;
2399 scan = scan->next) {
2400 KKASSERT(entry->wired_count == 0);
2401 entry->wired_count = 1;
2403 vm_map_unclip_range(map, entry, addr, addr + size,
2404 countp, MAP_CLIP_NO_HOLES);
2408 * Push any dirty cached pages in the address range to their pager.
2409 * If syncio is TRUE, dirty pages are written synchronously.
2410 * If invalidate is TRUE, any cached pages are freed as well.
2412 * This routine is called by sys_msync()
2414 * Returns an error if any part of the specified range is not mapped.
2419 vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end,
2420 boolean_t syncio, boolean_t invalidate)
2422 vm_map_entry_t current;
2423 vm_map_entry_t entry;
2426 vm_ooffset_t offset;
2428 vm_map_lock_read(map);
2429 VM_MAP_RANGE_CHECK(map, start, end);
2430 if (!vm_map_lookup_entry(map, start, &entry)) {
2431 vm_map_unlock_read(map);
2432 return (KERN_INVALID_ADDRESS);
2435 * Make a first pass to check for holes.
2437 for (current = entry; current->start < end; current = current->next) {
2438 if (current->maptype == VM_MAPTYPE_SUBMAP) {
2439 vm_map_unlock_read(map);
2440 return (KERN_INVALID_ARGUMENT);
2442 if (end > current->end &&
2443 (current->next == &map->header ||
2444 current->end != current->next->start)) {
2445 vm_map_unlock_read(map);
2446 return (KERN_INVALID_ADDRESS);
2451 pmap_remove(vm_map_pmap(map), start, end);
2454 * Make a second pass, cleaning/uncaching pages from the indicated
2457 * Hold vm_token to avoid blocking in vm_object_reference()
2459 lwkt_gettoken(&vm_token);
2460 lwkt_gettoken(&vmobj_token);
2462 for (current = entry; current->start < end; current = current->next) {
2463 offset = current->offset + (start - current->start);
2464 size = (end <= current->end ? end : current->end) - start;
2465 if (current->maptype == VM_MAPTYPE_SUBMAP) {
2467 vm_map_entry_t tentry;
2470 smap = current->object.sub_map;
2471 vm_map_lock_read(smap);
2472 vm_map_lookup_entry(smap, offset, &tentry);
2473 tsize = tentry->end - offset;
2476 object = tentry->object.vm_object;
2477 offset = tentry->offset + (offset - tentry->start);
2478 vm_map_unlock_read(smap);
2480 object = current->object.vm_object;
2483 * Note that there is absolutely no sense in writing out
2484 * anonymous objects, so we track down the vnode object
2486 * We invalidate (remove) all pages from the address space
2487 * anyway, for semantic correctness.
2489 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2490 * may start out with a NULL object.
2492 while (object && object->backing_object) {
2493 offset += object->backing_object_offset;
2494 object = object->backing_object;
2495 if (object->size < OFF_TO_IDX( offset + size))
2496 size = IDX_TO_OFF(object->size) - offset;
2498 if (object && (object->type == OBJT_VNODE) &&
2499 (current->protection & VM_PROT_WRITE) &&
2500 (object->flags & OBJ_NOMSYNC) == 0) {
2502 * Flush pages if writing is allowed, invalidate them
2503 * if invalidation requested. Pages undergoing I/O
2504 * will be ignored by vm_object_page_remove().
2506 * We cannot lock the vnode and then wait for paging
2507 * to complete without deadlocking against vm_fault.
2508 * Instead we simply call vm_object_page_remove() and
2509 * allow it to block internally on a page-by-page
2510 * basis when it encounters pages undergoing async
2515 vm_object_reference_locked(object);
2516 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY);
2517 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2518 flags |= invalidate ? OBJPC_INVAL : 0;
2521 * When operating on a virtual page table just
2522 * flush the whole object. XXX we probably ought
2525 switch(current->maptype) {
2526 case VM_MAPTYPE_NORMAL:
2527 vm_object_page_clean(object,
2529 OFF_TO_IDX(offset + size + PAGE_MASK),
2532 case VM_MAPTYPE_VPAGETABLE:
2533 vm_object_page_clean(object, 0, 0, flags);
2536 vn_unlock(((struct vnode *)object->handle));
2537 vm_object_deallocate_locked(object);
2539 if (object && invalidate &&
2540 ((object->type == OBJT_VNODE) ||
2541 (object->type == OBJT_DEVICE))) {
2543 (object->type == OBJT_DEVICE) ? FALSE : TRUE;
2544 vm_object_reference_locked(object);
2545 switch(current->maptype) {
2546 case VM_MAPTYPE_NORMAL:
2547 vm_object_page_remove(object,
2549 OFF_TO_IDX(offset + size + PAGE_MASK),
2552 case VM_MAPTYPE_VPAGETABLE:
2553 vm_object_page_remove(object, 0, 0, clean_only);
2556 vm_object_deallocate_locked(object);
2561 lwkt_reltoken(&vmobj_token);
2562 lwkt_reltoken(&vm_token);
2563 vm_map_unlock_read(map);
2565 return (KERN_SUCCESS);
2569 * Make the region specified by this entry pageable.
2571 * The vm_map must be exclusively locked.
2574 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2576 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2577 entry->wired_count = 0;
2578 vm_fault_unwire(map, entry);
2582 * Deallocate the given entry from the target map.
2584 * The vm_map must be exclusively locked.
2587 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
2589 vm_map_entry_unlink(map, entry);
2590 map->size -= entry->end - entry->start;
2592 switch(entry->maptype) {
2593 case VM_MAPTYPE_NORMAL:
2594 case VM_MAPTYPE_VPAGETABLE:
2595 vm_object_deallocate(entry->object.vm_object);
2601 vm_map_entry_dispose(map, entry, countp);
2605 * Deallocates the given address range from the target map.
2607 * The vm_map must be exclusively locked.
2610 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
2613 vm_map_entry_t entry;
2614 vm_map_entry_t first_entry;
2616 ASSERT_VM_MAP_LOCKED(map);
2619 * Find the start of the region, and clip it. Set entry to point
2620 * at the first record containing the requested address or, if no
2621 * such record exists, the next record with a greater address. The
2622 * loop will run from this point until a record beyond the termination
2623 * address is encountered.
2625 * map->hint must be adjusted to not point to anything we delete,
2626 * so set it to the entry prior to the one being deleted.
2628 * GGG see other GGG comment.
2630 if (vm_map_lookup_entry(map, start, &first_entry)) {
2631 entry = first_entry;
2632 vm_map_clip_start(map, entry, start, countp);
2633 map->hint = entry->prev; /* possible problem XXX */
2635 map->hint = first_entry; /* possible problem XXX */
2636 entry = first_entry->next;
2640 * If a hole opens up prior to the current first_free then
2641 * adjust first_free. As with map->hint, map->first_free
2642 * cannot be left set to anything we might delete.
2644 if (entry == &map->header) {
2645 map->first_free = &map->header;
2646 } else if (map->first_free->start >= start) {
2647 map->first_free = entry->prev;
2651 * Step through all entries in this region
2653 while ((entry != &map->header) && (entry->start < end)) {
2654 vm_map_entry_t next;
2656 vm_pindex_t offidxstart, offidxend, count;
2659 * If we hit an in-transition entry we have to sleep and
2660 * retry. It's easier (and not really slower) to just retry
2661 * since this case occurs so rarely and the hint is already
2662 * pointing at the right place. We have to reset the
2663 * start offset so as not to accidently delete an entry
2664 * another process just created in vacated space.
2666 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2667 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2668 start = entry->start;
2669 ++mycpu->gd_cnt.v_intrans_coll;
2670 ++mycpu->gd_cnt.v_intrans_wait;
2671 vm_map_transition_wait(map);
2674 vm_map_clip_end(map, entry, end, countp);
2680 offidxstart = OFF_TO_IDX(entry->offset);
2681 count = OFF_TO_IDX(e - s);
2682 object = entry->object.vm_object;
2685 * Unwire before removing addresses from the pmap; otherwise,
2686 * unwiring will put the entries back in the pmap.
2688 if (entry->wired_count != 0)
2689 vm_map_entry_unwire(map, entry);
2691 offidxend = offidxstart + count;
2694 * Hold vm_token when manipulating vm_objects,
2696 * Hold vmobj_token when potentially adding or removing
2697 * objects (collapse requires both).
2699 lwkt_gettoken(&vm_token);
2700 lwkt_gettoken(&vmobj_token);
2702 if (object == &kernel_object) {
2703 vm_object_page_remove(object, offidxstart,
2706 pmap_remove(map->pmap, s, e);
2708 if (object != NULL &&
2709 object->ref_count != 1 &&
2710 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) ==
2712 (object->type == OBJT_DEFAULT ||
2713 object->type == OBJT_SWAP)) {
2714 vm_object_collapse(object);
2715 vm_object_page_remove(object, offidxstart,
2717 if (object->type == OBJT_SWAP) {
2718 swap_pager_freespace(object,
2722 if (offidxend >= object->size &&
2723 offidxstart < object->size) {
2724 object->size = offidxstart;
2728 lwkt_reltoken(&vmobj_token);
2729 lwkt_reltoken(&vm_token);
2732 * Delete the entry (which may delete the object) only after
2733 * removing all pmap entries pointing to its pages.
2734 * (Otherwise, its page frames may be reallocated, and any
2735 * modify bits will be set in the wrong object!)
2737 vm_map_entry_delete(map, entry, countp);
2740 return (KERN_SUCCESS);
2744 * Remove the given address range from the target map.
2745 * This is the exported form of vm_map_delete.
2750 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2755 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2757 VM_MAP_RANGE_CHECK(map, start, end);
2758 result = vm_map_delete(map, start, end, &count);
2760 vm_map_entry_release(count);
2766 * Assert that the target map allows the specified privilege on the
2767 * entire address region given. The entire region must be allocated.
2769 * The caller must specify whether the vm_map is already locked or not.
2772 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2773 vm_prot_t protection, boolean_t have_lock)
2775 vm_map_entry_t entry;
2776 vm_map_entry_t tmp_entry;
2779 if (have_lock == FALSE)
2780 vm_map_lock_read(map);
2782 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2783 if (have_lock == FALSE)
2784 vm_map_unlock_read(map);
2790 while (start < end) {
2791 if (entry == &map->header) {
2799 if (start < entry->start) {
2804 * Check protection associated with entry.
2807 if ((entry->protection & protection) != protection) {
2811 /* go to next entry */
2814 entry = entry->next;
2816 if (have_lock == FALSE)
2817 vm_map_unlock_read(map);
2822 * Split the pages in a map entry into a new object. This affords
2823 * easier removal of unused pages, and keeps object inheritance from
2824 * being a negative impact on memory usage.
2826 * The vm_map must be exclusively locked.
2829 vm_map_split(vm_map_entry_t entry)
2832 vm_object_t orig_object, new_object, source;
2834 vm_pindex_t offidxstart, offidxend, idx;
2836 vm_ooffset_t offset;
2838 orig_object = entry->object.vm_object;
2839 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2841 if (orig_object->ref_count <= 1)
2844 offset = entry->offset;
2848 offidxstart = OFF_TO_IDX(offset);
2849 offidxend = offidxstart + OFF_TO_IDX(e - s);
2850 size = offidxend - offidxstart;
2852 switch(orig_object->type) {
2854 new_object = default_pager_alloc(NULL, IDX_TO_OFF(size),
2858 new_object = swap_pager_alloc(NULL, IDX_TO_OFF(size),
2866 if (new_object == NULL)
2870 * vm_token required when manipulating vm_objects.
2872 lwkt_gettoken(&vm_token);
2873 lwkt_gettoken(&vmobj_token);
2875 source = orig_object->backing_object;
2876 if (source != NULL) {
2877 /* Referenced by new_object */
2878 vm_object_reference_locked(source);
2879 LIST_INSERT_HEAD(&source->shadow_head,
2880 new_object, shadow_list);
2881 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2882 new_object->backing_object_offset =
2883 orig_object->backing_object_offset +
2884 IDX_TO_OFF(offidxstart);
2885 new_object->backing_object = source;
2886 source->shadow_count++;
2887 source->generation++;
2890 for (idx = 0; idx < size; idx++) {
2894 m = vm_page_lookup(orig_object, offidxstart + idx);
2899 * We must wait for pending I/O to complete before we can
2902 * We do not have to VM_PROT_NONE the page as mappings should
2903 * not be changed by this operation.
2905 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2908 vm_page_rename(m, new_object, idx);
2909 /* page automatically made dirty by rename and cache handled */
2913 if (orig_object->type == OBJT_SWAP) {
2914 vm_object_pip_add(orig_object, 1);
2916 * copy orig_object pages into new_object
2917 * and destroy unneeded pages in
2920 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2921 vm_object_pip_wakeup(orig_object);
2925 * Wakeup the pages we played with. No spl protection is needed
2926 * for a simple wakeup.
2928 for (idx = 0; idx < size; idx++) {
2929 m = vm_page_lookup(new_object, idx);
2934 entry->object.vm_object = new_object;
2935 entry->offset = 0LL;
2936 vm_object_deallocate_locked(orig_object);
2937 lwkt_reltoken(&vmobj_token);
2938 lwkt_reltoken(&vm_token);
2942 * Copies the contents of the source entry to the destination
2943 * entry. The entries *must* be aligned properly.
2945 * The vm_map must be exclusively locked.
2946 * vm_token must be held
2949 vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2950 vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
2952 vm_object_t src_object;
2954 if (dst_entry->maptype == VM_MAPTYPE_SUBMAP)
2956 if (src_entry->maptype == VM_MAPTYPE_SUBMAP)
2959 ASSERT_LWKT_TOKEN_HELD(&vm_token);
2960 lwkt_gettoken(&vmobj_token); /* required for collapse */
2962 if (src_entry->wired_count == 0) {
2964 * If the source entry is marked needs_copy, it is already
2967 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2968 pmap_protect(src_map->pmap,
2971 src_entry->protection & ~VM_PROT_WRITE);
2975 * Make a copy of the object.
2977 if ((src_object = src_entry->object.vm_object) != NULL) {
2978 if ((src_object->handle == NULL) &&
2979 (src_object->type == OBJT_DEFAULT ||
2980 src_object->type == OBJT_SWAP)) {
2981 vm_object_collapse(src_object);
2982 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2983 vm_map_split(src_entry);
2984 src_object = src_entry->object.vm_object;
2988 vm_object_reference_locked(src_object);
2989 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2990 dst_entry->object.vm_object = src_object;
2991 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2992 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2993 dst_entry->offset = src_entry->offset;
2995 dst_entry->object.vm_object = NULL;
2996 dst_entry->offset = 0;
2999 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3000 dst_entry->end - dst_entry->start, src_entry->start);
3003 * Of course, wired down pages can't be set copy-on-write.
3004 * Cause wired pages to be copied into the new map by
3005 * simulating faults (the new pages are pageable)
3007 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
3009 lwkt_reltoken(&vmobj_token);
3014 * Create a new process vmspace structure and vm_map
3015 * based on those of an existing process. The new map
3016 * is based on the old map, according to the inheritance
3017 * values on the regions in that map.
3019 * The source map must not be locked.
3023 vmspace_fork(struct vmspace *vm1)
3025 struct vmspace *vm2;
3026 vm_map_t old_map = &vm1->vm_map;
3028 vm_map_entry_t old_entry;
3029 vm_map_entry_t new_entry;
3033 lwkt_gettoken(&vm_token);
3034 lwkt_gettoken(&vmspace_token);
3035 lwkt_gettoken(&vmobj_token);
3036 vm_map_lock(old_map);
3037 old_map->infork = 1;
3040 * XXX Note: upcalls are not copied.
3042 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3043 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
3044 (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
3045 new_map = &vm2->vm_map; /* XXX */
3046 new_map->timestamp = 1;
3048 vm_map_lock(new_map);
3051 old_entry = old_map->header.next;
3052 while (old_entry != &old_map->header) {
3054 old_entry = old_entry->next;
3057 count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
3059 old_entry = old_map->header.next;
3060 while (old_entry != &old_map->header) {
3061 if (old_entry->maptype == VM_MAPTYPE_SUBMAP)
3062 panic("vm_map_fork: encountered a submap");
3064 switch (old_entry->inheritance) {
3065 case VM_INHERIT_NONE:
3067 case VM_INHERIT_SHARE:
3069 * Clone the entry, creating the shared object if
3072 object = old_entry->object.vm_object;
3073 if (object == NULL) {
3074 vm_map_entry_allocate_object(old_entry);
3075 object = old_entry->object.vm_object;
3079 * Add the reference before calling vm_map_entry_shadow
3080 * to insure that a shadow object is created.
3082 vm_object_reference_locked(object);
3083 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3084 vm_map_entry_shadow(old_entry);
3085 /* Transfer the second reference too. */
3086 vm_object_reference_locked(
3087 old_entry->object.vm_object);
3088 vm_object_deallocate_locked(object);
3089 object = old_entry->object.vm_object;
3091 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3094 * Clone the entry, referencing the shared object.
3096 new_entry = vm_map_entry_create(new_map, &count);
3097 *new_entry = *old_entry;
3098 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3099 new_entry->wired_count = 0;
3102 * Insert the entry into the new map -- we know we're
3103 * inserting at the end of the new map.
3106 vm_map_entry_link(new_map, new_map->header.prev,
3110 * Update the physical map
3112 pmap_copy(new_map->pmap, old_map->pmap,
3114 (old_entry->end - old_entry->start),
3117 case VM_INHERIT_COPY:
3119 * Clone the entry and link into the map.
3121 new_entry = vm_map_entry_create(new_map, &count);
3122 *new_entry = *old_entry;
3123 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3124 new_entry->wired_count = 0;
3125 new_entry->object.vm_object = NULL;
3126 vm_map_entry_link(new_map, new_map->header.prev,
3128 vm_map_copy_entry(old_map, new_map, old_entry,
3132 old_entry = old_entry->next;
3135 new_map->size = old_map->size;
3136 old_map->infork = 0;
3137 vm_map_unlock(old_map);
3138 vm_map_unlock(new_map);
3139 vm_map_entry_release(count);
3141 lwkt_reltoken(&vmobj_token);
3142 lwkt_reltoken(&vmspace_token);
3143 lwkt_reltoken(&vm_token);
3149 * Create an auto-grow stack entry
3154 vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3155 int flags, vm_prot_t prot, vm_prot_t max, int cow)
3157 vm_map_entry_t prev_entry;
3158 vm_map_entry_t new_stack_entry;
3159 vm_size_t init_ssize;
3162 vm_offset_t tmpaddr;
3164 cow |= MAP_IS_STACK;
3166 if (max_ssize < sgrowsiz)
3167 init_ssize = max_ssize;
3169 init_ssize = sgrowsiz;
3171 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3175 * Find space for the mapping
3177 if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
3178 if (vm_map_findspace(map, addrbos, max_ssize, 1,
3181 vm_map_entry_release(count);
3182 return (KERN_NO_SPACE);
3187 /* If addr is already mapped, no go */
3188 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3190 vm_map_entry_release(count);
3191 return (KERN_NO_SPACE);
3195 /* XXX already handled by kern_mmap() */
3196 /* If we would blow our VMEM resource limit, no go */
3197 if (map->size + init_ssize >
3198 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
3200 vm_map_entry_release(count);
3201 return (KERN_NO_SPACE);
3206 * If we can't accomodate max_ssize in the current mapping,
3207 * no go. However, we need to be aware that subsequent user
3208 * mappings might map into the space we have reserved for
3209 * stack, and currently this space is not protected.
3211 * Hopefully we will at least detect this condition
3212 * when we try to grow the stack.
3214 if ((prev_entry->next != &map->header) &&
3215 (prev_entry->next->start < addrbos + max_ssize)) {
3217 vm_map_entry_release(count);
3218 return (KERN_NO_SPACE);
3222 * We initially map a stack of only init_ssize. We will
3223 * grow as needed later. Since this is to be a grow
3224 * down stack, we map at the top of the range.
3226 * Note: we would normally expect prot and max to be
3227 * VM_PROT_ALL, and cow to be 0. Possibly we should
3228 * eliminate these as input parameters, and just
3229 * pass these values here in the insert call.
3231 rv = vm_map_insert(map, &count,
3232 NULL, 0, addrbos + max_ssize - init_ssize,
3233 addrbos + max_ssize,
3238 /* Now set the avail_ssize amount */
3239 if (rv == KERN_SUCCESS) {
3240 if (prev_entry != &map->header)
3241 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
3242 new_stack_entry = prev_entry->next;
3243 if (new_stack_entry->end != addrbos + max_ssize ||
3244 new_stack_entry->start != addrbos + max_ssize - init_ssize)
3245 panic ("Bad entry start/end for new stack entry");
3247 new_stack_entry->aux.avail_ssize = max_ssize - init_ssize;
3251 vm_map_entry_release(count);
3256 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3257 * desired address is already mapped, or if we successfully grow
3258 * the stack. Also returns KERN_SUCCESS if addr is outside the
3259 * stack range (this is strange, but preserves compatibility with
3260 * the grow function in vm_machdep.c).
3265 vm_map_growstack (struct proc *p, vm_offset_t addr)
3267 vm_map_entry_t prev_entry;
3268 vm_map_entry_t stack_entry;
3269 vm_map_entry_t new_stack_entry;
3270 struct vmspace *vm = p->p_vmspace;
3271 vm_map_t map = &vm->vm_map;
3274 int rv = KERN_SUCCESS;
3276 int use_read_lock = 1;
3279 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3282 vm_map_lock_read(map);
3286 /* If addr is already in the entry range, no need to grow.*/
3287 if (vm_map_lookup_entry(map, addr, &prev_entry))
3290 if ((stack_entry = prev_entry->next) == &map->header)
3292 if (prev_entry == &map->header)
3293 end = stack_entry->start - stack_entry->aux.avail_ssize;
3295 end = prev_entry->end;
3298 * This next test mimics the old grow function in vm_machdep.c.
3299 * It really doesn't quite make sense, but we do it anyway
3300 * for compatibility.
3302 * If not growable stack, return success. This signals the
3303 * caller to proceed as he would normally with normal vm.
3305 if (stack_entry->aux.avail_ssize < 1 ||
3306 addr >= stack_entry->start ||
3307 addr < stack_entry->start - stack_entry->aux.avail_ssize) {
3311 /* Find the minimum grow amount */
3312 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
3313 if (grow_amount > stack_entry->aux.avail_ssize) {
3319 * If there is no longer enough space between the entries
3320 * nogo, and adjust the available space. Note: this
3321 * should only happen if the user has mapped into the
3322 * stack area after the stack was created, and is
3323 * probably an error.
3325 * This also effectively destroys any guard page the user
3326 * might have intended by limiting the stack size.
3328 if (grow_amount > stack_entry->start - end) {
3329 if (use_read_lock && vm_map_lock_upgrade(map)) {
3334 stack_entry->aux.avail_ssize = stack_entry->start - end;
3339 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
3341 /* If this is the main process stack, see if we're over the
3344 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3345 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3350 /* Round up the grow amount modulo SGROWSIZ */
3351 grow_amount = roundup (grow_amount, sgrowsiz);
3352 if (grow_amount > stack_entry->aux.avail_ssize) {
3353 grow_amount = stack_entry->aux.avail_ssize;
3355 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3356 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3357 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
3361 /* If we would blow our VMEM resource limit, no go */
3362 if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
3367 if (use_read_lock && vm_map_lock_upgrade(map)) {
3373 /* Get the preliminary new entry start value */
3374 addr = stack_entry->start - grow_amount;
3376 /* If this puts us into the previous entry, cut back our growth
3377 * to the available space. Also, see the note above.
3380 stack_entry->aux.avail_ssize = stack_entry->start - end;
3384 rv = vm_map_insert(map, &count,
3385 NULL, 0, addr, stack_entry->start,
3387 VM_PROT_ALL, VM_PROT_ALL,
3390 /* Adjust the available stack space by the amount we grew. */
3391 if (rv == KERN_SUCCESS) {
3392 if (prev_entry != &map->header)
3393 vm_map_clip_end(map, prev_entry, addr, &count);
3394 new_stack_entry = prev_entry->next;
3395 if (new_stack_entry->end != stack_entry->start ||
3396 new_stack_entry->start != addr)
3397 panic ("Bad stack grow start/end in new stack entry");
3399 new_stack_entry->aux.avail_ssize =
3400 stack_entry->aux.avail_ssize -
3401 (new_stack_entry->end - new_stack_entry->start);
3403 vm->vm_ssize += btoc(new_stack_entry->end -
3404 new_stack_entry->start);
3410 vm_map_unlock_read(map);
3413 vm_map_entry_release(count);
3418 * Unshare the specified VM space for exec. If other processes are
3419 * mapped to it, then create a new one. The new vmspace is null.
3424 vmspace_exec(struct proc *p, struct vmspace *vmcopy)
3426 struct vmspace *oldvmspace = p->p_vmspace;
3427 struct vmspace *newvmspace;
3428 vm_map_t map = &p->p_vmspace->vm_map;
3431 * If we are execing a resident vmspace we fork it, otherwise
3432 * we create a new vmspace. Note that exitingcnt and upcalls
3433 * are not copied to the new vmspace.
3435 lwkt_gettoken(&vmspace_token);
3437 newvmspace = vmspace_fork(vmcopy);
3439 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3440 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
3441 (caddr_t)&oldvmspace->vm_endcopy -
3442 (caddr_t)&oldvmspace->vm_startcopy);
3446 * Finish initializing the vmspace before assigning it
3447 * to the process. The vmspace will become the current vmspace
3450 pmap_pinit2(vmspace_pmap(newvmspace));
3451 pmap_replacevm(p, newvmspace, 0);
3452 sysref_put(&oldvmspace->vm_sysref);
3453 lwkt_reltoken(&vmspace_token);
3457 * Unshare the specified VM space for forcing COW. This
3458 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3460 * The exitingcnt test is not strictly necessary but has been
3461 * included for code sanity (to make the code a bit more deterministic).
3464 vmspace_unshare(struct proc *p)
3466 struct vmspace *oldvmspace = p->p_vmspace;
3467 struct vmspace *newvmspace;
3469 lwkt_gettoken(&vmspace_token);
3470 if (oldvmspace->vm_sysref.refcnt == 1 && oldvmspace->vm_exitingcnt == 0)
3472 newvmspace = vmspace_fork(oldvmspace);
3473 pmap_pinit2(vmspace_pmap(newvmspace));
3474 pmap_replacevm(p, newvmspace, 0);
3475 sysref_put(&oldvmspace->vm_sysref);
3476 lwkt_reltoken(&vmspace_token);
3480 * Finds the VM object, offset, and protection for a given virtual address
3481 * in the specified map, assuming a page fault of the type specified.
3483 * Leaves the map in question locked for read; return values are guaranteed
3484 * until a vm_map_lookup_done call is performed. Note that the map argument
3485 * is in/out; the returned map must be used in the call to vm_map_lookup_done.
3487 * A handle (out_entry) is returned for use in vm_map_lookup_done, to make
3490 * If a lookup is requested with "write protection" specified, the map may
3491 * be changed to perform virtual copying operations, although the data
3492 * referenced will remain the same.
3497 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3499 vm_prot_t fault_typea,
3500 vm_map_entry_t *out_entry, /* OUT */
3501 vm_object_t *object, /* OUT */
3502 vm_pindex_t *pindex, /* OUT */
3503 vm_prot_t *out_prot, /* OUT */
3504 boolean_t *wired) /* OUT */
3506 vm_map_entry_t entry;
3507 vm_map_t map = *var_map;
3509 vm_prot_t fault_type = fault_typea;
3510 int use_read_lock = 1;
3511 int rv = KERN_SUCCESS;
3515 vm_map_lock_read(map);
3520 * If the map has an interesting hint, try it before calling full
3521 * blown lookup routine.
3526 if ((entry == &map->header) ||
3527 (vaddr < entry->start) || (vaddr >= entry->end)) {
3528 vm_map_entry_t tmp_entry;
3531 * Entry was either not a valid hint, or the vaddr was not
3532 * contained in the entry, so do a full lookup.
3534 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
3535 rv = KERN_INVALID_ADDRESS;
3546 if (entry->maptype == VM_MAPTYPE_SUBMAP) {
3547 vm_map_t old_map = map;
3549 *var_map = map = entry->object.sub_map;
3551 vm_map_unlock_read(old_map);
3553 vm_map_unlock(old_map);
3559 * Check whether this task is allowed to have this page.
3560 * Note the special case for MAP_ENTRY_COW
3561 * pages with an override. This is to implement a forced
3562 * COW for debuggers.
3565 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3566 prot = entry->max_protection;
3568 prot = entry->protection;
3570 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3571 if ((fault_type & prot) != fault_type) {
3572 rv = KERN_PROTECTION_FAILURE;
3576 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3577 (entry->eflags & MAP_ENTRY_COW) &&
3578 (fault_type & VM_PROT_WRITE) &&
3579 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3580 rv = KERN_PROTECTION_FAILURE;
3585 * If this page is not pageable, we have to get it for all possible
3588 *wired = (entry->wired_count != 0);
3590 prot = fault_type = entry->protection;
3593 * Virtual page tables may need to update the accessed (A) bit
3594 * in a page table entry. Upgrade the fault to a write fault for
3595 * that case if the map will support it. If the map does not support
3596 * it the page table entry simply will not be updated.
3598 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
3599 if (prot & VM_PROT_WRITE)
3600 fault_type |= VM_PROT_WRITE;
3604 * If the entry was copy-on-write, we either ...
3606 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3608 * If we want to write the page, we may as well handle that
3609 * now since we've got the map locked.
3611 * If we don't need to write the page, we just demote the
3612 * permissions allowed.
3615 if (fault_type & VM_PROT_WRITE) {
3617 * Make a new object, and place it in the object
3618 * chain. Note that no new references have appeared
3619 * -- one just moved from the map to the new
3623 if (use_read_lock && vm_map_lock_upgrade(map)) {
3629 vm_map_entry_shadow(entry);
3632 * We're attempting to read a copy-on-write page --
3633 * don't allow writes.
3636 prot &= ~VM_PROT_WRITE;
3641 * Create an object if necessary.
3643 if (entry->object.vm_object == NULL &&
3645 if (use_read_lock && vm_map_lock_upgrade(map)) {
3650 vm_map_entry_allocate_object(entry);
3654 * Return the object/offset from this entry. If the entry was
3655 * copy-on-write or empty, it has been fixed up.
3658 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3659 *object = entry->object.vm_object;
3662 * Return whether this is the only map sharing this data. On
3663 * success we return with a read lock held on the map. On failure
3664 * we return with the map unlocked.
3668 if (rv == KERN_SUCCESS) {
3669 if (use_read_lock == 0)
3670 vm_map_lock_downgrade(map);
3671 } else if (use_read_lock) {
3672 vm_map_unlock_read(map);
3680 * Releases locks acquired by a vm_map_lookup()
3681 * (according to the handle returned by that lookup).
3683 * No other requirements.
3686 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count)
3689 * Unlock the main-level map
3691 vm_map_unlock_read(map);
3693 vm_map_entry_release(count);
3696 #include "opt_ddb.h"
3698 #include <sys/kernel.h>
3700 #include <ddb/ddb.h>
3705 DB_SHOW_COMMAND(map, vm_map_print)
3708 /* XXX convert args. */
3709 vm_map_t map = (vm_map_t)addr;
3710 boolean_t full = have_addr;
3712 vm_map_entry_t entry;
3714 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3716 (void *)map->pmap, map->nentries, map->timestamp);
3719 if (!full && db_indent)
3723 for (entry = map->header.next; entry != &map->header;
3724 entry = entry->next) {
3725 db_iprintf("map entry %p: start=%p, end=%p\n",
3726 (void *)entry, (void *)entry->start, (void *)entry->end);
3729 static char *inheritance_name[4] =
3730 {"share", "copy", "none", "donate_copy"};
3732 db_iprintf(" prot=%x/%x/%s",
3734 entry->max_protection,
3735 inheritance_name[(int)(unsigned char)entry->inheritance]);
3736 if (entry->wired_count != 0)
3737 db_printf(", wired");
3739 if (entry->maptype == VM_MAPTYPE_SUBMAP) {
3740 /* XXX no %qd in kernel. Truncate entry->offset. */
3741 db_printf(", share=%p, offset=0x%lx\n",
3742 (void *)entry->object.sub_map,
3743 (long)entry->offset);
3745 if ((entry->prev == &map->header) ||
3746 (entry->prev->object.sub_map !=
3747 entry->object.sub_map)) {
3749 vm_map_print((db_expr_t)(intptr_t)
3750 entry->object.sub_map,
3755 /* XXX no %qd in kernel. Truncate entry->offset. */
3756 db_printf(", object=%p, offset=0x%lx",
3757 (void *)entry->object.vm_object,
3758 (long)entry->offset);
3759 if (entry->eflags & MAP_ENTRY_COW)
3760 db_printf(", copy (%s)",
3761 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3765 if ((entry->prev == &map->header) ||
3766 (entry->prev->object.vm_object !=
3767 entry->object.vm_object)) {
3769 vm_object_print((db_expr_t)(intptr_t)
3770 entry->object.vm_object,
3785 DB_SHOW_COMMAND(procvm, procvm)
3790 p = (struct proc *) addr;
3795 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3796 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3797 (void *)vmspace_pmap(p->p_vmspace));
3799 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);