FreeBSD/Linux Kernel Cross Reference
sys/uvm/uvm_map.c
1 /* $NetBSD: uvm_map.c,v 1.186.2.2 2005/05/01 11:05:06 tron Exp $ */
2
3 /*
4 * Copyright (c) 1997 Charles D. Cranor and Washington University.
5 * Copyright (c) 1991, 1993, The Regents of the University of California.
6 *
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * The Mach Operating System project at Carnegie-Mellon University.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by Charles D. Cranor,
23 * Washington University, the University of California, Berkeley and
24 * its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * SUCH DAMAGE.
40 *
41 * @(#)vm_map.c 8.3 (Berkeley) 1/12/94
42 * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
43 *
44 *
45 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46 * All rights reserved.
47 *
48 * Permission to use, copy, modify and distribute this software and
49 * its documentation is hereby granted, provided that both the copyright
50 * notice and this permission notice appear in all copies of the
51 * software, derivative works or modified versions, and any portions
52 * thereof, and that both notices appear in supporting documentation.
53 *
54 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57 *
58 * Carnegie Mellon requests users of this software to return to
59 *
60 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
61 * School of Computer Science
62 * Carnegie Mellon University
63 * Pittsburgh PA 15213-3890
64 *
65 * any improvements or extensions that they make and grant Carnegie the
66 * rights to redistribute these changes.
67 */
68
69 /*
70 * uvm_map.c: uvm map operations
71 */
72
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.186.2.2 2005/05/01 11:05:06 tron Exp $");
75
76 #include "opt_ddb.h"
77 #include "opt_uvmhist.h"
78 #include "opt_uvm.h"
79 #include "opt_sysv.h"
80
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mman.h>
84 #include <sys/proc.h>
85 #include <sys/malloc.h>
86 #include <sys/pool.h>
87 #include <sys/kernel.h>
88 #include <sys/mount.h>
89 #include <sys/vnode.h>
90
91 #ifdef SYSVSHM
92 #include <sys/shm.h>
93 #endif
94
95 #define UVM_MAP
96 #include <uvm/uvm.h>
97 #undef RB_AUGMENT
98 #define RB_AUGMENT(x) uvm_rb_augment(x)
99
100 #ifdef DDB
101 #include <uvm/uvm_ddb.h>
102 #endif
103
104 #ifndef UVMMAP_NOCOUNTERS
105 #include <sys/device.h>
106 struct evcnt map_ubackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
107 "uvmmap", "ubackmerge");
108 struct evcnt map_uforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
109 "uvmmap", "uforwmerge");
110 struct evcnt map_ubimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
111 "uvmmap", "ubimerge");
112 struct evcnt map_unomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
113 "uvmmap", "unomerge");
114 struct evcnt map_kbackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
115 "uvmmap", "kbackmerge");
116 struct evcnt map_kforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
117 "uvmmap", "kforwmerge");
118 struct evcnt map_kbimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
119 "uvmmap", "kbimerge");
120 struct evcnt map_knomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
121 "uvmmap", "knomerge");
122 struct evcnt uvm_map_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
123 "uvmmap", "map_call");
124 struct evcnt uvm_mlk_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
125 "uvmmap", "mlk_call");
126 struct evcnt uvm_mlk_hint = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
127 "uvmmap", "mlk_hint");
128
129 EVCNT_ATTACH_STATIC(map_ubackmerge);
130 EVCNT_ATTACH_STATIC(map_uforwmerge);
131 EVCNT_ATTACH_STATIC(map_ubimerge);
132 EVCNT_ATTACH_STATIC(map_unomerge);
133 EVCNT_ATTACH_STATIC(map_kbackmerge);
134 EVCNT_ATTACH_STATIC(map_kforwmerge);
135 EVCNT_ATTACH_STATIC(map_kbimerge);
136 EVCNT_ATTACH_STATIC(map_knomerge);
137 EVCNT_ATTACH_STATIC(uvm_map_call);
138 EVCNT_ATTACH_STATIC(uvm_mlk_call);
139 EVCNT_ATTACH_STATIC(uvm_mlk_hint);
140
141 #define UVMCNT_INCR(ev) ev.ev_count++
142 #define UVMCNT_DECR(ev) ev.ev_count--
143 #else
144 #define UVMCNT_INCR(ev)
145 #define UVMCNT_DECR(ev)
146 #endif
147
148 const char vmmapbsy[] = "vmmapbsy";
149
150 /*
151 * pool for vmspace structures.
152 */
153
154 POOL_INIT(uvm_vmspace_pool, sizeof(struct vmspace), 0, 0, 0, "vmsppl",
155 &pool_allocator_nointr);
156
157 /*
158 * pool for dynamically-allocated map entries.
159 */
160
161 POOL_INIT(uvm_map_entry_pool, sizeof(struct vm_map_entry), 0, 0, 0, "vmmpepl",
162 &pool_allocator_nointr);
163
164 MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures");
165 MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap");
166
167 #ifdef PMAP_GROWKERNEL
168 /*
169 * This global represents the end of the kernel virtual address
170 * space. If we want to exceed this, we must grow the kernel
171 * virtual address space dynamically.
172 *
173 * Note, this variable is locked by kernel_map's lock.
174 */
175 vaddr_t uvm_maxkaddr;
176 #endif
177
178 /*
179 * macros
180 */
181
182 /*
183 * VM_MAP_USE_KMAPENT: determine if uvm_kmapent_alloc/free is used
184 * for the vm_map.
185 */
186 extern struct vm_map *pager_map; /* XXX */
187 #define VM_MAP_USE_KMAPENT(map) \
188 (((map)->flags & VM_MAP_INTRSAFE) || (map) == kernel_map)
189
190 /*
191 * uvm_map_entry_link: insert entry into a map
192 *
193 * => map must be locked
194 */
195 #define uvm_map_entry_link(map, after_where, entry) do { \
196 KASSERT(entry->start < entry->end); \
197 (map)->nentries++; \
198 (entry)->prev = (after_where); \
199 (entry)->next = (after_where)->next; \
200 (entry)->prev->next = (entry); \
201 (entry)->next->prev = (entry); \
202 uvm_rb_insert((map), (entry)); \
203 } while (/*CONSTCOND*/ 0)
204
205 /*
206 * uvm_map_entry_unlink: remove entry from a map
207 *
208 * => map must be locked
209 */
210 #define uvm_map_entry_unlink(map, entry) do { \
211 (map)->nentries--; \
212 (entry)->next->prev = (entry)->prev; \
213 (entry)->prev->next = (entry)->next; \
214 uvm_rb_remove((map), (entry)); \
215 } while (/*CONSTCOND*/ 0)
216
217 /*
218 * SAVE_HINT: saves the specified entry as the hint for future lookups.
219 *
220 * => map need not be locked (protected by hint_lock).
221 */
222 #define SAVE_HINT(map,check,value) do { \
223 simple_lock(&(map)->hint_lock); \
224 if ((map)->hint == (check)) \
225 (map)->hint = (value); \
226 simple_unlock(&(map)->hint_lock); \
227 } while (/*CONSTCOND*/ 0)
228
229 /*
230 * VM_MAP_RANGE_CHECK: check and correct range
231 *
232 * => map must at least be read locked
233 */
234
235 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
236 if (start < vm_map_min(map)) \
237 start = vm_map_min(map); \
238 if (end > vm_map_max(map)) \
239 end = vm_map_max(map); \
240 if (start > end) \
241 start = end; \
242 } while (/*CONSTCOND*/ 0)
243
244 /*
245 * local prototypes
246 */
247
248 static struct vm_map_entry *
249 uvm_mapent_alloc(struct vm_map *, int);
250 static struct vm_map_entry *
251 uvm_mapent_alloc_split(struct vm_map *,
252 const struct vm_map_entry *, int,
253 struct uvm_mapent_reservation *);
254 static void uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *);
255 static void uvm_mapent_free(struct vm_map_entry *);
256 static struct vm_map_entry *
257 uvm_kmapent_alloc(struct vm_map *, int);
258 static void uvm_kmapent_free(struct vm_map_entry *);
259 static void uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *);
260 static void uvm_map_reference_amap(struct vm_map_entry *, int);
261 static int uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int,
262 struct vm_map_entry *);
263 static void uvm_map_unreference_amap(struct vm_map_entry *, int);
264
265 int _uvm_tree_sanity(struct vm_map *, const char *);
266 static vsize_t uvm_rb_subtree_space(const struct vm_map_entry *);
267
268 static __inline int
269 uvm_compare(const struct vm_map_entry *a, const struct vm_map_entry *b)
270 {
271
272 if (a->start < b->start)
273 return (-1);
274 else if (a->start > b->start)
275 return (1);
276
277 return (0);
278 }
279
280 static __inline void
281 uvm_rb_augment(struct vm_map_entry *entry)
282 {
283
284 entry->space = uvm_rb_subtree_space(entry);
285 }
286
287 RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
288
289 RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
290
291 static __inline vsize_t
292 uvm_rb_space(const struct vm_map *map, const struct vm_map_entry *entry)
293 {
294 /* XXX map is not used */
295
296 KASSERT(entry->next != NULL);
297 return entry->next->start - entry->end;
298 }
299
300 static vsize_t
301 uvm_rb_subtree_space(const struct vm_map_entry *entry)
302 {
303 vaddr_t space, tmp;
304
305 space = entry->ownspace;
306 if (RB_LEFT(entry, rb_entry)) {
307 tmp = RB_LEFT(entry, rb_entry)->space;
308 if (tmp > space)
309 space = tmp;
310 }
311
312 if (RB_RIGHT(entry, rb_entry)) {
313 tmp = RB_RIGHT(entry, rb_entry)->space;
314 if (tmp > space)
315 space = tmp;
316 }
317
318 return (space);
319 }
320
321 static __inline void
322 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry)
323 {
324 /* We need to traverse to the very top */
325 do {
326 entry->ownspace = uvm_rb_space(map, entry);
327 entry->space = uvm_rb_subtree_space(entry);
328 } while ((entry = RB_PARENT(entry, rb_entry)) != NULL);
329 }
330
331 static __inline void
332 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry)
333 {
334 vaddr_t space = uvm_rb_space(map, entry);
335 struct vm_map_entry *tmp;
336
337 entry->ownspace = entry->space = space;
338 tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry);
339 #ifdef DIAGNOSTIC
340 if (tmp != NULL)
341 panic("uvm_rb_insert: duplicate entry?");
342 #endif
343 uvm_rb_fixup(map, entry);
344 if (entry->prev != &map->header)
345 uvm_rb_fixup(map, entry->prev);
346 }
347
348 static __inline void
349 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry)
350 {
351 struct vm_map_entry *parent;
352
353 parent = RB_PARENT(entry, rb_entry);
354 RB_REMOVE(uvm_tree, &(map)->rbhead, entry);
355 if (entry->prev != &map->header)
356 uvm_rb_fixup(map, entry->prev);
357 if (parent)
358 uvm_rb_fixup(map, parent);
359 }
360
361 #ifdef DEBUG
362 int uvm_debug_check_rbtree = 0;
363 #define uvm_tree_sanity(x,y) \
364 if (uvm_debug_check_rbtree) \
365 _uvm_tree_sanity(x,y)
366 #else
367 #define uvm_tree_sanity(x,y)
368 #endif
369
370 int
371 _uvm_tree_sanity(struct vm_map *map, const char *name)
372 {
373 struct vm_map_entry *tmp, *trtmp;
374 int n = 0, i = 1;
375
376 RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
377 if (tmp->ownspace != uvm_rb_space(map, tmp)) {
378 printf("%s: %d/%d ownspace %lx != %lx %s\n",
379 name, n + 1, map->nentries,
380 (ulong)tmp->ownspace, (ulong)uvm_rb_space(map, tmp),
381 tmp->next == &map->header ? "(last)" : "");
382 goto error;
383 }
384 }
385 trtmp = NULL;
386 RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
387 if (tmp->space != uvm_rb_subtree_space(tmp)) {
388 printf("%s: space %lx != %lx\n",
389 name, (ulong)tmp->space,
390 (ulong)uvm_rb_subtree_space(tmp));
391 goto error;
392 }
393 if (trtmp != NULL && trtmp->start >= tmp->start) {
394 printf("%s: corrupt: 0x%lx >= 0x%lx\n",
395 name, trtmp->start, tmp->start);
396 goto error;
397 }
398 n++;
399
400 trtmp = tmp;
401 }
402
403 if (n != map->nentries) {
404 printf("%s: nentries: %d vs %d\n",
405 name, n, map->nentries);
406 goto error;
407 }
408
409 for (tmp = map->header.next; tmp && tmp != &map->header;
410 tmp = tmp->next, i++) {
411 trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp);
412 if (trtmp != tmp) {
413 printf("%s: lookup: %d: %p - %p: %p\n",
414 name, i, tmp, trtmp,
415 RB_PARENT(tmp, rb_entry));
416 goto error;
417 }
418 }
419
420 return (0);
421 error:
422 #ifdef DDB
423 /* handy breakpoint location for error case */
424 __asm(".globl treesanity_label\ntreesanity_label:");
425 #endif
426 return (-1);
427 }
428
429 /*
430 * local inlines
431 */
432
433 static __inline struct vm_map *uvm_kmapent_map(struct vm_map_entry *);
434
435 /*
436 * uvm_mapent_alloc: allocate a map entry
437 */
438
439 static __inline struct vm_map_entry *
440 uvm_mapent_alloc(struct vm_map *map, int flags)
441 {
442 struct vm_map_entry *me;
443 int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
444 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
445
446 if (VM_MAP_USE_KMAPENT(map)) {
447 me = uvm_kmapent_alloc(map, flags);
448 } else {
449 me = pool_get(&uvm_map_entry_pool, pflags);
450 if (__predict_false(me == NULL))
451 return NULL;
452 me->flags = 0;
453 }
454
455 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me,
456 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0);
457 return (me);
458 }
459
460 /*
461 * uvm_mapent_alloc_split: allocate a map entry for clipping.
462 */
463
464 static __inline struct vm_map_entry *
465 uvm_mapent_alloc_split(struct vm_map *map,
466 const struct vm_map_entry *old_entry, int flags,
467 struct uvm_mapent_reservation *umr)
468 {
469 struct vm_map_entry *me;
470
471 KASSERT(!VM_MAP_USE_KMAPENT(map) ||
472 (old_entry->flags & UVM_MAP_QUANTUM) || !UMR_EMPTY(umr));
473
474 if (old_entry->flags & UVM_MAP_QUANTUM) {
475 int s;
476 struct vm_map_kernel *vmk = vm_map_to_kernel(map);
477
478 s = splvm();
479 simple_lock(&uvm.kentry_lock);
480 me = vmk->vmk_merged_entries;
481 KASSERT(me);
482 vmk->vmk_merged_entries = me->next;
483 simple_unlock(&uvm.kentry_lock);
484 splx(s);
485 KASSERT(me->flags & UVM_MAP_QUANTUM);
486 } else {
487 me = uvm_mapent_alloc(map, flags);
488 }
489
490 return me;
491 }
492
493 /*
494 * uvm_mapent_free: free map entry
495 */
496
497 static __inline void
498 uvm_mapent_free(struct vm_map_entry *me)
499 {
500 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
501
502 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
503 me, me->flags, 0, 0);
504 if (me->flags & UVM_MAP_KERNEL) {
505 uvm_kmapent_free(me);
506 } else {
507 pool_put(&uvm_map_entry_pool, me);
508 }
509 }
510
511 /*
512 * uvm_mapent_free_merge: free merged map entry
513 *
514 * => keep the entry if needed.
515 * => caller shouldn't hold map locked.
516 */
517
518 static __inline void
519 uvm_mapent_free_merged(struct vm_map *map, struct vm_map_entry *me)
520 {
521
522 KASSERT(!(me->flags & UVM_MAP_KERNEL) || uvm_kmapent_map(me) == map);
523
524 if (me->flags & UVM_MAP_QUANTUM) {
525 /*
526 * keep this entry for later splitting.
527 */
528 struct vm_map_kernel *vmk;
529 int s;
530
531 KASSERT(VM_MAP_IS_KERNEL(map));
532 KASSERT(!VM_MAP_USE_KMAPENT(map) ||
533 (me->flags & UVM_MAP_KERNEL));
534
535 vmk = vm_map_to_kernel(map);
536 s = splvm();
537 simple_lock(&uvm.kentry_lock);
538 me->next = vmk->vmk_merged_entries;
539 vmk->vmk_merged_entries = me;
540 simple_unlock(&uvm.kentry_lock);
541 splx(s);
542 } else {
543 uvm_mapent_free(me);
544 }
545 }
546
547 /*
548 * uvm_mapent_copy: copy a map entry, preserving flags
549 */
550
551 static __inline void
552 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst)
553 {
554
555 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
556 ((char *)src));
557 }
558
559 /*
560 * uvm_map_entry_unwire: unwire a map entry
561 *
562 * => map should be locked by caller
563 */
564
565 static __inline void
566 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry)
567 {
568
569 entry->wired_count = 0;
570 uvm_fault_unwire_locked(map, entry->start, entry->end);
571 }
572
573
574 /*
575 * wrapper for calling amap_ref()
576 */
577 static __inline void
578 uvm_map_reference_amap(struct vm_map_entry *entry, int flags)
579 {
580
581 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
582 (entry->end - entry->start) >> PAGE_SHIFT, flags);
583 }
584
585
586 /*
587 * wrapper for calling amap_unref()
588 */
589 static __inline void
590 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags)
591 {
592
593 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
594 (entry->end - entry->start) >> PAGE_SHIFT, flags);
595 }
596
597
598 /*
599 * uvm_map_init: init mapping system at boot time. note that we allocate
600 * and init the static pool of struct vm_map_entry *'s for the kernel here.
601 */
602
603 void
604 uvm_map_init(void)
605 {
606 #if defined(UVMHIST)
607 static struct uvm_history_ent maphistbuf[100];
608 static struct uvm_history_ent pdhistbuf[100];
609 #endif
610
611 /*
612 * first, init logging system.
613 */
614
615 UVMHIST_FUNC("uvm_map_init");
616 UVMHIST_INIT_STATIC(maphist, maphistbuf);
617 UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
618 UVMHIST_CALLED(maphist);
619 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
620
621 /*
622 * initialize the global lock for kernel map entry.
623 *
624 * XXX is it worth to have per-map lock instead?
625 */
626
627 simple_lock_init(&uvm.kentry_lock);
628 }
629
630 /*
631 * clippers
632 */
633
634 /*
635 * uvm_map_clip_start: ensure that the entry begins at or after
636 * the starting address, if it doesn't we split the entry.
637 *
638 * => caller should use UVM_MAP_CLIP_START macro rather than calling
639 * this directly
640 * => map must be locked by caller
641 */
642
643 void
644 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry,
645 vaddr_t start, struct uvm_mapent_reservation *umr)
646 {
647 struct vm_map_entry *new_entry;
648 vaddr_t new_adj;
649
650 /* uvm_map_simplify_entry(map, entry); */ /* XXX */
651
652 uvm_tree_sanity(map, "clip_start entry");
653
654 /*
655 * Split off the front portion. note that we must insert the new
656 * entry BEFORE this one, so that this entry has the specified
657 * starting address.
658 */
659 new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
660 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
661
662 new_entry->end = start;
663 new_adj = start - new_entry->start;
664 if (entry->object.uvm_obj)
665 entry->offset += new_adj; /* shift start over */
666
667 /* Does not change order for the RB tree */
668 entry->start = start;
669
670 if (new_entry->aref.ar_amap) {
671 amap_splitref(&new_entry->aref, &entry->aref, new_adj);
672 }
673
674 uvm_map_entry_link(map, entry->prev, new_entry);
675
676 if (UVM_ET_ISSUBMAP(entry)) {
677 /* ... unlikely to happen, but play it safe */
678 uvm_map_reference(new_entry->object.sub_map);
679 } else {
680 if (UVM_ET_ISOBJ(entry) &&
681 entry->object.uvm_obj->pgops &&
682 entry->object.uvm_obj->pgops->pgo_reference)
683 entry->object.uvm_obj->pgops->pgo_reference(
684 entry->object.uvm_obj);
685 }
686
687 uvm_tree_sanity(map, "clip_start leave");
688 }
689
690 /*
691 * uvm_map_clip_end: ensure that the entry ends at or before
692 * the ending address, if it does't we split the reference
693 *
694 * => caller should use UVM_MAP_CLIP_END macro rather than calling
695 * this directly
696 * => map must be locked by caller
697 */
698
699 void
700 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end,
701 struct uvm_mapent_reservation *umr)
702 {
703 struct vm_map_entry * new_entry;
704 vaddr_t new_adj; /* #bytes we move start forward */
705
706 uvm_tree_sanity(map, "clip_end entry");
707
708 /*
709 * Create a new entry and insert it
710 * AFTER the specified entry
711 */
712 new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
713 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
714
715 new_entry->start = entry->end = end;
716 new_adj = end - entry->start;
717 if (new_entry->object.uvm_obj)
718 new_entry->offset += new_adj;
719
720 if (entry->aref.ar_amap)
721 amap_splitref(&entry->aref, &new_entry->aref, new_adj);
722
723 uvm_rb_fixup(map, entry);
724
725 uvm_map_entry_link(map, entry, new_entry);
726
727 if (UVM_ET_ISSUBMAP(entry)) {
728 /* ... unlikely to happen, but play it safe */
729 uvm_map_reference(new_entry->object.sub_map);
730 } else {
731 if (UVM_ET_ISOBJ(entry) &&
732 entry->object.uvm_obj->pgops &&
733 entry->object.uvm_obj->pgops->pgo_reference)
734 entry->object.uvm_obj->pgops->pgo_reference(
735 entry->object.uvm_obj);
736 }
737
738 uvm_tree_sanity(map, "clip_end leave");
739 }
740
741
742 /*
743 * M A P - m a i n e n t r y p o i n t
744 */
745 /*
746 * uvm_map: establish a valid mapping in a map
747 *
748 * => assume startp is page aligned.
749 * => assume size is a multiple of PAGE_SIZE.
750 * => assume sys_mmap provides enough of a "hint" to have us skip
751 * over text/data/bss area.
752 * => map must be unlocked (we will lock it)
753 * => <uobj,uoffset> value meanings (4 cases):
754 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER
755 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER
756 * [3] <uobj,uoffset> == normal mapping
757 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA
758 *
759 * case [4] is for kernel mappings where we don't know the offset until
760 * we've found a virtual address. note that kernel object offsets are
761 * always relative to vm_map_min(kernel_map).
762 *
763 * => if `align' is non-zero, we align the virtual address to the specified
764 * alignment.
765 * this is provided as a mechanism for large pages.
766 *
767 * => XXXCDC: need way to map in external amap?
768 */
769
770 int
771 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size,
772 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags)
773 {
774 struct uvm_map_args args;
775 struct vm_map_entry *new_entry;
776 int error;
777
778 KASSERT((flags & UVM_FLAG_QUANTUM) == 0 || VM_MAP_IS_KERNEL(map));
779
780 /*
781 * for pager_map, allocate the new entry first to avoid sleeping
782 * for memory while we have the map locked.
783 *
784 * besides, because we allocates entries for in-kernel maps
785 * a bit differently (cf. uvm_kmapent_alloc/free), we need to
786 * allocate them before locking the map.
787 */
788
789 new_entry = NULL;
790 if (VM_MAP_USE_KMAPENT(map) || (flags & UVM_FLAG_QUANTUM) ||
791 map == pager_map) {
792 new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
793 if (__predict_false(new_entry == NULL))
794 return ENOMEM;
795 if (flags & UVM_FLAG_QUANTUM)
796 new_entry->flags |= UVM_MAP_QUANTUM;
797 }
798 if (map == pager_map)
799 flags |= UVM_FLAG_NOMERGE;
800
801 error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align,
802 flags, &args);
803 if (!error) {
804 error = uvm_map_enter(map, &args, new_entry);
805 *startp = args.uma_start;
806 } else if (new_entry) {
807 uvm_mapent_free(new_entry);
808 }
809
810 return error;
811 }
812
813 int
814 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size,
815 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags,
816 struct uvm_map_args *args)
817 {
818 struct vm_map_entry *prev_entry;
819 vm_prot_t prot = UVM_PROTECTION(flags);
820 vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
821
822 UVMHIST_FUNC("uvm_map_prepare");
823 UVMHIST_CALLED(maphist);
824
825 UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
826 map, start, size, flags);
827 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0);
828
829 /*
830 * detect a popular device driver bug.
831 */
832
833 KASSERT(doing_shutdown || curlwp != NULL ||
834 (map->flags & VM_MAP_INTRSAFE));
835
836 /*
837 * zero-sized mapping doesn't make any sense.
838 */
839 KASSERT(size > 0);
840
841 KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0);
842
843 uvm_tree_sanity(map, "map entry");
844
845 /*
846 * check sanity of protection code
847 */
848
849 if ((prot & maxprot) != prot) {
850 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x",
851 prot, maxprot,0,0);
852 return EACCES;
853 }
854
855 /*
856 * figure out where to put new VM range
857 */
858
859 retry:
860 if (vm_map_lock_try(map) == FALSE) {
861 if (flags & UVM_FLAG_TRYLOCK) {
862 return EAGAIN;
863 }
864 vm_map_lock(map); /* could sleep here */
865 }
866 if ((prev_entry = uvm_map_findspace(map, start, size, &start,
867 uobj, uoffset, align, flags)) == NULL) {
868 unsigned int timestamp;
869
870 if ((flags & UVM_FLAG_WAITVA) == 0) {
871 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",
872 0,0,0,0);
873 vm_map_unlock(map);
874 return ENOMEM;
875 }
876 timestamp = map->timestamp;
877 UVMHIST_LOG(maphist,"waiting va timestamp=0x%x",
878 timestamp,0,0,0);
879 simple_lock(&map->flags_lock);
880 map->flags |= VM_MAP_WANTVA;
881 simple_unlock(&map->flags_lock);
882 vm_map_unlock(map);
883
884 /*
885 * wait until someone does unmap.
886 * XXX fragile locking
887 */
888
889 simple_lock(&map->flags_lock);
890 while ((map->flags & VM_MAP_WANTVA) != 0 &&
891 map->timestamp == timestamp) {
892 ltsleep(&map->header, PVM, "vmmapva", 0,
893 &map->flags_lock);
894 }
895 simple_unlock(&map->flags_lock);
896 goto retry;
897 }
898
899 #ifdef PMAP_GROWKERNEL
900 /*
901 * If the kernel pmap can't map the requested space,
902 * then allocate more resources for it.
903 */
904 if (map == kernel_map && uvm_maxkaddr < (start + size))
905 uvm_maxkaddr = pmap_growkernel(start + size);
906 #endif
907
908 UVMCNT_INCR(uvm_map_call);
909
910 /*
911 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
912 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in
913 * either case we want to zero it before storing it in the map entry
914 * (because it looks strange and confusing when debugging...)
915 *
916 * if uobj is not null
917 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
918 * and we do not need to change uoffset.
919 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
920 * now (based on the starting address of the map). this case is
921 * for kernel object mappings where we don't know the offset until
922 * the virtual address is found (with uvm_map_findspace). the
923 * offset is the distance we are from the start of the map.
924 */
925
926 if (uobj == NULL) {
927 uoffset = 0;
928 } else {
929 if (uoffset == UVM_UNKNOWN_OFFSET) {
930 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
931 uoffset = start - vm_map_min(kernel_map);
932 }
933 }
934
935 args->uma_flags = flags;
936 args->uma_prev = prev_entry;
937 args->uma_start = start;
938 args->uma_size = size;
939 args->uma_uobj = uobj;
940 args->uma_uoffset = uoffset;
941
942 return 0;
943 }
944
945 int
946 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args,
947 struct vm_map_entry *new_entry)
948 {
949 struct vm_map_entry *prev_entry = args->uma_prev;
950 struct vm_map_entry *dead = NULL;
951
952 const uvm_flag_t flags = args->uma_flags;
953 const vm_prot_t prot = UVM_PROTECTION(flags);
954 const vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
955 const vm_inherit_t inherit = UVM_INHERIT(flags);
956 const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
957 AMAP_EXTEND_NOWAIT : 0;
958 const int advice = UVM_ADVICE(flags);
959 const int meflagmask = UVM_MAP_NOMERGE | UVM_MAP_QUANTUM;
960 const int meflagval = (flags & UVM_FLAG_QUANTUM) ?
961 UVM_MAP_QUANTUM : 0;
962
963 vaddr_t start = args->uma_start;
964 vsize_t size = args->uma_size;
965 struct uvm_object *uobj = args->uma_uobj;
966 voff_t uoffset = args->uma_uoffset;
967
968 const int kmap = (vm_map_pmap(map) == pmap_kernel());
969 int merged = 0;
970 int error;
971 int newetype;
972
973 UVMHIST_FUNC("uvm_map_enter");
974 UVMHIST_CALLED(maphist);
975
976 UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
977 map, start, size, flags);
978 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0);
979
980 if (flags & UVM_FLAG_QUANTUM) {
981 KASSERT(new_entry);
982 KASSERT(new_entry->flags & UVM_MAP_QUANTUM);
983 }
984
985 if (uobj)
986 newetype = UVM_ET_OBJ;
987 else
988 newetype = 0;
989
990 if (flags & UVM_FLAG_COPYONW) {
991 newetype |= UVM_ET_COPYONWRITE;
992 if ((flags & UVM_FLAG_OVERLAY) == 0)
993 newetype |= UVM_ET_NEEDSCOPY;
994 }
995
996 /*
997 * try and insert in map by extending previous entry, if possible.
998 * XXX: we don't try and pull back the next entry. might be useful
999 * for a stack, but we are currently allocating our stack in advance.
1000 */
1001
1002 if (flags & UVM_FLAG_NOMERGE)
1003 goto nomerge;
1004
1005 if (prev_entry->etype == newetype &&
1006 prev_entry->end == start &&
1007 prev_entry != &map->header &&
1008 prev_entry->object.uvm_obj == uobj) {
1009
1010 if ((prev_entry->flags & meflagmask) != meflagval)
1011 goto forwardmerge;
1012
1013 if (uobj && prev_entry->offset +
1014 (prev_entry->end - prev_entry->start) != uoffset)
1015 goto forwardmerge;
1016
1017 if (prev_entry->protection != prot ||
1018 prev_entry->max_protection != maxprot)
1019 goto forwardmerge;
1020
1021 if (prev_entry->inheritance != inherit ||
1022 prev_entry->advice != advice)
1023 goto forwardmerge;
1024
1025 /* wiring status must match (new area is unwired) */
1026 if (VM_MAPENT_ISWIRED(prev_entry))
1027 goto forwardmerge;
1028
1029 /*
1030 * can't extend a shared amap. note: no need to lock amap to
1031 * look at refs since we don't care about its exact value.
1032 * if it is one (i.e. we have only reference) it will stay there
1033 */
1034
1035 if (prev_entry->aref.ar_amap &&
1036 amap_refs(prev_entry->aref.ar_amap) != 1) {
1037 goto forwardmerge;
1038 }
1039
1040 if (prev_entry->aref.ar_amap) {
1041 error = amap_extend(prev_entry, size,
1042 amapwaitflag | AMAP_EXTEND_FORWARDS);
1043 if (error)
1044 goto done;
1045 }
1046
1047 if (kmap)
1048 UVMCNT_INCR(map_kbackmerge);
1049 else
1050 UVMCNT_INCR(map_ubackmerge);
1051 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0);
1052
1053 /*
1054 * drop our reference to uobj since we are extending a reference
1055 * that we already have (the ref count can not drop to zero).
1056 */
1057
1058 if (uobj && uobj->pgops->pgo_detach)
1059 uobj->pgops->pgo_detach(uobj);
1060
1061 prev_entry->end += size;
1062 uvm_rb_fixup(map, prev_entry);
1063
1064 uvm_tree_sanity(map, "map backmerged");
1065
1066 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
1067 merged++;
1068 }
1069
1070 forwardmerge:
1071 if (prev_entry->next->etype == newetype &&
1072 prev_entry->next->start == (start + size) &&
1073 prev_entry->next != &map->header &&
1074 prev_entry->next->object.uvm_obj == uobj) {
1075
1076 if ((prev_entry->next->flags & meflagmask) != meflagval)
1077 goto nomerge;
1078
1079 if (uobj && prev_entry->next->offset != uoffset + size)
1080 goto nomerge;
1081
1082 if (prev_entry->next->protection != prot ||
1083 prev_entry->next->max_protection != maxprot)
1084 goto nomerge;
1085
1086 if (prev_entry->next->inheritance != inherit ||
1087 prev_entry->next->advice != advice)
1088 goto nomerge;
1089
1090 /* wiring status must match (new area is unwired) */
1091 if (VM_MAPENT_ISWIRED(prev_entry->next))
1092 goto nomerge;
1093
1094 /*
1095 * can't extend a shared amap. note: no need to lock amap to
1096 * look at refs since we don't care about its exact value.
1097 * if it is one (i.e. we have only reference) it will stay there.
1098 *
1099 * note that we also can't merge two amaps, so if we
1100 * merged with the previous entry which has an amap,
1101 * and the next entry also has an amap, we give up.
1102 *
1103 * Interesting cases:
1104 * amap, new, amap -> give up second merge (single fwd extend)
1105 * amap, new, none -> double forward extend (extend again here)
1106 * none, new, amap -> double backward extend (done here)
1107 * uobj, new, amap -> single backward extend (done here)
1108 *
1109 * XXX should we attempt to deal with someone refilling
1110 * the deallocated region between two entries that are
1111 * backed by the same amap (ie, arefs is 2, "prev" and
1112 * "next" refer to it, and adding this allocation will
1113 * close the hole, thus restoring arefs to 1 and
1114 * deallocating the "next" vm_map_entry)? -- @@@
1115 */
1116
1117 if (prev_entry->next->aref.ar_amap &&
1118 (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
1119 (merged && prev_entry->aref.ar_amap))) {
1120 goto nomerge;
1121 }
1122
1123 if (merged) {
1124 /*
1125 * Try to extend the amap of the previous entry to
1126 * cover the next entry as well. If it doesn't work
1127 * just skip on, don't actually give up, since we've
1128 * already completed the back merge.
1129 */
1130 if (prev_entry->aref.ar_amap) {
1131 if (amap_extend(prev_entry,
1132 prev_entry->next->end -
1133 prev_entry->next->start,
1134 amapwaitflag | AMAP_EXTEND_FORWARDS))
1135 goto nomerge;
1136 }
1137
1138 /*
1139 * Try to extend the amap of the *next* entry
1140 * back to cover the new allocation *and* the
1141 * previous entry as well (the previous merge
1142 * didn't have an amap already otherwise we
1143 * wouldn't be checking here for an amap). If
1144 * it doesn't work just skip on, again, don't
1145 * actually give up, since we've already
1146 * completed the back merge.
1147 */
1148 else if (prev_entry->next->aref.ar_amap) {
1149 if (amap_extend(prev_entry->next,
1150 prev_entry->end -
1151 prev_entry->start,
1152 amapwaitflag | AMAP_EXTEND_BACKWARDS))
1153 goto nomerge;
1154 }
1155 } else {
1156 /*
1157 * Pull the next entry's amap backwards to cover this
1158 * new allocation.
1159 */
1160 if (prev_entry->next->aref.ar_amap) {
1161 error = amap_extend(prev_entry->next, size,
1162 amapwaitflag | AMAP_EXTEND_BACKWARDS);
1163 if (error)
1164 goto done;
1165 }
1166 }
1167
1168 if (merged) {
1169 if (kmap) {
1170 UVMCNT_DECR(map_kbackmerge);
1171 UVMCNT_INCR(map_kbimerge);
1172 } else {
1173 UVMCNT_DECR(map_ubackmerge);
1174 UVMCNT_INCR(map_ubimerge);
1175 }
1176 } else {
1177 if (kmap)
1178 UVMCNT_INCR(map_kforwmerge);
1179 else
1180 UVMCNT_INCR(map_uforwmerge);
1181 }
1182 UVMHIST_LOG(maphist," starting forward merge", 0, 0, 0, 0);
1183
1184 /*
1185 * drop our reference to uobj since we are extending a reference
1186 * that we already have (the ref count can not drop to zero).
1187 * (if merged, we've already detached)
1188 */
1189 if (uobj && uobj->pgops->pgo_detach && !merged)
1190 uobj->pgops->pgo_detach(uobj);
1191
1192 if (merged) {
1193 dead = prev_entry->next;
1194 prev_entry->end = dead->end;
1195 uvm_map_entry_unlink(map, dead);
1196 if (dead->aref.ar_amap != NULL) {
1197 prev_entry->aref = dead->aref;
1198 dead->aref.ar_amap = NULL;
1199 }
1200 } else {
1201 prev_entry->next->start -= size;
1202 if (prev_entry != &map->header)
1203 uvm_rb_fixup(map, prev_entry);
1204 if (uobj)
1205 prev_entry->next->offset = uoffset;
1206 }
1207
1208 uvm_tree_sanity(map, "map forwardmerged");
1209
1210 UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
1211 merged++;
1212 }
1213
1214 nomerge:
1215 if (!merged) {
1216 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0);
1217 if (kmap)
1218 UVMCNT_INCR(map_knomerge);
1219 else
1220 UVMCNT_INCR(map_unomerge);
1221
1222 /*
1223 * allocate new entry and link it in.
1224 */
1225
1226 if (new_entry == NULL) {
1227 new_entry = uvm_mapent_alloc(map,
1228 (flags & UVM_FLAG_NOWAIT));
1229 if (__predict_false(new_entry == NULL)) {
1230 error = ENOMEM;
1231 goto done;
1232 }
1233 }
1234 new_entry->start = start;
1235 new_entry->end = new_entry->start + size;
1236 new_entry->object.uvm_obj = uobj;
1237 new_entry->offset = uoffset;
1238
1239 new_entry->etype = newetype;
1240
1241 if (flags & UVM_FLAG_NOMERGE) {
1242 new_entry->flags |= UVM_MAP_NOMERGE;
1243 }
1244
1245 new_entry->protection = prot;
1246 new_entry->max_protection = maxprot;
1247 new_entry->inheritance = inherit;
1248 new_entry->wired_count = 0;
1249 new_entry->advice = advice;
1250 if (flags & UVM_FLAG_OVERLAY) {
1251
1252 /*
1253 * to_add: for BSS we overallocate a little since we
1254 * are likely to extend
1255 */
1256
1257 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
1258 UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
1259 struct vm_amap *amap = amap_alloc(size, to_add,
1260 (flags & UVM_FLAG_NOWAIT) ? M_NOWAIT : M_WAITOK);
1261 if (__predict_false(amap == NULL)) {
1262 error = ENOMEM;
1263 goto done;
1264 }
1265 new_entry->aref.ar_pageoff = 0;
1266 new_entry->aref.ar_amap = amap;
1267 } else {
1268 new_entry->aref.ar_pageoff = 0;
1269 new_entry->aref.ar_amap = NULL;
1270 }
1271 uvm_map_entry_link(map, prev_entry, new_entry);
1272
1273 /*
1274 * Update the free space hint
1275 */
1276
1277 if ((map->first_free == prev_entry) &&
1278 (prev_entry->end >= new_entry->start))
1279 map->first_free = new_entry;
1280
1281 new_entry = NULL;
1282 }
1283
1284 map->size += size;
1285
1286 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1287
1288 error = 0;
1289 done:
1290 vm_map_unlock(map);
1291 if (new_entry) {
1292 if (error == 0) {
1293 KDASSERT(merged);
1294 uvm_mapent_free_merged(map, new_entry);
1295 } else {
1296 uvm_mapent_free(new_entry);
1297 }
1298 }
1299 if (dead) {
1300 KDASSERT(merged);
1301 uvm_mapent_free_merged(map, dead);
1302 }
1303 return error;
1304 }
1305
1306 /*
1307 * uvm_map_lookup_entry: find map entry at or before an address
1308 *
1309 * => map must at least be read-locked by caller
1310 * => entry is returned in "entry"
1311 * => return value is true if address is in the returned entry
1312 */
1313
1314 boolean_t
1315 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address,
1316 struct vm_map_entry **entry /* OUT */)
1317 {
1318 struct vm_map_entry *cur;
1319 boolean_t use_tree = FALSE;
1320 UVMHIST_FUNC("uvm_map_lookup_entry");
1321 UVMHIST_CALLED(maphist);
1322
1323 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
1324 map, address, entry, 0);
1325
1326 /*
1327 * start looking either from the head of the
1328 * list, or from the hint.
1329 */
1330
1331 simple_lock(&map->hint_lock);
1332 cur = map->hint;
1333 simple_unlock(&map->hint_lock);
1334
1335 if (cur == &map->header)
1336 cur = cur->next;
1337
1338 UVMCNT_INCR(uvm_mlk_call);
1339 if (address >= cur->start) {
1340
1341 /*
1342 * go from hint to end of list.
1343 *
1344 * but first, make a quick check to see if
1345 * we are already looking at the entry we
1346 * want (which is usually the case).
1347 * note also that we don't need to save the hint
1348 * here... it is the same hint (unless we are
1349 * at the header, in which case the hint didn't
1350 * buy us anything anyway).
1351 */
1352
1353 if (cur != &map->header && cur->end > address) {
1354 UVMCNT_INCR(uvm_mlk_hint);
1355 *entry = cur;
1356 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
1357 cur, 0, 0, 0);
1358 return (TRUE);
1359 }
1360
1361 if (map->nentries > 30)
1362 use_tree = TRUE;
1363 } else {
1364
1365 /*
1366 * invalid hint. use tree.
1367 */
1368 use_tree = TRUE;
1369 }
1370
1371 uvm_tree_sanity(map, __func__);
1372
1373 if (use_tree) {
1374 struct vm_map_entry *prev = &map->header;
1375 cur = RB_ROOT(&map->rbhead);
1376
1377 /*
1378 * Simple lookup in the tree. Happens when the hint is
1379 * invalid, or nentries reach a threshold.
1380 */
1381 while (cur) {
1382 if (address >= cur->start) {
1383 if (address < cur->end) {
1384 *entry = cur;
1385 goto got;
1386 }
1387 prev = cur;
1388 cur = RB_RIGHT(cur, rb_entry);
1389 } else
1390 cur = RB_LEFT(cur, rb_entry);
1391 }
1392 *entry = prev;
1393 goto failed;
1394 }
1395
1396 /*
1397 * search linearly
1398 */
1399
1400 while (cur != &map->header) {
1401 if (cur->end > address) {
1402 if (address >= cur->start) {
1403 /*
1404 * save this lookup for future
1405 * hints, and return
1406 */
1407
1408 *entry = cur;
1409 got:
1410 SAVE_HINT(map, map->hint, *entry);
1411 UVMHIST_LOG(maphist,"<- search got it (0x%x)",
1412 cur, 0, 0, 0);
1413 KDASSERT((*entry)->start <= address);
1414 KDASSERT(address < (*entry)->end);
1415 return (TRUE);
1416 }
1417 break;
1418 }
1419 cur = cur->next;
1420 }
1421 *entry = cur->prev;
1422 failed:
1423 SAVE_HINT(map, map->hint, *entry);
1424 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
1425 KDASSERT((*entry) == &map->header || (*entry)->end <= address);
1426 KDASSERT((*entry)->next == &map->header ||
1427 address < (*entry)->next->start);
1428 return (FALSE);
1429 }
1430
1431 /*
1432 * See if the range between start and start + length fits in the gap
1433 * entry->next->start and entry->end. Returns 1 if fits, 0 if doesn't
1434 * fit, and -1 address wraps around.
1435 */
1436 static __inline int
1437 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset,
1438 vsize_t align, int topdown, struct vm_map_entry *entry)
1439 {
1440 vaddr_t end;
1441
1442 #ifdef PMAP_PREFER
1443 /*
1444 * push start address forward as needed to avoid VAC alias problems.
1445 * we only do this if a valid offset is specified.
1446 */
1447
1448 if (uoffset != UVM_UNKNOWN_OFFSET)
1449 PMAP_PREFER(uoffset, start, length, topdown);
1450 #endif
1451 if (align != 0) {
1452 if ((*start & (align - 1)) != 0) {
1453 if (topdown)
1454 *start &= ~(align - 1);
1455 else
1456 *start = roundup(*start, align);
1457 }
1458 /*
1459 * XXX Should we PMAP_PREFER() here again?
1460 * eh...i think we're okay
1461 */
1462 }
1463
1464 /*
1465 * Find the end of the proposed new region. Be sure we didn't
1466 * wrap around the address; if so, we lose. Otherwise, if the
1467 * proposed new region fits before the next entry, we win.
1468 */
1469
1470 end = *start + length;
1471 if (end < *start)
1472 return (-1);
1473
1474 if (entry->next->start >= end && *start >= entry->end)
1475 return (1);
1476
1477 return (0);
1478 }
1479
1480 /*
1481 * uvm_map_findspace: find "length" sized space in "map".
1482 *
1483 * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is
1484 * set in "flags" (in which case we insist on using "hint").
1485 * => "result" is VA returned
1486 * => uobj/uoffset are to be used to handle VAC alignment, if required
1487 * => if "align" is non-zero, we attempt to align to that value.
1488 * => caller must at least have read-locked map
1489 * => returns NULL on failure, or pointer to prev. map entry if success
1490 * => note this is a cross between the old vm_map_findspace and vm_map_find
1491 */
1492
1493 struct vm_map_entry *
1494 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length,
1495 vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset,
1496 vsize_t align, int flags)
1497 {
1498 struct vm_map_entry *entry;
1499 struct vm_map_entry *child, *prev, *tmp;
1500 vaddr_t orig_hint;
1501 const int topdown = map->flags & VM_MAP_TOPDOWN;
1502 UVMHIST_FUNC("uvm_map_findspace");
1503 UVMHIST_CALLED(maphist);
1504
1505 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
1506 map, hint, length, flags);
1507 KASSERT((align & (align - 1)) == 0);
1508 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1509
1510 uvm_tree_sanity(map, "map_findspace entry");
1511
1512 /*
1513 * remember the original hint. if we are aligning, then we
1514 * may have to try again with no alignment constraint if
1515 * we fail the first time.
1516 */
1517
1518 orig_hint = hint;
1519 if (hint < vm_map_min(map)) { /* check ranges ... */
1520 if (flags & UVM_FLAG_FIXED) {
1521 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1522 return (NULL);
1523 }
1524 hint = vm_map_min(map);
1525 }
1526 if (hint > vm_map_max(map)) {
1527 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
1528 hint, vm_map_min(map), vm_map_max(map), 0);
1529 return (NULL);
1530 }
1531
1532 /*
1533 * Look for the first possible address; if there's already
1534 * something at this address, we have to start after it.
1535 */
1536
1537 /*
1538 * @@@: there are four, no, eight cases to consider.
1539 *
1540 * 0: found, fixed, bottom up -> fail
1541 * 1: found, fixed, top down -> fail
1542 * 2: found, not fixed, bottom up -> start after entry->end,
1543 * loop up
1544 * 3: found, not fixed, top down -> start before entry->start,
1545 * loop down
1546 * 4: not found, fixed, bottom up -> check entry->next->start, fail
1547 * 5: not found, fixed, top down -> check entry->next->start, fail
1548 * 6: not found, not fixed, bottom up -> check entry->next->start,
1549 * loop up
1550 * 7: not found, not fixed, top down -> check entry->next->start,
1551 * loop down
1552 *
1553 * as you can see, it reduces to roughly five cases, and that
1554 * adding top down mapping only adds one unique case (without
1555 * it, there would be four cases).
1556 */
1557
1558 if ((flags & UVM_FLAG_FIXED) == 0 && hint == vm_map_min(map)) {
1559 entry = map->first_free;
1560 } else {
1561 if (uvm_map_lookup_entry(map, hint, &entry)) {
1562 /* "hint" address already in use ... */
1563 if (flags & UVM_FLAG_FIXED) {
1564 UVMHIST_LOG(maphist, "<- fixed & VA in use",
1565 0, 0, 0, 0);
1566 return (NULL);
1567 }
1568 if (topdown)
1569 /* Start from lower gap. */
1570 entry = entry->prev;
1571 } else if (flags & UVM_FLAG_FIXED) {
1572 if (entry->next->start >= hint + length &&
1573 hint + length > hint)
1574 goto found;
1575
1576 /* "hint" address is gap but too small */
1577 UVMHIST_LOG(maphist, "<- fixed mapping failed",
1578 0, 0, 0, 0);
1579 return (NULL); /* only one shot at it ... */
1580 } else {
1581 /*
1582 * See if given hint fits in this gap.
1583 */
1584 switch (uvm_map_space_avail(&hint, length,
1585 uoffset, align, topdown, entry)) {
1586 case 1:
1587 goto found;
1588 case -1:
1589 goto wraparound;
1590 }
1591
1592 if (topdown) {
1593 /*
1594 * Still there is a chance to fit
1595 * if hint > entry->end.
1596 */
1597 } else {
1598 /* Start from higher gap. */
1599 entry = entry->next;
1600 if (entry == &map->header)
1601 goto notfound;
1602 goto nextgap;
1603 }
1604 }
1605 }
1606
1607 /*
1608 * Note that all UVM_FLAGS_FIXED case is already handled.
1609 */
1610 KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1611
1612 /* Try to find the space in the red-black tree */
1613
1614 /* Check slot before any entry */
1615 hint = topdown ? entry->next->start - length : entry->end;
1616 switch (uvm_map_space_avail(&hint, length, uoffset, align,
1617 topdown, entry)) {
1618 case 1:
1619 goto found;
1620 case -1:
1621 goto wraparound;
1622 }
1623
1624 nextgap:
1625 KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1626 /* If there is not enough space in the whole tree, we fail */
1627 tmp = RB_ROOT(&map->rbhead);
1628 if (tmp == NULL || tmp->space < length)
1629 goto notfound;
1630
1631 prev = NULL; /* previous candidate */
1632
1633 /* Find an entry close to hint that has enough space */
1634 for (; tmp;) {
1635 KASSERT(tmp->next->start == tmp->end + tmp->ownspace);
1636 if (topdown) {
1637 if (tmp->next->start < hint + length &&
1638 (prev == NULL || tmp->end > prev->end)) {
1639 if (tmp->ownspace >= length)
1640 prev = tmp;
1641 else if ((child = RB_LEFT(tmp, rb_entry))
1642 != NULL && child->space >= length)
1643 prev = tmp;
1644 }
1645 } else {
1646 if (tmp->end >= hint &&
1647 (prev == NULL || tmp->end < prev->end)) {
1648 if (tmp->ownspace >= length)
1649 prev = tmp;
1650 else if ((child = RB_RIGHT(tmp, rb_entry))
1651 != NULL && child->space >= length)
1652 prev = tmp;
1653 }
1654 }
1655 if (tmp->next->start < hint + length)
1656 child = RB_RIGHT(tmp, rb_entry);
1657 else if (tmp->end > hint)
1658 child = RB_LEFT(tmp, rb_entry);
1659 else {
1660 if (tmp->ownspace >= length)
1661 break;
1662 if (topdown)
1663 child = RB_LEFT(tmp, rb_entry);
1664 else
1665 child = RB_RIGHT(tmp, rb_entry);
1666 }
1667 if (child == NULL || child->space < length)
1668 break;
1669 tmp = child;
1670 }
1671
1672 if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) {
1673 /*
1674 * Check if the entry that we found satifies the
1675 * space requirement
1676 */
1677 if (topdown) {
1678 if (hint > tmp->next->start - length)
1679 hint = tmp->next->start - length;
1680 } else {
1681 if (hint < tmp->end)
1682 hint = tmp->end;
1683 }
1684 switch (uvm_map_space_avail(&hint, length, uoffset, align,
1685 topdown, tmp)) {
1686 case 1:
1687 entry = tmp;
1688 goto found;
1689 case -1:
1690 goto wraparound;
1691 }
1692 if (tmp->ownspace >= length)
1693 goto listsearch;
1694 }
1695 if (prev == NULL)
1696 goto notfound;
1697
1698 if (topdown) {
1699 KASSERT(orig_hint >= prev->next->start - length ||
1700 prev->next->start - length > prev->next->start);
1701 hint = prev->next->start - length;
1702 } else {
1703 KASSERT(orig_hint <= prev->end);
1704 hint = prev->end;
1705 }
1706 switch (uvm_map_space_avail(&hint, length, uoffset, align,
1707 topdown, prev)) {
1708 case 1:
1709 entry = prev;
1710 goto found;
1711 case -1:
1712 goto wraparound;
1713 }
1714 if (prev->ownspace >= length)
1715 goto listsearch;
1716
1717 if (topdown)
1718 tmp = RB_LEFT(prev, rb_entry);
1719 else
1720 tmp = RB_RIGHT(prev, rb_entry);
1721 for (;;) {
1722 KASSERT(tmp && tmp->space >= length);
1723 if (topdown)
1724 child = RB_RIGHT(tmp, rb_entry);
1725 else
1726 child = RB_LEFT(tmp, rb_entry);
1727 if (child && child->space >= length) {
1728 tmp = child;
1729 continue;
1730 }
1731 if (tmp->ownspace >= length)
1732 break;
1733 if (topdown)
1734 tmp = RB_LEFT(tmp, rb_entry);
1735 else
1736 tmp = RB_RIGHT(tmp, rb_entry);
1737 }
1738
1739 if (topdown) {
1740 KASSERT(orig_hint >= tmp->next->start - length ||
1741 tmp->next->start - length > tmp->next->start);
1742 hint = tmp->next->start - length;
1743 } else {
1744 KASSERT(orig_hint <= tmp->end);
1745 hint = tmp->end;
1746 }
1747 switch (uvm_map_space_avail(&hint, length, uoffset, align,
1748 topdown, tmp)) {
1749 case 1:
1750 entry = tmp;
1751 goto found;
1752 case -1:
1753 goto wraparound;
1754 }
1755
1756 /*
1757 * The tree fails to find an entry because of offset or alignment
1758 * restrictions. Search the list instead.
1759 */
1760 listsearch:
1761 /*
1762 * Look through the rest of the map, trying to fit a new region in
1763 * the gap between existing regions, or after the very last region.
1764 * note: entry->end = base VA of current gap,
1765 * entry->next->start = VA of end of current gap
1766 */
1767
1768 for (;;) {
1769 /* Update hint for current gap. */
1770 hint = topdown ? entry->next->start - length : entry->end;
1771
1772 /* See if it fits. */
1773 switch (uvm_map_space_avail(&hint, length, uoffset, align,
1774 topdown, entry)) {
1775 case 1:
1776 goto found;
1777 case -1:
1778 goto wraparound;
1779 }
1780
1781 /* Advance to next/previous gap */
1782 if (topdown) {
1783 if (entry == &map->header) {
1784 UVMHIST_LOG(maphist, "<- failed (off start)",
1785 0,0,0,0);
1786 goto notfound;
1787 }
1788 entry = entry->prev;
1789 } else {
1790 entry = entry->next;
1791 if (entry == &map->header) {
1792 UVMHIST_LOG(maphist, "<- failed (off end)",
1793 0,0,0,0);
1794 goto notfound;
1795 }
1796 }
1797 }
1798
1799 found:
1800 SAVE_HINT(map, map->hint, entry);
1801 *result = hint;
1802 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0);
1803 KASSERT( topdown || hint >= orig_hint);
1804 KASSERT(!topdown || hint <= orig_hint);
1805 KASSERT(entry->end <= hint);
1806 KASSERT(hint + length <= entry->next->start);
1807 return (entry);
1808
1809 wraparound:
1810 UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0);
1811
1812 return (NULL);
1813
1814 notfound:
1815 UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0);
1816
1817 return (NULL);
1818 }
1819
1820 /*
1821 * U N M A P - m a i n h e l p e r f u n c t i o n s
1822 */
1823
1824 /*
1825 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
1826 *
1827 * => caller must check alignment and size
1828 * => map must be locked by caller
1829 * => we return a list of map entries that we've remove from the map
1830 * in "entry_list"
1831 */
1832
1833 void
1834 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end,
1835 struct vm_map_entry **entry_list /* OUT */,
1836 struct uvm_mapent_reservation *umr)
1837 {
1838 struct vm_map_entry *entry, *first_entry, *next;
1839 vaddr_t len;
1840 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
1841
1842 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
1843 map, start, end, 0);
1844 VM_MAP_RANGE_CHECK(map, start, end);
1845
1846 uvm_tree_sanity(map, "unmap_remove entry");
1847
1848 /*
1849 * find first entry
1850 */
1851
1852 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
1853 /* clip and go... */
1854 entry = first_entry;
1855 UVM_MAP_CLIP_START(map, entry, start, umr);
1856 /* critical! prevents stale hint */
1857 SAVE_HINT(map, entry, entry->prev);
1858 } else {
1859 entry = first_entry->next;
1860 }
1861
1862 /*
1863 * Save the free space hint
1864 */
1865
1866 if (map->first_free->start >= start)
1867 map->first_free = entry->prev;
1868
1869 /*
1870 * note: we now re-use first_entry for a different task. we remove
1871 * a number of map entries from the map and save them in a linked
1872 * list headed by "first_entry". once we remove them from the map
1873 * the caller should unlock the map and drop the references to the
1874 * backing objects [c.f. uvm_unmap_detach]. the object is to
1875 * separate unmapping from reference dropping. why?
1876 * [1] the map has to be locked for unmapping
1877 * [2] the map need not be locked for reference dropping
1878 * [3] dropping references may trigger pager I/O, and if we hit
1879 * a pager that does synchronous I/O we may have to wait for it.
1880 * [4] we would like all waiting for I/O to occur with maps unlocked
1881 * so that we don't block other threads.
1882 */
1883
1884 first_entry = NULL;
1885 *entry_list = NULL;
1886
1887 /*
1888 * break up the area into map entry sized regions and unmap. note
1889 * that all mappings have to be removed before we can even consider
1890 * dropping references to amaps or VM objects (otherwise we could end
1891 * up with a mapping to a page on the free list which would be very bad)
1892 */
1893
1894 while ((entry != &map->header) && (entry->start < end)) {
1895 KASSERT((entry->flags & UVM_MAP_FIRST) == 0);
1896
1897 UVM_MAP_CLIP_END(map, entry, end, umr);
1898 next = entry->next;
1899 len = entry->end - entry->start;
1900
1901 /*
1902 * unwire before removing addresses from the pmap; otherwise
1903 * unwiring will put the entries back into the pmap (XXX).
1904 */
1905
1906 if (VM_MAPENT_ISWIRED(entry)) {
1907 uvm_map_entry_unwire(map, entry);
1908 }
1909 if ((map->flags & VM_MAP_PAGEABLE) == 0) {
1910
1911 /*
1912 * if the map is non-pageable, any pages mapped there
1913 * must be wired and entered with pmap_kenter_pa(),
1914 * and we should free any such pages immediately.
1915 * this is mostly used for kmem_map and mb_map.
1916 */
1917
1918 if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
1919 uvm_km_pgremove_intrsafe(entry->start,
1920 entry->end);
1921 pmap_kremove(entry->start, len);
1922 }
1923 } else if (UVM_ET_ISOBJ(entry) &&
1924 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
1925 KASSERT(vm_map_pmap(map) == pmap_kernel());
1926
1927 /*
1928 * note: kernel object mappings are currently used in
1929 * two ways:
1930 * [1] "normal" mappings of pages in the kernel object
1931 * [2] uvm_km_valloc'd allocations in which we
1932 * pmap_enter in some non-kernel-object page
1933 * (e.g. vmapbuf).
1934 *
1935 * for case [1], we need to remove the mapping from
1936 * the pmap and then remove the page from the kernel
1937 * object (because, once pages in a kernel object are
1938 * unmapped they are no longer needed, unlike, say,
1939 * a vnode where you might want the data to persist
1940 * until flushed out of a queue).
1941 *
1942 * for case [2], we need to remove the mapping from
1943 * the pmap. there shouldn't be any pages at the
1944 * specified offset in the kernel object [but it
1945 * doesn't hurt to call uvm_km_pgremove just to be
1946 * safe?]
1947 *
1948 * uvm_km_pgremove currently does the following:
1949 * for pages in the kernel object in range:
1950 * - drops the swap slot
1951 * - uvm_pagefree the page
1952 */
1953
1954 /*
1955 * remove mappings from pmap and drop the pages
1956 * from the object. offsets are always relative
1957 * to vm_map_min(kernel_map).
1958 */
1959
1960 pmap_remove(pmap_kernel(), entry->start,
1961 entry->start + len);
1962 uvm_km_pgremove(entry->object.uvm_obj,
1963 entry->start - vm_map_min(kernel_map),
1964 entry->end - vm_map_min(kernel_map));
1965
1966 /*
1967 * null out kernel_object reference, we've just
1968 * dropped it
1969 */
1970
1971 entry->etype &= ~UVM_ET_OBJ;
1972 entry->object.uvm_obj = NULL;
1973 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
1974
1975 /*
1976 * remove mappings the standard way.
1977 */
1978
1979 pmap_remove(map->pmap, entry->start, entry->end);
1980 }
1981
1982 #if defined(DEBUG)
1983 if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
1984
1985 /*
1986 * check if there's remaining mapping,
1987 * which is a bug in caller.
1988 */
1989
1990 vaddr_t va;
1991 for (va = entry->start; va < entry->end;
1992 va += PAGE_SIZE) {
1993 if (pmap_extract(vm_map_pmap(map), va, NULL)) {
1994 panic("uvm_unmap_remove: has mapping");
1995 }
1996 }
1997 }
1998 #endif /* defined(DEBUG) */
1999
2000 /*
2001 * remove entry from map and put it on our list of entries
2002 * that we've nuked. then go to next entry.
2003 */
2004
2005 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0);
2006
2007 /* critical! prevents stale hint */
2008 SAVE_HINT(map, entry, entry->prev);
2009
2010 uvm_map_entry_unlink(map, entry);
2011 KASSERT(map->size >= len);
2012 map->size -= len;
2013 entry->prev = NULL;
2014 entry->next = first_entry;
2015 first_entry = entry;
2016 entry = next;
2017 }
2018 if ((map->flags & VM_MAP_DYING) == 0) {
2019 pmap_update(vm_map_pmap(map));
2020 }
2021
2022 uvm_tree_sanity(map, "unmap_remove leave");
2023
2024 /*
2025 * now we've cleaned up the map and are ready for the caller to drop
2026 * references to the mapped objects.
2027 */
2028
2029 *entry_list = first_entry;
2030 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
2031
2032 simple_lock(&map->flags_lock);
2033 if (map->flags & VM_MAP_WANTVA) {
2034 map->flags &= ~VM_MAP_WANTVA;
2035 wakeup(&map->header);
2036 }
2037 simple_unlock(&map->flags_lock);
2038 }
2039
2040 /*
2041 * uvm_unmap_detach: drop references in a chain of map entries
2042 *
2043 * => we will free the map entries as we traverse the list.
2044 */
2045
2046 void
2047 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags)
2048 {
2049 struct vm_map_entry *next_entry;
2050 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
2051
2052 while (first_entry) {
2053 KASSERT(!VM_MAPENT_ISWIRED(first_entry));
2054 UVMHIST_LOG(maphist,
2055 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
2056 first_entry, first_entry->aref.ar_amap,
2057 first_entry->object.uvm_obj,
2058 UVM_ET_ISSUBMAP(first_entry));
2059
2060 /*
2061 * drop reference to amap, if we've got one
2062 */
2063
2064 if (first_entry->aref.ar_amap)
2065 uvm_map_unreference_amap(first_entry, flags);
2066
2067 /*
2068 * drop reference to our backing object, if we've got one
2069 */
2070
2071 KASSERT(!UVM_ET_ISSUBMAP(first_entry));
2072 if (UVM_ET_ISOBJ(first_entry) &&
2073 first_entry->object.uvm_obj->pgops->pgo_detach) {
2074 (*first_entry->object.uvm_obj->pgops->pgo_detach)
2075 (first_entry->object.uvm_obj);
2076 }
2077 next_entry = first_entry->next;
2078 uvm_mapent_free(first_entry);
2079 first_entry = next_entry;
2080 }
2081 UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
2082 }
2083
2084 /*
2085 * E X T R A C T I O N F U N C T I O N S
2086 */
2087
2088 /*
2089 * uvm_map_reserve: reserve space in a vm_map for future use.
2090 *
2091 * => we reserve space in a map by putting a dummy map entry in the
2092 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
2093 * => map should be unlocked (we will write lock it)
2094 * => we return true if we were able to reserve space
2095 * => XXXCDC: should be inline?
2096 */
2097
2098 int
2099 uvm_map_reserve(struct vm_map *map, vsize_t size,
2100 vaddr_t offset /* hint for pmap_prefer */,
2101 vsize_t align /* alignment hint */,
2102 vaddr_t *raddr /* IN:hint, OUT: reserved VA */)
2103 {
2104 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
2105
2106 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
2107 map,size,offset,raddr);
2108
2109 size = round_page(size);
2110 if (*raddr < vm_map_min(map))
2111 *raddr = vm_map_min(map); /* hint */
2112
2113 /*
2114 * reserve some virtual space.
2115 */
2116
2117 if (uvm_map(map, raddr, size, NULL, offset, 0,
2118 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
2119 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) {
2120 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
2121 return (FALSE);
2122 }
2123
2124 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
2125 return (TRUE);
2126 }
2127
2128 /*
2129 * uvm_map_replace: replace a reserved (blank) area of memory with
2130 * real mappings.
2131 *
2132 * => caller must WRITE-LOCK the map
2133 * => we return TRUE if replacement was a success
2134 * => we expect the newents chain to have nnewents entrys on it and
2135 * we expect newents->prev to point to the last entry on the list
2136 * => note newents is allowed to be NULL
2137 */
2138
2139 int
2140 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end,
2141 struct vm_map_entry *newents, int nnewents)
2142 {
2143 struct vm_map_entry *oldent, *last;
2144
2145 uvm_tree_sanity(map, "map_replace entry");
2146
2147 /*
2148 * first find the blank map entry at the specified address
2149 */
2150
2151 if (!uvm_map_lookup_entry(map, start, &oldent)) {
2152 return (FALSE);
2153 }
2154
2155 /*
2156 * check to make sure we have a proper blank entry
2157 */
2158
2159 if (oldent->start != start || oldent->end != end ||
2160 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
2161 return (FALSE);
2162 }
2163
2164 #ifdef DIAGNOSTIC
2165
2166 /*
2167 * sanity check the newents chain
2168 */
2169
2170 {
2171 struct vm_map_entry *tmpent = newents;
2172 int nent = 0;
2173 vaddr_t cur = start;
2174
2175 while (tmpent) {
2176 nent++;
2177 if (tmpent->start < cur)
2178 panic("uvm_map_replace1");
2179 if (tmpent->start > tmpent->end || tmpent->end > end) {
2180 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
2181 tmpent->start, tmpent->end, end);
2182 panic("uvm_map_replace2");
2183 }
2184 cur = tmpent->end;
2185 if (tmpent->next) {
2186 if (tmpent->next->prev != tmpent)
2187 panic("uvm_map_replace3");
2188 } else {
2189 if (newents->prev != tmpent)
2190 panic("uvm_map_replace4");
2191 }
2192 tmpent = tmpent->next;
2193 }
2194 if (nent != nnewents)
2195 panic("uvm_map_replace5");
2196 }
2197 #endif
2198
2199 /*
2200 * map entry is a valid blank! replace it. (this does all the
2201 * work of map entry link/unlink...).
2202 */
2203
2204 if (newents) {
2205 last = newents->prev;
2206
2207 /* critical: flush stale hints out of map */
2208 SAVE_HINT(map, map->hint, newents);
2209 if (map->first_free == oldent)
2210 map->first_free = last;
2211
2212 last->next = oldent->next;
2213 last->next->prev = last;
2214
2215 /* Fix RB tree */
2216 uvm_rb_remove(map, oldent);
2217
2218 newents->prev = oldent->prev;
2219 newents->prev->next = newents;
2220 map->nentries = map->nentries + (nnewents - 1);
2221
2222 /* Fixup the RB tree */
2223 {
2224 int i;
2225 struct vm_map_entry *tmp;
2226
2227 tmp = newents;
2228 for (i = 0; i < nnewents && tmp; i++) {
2229 uvm_rb_insert(map, tmp);
2230 tmp = tmp->next;
2231 }
2232 }
2233 } else {
2234
2235 /* critical: flush stale hints out of map */
2236 SAVE_HINT(map, map->hint, oldent->prev);
2237 if (map->first_free == oldent)
2238 map->first_free = oldent->prev;
2239
2240 /* NULL list of new entries: just remove the old one */
2241 uvm_map_entry_unlink(map, oldent);
2242 }
2243
2244 uvm_tree_sanity(map, "map_replace leave");
2245
2246 /*
2247 * now we can free the old blank entry, unlock the map and return.
2248 */
2249
2250 uvm_mapent_free(oldent);
2251 return (TRUE);
2252 }
2253
2254 /*
2255 * uvm_map_extract: extract a mapping from a map and put it somewhere
2256 * (maybe removing the old mapping)
2257 *
2258 * => maps should be unlocked (we will write lock them)
2259 * => returns 0 on success, error code otherwise
2260 * => start must be page aligned
2261 * => len must be page sized
2262 * => flags:
2263 * UVM_EXTRACT_REMOVE: remove mappings from srcmap
2264 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
2265 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
2266 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
2267 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
2268 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
2269 * be used from within the kernel in a kernel level map <<<
2270 */
2271
2272 int
2273 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len,
2274 struct vm_map *dstmap, vaddr_t *dstaddrp, int flags)
2275 {
2276 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge;
2277 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
2278 *deadentry, *oldentry;
2279 vsize_t elen;
2280 int nchain, error, copy_ok;
2281 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
2282
2283 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
2284 len,0);
2285 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
2286
2287 uvm_tree_sanity(srcmap, "map_extract src enter");
2288 uvm_tree_sanity(dstmap, "map_extract dst enter");
2289
2290 /*
2291 * step 0: sanity check: start must be on a page boundary, length
2292 * must be page sized. can't ask for CONTIG/QREF if you asked for
2293 * REMOVE.
2294 */
2295
2296 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
2297 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
2298 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
2299
2300 /*
2301 * step 1: reserve space in the target map for the extracted area
2302 */
2303
2304 dstaddr = vm_map_min(dstmap);
2305 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE)
2306 return (ENOMEM);
2307 *dstaddrp = dstaddr; /* pass address back to caller */
2308 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0);
2309
2310 /*
2311 * step 2: setup for the extraction process loop by init'ing the
2312 * map entry chain, locking src map, and looking up the first useful
2313 * entry in the map.
2314 */
2315
2316 end = start + len;
2317 newend = dstaddr + len;
2318 chain = endchain = NULL;
2319 nchain = 0;
2320 vm_map_lock(srcmap);
2321
2322 if (uvm_map_lookup_entry(srcmap, start, &entry)) {
2323
2324 /* "start" is within an entry */
2325 if (flags & UVM_EXTRACT_QREF) {
2326
2327 /*
2328 * for quick references we don't clip the entry, so
2329 * the entry may map space "before" the starting
2330 * virtual address... this is the "fudge" factor
2331 * (which can be non-zero only the first time
2332 * through the "while" loop in step 3).
2333 */
2334
2335 fudge = start - entry->start;
2336 } else {
2337
2338 /*
2339 * normal reference: we clip the map to fit (thus
2340 * fudge is zero)
2341 */
2342
2343 UVM_MAP_CLIP_START(srcmap, entry, start, NULL);
2344 SAVE_HINT(srcmap, srcmap->hint, entry->prev);
2345 fudge = 0;
2346 }
2347 } else {
2348
2349 /* "start" is not within an entry ... skip to next entry */
2350 if (flags & UVM_EXTRACT_CONTIG) {
2351 error = EINVAL;
2352 goto bad; /* definite hole here ... */
2353 }
2354
2355 entry = entry->next;
2356 fudge = 0;
2357 }
2358
2359 /* save values from srcmap for step 6 */
2360 orig_entry = entry;
2361 orig_fudge = fudge;
2362
2363 /*
2364 * step 3: now start looping through the map entries, extracting
2365 * as we go.
2366 */
2367
2368 while (entry->start < end && entry != &srcmap->header) {
2369
2370 /* if we are not doing a quick reference, clip it */
2371 if ((flags & UVM_EXTRACT_QREF) == 0)
2372 UVM_MAP_CLIP_END(srcmap, entry, end, NULL);
2373
2374 /* clear needs_copy (allow chunking) */
2375 if (UVM_ET_ISNEEDSCOPY(entry)) {
2376 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
2377 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */
2378 error = ENOMEM;
2379 goto bad;
2380 }
2381
2382 /* amap_copy could clip (during chunk)! update fudge */
2383 if (fudge) {
2384 fudge = start - entry->start;
2385 orig_fudge = fudge;
2386 }
2387 }
2388
2389 /* calculate the offset of this from "start" */
2390 oldoffset = (entry->start + fudge) - start;
2391
2392 /* allocate a new map entry */
2393 newentry = uvm_mapent_alloc(dstmap, 0);
2394 if (newentry == NULL) {
2395 error = ENOMEM;
2396 goto bad;
2397 }
2398
2399 /* set up new map entry */
2400 newentry->next = NULL;
2401 newentry->prev = endchain;
2402 newentry->start = dstaddr + oldoffset;
2403 newentry->end =
2404 newentry->start + (entry->end - (entry->start + fudge));
2405 if (newentry->end > newend || newentry->end < newentry->start)
2406 newentry->end = newend;
2407 newentry->object.uvm_obj = entry->object.uvm_obj;
2408 if (newentry->object.uvm_obj) {
2409 if (newentry->object.uvm_obj->pgops->pgo_reference)
2410 newentry->object.uvm_obj->pgops->
2411 pgo_reference(newentry->object.uvm_obj);
2412 newentry->offset = entry->offset + fudge;
2413 } else {
2414 newentry->offset = 0;
2415 }
2416 newentry->etype = entry->etype;
2417 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
2418 entry->max_protection : entry->protection;
2419 newentry->max_protection = entry->max_protection;
2420 newentry->inheritance = entry->inheritance;
2421 newentry->wired_count = 0;
2422 newentry->aref.ar_amap = entry->aref.ar_amap;
2423 if (newentry->aref.ar_amap) {
2424 newentry->aref.ar_pageoff =
2425 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
2426 uvm_map_reference_amap(newentry, AMAP_SHARED |
2427 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
2428 } else {
2429 newentry->aref.ar_pageoff = 0;
2430 }
2431 newentry->advice = entry->advice;
2432
2433 /* now link it on the chain */
2434 nchain++;
2435 if (endchain == NULL) {
2436 chain = endchain = newentry;
2437 } else {
2438 endchain->next = newentry;
2439 endchain = newentry;
2440 }
2441
2442 /* end of 'while' loop! */
2443 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
2444 (entry->next == &srcmap->header ||
2445 entry->next->start != entry->end)) {
2446 error = EINVAL;
2447 goto bad;
2448 }
2449 entry = entry->next;
2450 fudge = 0;
2451 }
2452
2453 /*
2454 * step 4: close off chain (in format expected by uvm_map_replace)
2455 */
2456
2457 if (chain)
2458 chain->prev = endchain;
2459
2460 /*
2461 * step 5: attempt to lock the dest map so we can pmap_copy.
2462 * note usage of copy_ok:
2463 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
2464 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
2465 */
2466
2467 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
2468 copy_ok = 1;
2469 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2470 nchain)) {
2471 if (srcmap != dstmap)
2472 vm_map_unlock(dstmap);
2473 error = EIO;
2474 goto bad;
2475 }
2476 } else {
2477 copy_ok = 0;
2478 /* replace defered until step 7 */
2479 }
2480
2481 /*
2482 * step 6: traverse the srcmap a second time to do the following:
2483 * - if we got a lock on the dstmap do pmap_copy
2484 * - if UVM_EXTRACT_REMOVE remove the entries
2485 * we make use of orig_entry and orig_fudge (saved in step 2)
2486 */
2487
2488 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
2489
2490 /* purge possible stale hints from srcmap */
2491 if (flags & UVM_EXTRACT_REMOVE) {
2492 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
2493 if (srcmap->first_free->start >= start)
2494 srcmap->first_free = orig_entry->prev;
2495 }
2496
2497 entry = orig_entry;
2498 fudge = orig_fudge;
2499 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */
2500
2501 while (entry->start < end && entry != &srcmap->header) {
2502 if (copy_ok) {
2503 oldoffset = (entry->start + fudge) - start;
2504 elen = MIN(end, entry->end) -
2505 (entry->start + fudge);
2506 pmap_copy(dstmap->pmap, srcmap->pmap,
2507 dstaddr + oldoffset, elen,
2508 entry->start + fudge);
2509 }
2510
2511 /* we advance "entry" in the following if statement */
2512 if (flags & UVM_EXTRACT_REMOVE) {
2513 pmap_remove(srcmap->pmap, entry->start,
2514 entry->end);
2515 oldentry = entry; /* save entry */
2516 entry = entry->next; /* advance */
2517 uvm_map_entry_unlink(srcmap, oldentry);
2518 /* add to dead list */
2519 oldentry->next = deadentry;
2520 deadentry = oldentry;
2521 } else {
2522 entry = entry->next; /* advance */
2523 }
2524
2525 /* end of 'while' loop */
2526 fudge = 0;
2527 }
2528 pmap_update(srcmap->pmap);
2529
2530 /*
2531 * unlock dstmap. we will dispose of deadentry in
2532 * step 7 if needed
2533 */
2534
2535 if (copy_ok && srcmap != dstmap)
2536 vm_map_unlock(dstmap);
2537
2538 } else {
2539 deadentry = NULL;
2540 }
2541
2542 /*
2543 * step 7: we are done with the source map, unlock. if copy_ok
2544 * is 0 then we have not replaced the dummy mapping in dstmap yet
2545 * and we need to do so now.
2546 */
2547
2548 vm_map_unlock(srcmap);
2549 if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
2550 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */
2551
2552 /* now do the replacement if we didn't do it in step 5 */
2553 if (copy_ok == 0) {
2554 vm_map_lock(dstmap);
2555 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2556 nchain);
2557 vm_map_unlock(dstmap);
2558
2559 if (error == FALSE) {
2560 error = EIO;
2561 goto bad2;
2562 }
2563 }
2564
2565 uvm_tree_sanity(srcmap, "map_extract src leave");
2566 uvm_tree_sanity(dstmap, "map_extract dst leave");
2567
2568 return (0);
2569
2570 /*
2571 * bad: failure recovery
2572 */
2573 bad:
2574 vm_map_unlock(srcmap);
2575 bad2: /* src already unlocked */
2576 if (chain)
2577 uvm_unmap_detach(chain,
2578 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
2579
2580 uvm_tree_sanity(srcmap, "map_extract src err leave");
2581 uvm_tree_sanity(dstmap, "map_extract dst err leave");
2582
2583 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */
2584 return (error);
2585 }
2586
2587 /* end of extraction functions */
2588
2589 /*
2590 * uvm_map_submap: punch down part of a map into a submap
2591 *
2592 * => only the kernel_map is allowed to be submapped
2593 * => the purpose of submapping is to break up the locking granularity
2594 * of a larger map
2595 * => the range specified must have been mapped previously with a uvm_map()
2596 * call [with uobj==NULL] to create a blank map entry in the main map.
2597 * [And it had better still be blank!]
2598 * => maps which contain submaps should never be copied or forked.
2599 * => to remove a submap, use uvm_unmap() on the main map
2600 * and then uvm_map_deallocate() the submap.
2601 * => main map must be unlocked.
2602 * => submap must have been init'd and have a zero reference count.
2603 * [need not be locked as we don't actually reference it]
2604 */
2605
2606 int
2607 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end,
2608 struct vm_map *submap)
2609 {
2610 struct vm_map_entry *entry;
2611 struct uvm_mapent_reservation umr;
2612 int error;
2613
2614 uvm_mapent_reserve(map, &umr, 2, 0);
2615
2616 vm_map_lock(map);
2617 VM_MAP_RANGE_CHECK(map, start, end);
2618
2619 if (uvm_map_lookup_entry(map, start, &entry)) {
2620 UVM_MAP_CLIP_START(map, entry, start, &umr);
2621 UVM_MAP_CLIP_END(map, entry, end, &umr); /* to be safe */
2622 } else {
2623 entry = NULL;
2624 }
2625
2626 if (entry != NULL &&
2627 entry->start == start && entry->end == end &&
2628 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
2629 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
2630 entry->etype |= UVM_ET_SUBMAP;
2631 entry->object.sub_map = submap;
2632 entry->offset = 0;
2633 uvm_map_reference(submap);
2634 error = 0;
2635 } else {
2636 error = EINVAL;
2637 }
2638 vm_map_unlock(map);
2639
2640 uvm_mapent_unreserve(map, &umr);
2641
2642 return error;
2643 }
2644
2645 /*
2646 * uvm_map_setup_kernel: init in-kernel map
2647 *
2648 * => map must not be in service yet.
2649 */
2650
2651 void
2652 uvm_map_setup_kernel(struct vm_map_kernel *map,
2653 vaddr_t min, vaddr_t max, int flags)
2654 {
2655
2656 uvm_map_setup(&map->vmk_map, min, max, flags);
2657
2658 LIST_INIT(&map->vmk_kentry_free);
2659 map->vmk_merged_entries = NULL;
2660 }
2661
2662
2663 /*
2664 * uvm_map_protect: change map protection
2665 *
2666 * => set_max means set max_protection.
2667 * => map must be unlocked.
2668 */
2669
2670 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \
2671 ~VM_PROT_WRITE : VM_PROT_ALL)
2672
2673 int
2674 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end,
2675 vm_prot_t new_prot, boolean_t set_max)
2676 {
2677 struct vm_map_entry *current, *entry;
2678 int error = 0;
2679 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
2680 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
2681 map, start, end, new_prot);
2682
2683 vm_map_lock(map);
2684 VM_MAP_RANGE_CHECK(map, start, end);
2685 if (uvm_map_lookup_entry(map, start, &entry)) {
2686 UVM_MAP_CLIP_START(map, entry, start, NULL);
2687 } else {
2688 entry = entry->next;
2689 }
2690
2691 /*
2692 * make a first pass to check for protection violations.
2693 */
2694
2695 current = entry;
2696 while ((current != &map->header) && (current->start < end)) {
2697 if (UVM_ET_ISSUBMAP(current)) {
2698 error = EINVAL;
2699 goto out;
2700 }
2701 if ((new_prot & current->max_protection) != new_prot) {
2702 error = EACCES;
2703 goto out;
2704 }
2705 /*
2706 * Don't allow VM_PROT_EXECUTE to be set on entries that
2707 * point to vnodes that are associated with a NOEXEC file
2708 * system.
2709 */
2710 if (UVM_ET_ISOBJ(current) &&
2711 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
2712 struct vnode *vp =
2713 (struct vnode *) current->object.uvm_obj;
2714
2715 if ((new_prot & VM_PROT_EXECUTE) != 0 &&
2716 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
2717 error = EACCES;
2718 goto out;
2719 }
2720 }
2721 current = current->next;
2722 }
2723
2724 /* go back and fix up protections (no need to clip this time). */
2725
2726 current = entry;
2727 while ((current != &map->header) && (current->start < end)) {
2728 vm_prot_t old_prot;
2729
2730 UVM_MAP_CLIP_END(map, current, end, NULL);
2731 old_prot = current->protection;
2732 if (set_max)
2733 current->protection =
2734 (current->max_protection = new_prot) & old_prot;
2735 else
2736 current->protection = new_prot;
2737
2738 /*
2739 * update physical map if necessary. worry about copy-on-write
2740 * here -- CHECK THIS XXX
2741 */
2742
2743 if (current->protection != old_prot) {
2744 /* update pmap! */
2745 pmap_protect(map->pmap, current->start, current->end,
2746 current->protection & MASK(entry));
2747
2748 /*
2749 * If this entry points at a vnode, and the
2750 * protection includes VM_PROT_EXECUTE, mark
2751 * the vnode as VEXECMAP.
2752 */
2753 if (UVM_ET_ISOBJ(current)) {
2754 struct uvm_object *uobj =
2755 current->object.uvm_obj;
2756
2757 if (UVM_OBJ_IS_VNODE(uobj) &&
2758 (current->protection & VM_PROT_EXECUTE))
2759 vn_markexec((struct vnode *) uobj);
2760 }
2761 }
2762
2763 /*
2764 * If the map is configured to lock any future mappings,
2765 * wire this entry now if the old protection was VM_PROT_NONE
2766 * and the new protection is not VM_PROT_NONE.
2767 */
2768
2769 if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
2770 VM_MAPENT_ISWIRED(entry) == 0 &&
2771 old_prot == VM_PROT_NONE &&
2772 new_prot != VM_PROT_NONE) {
2773 if (uvm_map_pageable(map, entry->start,
2774 entry->end, FALSE,
2775 UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
2776
2777 /*
2778 * If locking the entry fails, remember the
2779 * error if it's the first one. Note we
2780 * still continue setting the protection in
2781 * the map, but will return the error
2782 * condition regardless.
2783 *
2784 * XXX Ignore what the actual error is,
2785 * XXX just call it a resource shortage
2786 * XXX so that it doesn't get confused
2787 * XXX what uvm_map_protect() itself would
2788 * XXX normally return.
2789 */
2790
2791 error = ENOMEM;
2792 }
2793 }
2794 current = current->next;
2795 }
2796 pmap_update(map->pmap);
2797
2798 out:
2799 vm_map_unlock(map);
2800
2801 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
2802 return error;
2803 }
2804
2805 #undef MASK
2806
2807 /*
2808 * uvm_map_inherit: set inheritance code for range of addrs in map.
2809 *
2810 * => map must be unlocked
2811 * => note that the inherit code is used during a "fork". see fork
2812 * code for details.
2813 */
2814
2815 int
2816 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end,
2817 vm_inherit_t new_inheritance)
2818 {
2819 struct vm_map_entry *entry, *temp_entry;
2820 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
2821 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
2822 map, start, end, new_inheritance);
2823
2824 switch (new_inheritance) {
2825 case MAP_INHERIT_NONE:
2826 case MAP_INHERIT_COPY:
2827 case MAP_INHERIT_SHARE:
2828 break;
2829 default:
2830 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2831 return EINVAL;
2832 }
2833
2834 vm_map_lock(map);
2835 VM_MAP_RANGE_CHECK(map, start, end);
2836 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2837 entry = temp_entry;
2838 UVM_MAP_CLIP_START(map, entry, start, NULL);
2839 } else {
2840 entry = temp_entry->next;
2841 }
2842 while ((entry != &map->header) && (entry->start < end)) {
2843 UVM_MAP_CLIP_END(map, entry, end, NULL);
2844 entry->inheritance = new_inheritance;
2845 entry = entry->next;
2846 }
2847 vm_map_unlock(map);
2848 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2849 return 0;
2850 }
2851
2852 /*
2853 * uvm_map_advice: set advice code for range of addrs in map.
2854 *
2855 * => map must be unlocked
2856 */
2857
2858 int
2859 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice)
2860 {
2861 struct vm_map_entry *entry, *temp_entry;
2862 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
2863 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
2864 map, start, end, new_advice);
2865
2866 vm_map_lock(map);
2867 VM_MAP_RANGE_CHECK(map, start, end);
2868 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2869 entry = temp_entry;
2870 UVM_MAP_CLIP_START(map, entry, start, NULL);
2871 } else {
2872 entry = temp_entry->next;
2873 }
2874
2875 /*
2876 * XXXJRT: disallow holes?
2877 */
2878
2879 while ((entry != &map->header) && (entry->start < end)) {
2880 UVM_MAP_CLIP_END(map, entry, end, NULL);
2881
2882 switch (new_advice) {
2883 case MADV_NORMAL:
2884 case MADV_RANDOM:
2885 case MADV_SEQUENTIAL:
2886 /* nothing special here */
2887 break;
2888
2889 default:
2890 vm_map_unlock(map);
2891 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2892 return EINVAL;
2893 }
2894 entry->advice = new_advice;
2895 entry = entry->next;
2896 }
2897
2898 vm_map_unlock(map);
2899 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2900 return 0;
2901 }
2902
2903 /*
2904 * uvm_map_pageable: sets the pageability of a range in a map.
2905 *
2906 * => wires map entries. should not be used for transient page locking.
2907 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
2908 * => regions sepcified as not pageable require lock-down (wired) memory
2909 * and page tables.
2910 * => map must never be read-locked
2911 * => if islocked is TRUE, map is already write-locked
2912 * => we always unlock the map, since we must downgrade to a read-lock
2913 * to call uvm_fault_wire()
2914 * => XXXCDC: check this and try and clean it up.
2915 */
2916
2917 int
2918 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end,
2919 boolean_t new_pageable, int lockflags)
2920 {
2921 struct vm_map_entry *entry, *start_entry, *failed_entry;
2922 int rv;
2923 #ifdef DIAGNOSTIC
2924 u_int timestamp_save;
2925 #endif
2926 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
2927 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
2928 map, start, end, new_pageable);
2929 KASSERT(map->flags & VM_MAP_PAGEABLE);
2930
2931 if ((lockflags & UVM_LK_ENTER) == 0)
2932 vm_map_lock(map);
2933 VM_MAP_RANGE_CHECK(map, start, end);
2934
2935 /*
2936 * only one pageability change may take place at one time, since
2937 * uvm_fault_wire assumes it will be called only once for each
2938 * wiring/unwiring. therefore, we have to make sure we're actually
2939 * changing the pageability for the entire region. we do so before
2940 * making any changes.
2941 */
2942
2943 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
2944 if ((lockflags & UVM_LK_EXIT) == 0)
2945 vm_map_unlock(map);
2946
2947 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
2948 return EFAULT;
2949 }
2950 entry = start_entry;
2951
2952 /*
2953 * handle wiring and unwiring separately.
2954 */
2955
2956 if (new_pageable) { /* unwire */
2957 UVM_MAP_CLIP_START(map, entry, start, NULL);
2958
2959 /*
2960 * unwiring. first ensure that the range to be unwired is
2961 * really wired down and that there are no holes.
2962 */
2963
2964 while ((entry != &map->header) && (entry->start < end)) {
2965 if (entry->wired_count == 0 ||
2966 (entry->end < end &&
2967 (entry->next == &map->header ||
2968 entry->next->start > entry->end))) {
2969 if ((lockflags & UVM_LK_EXIT) == 0)
2970 vm_map_unlock(map);
2971 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
2972 return EINVAL;
2973 }
2974 entry = entry->next;
2975 }
2976
2977 /*
2978 * POSIX 1003.1b - a single munlock call unlocks a region,
2979 * regardless of the number of mlock calls made on that
2980 * region.
2981 */
2982
2983 entry = start_entry;
2984 while ((entry != &map->header) && (entry->start < end)) {
2985 UVM_MAP_CLIP_END(map, entry, end, NULL);
2986 if (VM_MAPENT_ISWIRED(entry))
2987 uvm_map_entry_unwire(map, entry);
2988 entry = entry->next;
2989 }
2990 if ((lockflags & UVM_LK_EXIT) == 0)
2991 vm_map_unlock(map);
2992 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2993 return 0;
2994 }
2995
2996 /*
2997 * wire case: in two passes [XXXCDC: ugly block of code here]
2998 *
2999 * 1: holding the write lock, we create any anonymous maps that need
3000 * to be created. then we clip each map entry to the region to
3001 * be wired and increment its wiring count.
3002 *
3003 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
3004 * in the pages for any newly wired area (wired_count == 1).
3005 *
3006 * downgrading to a read lock for uvm_fault_wire avoids a possible
3007 * deadlock with another thread that may have faulted on one of
3008 * the pages to be wired (it would mark the page busy, blocking
3009 * us, then in turn block on the map lock that we hold). because
3010 * of problems in the recursive lock package, we cannot upgrade
3011 * to a write lock in vm_map_lookup. thus, any actions that
3012 * require the write lock must be done beforehand. because we
3013 * keep the read lock on the map, the copy-on-write status of the
3014 * entries we modify here cannot change.
3015 */
3016
3017 while ((entry != &map->header) && (entry->start < end)) {
3018 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3019
3020 /*
3021 * perform actions of vm_map_lookup that need the
3022 * write lock on the map: create an anonymous map
3023 * for a copy-on-write region, or an anonymous map
3024 * for a zero-fill region. (XXXCDC: submap case
3025 * ok?)
3026 */
3027
3028 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
3029 if (UVM_ET_ISNEEDSCOPY(entry) &&
3030 ((entry->max_protection & VM_PROT_WRITE) ||
3031 (entry->object.uvm_obj == NULL))) {
3032 amap_copy(map, entry, M_WAITOK, TRUE,
3033 start, end);
3034 /* XXXCDC: wait OK? */
3035 }
3036 }
3037 }
3038 UVM_MAP_CLIP_START(map, entry, start, NULL);
3039 UVM_MAP_CLIP_END(map, entry, end, NULL);
3040 entry->wired_count++;
3041
3042 /*
3043 * Check for holes
3044 */
3045
3046 if (entry->protection == VM_PROT_NONE ||
3047 (entry->end < end &&
3048 (entry->next == &map->header ||
3049 entry->next->start > entry->end))) {
3050
3051 /*
3052 * found one. amap creation actions do not need to
3053 * be undone, but the wired counts need to be restored.
3054 */
3055
3056 while (entry != &map->header && entry->end > start) {
3057 entry->wired_count--;
3058 entry = entry->prev;
3059 }
3060 if ((lockflags & UVM_LK_EXIT) == 0)
3061 vm_map_unlock(map);
3062 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
3063 return EINVAL;
3064 }
3065 entry = entry->next;
3066 }
3067
3068 /*
3069 * Pass 2.
3070 */
3071
3072 #ifdef DIAGNOSTIC
3073 timestamp_save = map->timestamp;
3074 #endif
3075 vm_map_busy(map);
3076 vm_map_downgrade(map);
3077
3078 rv = 0;
3079 entry = start_entry;
3080 while (entry != &map->header && entry->start < end) {
3081 if (entry->wired_count == 1) {
3082 rv = uvm_fault_wire(map, entry->start, entry->end,
3083 VM_FAULT_WIREMAX, entry->max_protection);
3084 if (rv) {
3085
3086 /*
3087 * wiring failed. break out of the loop.
3088 * we'll clean up the map below, once we
3089 * have a write lock again.
3090 */
3091
3092 break;
3093 }
3094 }
3095 entry = entry->next;
3096 }
3097
3098 if (rv) { /* failed? */
3099
3100 /*
3101 * Get back to an exclusive (write) lock.
3102 */
3103
3104 vm_map_upgrade(map);
3105 vm_map_unbusy(map);
3106
3107 #ifdef DIAGNOSTIC
3108 if (timestamp_save != map->timestamp)
3109 panic("uvm_map_pageable: stale map");
3110 #endif
3111
3112 /*
3113 * first drop the wiring count on all the entries
3114 * which haven't actually been wired yet.
3115 */
3116
3117 failed_entry = entry;
3118 while (entry != &map->header && entry->start < end) {
3119 entry->wired_count--;
3120 entry = entry->next;
3121 }
3122
3123 /*
3124 * now, unwire all the entries that were successfully
3125 * wired above.
3126 */
3127
3128 entry = start_entry;
3129 while (entry != failed_entry) {
3130 entry->wired_count--;
3131 if (VM_MAPENT_ISWIRED(entry) == 0)
3132 uvm_map_entry_unwire(map, entry);
3133 entry = entry->next;
3134 }
3135 if ((lockflags & UVM_LK_EXIT) == 0)
3136 vm_map_unlock(map);
3137 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
3138 return (rv);
3139 }
3140
3141 /* We are holding a read lock here. */
3142 if ((lockflags & UVM_LK_EXIT) == 0) {
3143 vm_map_unbusy(map);
3144 vm_map_unlock_read(map);
3145 } else {
3146
3147 /*
3148 * Get back to an exclusive (write) lock.
3149 */
3150
3151 vm_map_upgrade(map);
3152 vm_map_unbusy(map);
3153 }
3154
3155 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3156 return 0;
3157 }
3158
3159 /*
3160 * uvm_map_pageable_all: special case of uvm_map_pageable - affects
3161 * all mapped regions.
3162 *
3163 * => map must not be locked.
3164 * => if no flags are specified, all regions are unwired.
3165 * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
3166 */
3167
3168 int
3169 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit)
3170 {
3171 struct vm_map_entry *entry, *failed_entry;
3172 vsize_t size;
3173 int rv;
3174 #ifdef DIAGNOSTIC
3175 u_int timestamp_save;
3176 #endif
3177 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
3178 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
3179
3180 KASSERT(map->flags & VM_MAP_PAGEABLE);
3181
3182 vm_map_lock(map);
3183
3184 /*
3185 * handle wiring and unwiring separately.
3186 */
3187
3188 if (flags == 0) { /* unwire */
3189
3190 /*
3191 * POSIX 1003.1b -- munlockall unlocks all regions,
3192 * regardless of how many times mlockall has been called.
3193 */
3194
3195 for (entry = map->header.next; entry != &map->header;
3196 entry = entry->next) {
3197 if (VM_MAPENT_ISWIRED(entry))
3198 uvm_map_entry_unwire(map, entry);
3199 }
3200 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3201 vm_map_unlock(map);
3202 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3203 return 0;
3204 }
3205
3206 if (flags & MCL_FUTURE) {
3207
3208 /*
3209 * must wire all future mappings; remember this.
3210 */
3211
3212 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
3213 }
3214
3215 if ((flags & MCL_CURRENT) == 0) {
3216
3217 /*
3218 * no more work to do!
3219 */
3220
3221 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
3222 vm_map_unlock(map);
3223 return 0;
3224 }
3225
3226 /*
3227 * wire case: in three passes [XXXCDC: ugly block of code here]
3228 *
3229 * 1: holding the write lock, count all pages mapped by non-wired
3230 * entries. if this would cause us to go over our limit, we fail.
3231 *
3232 * 2: still holding the write lock, we create any anonymous maps that
3233 * need to be created. then we increment its wiring count.
3234 *
3235 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
3236 * in the pages for any newly wired area (wired_count == 1).
3237 *
3238 * downgrading to a read lock for uvm_fault_wire avoids a possible
3239 * deadlock with another thread that may have faulted on one of
3240 * the pages to be wired (it would mark the page busy, blocking
3241 * us, then in turn block on the map lock that we hold). because
3242 * of problems in the recursive lock package, we cannot upgrade
3243 * to a write lock in vm_map_lookup. thus, any actions that
3244 * require the write lock must be done beforehand. because we
3245 * keep the read lock on the map, the copy-on-write status of the
3246 * entries we modify here cannot change.
3247 */
3248
3249 for (size = 0, entry = map->header.next; entry != &map->header;
3250 entry = entry->next) {
3251 if (entry->protection != VM_PROT_NONE &&
3252 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3253 size += entry->end - entry->start;
3254 }
3255 }
3256
3257 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
3258 vm_map_unlock(map);
3259 return ENOMEM;
3260 }
3261
3262 if (limit != 0 &&
3263 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
3264 vm_map_unlock(map);
3265 return ENOMEM;
3266 }
3267
3268 /*
3269 * Pass 2.
3270 */
3271
3272 for (entry = map->header.next; entry != &map->header;
3273 entry = entry->next) {
3274 if (entry->protection == VM_PROT_NONE)
3275 continue;
3276 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3277
3278 /*
3279 * perform actions of vm_map_lookup that need the
3280 * write lock on the map: create an anonymous map
3281 * for a copy-on-write region, or an anonymous map
3282 * for a zero-fill region. (XXXCDC: submap case
3283 * ok?)
3284 */
3285
3286 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
3287 if (UVM_ET_ISNEEDSCOPY(entry) &&
3288 ((entry->max_protection & VM_PROT_WRITE) ||
3289 (entry->object.uvm_obj == NULL))) {
3290 amap_copy(map, entry, M_WAITOK, TRUE,
3291 entry->start, entry->end);
3292 /* XXXCDC: wait OK? */
3293 }
3294 }
3295 }
3296 entry->wired_count++;
3297 }
3298
3299 /*
3300 * Pass 3.
3301 */
3302
3303 #ifdef DIAGNOSTIC
3304 timestamp_save = map->timestamp;
3305 #endif
3306 vm_map_busy(map);
3307 vm_map_downgrade(map);
3308
3309 rv = 0;
3310 for (entry = map->header.next; entry != &map->header;
3311 entry = entry->next) {
3312 if (entry->wired_count == 1) {
3313 rv = uvm_fault_wire(map, entry->start, entry->end,
3314 VM_FAULT_WIREMAX, entry->max_protection);
3315 if (rv) {
3316
3317 /*
3318 * wiring failed. break out of the loop.
3319 * we'll clean up the map below, once we
3320 * have a write lock again.
3321 */
3322
3323 break;
3324 }
3325 }
3326 }
3327
3328 if (rv) {
3329
3330 /*
3331 * Get back an exclusive (write) lock.
3332 */
3333
3334 vm_map_upgrade(map);
3335 vm_map_unbusy(map);
3336
3337 #ifdef DIAGNOSTIC
3338 if (timestamp_save != map->timestamp)
3339 panic("uvm_map_pageable_all: stale map");
3340 #endif
3341
3342 /*
3343 * first drop the wiring count on all the entries
3344 * which haven't actually been wired yet.
3345 *
3346 * Skip VM_PROT_NONE entries like we did above.
3347 */
3348
3349 failed_entry = entry;
3350 for (/* nothing */; entry != &map->header;
3351 entry = entry->next) {
3352 if (entry->protection == VM_PROT_NONE)
3353 continue;
3354 entry->wired_count--;
3355 }
3356
3357 /*
3358 * now, unwire all the entries that were successfully
3359 * wired above.
3360 *
3361 * Skip VM_PROT_NONE entries like we did above.
3362 */
3363
3364 for (entry = map->header.next; entry != failed_entry;
3365 entry = entry->next) {
3366 if (entry->protection == VM_PROT_NONE)
3367 continue;
3368 entry->wired_count--;
3369 if (VM_MAPENT_ISWIRED(entry))
3370 uvm_map_entry_unwire(map, entry);
3371 }
3372 vm_map_unlock(map);
3373 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
3374 return (rv);
3375 }
3376
3377 /* We are holding a read lock here. */
3378 vm_map_unbusy(map);
3379 vm_map_unlock_read(map);
3380
3381 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3382 return 0;
3383 }
3384
3385 /*
3386 * uvm_map_clean: clean out a map range
3387 *
3388 * => valid flags:
3389 * if (flags & PGO_CLEANIT): dirty pages are cleaned first
3390 * if (flags & PGO_SYNCIO): dirty pages are written synchronously
3391 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
3392 * if (flags & PGO_FREE): any cached pages are freed after clean
3393 * => returns an error if any part of the specified range isn't mapped
3394 * => never a need to flush amap layer since the anonymous memory has
3395 * no permanent home, but may deactivate pages there
3396 * => called from sys_msync() and sys_madvise()
3397 * => caller must not write-lock map (read OK).
3398 * => we may sleep while cleaning if SYNCIO [with map read-locked]
3399 */
3400
3401 int
3402 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
3403 {
3404 struct vm_map_entry *current, *entry;
3405 struct uvm_object *uobj;
3406 struct vm_amap *amap;
3407 struct vm_anon *anon;
3408 struct vm_page *pg;
3409 vaddr_t offset;
3410 vsize_t size;
3411 voff_t uoff;
3412 int error, refs;
3413 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
3414
3415 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
3416 map, start, end, flags);
3417 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
3418 (PGO_FREE|PGO_DEACTIVATE));
3419
3420 vm_map_lock_read(map);
3421 VM_MAP_RANGE_CHECK(map, start, end);
3422 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
3423 vm_map_unlock_read(map);
3424 return EFAULT;
3425 }
3426
3427 /*
3428 * Make a first pass to check for holes and wiring problems.
3429 */
3430
3431 for (current = entry; current->start < end; current = current->next) {
3432 if (UVM_ET_ISSUBMAP(current)) {
3433 vm_map_unlock_read(map);
3434 return EINVAL;
3435 }
3436 if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) {
3437 vm_map_unlock_read(map);
3438 return EBUSY;
3439 }
3440 if (end <= current->end) {
3441 break;
3442 }
3443 if (current->end != current->next->start) {
3444 vm_map_unlock_read(map);
3445 return EFAULT;
3446 }
3447 }
3448
3449 error = 0;
3450 for (current = entry; start < end; current = current->next) {
3451 amap = current->aref.ar_amap; /* top layer */
3452 uobj = current->object.uvm_obj; /* bottom layer */
3453 KASSERT(start >= current->start);
3454
3455 /*
3456 * No amap cleaning necessary if:
3457 *
3458 * (1) There's no amap.
3459 *
3460 * (2) We're not deactivating or freeing pages.
3461 */
3462
3463 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
3464 goto flush_object;
3465
3466 amap_lock(amap);
3467 offset = start - current->start;
3468 size = MIN(end, current->end) - start;
3469 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
3470 anon = amap_lookup(¤t->aref, offset);
3471 if (anon == NULL)
3472 continue;
3473
3474 simple_lock(&anon->an_lock);
3475 pg = anon->u.an_page;
3476 if (pg == NULL) {
3477 simple_unlock(&anon->an_lock);
3478 continue;
3479 }
3480
3481 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
3482
3483 /*
3484 * In these first 3 cases, we just deactivate the page.
3485 */
3486
3487 case PGO_CLEANIT|PGO_FREE:
3488 case PGO_CLEANIT|PGO_DEACTIVATE:
3489 case PGO_DEACTIVATE:
3490 deactivate_it:
3491 /*
3492 * skip the page if it's loaned or wired,
3493 * since it shouldn't be on a paging queue
3494 * at all in these cases.
3495 */
3496
3497 uvm_lock_pageq();
3498 if (pg->loan_count != 0 ||
3499 pg->wire_count != 0) {
3500 uvm_unlock_pageq();
3501 simple_unlock(&anon->an_lock);
3502 continue;
3503 }
3504 KASSERT(pg->uanon == anon);
3505 pmap_clear_reference(pg);
3506 uvm_pagedeactivate(pg);
3507 uvm_unlock_pageq();
3508 simple_unlock(&anon->an_lock);
3509 continue;
3510
3511 case PGO_FREE:
3512
3513 /*
3514 * If there are multiple references to
3515 * the amap, just deactivate the page.
3516 */
3517
3518 if (amap_refs(amap) > 1)
3519 goto deactivate_it;
3520
3521 /* skip the page if it's wired */
3522 if (pg->wire_count != 0) {
3523 simple_unlock(&anon->an_lock);
3524 continue;
3525 }
3526 amap_unadd(¤t->aref, offset);
3527 refs = --anon->an_ref;
3528 simple_unlock(&anon->an_lock);
3529 if (refs == 0)
3530 uvm_anfree(anon);
3531 continue;
3532 }
3533 }
3534 amap_unlock(amap);
3535
3536 flush_object:
3537 /*
3538 * flush pages if we've got a valid backing object.
3539 * note that we must always clean object pages before
3540 * freeing them since otherwise we could reveal stale
3541 * data from files.
3542 */
3543
3544 uoff = current->offset + (start - current->start);
3545 size = MIN(end, current->end) - start;
3546 if (uobj != NULL) {
3547 simple_lock(&uobj->vmobjlock);
3548 if (uobj->pgops->pgo_put != NULL)
3549 error = (uobj->pgops->pgo_put)(uobj, uoff,
3550 uoff + size, flags | PGO_CLEANIT);
3551 else
3552 error = 0;
3553 }
3554 start += size;
3555 }
3556 vm_map_unlock_read(map);
3557 return (error);
3558 }
3559
3560
3561 /*
3562 * uvm_map_checkprot: check protection in map
3563 *
3564 * => must allow specified protection in a fully allocated region.
3565 * => map must be read or write locked by caller.
3566 */
3567
3568 boolean_t
3569 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end,
3570 vm_prot_t protection)
3571 {
3572 struct vm_map_entry *entry;
3573 struct vm_map_entry *tmp_entry;
3574
3575 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
3576 return (FALSE);
3577 }
3578 entry = tmp_entry;
3579 while (start < end) {
3580 if (entry == &map->header) {
3581 return (FALSE);
3582 }
3583
3584 /*
3585 * no holes allowed
3586 */
3587
3588 if (start < entry->start) {
3589 return (FALSE);
3590 }
3591
3592 /*
3593 * check protection associated with entry
3594 */
3595
3596 if ((entry->protection & protection) != protection) {
3597 return (FALSE);
3598 }
3599 start = entry->end;
3600 entry = entry->next;
3601 }
3602 return (TRUE);
3603 }
3604
3605 /*
3606 * uvmspace_alloc: allocate a vmspace structure.
3607 *
3608 * - structure includes vm_map and pmap
3609 * - XXX: no locking on this structure
3610 * - refcnt set to 1, rest must be init'd by caller
3611 */
3612 struct vmspace *
3613 uvmspace_alloc(vaddr_t min, vaddr_t max)
3614 {
3615 struct vmspace *vm;
3616 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
3617
3618 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
3619 uvmspace_init(vm, NULL, min, max);
3620 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
3621 return (vm);
3622 }
3623
3624 /*
3625 * uvmspace_init: initialize a vmspace structure.
3626 *
3627 * - XXX: no locking on this structure
3628 * - refcnt set to 1, rest must be init'd by caller
3629 */
3630 void
3631 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t min, vaddr_t max)
3632 {
3633 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
3634
3635 memset(vm, 0, sizeof(*vm));
3636 uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE
3637 #ifdef __USING_TOPDOWN_VM
3638 | VM_MAP_TOPDOWN
3639 #endif
3640 );
3641 if (pmap)
3642 pmap_reference(pmap);
3643 else
3644 pmap = pmap_create();
3645 vm->vm_map.pmap = pmap;
3646 vm->vm_refcnt = 1;
3647 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3648 }
3649
3650 /*
3651 * uvmspace_share: share a vmspace between two processes
3652 *
3653 * - used for vfork, threads(?)
3654 */
3655
3656 void
3657 uvmspace_share(struct proc *p1, struct proc *p2)
3658 {
3659 struct simplelock *slock = &p1->p_vmspace->vm_map.ref_lock;
3660
3661 p2->p_vmspace = p1->p_vmspace;
3662 simple_lock(slock);
3663 p1->p_vmspace->vm_refcnt++;
3664 simple_unlock(slock);
3665 }
3666
3667 /*
3668 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
3669 *
3670 * - XXX: no locking on vmspace
3671 */
3672
3673 void
3674 uvmspace_unshare(struct lwp *l)
3675 {
3676 struct proc *p = l->l_proc;
3677 struct vmspace *nvm, *ovm = p->p_vmspace;
3678
3679 if (ovm->vm_refcnt == 1)
3680 /* nothing to do: vmspace isn't shared in the first place */
3681 return;
3682
3683 /* make a new vmspace, still holding old one */
3684 nvm = uvmspace_fork(ovm);
3685
3686 pmap_deactivate(l); /* unbind old vmspace */
3687 p->p_vmspace = nvm;
3688 pmap_activate(l); /* switch to new vmspace */
3689
3690 uvmspace_free(ovm); /* drop reference to old vmspace */
3691 }
3692
3693 /*
3694 * uvmspace_exec: the process wants to exec a new program
3695 */
3696
3697 void
3698 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end)
3699 {
3700 struct proc *p = l->l_proc;
3701 struct vmspace *nvm, *ovm = p->p_vmspace;
3702 struct vm_map *map = &ovm->vm_map;
3703
3704 #ifdef __sparc__
3705 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
3706 kill_user_windows(l); /* before stack addresses go away */
3707 #endif
3708
3709 /*
3710 * see if more than one process is using this vmspace...
3711 */
3712
3713 if (ovm->vm_refcnt == 1) {
3714
3715 /*
3716 * if p is the only process using its vmspace then we can safely
3717 * recycle that vmspace for the program that is being exec'd.
3718 */
3719
3720 #ifdef SYSVSHM
3721 /*
3722 * SYSV SHM semantics require us to kill all segments on an exec
3723 */
3724
3725 if (ovm->vm_shm)
3726 shmexit(ovm);
3727 #endif
3728
3729 /*
3730 * POSIX 1003.1b -- "lock future mappings" is revoked
3731 * when a process execs another program image.
3732 */
3733
3734 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3735
3736 /*
3737 * now unmap the old program
3738 */
3739
3740 pmap_remove_all(map->pmap);
3741 uvm_unmap(map, vm_map_min(map), vm_map_max(map));
3742 KASSERT(map->header.prev == &map->header);
3743 KASSERT(map->nentries == 0);
3744
3745 /*
3746 * resize the map
3747 */
3748
3749 vm_map_setmin(map, start);
3750 vm_map_setmax(map, end);
3751 } else {
3752
3753 /*
3754 * p's vmspace is being shared, so we can't reuse it for p since
3755 * it is still being used for others. allocate a new vmspace
3756 * for p
3757 */
3758
3759 nvm = uvmspace_alloc(start, end);
3760
3761 /*
3762 * install new vmspace and drop our ref to the old one.
3763 */
3764
3765 pmap_deactivate(l);
3766 p->p_vmspace = nvm;
3767 pmap_activate(l);
3768
3769 uvmspace_free(ovm);
3770 }
3771 }
3772
3773 /*
3774 * uvmspace_free: free a vmspace data structure
3775 */
3776
3777 void
3778 uvmspace_free(struct vmspace *vm)
3779 {
3780 struct vm_map_entry *dead_entries;
3781 struct vm_map *map = &vm->vm_map;
3782 int n;
3783
3784 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
3785
3786 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
3787 simple_lock(&map->ref_lock);
3788 n = --vm->vm_refcnt;
3789 simple_unlock(&map->ref_lock);
3790 if (n > 0)
3791 return;
3792
3793 /*
3794 * at this point, there should be no other references to the map.
3795 * delete all of the mappings, then destroy the pmap.
3796 */
3797
3798 map->flags |= VM_MAP_DYING;
3799 pmap_remove_all(map->pmap);
3800 #ifdef SYSVSHM
3801 /* Get rid of any SYSV shared memory segments. */
3802 if (vm->vm_shm != NULL)
3803 shmexit(vm);
3804 #endif
3805 if (map->nentries) {
3806 uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map),
3807 &dead_entries, NULL);
3808 if (dead_entries != NULL)
3809 uvm_unmap_detach(dead_entries, 0);
3810 }
3811 KASSERT(map->nentries == 0);
3812 KASSERT(map->size == 0);
3813 pmap_destroy(map->pmap);
3814 pool_put(&uvm_vmspace_pool, vm);
3815 }
3816
3817 /*
3818 * F O R K - m a i n e n t r y p o i n t
3819 */
3820 /*
3821 * uvmspace_fork: fork a process' main map
3822 *
3823 * => create a new vmspace for child process from parent.
3824 * => parent's map must not be locked.
3825 */
3826
3827 struct vmspace *
3828 uvmspace_fork(struct vmspace *vm1)
3829 {
3830 struct vmspace *vm2;
3831 struct vm_map *old_map = &vm1->vm_map;
3832 struct vm_map *new_map;
3833 struct vm_map_entry *old_entry;
3834 struct vm_map_entry *new_entry;
3835 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
3836
3837 vm_map_lock(old_map);
3838
3839 vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map));
3840 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
3841 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
3842 new_map = &vm2->vm_map; /* XXX */
3843
3844 old_entry = old_map->header.next;
3845 new_map->size = old_map->size;
3846
3847 /*
3848 * go entry-by-entry
3849 */
3850
3851 while (old_entry != &old_map->header) {
3852
3853 /*
3854 * first, some sanity checks on the old entry
3855 */
3856
3857 KASSERT(!UVM_ET_ISSUBMAP(old_entry));
3858 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
3859 !UVM_ET_ISNEEDSCOPY(old_entry));
3860
3861 switch (old_entry->inheritance) {
3862 case MAP_INHERIT_NONE:
3863
3864 /*
3865 * drop the mapping, modify size
3866 */
3867 new_map->size -= old_entry->end - old_entry->start;
3868 break;
3869
3870 case MAP_INHERIT_SHARE:
3871
3872 /*
3873 * share the mapping: this means we want the old and
3874 * new entries to share amaps and backing objects.
3875 */
3876 /*
3877 * if the old_entry needs a new amap (due to prev fork)
3878 * then we need to allocate it now so that we have
3879 * something we own to share with the new_entry. [in
3880 * other words, we need to clear needs_copy]
3881 */
3882
3883 if (UVM_ET_ISNEEDSCOPY(old_entry)) {
3884 /* get our own amap, clears needs_copy */
3885 amap_copy(old_map, old_entry, M_WAITOK, FALSE,
3886 0, 0);
3887 /* XXXCDC: WAITOK??? */
3888 }
3889
3890 new_entry = uvm_mapent_alloc(new_map, 0);
3891 /* old_entry -> new_entry */
3892 uvm_mapent_copy(old_entry, new_entry);
3893
3894 /* new pmap has nothing wired in it */
3895 new_entry->wired_count = 0;
3896
3897 /*
3898 * gain reference to object backing the map (can't
3899 * be a submap, already checked this case).
3900 */
3901
3902 if (new_entry->aref.ar_amap)
3903 uvm_map_reference_amap(new_entry, AMAP_SHARED);
3904
3905 if (new_entry->object.uvm_obj &&
3906 new_entry->object.uvm_obj->pgops->pgo_reference)
3907 new_entry->object.uvm_obj->
3908 pgops->pgo_reference(
3909 new_entry->object.uvm_obj);
3910
3911 /* insert entry at end of new_map's entry list */
3912 uvm_map_entry_link(new_map, new_map->header.prev,
3913 new_entry);
3914
3915 break;
3916
3917 case MAP_INHERIT_COPY:
3918
3919 /*
3920 * copy-on-write the mapping (using mmap's
3921 * MAP_PRIVATE semantics)
3922 *
3923 * allocate new_entry, adjust reference counts.
3924 * (note that new references are read-only).
3925 */
3926
3927 new_entry = uvm_mapent_alloc(new_map, 0);
3928 /* old_entry -> new_entry */
3929 uvm_mapent_copy(old_entry, new_entry);
3930
3931 if (new_entry->aref.ar_amap)
3932 uvm_map_reference_amap(new_entry, 0);
3933
3934 if (new_entry->object.uvm_obj &&
3935 new_entry->object.uvm_obj->pgops->pgo_reference)
3936 new_entry->object.uvm_obj->pgops->pgo_reference
3937 (new_entry->object.uvm_obj);
3938
3939 /* new pmap has nothing wired in it */
3940 new_entry->wired_count = 0;
3941
3942 new_entry->etype |=
3943 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3944 uvm_map_entry_link(new_map, new_map->header.prev,
3945 new_entry);
3946
3947 /*
3948 * the new entry will need an amap. it will either
3949 * need to be copied from the old entry or created
3950 * from scratch (if the old entry does not have an
3951 * amap). can we defer this process until later
3952 * (by setting "needs_copy") or do we need to copy
3953 * the amap now?
3954 *
3955 * we must copy the amap now if any of the following
3956 * conditions hold:
3957 * 1. the old entry has an amap and that amap is
3958 * being shared. this means that the old (parent)
3959 * process is sharing the amap with another
3960 * process. if we do not clear needs_copy here
3961 * we will end up in a situation where both the
3962 * parent and child process are refering to the
3963 * same amap with "needs_copy" set. if the
3964 * parent write-faults, the fault routine will
3965 * clear "needs_copy" in the parent by allocating
3966 * a new amap. this is wrong because the
3967 * parent is supposed to be sharing the old amap
3968 * and the new amap will break that.
3969 *
3970 * 2. if the old entry has an amap and a non-zero
3971 * wire count then we are going to have to call
3972 * amap_cow_now to avoid page faults in the
3973 * parent process. since amap_cow_now requires
3974 * "needs_copy" to be clear we might as well
3975 * clear it here as well.
3976 *
3977 */
3978
3979 if (old_entry->aref.ar_amap != NULL) {
3980 if ((amap_flags(old_entry->aref.ar_amap) &
3981 AMAP_SHARED) != 0 ||
3982 VM_MAPENT_ISWIRED(old_entry)) {
3983
3984 amap_copy(new_map, new_entry, M_WAITOK,
3985 FALSE, 0, 0);
3986 /* XXXCDC: M_WAITOK ... ok? */
3987 }
3988 }
3989
3990 /*
3991 * if the parent's entry is wired down, then the
3992 * parent process does not want page faults on
3993 * access to that memory. this means that we
3994 * cannot do copy-on-write because we can't write
3995 * protect the old entry. in this case we
3996 * resolve all copy-on-write faults now, using
3997 * amap_cow_now. note that we have already
3998 * allocated any needed amap (above).
3999 */
4000
4001 if (VM_MAPENT_ISWIRED(old_entry)) {
4002
4003 /*
4004 * resolve all copy-on-write faults now
4005 * (note that there is nothing to do if
4006 * the old mapping does not have an amap).
4007 */
4008 if (old_entry->aref.ar_amap)
4009 amap_cow_now(new_map, new_entry);
4010
4011 } else {
4012
4013 /*
4014 * setup mappings to trigger copy-on-write faults
4015 * we must write-protect the parent if it has
4016 * an amap and it is not already "needs_copy"...
4017 * if it is already "needs_copy" then the parent
4018 * has already been write-protected by a previous
4019 * fork operation.
4020 */
4021
4022 if (old_entry->aref.ar_amap &&
4023 !UVM_ET_ISNEEDSCOPY(old_entry)) {
4024 if (old_entry->max_protection & VM_PROT_WRITE) {
4025 pmap_protect(old_map->pmap,
4026 old_entry->start,
4027 old_entry->end,
4028 old_entry->protection &
4029 ~VM_PROT_WRITE);
4030 pmap_update(old_map->pmap);
4031 }
4032 old_entry->etype |= UVM_ET_NEEDSCOPY;
4033 }
4034 }
4035 break;
4036 } /* end of switch statement */
4037 old_entry = old_entry->next;
4038 }
4039
4040 vm_map_unlock(old_map);
4041
4042 #ifdef SYSVSHM
4043 if (vm1->vm_shm)
4044 shmfork(vm1, vm2);
4045 #endif
4046
4047 #ifdef PMAP_FORK
4048 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
4049 #endif
4050
4051 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
4052 return (vm2);
4053 }
4054
4055
4056 /*
4057 * in-kernel map entry allocation.
4058 */
4059
4060 int ukh_alloc, ukh_free;
4061 int uke_alloc, uke_free;
4062
4063 struct uvm_kmapent_hdr {
4064 LIST_ENTRY(uvm_kmapent_hdr) ukh_listq;
4065 int ukh_nused;
4066 struct vm_map_entry *ukh_freelist;
4067 struct vm_map *ukh_map;
4068 struct vm_map_entry ukh_entries[0];
4069 };
4070
4071 #define UVM_KMAPENT_CHUNK \
4072 ((PAGE_SIZE - sizeof(struct uvm_kmapent_hdr)) \
4073 / sizeof(struct vm_map_entry))
4074
4075 #define UVM_KHDR_FIND(entry) \
4076 ((struct uvm_kmapent_hdr *)(((vaddr_t)entry) & ~PAGE_MASK))
4077
4078 static __inline struct vm_map_entry *uvm_kmapent_get(struct uvm_kmapent_hdr *);
4079 static __inline void uvm_kmapent_put(struct uvm_kmapent_hdr *,
4080 struct vm_map_entry *);
4081
4082 static __inline struct vm_map *
4083 uvm_kmapent_map(struct vm_map_entry *entry)
4084 {
4085 const struct uvm_kmapent_hdr *ukh;
4086
4087 ukh = UVM_KHDR_FIND(entry);
4088 return ukh->ukh_map;
4089 }
4090
4091 static __inline struct vm_map_entry *
4092 uvm_kmapent_get(struct uvm_kmapent_hdr *ukh)
4093 {
4094 struct vm_map_entry *entry;
4095
4096 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4097 KASSERT(ukh->ukh_nused >= 0);
4098
4099 entry = ukh->ukh_freelist;
4100 if (entry) {
4101 KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
4102 == UVM_MAP_KERNEL);
4103 ukh->ukh_freelist = entry->next;
4104 ukh->ukh_nused++;
4105 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4106 } else {
4107 KASSERT(ukh->ukh_nused == UVM_KMAPENT_CHUNK);
4108 }
4109
4110 return entry;
4111 }
4112
4113 static __inline void
4114 uvm_kmapent_put(struct uvm_kmapent_hdr *ukh, struct vm_map_entry *entry)
4115 {
4116
4117 KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
4118 == UVM_MAP_KERNEL);
4119 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4120 KASSERT(ukh->ukh_nused > 0);
4121 KASSERT(ukh->ukh_freelist != NULL ||
4122 ukh->ukh_nused == UVM_KMAPENT_CHUNK);
4123 KASSERT(ukh->ukh_freelist == NULL ||
4124 ukh->ukh_nused < UVM_KMAPENT_CHUNK);
4125
4126 ukh->ukh_nused--;
4127 entry->next = ukh->ukh_freelist;
4128 ukh->ukh_freelist = entry;
4129 }
4130
4131 /*
4132 * uvm_kmapent_alloc: allocate a map entry for in-kernel map
4133 */
4134
4135 static struct vm_map_entry *
4136 uvm_kmapent_alloc(struct vm_map *map, int flags)
4137 {
4138 struct vm_page *pg;
4139 struct uvm_map_args args;
4140 struct uvm_kmapent_hdr *ukh;
4141 struct vm_map_entry *entry;
4142 uvm_flag_t mapflags = UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
4143 UVM_INH_NONE, UVM_ADV_RANDOM, flags | UVM_FLAG_NOMERGE);
4144 vaddr_t va;
4145 int error;
4146 int i;
4147 int s;
4148
4149 KDASSERT(UVM_KMAPENT_CHUNK > 2);
4150 KDASSERT(kernel_map != NULL);
4151 KASSERT(vm_map_pmap(map) == pmap_kernel());
4152
4153 uke_alloc++;
4154 entry = NULL;
4155 again:
4156 /*
4157 * try to grab an entry from freelist.
4158 */
4159 s = splvm();
4160 simple_lock(&uvm.kentry_lock);
4161 ukh = LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free);
4162 if (ukh) {
4163 entry = uvm_kmapent_get(ukh);
4164 if (ukh->ukh_nused == UVM_KMAPENT_CHUNK)
4165 LIST_REMOVE(ukh, ukh_listq);
4166 }
4167 simple_unlock(&uvm.kentry_lock);
4168 splx(s);
4169
4170 if (entry)
4171 return entry;
4172
4173 /*
4174 * there's no free entry for this vm_map.
4175 * now we need to allocate some vm_map_entry.
4176 * for simplicity, always allocate one page chunk of them at once.
4177 */
4178
4179 pg = uvm_pagealloc(NULL, 0, NULL, 0);
4180 if (__predict_false(pg == NULL)) {
4181 if (flags & UVM_FLAG_NOWAIT)
4182 return NULL;
4183 uvm_wait("kme_alloc");
4184 goto again;
4185 }
4186
4187 error = uvm_map_prepare(map, 0, PAGE_SIZE, NULL, 0, 0, mapflags, &args);
4188 if (error) {
4189 uvm_pagefree(pg);
4190 return NULL;
4191 }
4192
4193 va = args.uma_start;
4194
4195 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
4196 pmap_update(vm_map_pmap(map));
4197
4198 ukh = (void *)va;
4199
4200 /*
4201 * use the first entry for ukh itsself.
4202 */
4203
4204 entry = &ukh->ukh_entries[0];
4205 entry->flags = UVM_MAP_KERNEL | UVM_MAP_KMAPENT;
4206 error = uvm_map_enter(map, &args, entry);
4207 KASSERT(error == 0);
4208
4209 ukh->ukh_nused = UVM_KMAPENT_CHUNK;
4210 ukh->ukh_map = map;
4211 ukh->ukh_freelist = NULL;
4212 for (i = UVM_KMAPENT_CHUNK - 1; i >= 2; i--) {
4213 struct vm_map_entry *entry = &ukh->ukh_entries[i];
4214
4215 entry->flags = UVM_MAP_KERNEL;
4216 uvm_kmapent_put(ukh, entry);
4217 }
4218 KASSERT(ukh->ukh_nused == 2);
4219
4220 s = splvm();
4221 simple_lock(&uvm.kentry_lock);
4222 LIST_INSERT_HEAD(&vm_map_to_kernel(map)->vmk_kentry_free,
4223 ukh, ukh_listq);
4224 simple_unlock(&uvm.kentry_lock);
4225 splx(s);
4226
4227 /*
4228 * return second entry.
4229 */
4230
4231 entry = &ukh->ukh_entries[1];
4232 entry->flags = UVM_MAP_KERNEL;
4233 ukh_alloc++;
4234 return entry;
4235 }
4236
4237 /*
4238 * uvm_mapent_free: free map entry for in-kernel map
4239 */
4240
4241 static void
4242 uvm_kmapent_free(struct vm_map_entry *entry)
4243 {
4244 struct uvm_kmapent_hdr *ukh;
4245 struct vm_page *pg;
4246 struct vm_map *map;
4247 struct pmap *pmap;
4248 vaddr_t va;
4249 paddr_t pa;
4250 struct vm_map_entry *deadentry;
4251 int s;
4252
4253 uke_free++;
4254 ukh = UVM_KHDR_FIND(entry);
4255 map = ukh->ukh_map;
4256
4257 s = splvm();
4258 simple_lock(&uvm.kentry_lock);
4259 uvm_kmapent_put(ukh, entry);
4260 if (ukh->ukh_nused > 1) {
4261 if (ukh->ukh_nused == UVM_KMAPENT_CHUNK - 1)
4262 LIST_INSERT_HEAD(
4263 &vm_map_to_kernel(map)->vmk_kentry_free,
4264 ukh, ukh_listq);
4265 simple_unlock(&uvm.kentry_lock);
4266 splx(s);
4267 return;
4268 }
4269
4270 /*
4271 * now we can free this ukh.
4272 *
4273 * however, keep an empty ukh to avoid ping-pong.
4274 */
4275
4276 if (LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free) == ukh &&
4277 LIST_NEXT(ukh, ukh_listq) == NULL) {
4278 simple_unlock(&uvm.kentry_lock);
4279 splx(s);
4280 return;
4281 }
4282 LIST_REMOVE(ukh, ukh_listq);
4283 simple_unlock(&uvm.kentry_lock);
4284 splx(s);
4285
4286 KASSERT(ukh->ukh_nused == 1);
4287
4288 /*
4289 * remove map entry for ukh itsself.
4290 */
4291
4292 va = (vaddr_t)ukh;
4293 KASSERT((va & PAGE_MASK) == 0);
4294 uvm_unmap_remove(map, va, va + PAGE_SIZE, &deadentry, NULL);
4295 KASSERT(deadentry->flags & UVM_MAP_KERNEL);
4296 KASSERT(deadentry->flags & UVM_MAP_KMAPENT);
4297 KASSERT(deadentry->next == NULL);
4298 KASSERT(deadentry == &ukh->ukh_entries[0]);
4299
4300 /*
4301 * unmap the page from pmap and free it.
4302 */
4303
4304 pmap = vm_map_pmap(map);
4305 KASSERT(pmap == pmap_kernel());
4306 if (!pmap_extract(pmap, va, &pa))
4307 panic("%s: no mapping", __func__);
4308 pmap_kremove(va, PAGE_SIZE);
4309 pg = PHYS_TO_VM_PAGE(pa);
4310 uvm_pagefree(pg);
4311 ukh_free++;
4312 }
4313
4314 /*
4315 * map entry reservation
4316 */
4317
4318 /*
4319 * uvm_mapent_reserve: reserve map entries for clipping before locking map.
4320 *
4321 * => needed when unmapping entries allocated without UVM_FLAG_QUANTUM.
4322 * => caller shouldn't hold map locked.
4323 */
4324 int
4325 uvm_mapent_reserve(struct vm_map *map, struct uvm_mapent_reservation *umr,
4326 int nentries, int flags)
4327 {
4328
4329 umr->umr_nentries = 0;
4330
4331 if ((flags & UVM_FLAG_QUANTUM) != 0)
4332 return 0;
4333
4334 if (!VM_MAP_USE_KMAPENT(map))
4335 return 0;
4336
4337 while (nentries--) {
4338 struct vm_map_entry *ent;
4339 ent = uvm_kmapent_alloc(map, flags);
4340 if (!ent) {
4341 uvm_mapent_unreserve(map, umr);
4342 return ENOMEM;
4343 }
4344 UMR_PUTENTRY(umr, ent);
4345 }
4346
4347 return 0;
4348 }
4349
4350 /*
4351 * uvm_mapent_unreserve:
4352 *
4353 * => caller shouldn't hold map locked.
4354 * => never fail or sleep.
4355 */
4356 void
4357 uvm_mapent_unreserve(struct vm_map *map, struct uvm_mapent_reservation *umr)
4358 {
4359
4360 while (!UMR_EMPTY(umr))
4361 uvm_kmapent_free(UMR_GETENTRY(umr));
4362 }
4363
4364 #if defined(DDB)
4365
4366 /*
4367 * DDB hooks
4368 */
4369
4370 /*
4371 * uvm_map_printit: actually prints the map
4372 */
4373
4374 void
4375 uvm_map_printit(struct vm_map *map, boolean_t full,
4376 void (*pr)(const char *, ...))
4377 {
4378 struct vm_map_entry *entry;
4379
4380 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, vm_map_min(map),
4381 vm_map_max(map));
4382 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
4383 map->nentries, map->size, map->ref_count, map->timestamp,
4384 map->flags);
4385 (*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap,
4386 pmap_resident_count(map->pmap), pmap_wired_count(map->pmap));
4387 if (!full)
4388 return;
4389 for (entry = map->header.next; entry != &map->header;
4390 entry = entry->next) {
4391 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
4392 entry, entry->start, entry->end, entry->object.uvm_obj,
4393 (long long)entry->offset, entry->aref.ar_amap,
4394 entry->aref.ar_pageoff);
4395 (*pr)(
4396 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
4397 "wc=%d, adv=%d\n",
4398 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
4399 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
4400 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
4401 entry->protection, entry->max_protection,
4402 entry->inheritance, entry->wired_count, entry->advice);
4403 }
4404 }
4405
4406 /*
4407 * uvm_object_printit: actually prints the object
4408 */
4409
4410 void
4411 uvm_object_printit(struct uvm_object *uobj, boolean_t full,
4412 void (*pr)(const char *, ...))
4413 {
4414 struct vm_page *pg;
4415 int cnt = 0;
4416
4417 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
4418 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
4419 if (UVM_OBJ_IS_KERN_OBJECT(uobj))
4420 (*pr)("refs=<SYSTEM>\n");
4421 else
4422 (*pr)("refs=%d\n", uobj->uo_refs);
4423
4424 if (!full) {
4425 return;
4426 }
4427 (*pr)(" PAGES <pg,offset>:\n ");
4428 TAILQ_FOREACH(pg, &uobj->memq, listq) {
4429 cnt++;
4430 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
4431 if ((cnt % 3) == 0) {
4432 (*pr)("\n ");
4433 }
4434 }
4435 if ((cnt % 3) != 0) {
4436 (*pr)("\n");
4437 }
4438 }
4439
4440 /*
4441 * uvm_page_printit: actually print the page
4442 */
4443
4444 static const char page_flagbits[] =
4445 "\2\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY"
4446 "\11ZERO\15PAGER1";
4447 static const char page_pqflagbits[] =
4448 "\2\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ";
4449
4450 void
4451 uvm_page_printit(struct vm_page *pg, boolean_t full,
4452 void (*pr)(const char *, ...))
4453 {
4454 struct vm_page *tpg;
4455 struct uvm_object *uobj;
4456 struct pglist *pgl;
4457 char pgbuf[128];
4458 char pqbuf[128];
4459
4460 (*pr)("PAGE %p:\n", pg);
4461 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
4462 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
4463 (*pr)(" flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
4464 pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg));
4465 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
4466 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
4467 #if defined(UVM_PAGE_TRKOWN)
4468 if (pg->flags & PG_BUSY)
4469 (*pr)(" owning process = %d, tag=%s\n",
4470 pg->owner, pg->owner_tag);
4471 else
4472 (*pr)(" page not busy, no owner\n");
4473 #else
4474 (*pr)(" [page ownership tracking disabled]\n");
4475 #endif
4476
4477 if (!full)
4478 return;
4479
4480 /* cross-verify object/anon */
4481 if ((pg->pqflags & PQ_FREE) == 0) {
4482 if (pg->pqflags & PQ_ANON) {
4483 if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
4484 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n",
4485 (pg->uanon) ? pg->uanon->u.an_page : NULL);
4486 else
4487 (*pr)(" anon backpointer is OK\n");
4488 } else {
4489 uobj = pg->uobject;
4490 if (uobj) {
4491 (*pr)(" checking object list\n");
4492 TAILQ_FOREACH(tpg, &uobj->memq, listq) {
4493 if (tpg == pg) {
4494 break;
4495 }
4496 }
4497 if (tpg)
4498 (*pr)(" page found on object list\n");
4499 else
4500 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
4501 }
4502 }
4503 }
4504
4505 /* cross-verify page queue */
4506 if (pg->pqflags & PQ_FREE) {
4507 int fl = uvm_page_lookup_freelist(pg);
4508 int color = VM_PGCOLOR_BUCKET(pg);
4509 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
4510 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
4511 } else if (pg->pqflags & PQ_INACTIVE) {
4512 pgl = &uvm.page_inactive;
4513 } else if (pg->pqflags & PQ_ACTIVE) {
4514 pgl = &uvm.page_active;
4515 } else {
4516 pgl = NULL;
4517 }
4518
4519 if (pgl) {
4520 (*pr)(" checking pageq list\n");
4521 TAILQ_FOREACH(tpg, pgl, pageq) {
4522 if (tpg == pg) {
4523 break;
4524 }
4525 }
4526 if (tpg)
4527 (*pr)(" page found on pageq list\n");
4528 else
4529 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
4530 }
4531 }
4532 #endif
Cache object: 197cca8e404ffe463b06ae487b2aad48
|