uvm_map.c revision 1.118 1 /* $NetBSD: uvm_map.c,v 1.118 2002/03/08 20:48:47 thorpej 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 (at) 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 1.118 2002/03/08 20:48:47 thorpej Exp $");
75
76 #include "opt_ddb.h"
77 #include "opt_uvmhist.h"
78 #include "opt_sysv.h"
79
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/mman.h>
83 #include <sys/proc.h>
84 #include <sys/malloc.h>
85 #include <sys/pool.h>
86 #include <sys/kernel.h>
87 #include <sys/mount.h>
88 #include <sys/vnode.h>
89
90 #ifdef SYSVSHM
91 #include <sys/shm.h>
92 #endif
93
94 #define UVM_MAP
95 #include <uvm/uvm.h>
96
97 #ifdef DDB
98 #include <uvm/uvm_ddb.h>
99 #endif
100
101 extern struct vm_map *pager_map;
102
103 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
104 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
105 const char vmmapbsy[] = "vmmapbsy";
106
107 /*
108 * pool for vmspace structures.
109 */
110
111 struct pool uvm_vmspace_pool;
112
113 /*
114 * pool for dynamically-allocated map entries.
115 */
116
117 struct pool uvm_map_entry_pool;
118 struct pool uvm_map_entry_kmem_pool;
119
120 #ifdef PMAP_GROWKERNEL
121 /*
122 * This global represents the end of the kernel virtual address
123 * space. If we want to exceed this, we must grow the kernel
124 * virtual address space dynamically.
125 *
126 * Note, this variable is locked by kernel_map's lock.
127 */
128 vaddr_t uvm_maxkaddr;
129 #endif
130
131 /*
132 * macros
133 */
134
135 /*
136 * uvm_map_entry_link: insert entry into a map
137 *
138 * => map must be locked
139 */
140 #define uvm_map_entry_link(map, after_where, entry) do { \
141 (map)->nentries++; \
142 (entry)->prev = (after_where); \
143 (entry)->next = (after_where)->next; \
144 (entry)->prev->next = (entry); \
145 (entry)->next->prev = (entry); \
146 } while (0)
147
148 /*
149 * uvm_map_entry_unlink: remove entry from a map
150 *
151 * => map must be locked
152 */
153 #define uvm_map_entry_unlink(map, entry) do { \
154 (map)->nentries--; \
155 (entry)->next->prev = (entry)->prev; \
156 (entry)->prev->next = (entry)->next; \
157 } while (0)
158
159 /*
160 * SAVE_HINT: saves the specified entry as the hint for future lookups.
161 *
162 * => map need not be locked (protected by hint_lock).
163 */
164 #define SAVE_HINT(map,check,value) do { \
165 simple_lock(&(map)->hint_lock); \
166 if ((map)->hint == (check)) \
167 (map)->hint = (value); \
168 simple_unlock(&(map)->hint_lock); \
169 } while (0)
170
171 /*
172 * VM_MAP_RANGE_CHECK: check and correct range
173 *
174 * => map must at least be read locked
175 */
176
177 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
178 if (start < vm_map_min(map)) \
179 start = vm_map_min(map); \
180 if (end > vm_map_max(map)) \
181 end = vm_map_max(map); \
182 if (start > end) \
183 start = end; \
184 } while (0)
185
186 /*
187 * local prototypes
188 */
189
190 static struct vm_map_entry *uvm_mapent_alloc __P((struct vm_map *));
191 static void uvm_mapent_copy __P((struct vm_map_entry *, struct vm_map_entry *));
192 static void uvm_mapent_free __P((struct vm_map_entry *));
193 static void uvm_map_entry_unwire __P((struct vm_map *, struct vm_map_entry *));
194 static void uvm_map_reference_amap __P((struct vm_map_entry *, int));
195 static void uvm_map_unreference_amap __P((struct vm_map_entry *, int));
196
197 /*
198 * local inlines
199 */
200
201 /*
202 * uvm_mapent_alloc: allocate a map entry
203 */
204
205 static __inline struct vm_map_entry *
206 uvm_mapent_alloc(map)
207 struct vm_map *map;
208 {
209 struct vm_map_entry *me;
210 int s;
211 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
212
213 if (map->flags & VM_MAP_INTRSAFE || cold) {
214 s = splvm();
215 simple_lock(&uvm.kentry_lock);
216 me = uvm.kentry_free;
217 if (me) uvm.kentry_free = me->next;
218 simple_unlock(&uvm.kentry_lock);
219 splx(s);
220 if (me == NULL) {
221 panic("uvm_mapent_alloc: out of static map entries, "
222 "check MAX_KMAPENT (currently %d)",
223 MAX_KMAPENT);
224 }
225 me->flags = UVM_MAP_STATIC;
226 } else if (map == kernel_map) {
227 me = pool_get(&uvm_map_entry_kmem_pool, PR_WAITOK);
228 me->flags = UVM_MAP_KMEM;
229 } else {
230 me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
231 me->flags = 0;
232 }
233
234 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me,
235 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0);
236 return(me);
237 }
238
239 /*
240 * uvm_mapent_free: free map entry
241 */
242
243 static __inline void
244 uvm_mapent_free(me)
245 struct vm_map_entry *me;
246 {
247 int s;
248 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
249
250 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
251 me, me->flags, 0, 0);
252 if (me->flags & UVM_MAP_STATIC) {
253 s = splvm();
254 simple_lock(&uvm.kentry_lock);
255 me->next = uvm.kentry_free;
256 uvm.kentry_free = me;
257 simple_unlock(&uvm.kentry_lock);
258 splx(s);
259 } else if (me->flags & UVM_MAP_KMEM) {
260 pool_put(&uvm_map_entry_kmem_pool, me);
261 } else {
262 pool_put(&uvm_map_entry_pool, me);
263 }
264 }
265
266 /*
267 * uvm_mapent_copy: copy a map entry, preserving flags
268 */
269
270 static __inline void
271 uvm_mapent_copy(src, dst)
272 struct vm_map_entry *src;
273 struct vm_map_entry *dst;
274 {
275 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
276 ((char *)src));
277 }
278
279 /*
280 * uvm_map_entry_unwire: unwire a map entry
281 *
282 * => map should be locked by caller
283 */
284
285 static __inline void
286 uvm_map_entry_unwire(map, entry)
287 struct vm_map *map;
288 struct vm_map_entry *entry;
289 {
290 entry->wired_count = 0;
291 uvm_fault_unwire_locked(map, entry->start, entry->end);
292 }
293
294
295 /*
296 * wrapper for calling amap_ref()
297 */
298 static __inline void
299 uvm_map_reference_amap(entry, flags)
300 struct vm_map_entry *entry;
301 int flags;
302 {
303 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
304 (entry->end - entry->start) >> PAGE_SHIFT, flags);
305 }
306
307
308 /*
309 * wrapper for calling amap_unref()
310 */
311 static __inline void
312 uvm_map_unreference_amap(entry, flags)
313 struct vm_map_entry *entry;
314 int flags;
315 {
316 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
317 (entry->end - entry->start) >> PAGE_SHIFT, flags);
318 }
319
320
321 /*
322 * uvm_map_init: init mapping system at boot time. note that we allocate
323 * and init the static pool of struct vm_map_entry *'s for the kernel here.
324 */
325
326 void
327 uvm_map_init()
328 {
329 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
330 #if defined(UVMHIST)
331 static struct uvm_history_ent maphistbuf[100];
332 static struct uvm_history_ent pdhistbuf[100];
333 #endif
334 int lcv;
335
336 /*
337 * first, init logging system.
338 */
339
340 UVMHIST_FUNC("uvm_map_init");
341 UVMHIST_INIT_STATIC(maphist, maphistbuf);
342 UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
343 UVMHIST_CALLED(maphist);
344 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
345 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0,
346 "# uvm_map() successful calls", 0);
347 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
348 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
349 0);
350 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0);
351 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0);
352
353 /*
354 * now set up static pool of kernel map entrys ...
355 */
356
357 simple_lock_init(&uvm.kentry_lock);
358 uvm.kentry_free = NULL;
359 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
360 kernel_map_entry[lcv].next = uvm.kentry_free;
361 uvm.kentry_free = &kernel_map_entry[lcv];
362 }
363
364 /*
365 * initialize the map-related pools.
366 */
367 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
368 0, 0, 0, "vmsppl", &pool_allocator_nointr);
369 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
370 0, 0, 0, "vmmpepl", &pool_allocator_nointr);
371 pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry),
372 0, 0, 0, "vmmpekpl", NULL);
373 }
374
375 /*
376 * clippers
377 */
378
379 /*
380 * uvm_map_clip_start: ensure that the entry begins at or after
381 * the starting address, if it doesn't we split the entry.
382 *
383 * => caller should use UVM_MAP_CLIP_START macro rather than calling
384 * this directly
385 * => map must be locked by caller
386 */
387
388 void
389 uvm_map_clip_start(map, entry, start)
390 struct vm_map *map;
391 struct vm_map_entry *entry;
392 vaddr_t start;
393 {
394 struct vm_map_entry *new_entry;
395 vaddr_t new_adj;
396
397 /* uvm_map_simplify_entry(map, entry); */ /* XXX */
398
399 /*
400 * Split off the front portion. note that we must insert the new
401 * entry BEFORE this one, so that this entry has the specified
402 * starting address.
403 */
404
405 new_entry = uvm_mapent_alloc(map);
406 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
407
408 new_entry->end = start;
409 new_adj = start - new_entry->start;
410 if (entry->object.uvm_obj)
411 entry->offset += new_adj; /* shift start over */
412 entry->start = start;
413
414 if (new_entry->aref.ar_amap) {
415 amap_splitref(&new_entry->aref, &entry->aref, new_adj);
416 }
417
418 uvm_map_entry_link(map, entry->prev, new_entry);
419
420 if (UVM_ET_ISSUBMAP(entry)) {
421 /* ... unlikely to happen, but play it safe */
422 uvm_map_reference(new_entry->object.sub_map);
423 } else {
424 if (UVM_ET_ISOBJ(entry) &&
425 entry->object.uvm_obj->pgops &&
426 entry->object.uvm_obj->pgops->pgo_reference)
427 entry->object.uvm_obj->pgops->pgo_reference(
428 entry->object.uvm_obj);
429 }
430 }
431
432 /*
433 * uvm_map_clip_end: ensure that the entry ends at or before
434 * the ending address, if it does't we split the reference
435 *
436 * => caller should use UVM_MAP_CLIP_END macro rather than calling
437 * this directly
438 * => map must be locked by caller
439 */
440
441 void
442 uvm_map_clip_end(map, entry, end)
443 struct vm_map *map;
444 struct vm_map_entry *entry;
445 vaddr_t end;
446 {
447 struct vm_map_entry * new_entry;
448 vaddr_t new_adj; /* #bytes we move start forward */
449
450 /*
451 * Create a new entry and insert it
452 * AFTER the specified entry
453 */
454
455 new_entry = uvm_mapent_alloc(map);
456 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
457
458 new_entry->start = entry->end = end;
459 new_adj = end - entry->start;
460 if (new_entry->object.uvm_obj)
461 new_entry->offset += new_adj;
462
463 if (entry->aref.ar_amap)
464 amap_splitref(&entry->aref, &new_entry->aref, new_adj);
465
466 uvm_map_entry_link(map, entry, new_entry);
467
468 if (UVM_ET_ISSUBMAP(entry)) {
469 /* ... unlikely to happen, but play it safe */
470 uvm_map_reference(new_entry->object.sub_map);
471 } else {
472 if (UVM_ET_ISOBJ(entry) &&
473 entry->object.uvm_obj->pgops &&
474 entry->object.uvm_obj->pgops->pgo_reference)
475 entry->object.uvm_obj->pgops->pgo_reference(
476 entry->object.uvm_obj);
477 }
478 }
479
480
481 /*
482 * M A P - m a i n e n t r y p o i n t
483 */
484 /*
485 * uvm_map: establish a valid mapping in a map
486 *
487 * => assume startp is page aligned.
488 * => assume size is a multiple of PAGE_SIZE.
489 * => assume sys_mmap provides enough of a "hint" to have us skip
490 * over text/data/bss area.
491 * => map must be unlocked (we will lock it)
492 * => <uobj,uoffset> value meanings (4 cases):
493 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER
494 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER
495 * [3] <uobj,uoffset> == normal mapping
496 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA
497 *
498 * case [4] is for kernel mappings where we don't know the offset until
499 * we've found a virtual address. note that kernel object offsets are
500 * always relative to vm_map_min(kernel_map).
501 *
502 * => if `align' is non-zero, we try to align the virtual address to
503 * the specified alignment. this is only a hint; if we can't
504 * do it, the address will be unaligned. this is provided as
505 * a mechanism for large pages.
506 *
507 * => XXXCDC: need way to map in external amap?
508 */
509
510 int
511 uvm_map(map, startp, size, uobj, uoffset, align, flags)
512 struct vm_map *map;
513 vaddr_t *startp; /* IN/OUT */
514 vsize_t size;
515 struct uvm_object *uobj;
516 voff_t uoffset;
517 vsize_t align;
518 uvm_flag_t flags;
519 {
520 struct vm_map_entry *prev_entry, *new_entry;
521 vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
522 UVM_MAXPROTECTION(flags);
523 vm_inherit_t inherit = UVM_INHERIT(flags);
524 int advice = UVM_ADVICE(flags);
525 UVMHIST_FUNC("uvm_map");
526 UVMHIST_CALLED(maphist);
527
528 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
529 map, *startp, size, flags);
530 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0);
531
532 /*
533 * detect a popular device driver bug.
534 */
535
536 KASSERT(curproc != NULL || map->flags & VM_MAP_INTRSAFE);
537
538 /*
539 * check sanity of protection code
540 */
541
542 if ((prot & maxprot) != prot) {
543 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x",
544 prot, maxprot,0,0);
545 return EACCES;
546 }
547
548 /*
549 * for pager_map, allocate the new entry first to avoid sleeping
550 * for memory while we have the map locked.
551 */
552
553 new_entry = NULL;
554 if (map == pager_map) {
555 new_entry = uvm_mapent_alloc(map);
556 }
557
558 /*
559 * figure out where to put new VM range
560 */
561
562 if (vm_map_lock_try(map) == FALSE) {
563 if (flags & UVM_FLAG_TRYLOCK) {
564 if (new_entry) {
565 uvm_mapent_free(new_entry);
566 }
567 return EAGAIN;
568 }
569 vm_map_lock(map); /* could sleep here */
570 }
571 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
572 uobj, uoffset, align, flags)) == NULL) {
573 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
574 vm_map_unlock(map);
575 if (new_entry) {
576 uvm_mapent_free(new_entry);
577 }
578 return ENOMEM;
579 }
580
581 #ifdef PMAP_GROWKERNEL
582 {
583 /*
584 * If the kernel pmap can't map the requested space,
585 * then allocate more resources for it.
586 */
587 if (map == kernel_map && uvm_maxkaddr < (*startp + size))
588 uvm_maxkaddr = pmap_growkernel(*startp + size);
589 }
590 #endif
591
592 UVMCNT_INCR(uvm_map_call);
593
594 /*
595 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
596 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in
597 * either case we want to zero it before storing it in the map entry
598 * (because it looks strange and confusing when debugging...)
599 *
600 * if uobj is not null
601 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
602 * and we do not need to change uoffset.
603 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
604 * now (based on the starting address of the map). this case is
605 * for kernel object mappings where we don't know the offset until
606 * the virtual address is found (with uvm_map_findspace). the
607 * offset is the distance we are from the start of the map.
608 */
609
610 if (uobj == NULL) {
611 uoffset = 0;
612 } else {
613 if (uoffset == UVM_UNKNOWN_OFFSET) {
614 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
615 uoffset = *startp - vm_map_min(kernel_map);
616 }
617 }
618
619 /*
620 * try and insert in map by extending previous entry, if possible.
621 * XXX: we don't try and pull back the next entry. might be useful
622 * for a stack, but we are currently allocating our stack in advance.
623 */
624
625 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
626 prev_entry->end == *startp && prev_entry != &map->header &&
627 prev_entry->object.uvm_obj == uobj) {
628
629 if (uobj && prev_entry->offset +
630 (prev_entry->end - prev_entry->start) != uoffset)
631 goto nomerge;
632
633 if (UVM_ET_ISSUBMAP(prev_entry))
634 goto nomerge;
635
636 if (prev_entry->protection != prot ||
637 prev_entry->max_protection != maxprot)
638 goto nomerge;
639
640 if (prev_entry->inheritance != inherit ||
641 prev_entry->advice != advice)
642 goto nomerge;
643
644 /* wiring status must match (new area is unwired) */
645 if (VM_MAPENT_ISWIRED(prev_entry))
646 goto nomerge;
647
648 /*
649 * can't extend a shared amap. note: no need to lock amap to
650 * look at refs since we don't care about its exact value.
651 * if it is one (i.e. we have only reference) it will stay there
652 */
653
654 if (prev_entry->aref.ar_amap &&
655 amap_refs(prev_entry->aref.ar_amap) != 1) {
656 goto nomerge;
657 }
658
659 /* got it! */
660
661 UVMCNT_INCR(map_backmerge);
662 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0);
663
664 /*
665 * drop our reference to uobj since we are extending a reference
666 * that we already have (the ref count can not drop to zero).
667 */
668 if (uobj && uobj->pgops->pgo_detach)
669 uobj->pgops->pgo_detach(uobj);
670
671 if (prev_entry->aref.ar_amap) {
672 amap_extend(prev_entry, size);
673 }
674
675 prev_entry->end += size;
676 map->size += size;
677
678 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
679 vm_map_unlock(map);
680 if (new_entry) {
681 uvm_mapent_free(new_entry);
682 }
683 return 0;
684 }
685
686 nomerge:
687 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0);
688
689 /*
690 * check for possible forward merge (which we don't do) and count
691 * the number of times we missed a *possible* chance to merge more
692 */
693
694 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
695 prev_entry->next != &map->header &&
696 prev_entry->next->start == (*startp + size))
697 UVMCNT_INCR(map_forwmerge);
698
699 /*
700 * allocate new entry and link it in.
701 */
702
703 if (new_entry == NULL) {
704 new_entry = uvm_mapent_alloc(map);
705 }
706 new_entry->start = *startp;
707 new_entry->end = new_entry->start + size;
708 new_entry->object.uvm_obj = uobj;
709 new_entry->offset = uoffset;
710
711 if (uobj)
712 new_entry->etype = UVM_ET_OBJ;
713 else
714 new_entry->etype = 0;
715
716 if (flags & UVM_FLAG_COPYONW) {
717 new_entry->etype |= UVM_ET_COPYONWRITE;
718 if ((flags & UVM_FLAG_OVERLAY) == 0)
719 new_entry->etype |= UVM_ET_NEEDSCOPY;
720 }
721
722 new_entry->protection = prot;
723 new_entry->max_protection = maxprot;
724 new_entry->inheritance = inherit;
725 new_entry->wired_count = 0;
726 new_entry->advice = advice;
727 if (flags & UVM_FLAG_OVERLAY) {
728
729 /*
730 * to_add: for BSS we overallocate a little since we
731 * are likely to extend
732 */
733
734 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
735 UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
736 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK);
737 new_entry->aref.ar_pageoff = 0;
738 new_entry->aref.ar_amap = amap;
739 } else {
740 new_entry->aref.ar_pageoff = 0;
741 new_entry->aref.ar_amap = NULL;
742 }
743 uvm_map_entry_link(map, prev_entry, new_entry);
744 map->size += size;
745
746 /*
747 * Update the free space hint
748 */
749
750 if ((map->first_free == prev_entry) &&
751 (prev_entry->end >= new_entry->start))
752 map->first_free = new_entry;
753
754 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
755 vm_map_unlock(map);
756 return 0;
757 }
758
759 /*
760 * uvm_map_lookup_entry: find map entry at or before an address
761 *
762 * => map must at least be read-locked by caller
763 * => entry is returned in "entry"
764 * => return value is true if address is in the returned entry
765 */
766
767 boolean_t
768 uvm_map_lookup_entry(map, address, entry)
769 struct vm_map *map;
770 vaddr_t address;
771 struct vm_map_entry **entry; /* OUT */
772 {
773 struct vm_map_entry *cur;
774 struct vm_map_entry *last;
775 UVMHIST_FUNC("uvm_map_lookup_entry");
776 UVMHIST_CALLED(maphist);
777
778 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
779 map, address, entry, 0);
780
781 /*
782 * start looking either from the head of the
783 * list, or from the hint.
784 */
785
786 simple_lock(&map->hint_lock);
787 cur = map->hint;
788 simple_unlock(&map->hint_lock);
789
790 if (cur == &map->header)
791 cur = cur->next;
792
793 UVMCNT_INCR(uvm_mlk_call);
794 if (address >= cur->start) {
795
796 /*
797 * go from hint to end of list.
798 *
799 * but first, make a quick check to see if
800 * we are already looking at the entry we
801 * want (which is usually the case).
802 * note also that we don't need to save the hint
803 * here... it is the same hint (unless we are
804 * at the header, in which case the hint didn't
805 * buy us anything anyway).
806 */
807
808 last = &map->header;
809 if ((cur != last) && (cur->end > address)) {
810 UVMCNT_INCR(uvm_mlk_hint);
811 *entry = cur;
812 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
813 cur, 0, 0, 0);
814 return (TRUE);
815 }
816 } else {
817
818 /*
819 * go from start to hint, *inclusively*
820 */
821
822 last = cur->next;
823 cur = map->header.next;
824 }
825
826 /*
827 * search linearly
828 */
829
830 while (cur != last) {
831 if (cur->end > address) {
832 if (address >= cur->start) {
833 /*
834 * save this lookup for future
835 * hints, and return
836 */
837
838 *entry = cur;
839 SAVE_HINT(map, map->hint, cur);
840 UVMHIST_LOG(maphist,"<- search got it (0x%x)",
841 cur, 0, 0, 0);
842 return (TRUE);
843 }
844 break;
845 }
846 cur = cur->next;
847 }
848 *entry = cur->prev;
849 SAVE_HINT(map, map->hint, *entry);
850 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
851 return (FALSE);
852 }
853
854 /*
855 * uvm_map_findspace: find "length" sized space in "map".
856 *
857 * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is
858 * set (in which case we insist on using "hint").
859 * => "result" is VA returned
860 * => uobj/uoffset are to be used to handle VAC alignment, if required
861 * => if `align' is non-zero, we attempt to align to that value.
862 * => caller must at least have read-locked map
863 * => returns NULL on failure, or pointer to prev. map entry if success
864 * => note this is a cross between the old vm_map_findspace and vm_map_find
865 */
866
867 struct vm_map_entry *
868 uvm_map_findspace(map, hint, length, result, uobj, uoffset, align, flags)
869 struct vm_map *map;
870 vaddr_t hint;
871 vsize_t length;
872 vaddr_t *result; /* OUT */
873 struct uvm_object *uobj;
874 voff_t uoffset;
875 vsize_t align;
876 int flags;
877 {
878 struct vm_map_entry *entry, *next, *tmp;
879 vaddr_t end, orig_hint;
880 UVMHIST_FUNC("uvm_map_findspace");
881 UVMHIST_CALLED(maphist);
882
883 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
884 map, hint, length, flags);
885 KASSERT((align & (align - 1)) == 0);
886 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
887
888 /*
889 * remember the original hint. if we are aligning, then we
890 * may have to try again with no alignment constraint if
891 * we fail the first time.
892 */
893
894 orig_hint = hint;
895 if (hint < map->min_offset) { /* check ranges ... */
896 if (flags & UVM_FLAG_FIXED) {
897 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
898 return(NULL);
899 }
900 hint = map->min_offset;
901 }
902 if (hint > map->max_offset) {
903 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
904 hint, map->min_offset, map->max_offset, 0);
905 return(NULL);
906 }
907
908 /*
909 * Look for the first possible address; if there's already
910 * something at this address, we have to start after it.
911 */
912
913 if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) {
914 if ((entry = map->first_free) != &map->header)
915 hint = entry->end;
916 } else {
917 if (uvm_map_lookup_entry(map, hint, &tmp)) {
918 /* "hint" address already in use ... */
919 if (flags & UVM_FLAG_FIXED) {
920 UVMHIST_LOG(maphist,"<- fixed & VA in use",
921 0, 0, 0, 0);
922 return(NULL);
923 }
924 hint = tmp->end;
925 }
926 entry = tmp;
927 }
928
929 /*
930 * Look through the rest of the map, trying to fit a new region in
931 * the gap between existing regions, or after the very last region.
932 * note: entry->end = base VA of current gap,
933 * next->start = VA of end of current gap
934 */
935
936 for (;; hint = (entry = next)->end) {
937
938 /*
939 * Find the end of the proposed new region. Be sure we didn't
940 * go beyond the end of the map, or wrap around the address;
941 * if so, we lose. Otherwise, if this is the last entry, or
942 * if the proposed new region fits before the next entry, we
943 * win.
944 */
945
946 #ifdef PMAP_PREFER
947 /*
948 * push hint forward as needed to avoid VAC alias problems.
949 * we only do this if a valid offset is specified.
950 */
951
952 if ((flags & UVM_FLAG_FIXED) == 0 &&
953 uoffset != UVM_UNKNOWN_OFFSET)
954 PMAP_PREFER(uoffset, &hint);
955 #endif
956 if (align != 0) {
957 if ((hint & (align - 1)) != 0)
958 hint = roundup(hint, align);
959 /*
960 * XXX Should we PMAP_PREFER() here again?
961 */
962 }
963 end = hint + length;
964 if (end > map->max_offset || end < hint) {
965 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
966 if (align != 0) {
967 UVMHIST_LOG(maphist,
968 "calling recursively, no align",
969 0,0,0,0);
970 return (uvm_map_findspace(map, orig_hint,
971 length, result, uobj, uoffset, 0, flags));
972 }
973 return (NULL);
974 }
975 next = entry->next;
976 if (next == &map->header || next->start >= end)
977 break;
978 if (flags & UVM_FLAG_FIXED) {
979 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
980 return(NULL); /* only one shot at it ... */
981 }
982 }
983 SAVE_HINT(map, map->hint, entry);
984 *result = hint;
985 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0);
986 return (entry);
987 }
988
989 /*
990 * U N M A P - m a i n h e l p e r f u n c t i o n s
991 */
992
993 /*
994 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
995 *
996 * => caller must check alignment and size
997 * => map must be locked by caller
998 * => we return a list of map entries that we've remove from the map
999 * in "entry_list"
1000 */
1001
1002 void
1003 uvm_unmap_remove(map, start, end, entry_list)
1004 struct vm_map *map;
1005 vaddr_t start, end;
1006 struct vm_map_entry **entry_list; /* OUT */
1007 {
1008 struct vm_map_entry *entry, *first_entry, *next;
1009 vaddr_t len;
1010 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
1011
1012 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
1013 map, start, end, 0);
1014 VM_MAP_RANGE_CHECK(map, start, end);
1015
1016 /*
1017 * find first entry
1018 */
1019
1020 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
1021 /* clip and go... */
1022 entry = first_entry;
1023 UVM_MAP_CLIP_START(map, entry, start);
1024 /* critical! prevents stale hint */
1025 SAVE_HINT(map, entry, entry->prev);
1026 } else {
1027 entry = first_entry->next;
1028 }
1029
1030 /*
1031 * Save the free space hint
1032 */
1033
1034 if (map->first_free->start >= start)
1035 map->first_free = entry->prev;
1036
1037 /*
1038 * note: we now re-use first_entry for a different task. we remove
1039 * a number of map entries from the map and save them in a linked
1040 * list headed by "first_entry". once we remove them from the map
1041 * the caller should unlock the map and drop the references to the
1042 * backing objects [c.f. uvm_unmap_detach]. the object is to
1043 * separate unmapping from reference dropping. why?
1044 * [1] the map has to be locked for unmapping
1045 * [2] the map need not be locked for reference dropping
1046 * [3] dropping references may trigger pager I/O, and if we hit
1047 * a pager that does synchronous I/O we may have to wait for it.
1048 * [4] we would like all waiting for I/O to occur with maps unlocked
1049 * so that we don't block other threads.
1050 */
1051
1052 first_entry = NULL;
1053 *entry_list = NULL;
1054
1055 /*
1056 * break up the area into map entry sized regions and unmap. note
1057 * that all mappings have to be removed before we can even consider
1058 * dropping references to amaps or VM objects (otherwise we could end
1059 * up with a mapping to a page on the free list which would be very bad)
1060 */
1061
1062 while ((entry != &map->header) && (entry->start < end)) {
1063 UVM_MAP_CLIP_END(map, entry, end);
1064 next = entry->next;
1065 len = entry->end - entry->start;
1066
1067 /*
1068 * unwire before removing addresses from the pmap; otherwise
1069 * unwiring will put the entries back into the pmap (XXX).
1070 */
1071
1072 if (VM_MAPENT_ISWIRED(entry)) {
1073 uvm_map_entry_unwire(map, entry);
1074 }
1075 if ((map->flags & VM_MAP_PAGEABLE) == 0) {
1076
1077 /*
1078 * if the map is non-pageable, any pages mapped there
1079 * must be wired and entered with pmap_kenter_pa(),
1080 * and we should free any such pages immediately.
1081 * this is mostly used for kmem_map and mb_map.
1082 */
1083
1084 uvm_km_pgremove_intrsafe(entry->start, entry->end);
1085 pmap_kremove(entry->start, len);
1086 } else if (UVM_ET_ISOBJ(entry) &&
1087 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
1088 KASSERT(vm_map_pmap(map) == pmap_kernel());
1089
1090 /*
1091 * note: kernel object mappings are currently used in
1092 * two ways:
1093 * [1] "normal" mappings of pages in the kernel object
1094 * [2] uvm_km_valloc'd allocations in which we
1095 * pmap_enter in some non-kernel-object page
1096 * (e.g. vmapbuf).
1097 *
1098 * for case [1], we need to remove the mapping from
1099 * the pmap and then remove the page from the kernel
1100 * object (because, once pages in a kernel object are
1101 * unmapped they are no longer needed, unlike, say,
1102 * a vnode where you might want the data to persist
1103 * until flushed out of a queue).
1104 *
1105 * for case [2], we need to remove the mapping from
1106 * the pmap. there shouldn't be any pages at the
1107 * specified offset in the kernel object [but it
1108 * doesn't hurt to call uvm_km_pgremove just to be
1109 * safe?]
1110 *
1111 * uvm_km_pgremove currently does the following:
1112 * for pages in the kernel object in range:
1113 * - drops the swap slot
1114 * - uvm_pagefree the page
1115 */
1116
1117 /*
1118 * remove mappings from pmap and drop the pages
1119 * from the object. offsets are always relative
1120 * to vm_map_min(kernel_map).
1121 */
1122
1123 pmap_remove(pmap_kernel(), entry->start,
1124 entry->start + len);
1125 uvm_km_pgremove(entry->object.uvm_obj,
1126 entry->start - vm_map_min(kernel_map),
1127 entry->end - vm_map_min(kernel_map));
1128
1129 /*
1130 * null out kernel_object reference, we've just
1131 * dropped it
1132 */
1133
1134 entry->etype &= ~UVM_ET_OBJ;
1135 entry->object.uvm_obj = NULL;
1136 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
1137
1138 /*
1139 * remove mappings the standard way.
1140 */
1141
1142 pmap_remove(map->pmap, entry->start, entry->end);
1143 }
1144
1145 /*
1146 * remove entry from map and put it on our list of entries
1147 * that we've nuked. then go to next entry.
1148 */
1149
1150 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0);
1151
1152 /* critical! prevents stale hint */
1153 SAVE_HINT(map, entry, entry->prev);
1154
1155 uvm_map_entry_unlink(map, entry);
1156 map->size -= len;
1157 entry->next = first_entry;
1158 first_entry = entry;
1159 entry = next;
1160 }
1161 pmap_update(vm_map_pmap(map));
1162
1163 /*
1164 * now we've cleaned up the map and are ready for the caller to drop
1165 * references to the mapped objects.
1166 */
1167
1168 *entry_list = first_entry;
1169 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1170 }
1171
1172 /*
1173 * uvm_unmap_detach: drop references in a chain of map entries
1174 *
1175 * => we will free the map entries as we traverse the list.
1176 */
1177
1178 void
1179 uvm_unmap_detach(first_entry, flags)
1180 struct vm_map_entry *first_entry;
1181 int flags;
1182 {
1183 struct vm_map_entry *next_entry;
1184 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1185
1186 while (first_entry) {
1187 KASSERT(!VM_MAPENT_ISWIRED(first_entry));
1188 UVMHIST_LOG(maphist,
1189 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1190 first_entry, first_entry->aref.ar_amap,
1191 first_entry->object.uvm_obj,
1192 UVM_ET_ISSUBMAP(first_entry));
1193
1194 /*
1195 * drop reference to amap, if we've got one
1196 */
1197
1198 if (first_entry->aref.ar_amap)
1199 uvm_map_unreference_amap(first_entry, flags);
1200
1201 /*
1202 * drop reference to our backing object, if we've got one
1203 */
1204
1205 if (UVM_ET_ISSUBMAP(first_entry)) {
1206 /* ... unlikely to happen, but play it safe */
1207 uvm_map_deallocate(first_entry->object.sub_map);
1208 } else {
1209 if (UVM_ET_ISOBJ(first_entry) &&
1210 first_entry->object.uvm_obj->pgops->pgo_detach)
1211 first_entry->object.uvm_obj->pgops->
1212 pgo_detach(first_entry->object.uvm_obj);
1213 }
1214 next_entry = first_entry->next;
1215 uvm_mapent_free(first_entry);
1216 first_entry = next_entry;
1217 }
1218 UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1219 }
1220
1221 /*
1222 * E X T R A C T I O N F U N C T I O N S
1223 */
1224
1225 /*
1226 * uvm_map_reserve: reserve space in a vm_map for future use.
1227 *
1228 * => we reserve space in a map by putting a dummy map entry in the
1229 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1230 * => map should be unlocked (we will write lock it)
1231 * => we return true if we were able to reserve space
1232 * => XXXCDC: should be inline?
1233 */
1234
1235 int
1236 uvm_map_reserve(map, size, offset, align, raddr)
1237 struct vm_map *map;
1238 vsize_t size;
1239 vaddr_t offset; /* hint for pmap_prefer */
1240 vsize_t align; /* alignment hint */
1241 vaddr_t *raddr; /* IN:hint, OUT: reserved VA */
1242 {
1243 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1244
1245 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1246 map,size,offset,raddr);
1247
1248 size = round_page(size);
1249 if (*raddr < vm_map_min(map))
1250 *raddr = vm_map_min(map); /* hint */
1251
1252 /*
1253 * reserve some virtual space.
1254 */
1255
1256 if (uvm_map(map, raddr, size, NULL, offset, 0,
1257 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1258 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) {
1259 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1260 return (FALSE);
1261 }
1262
1263 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1264 return (TRUE);
1265 }
1266
1267 /*
1268 * uvm_map_replace: replace a reserved (blank) area of memory with
1269 * real mappings.
1270 *
1271 * => caller must WRITE-LOCK the map
1272 * => we return TRUE if replacement was a success
1273 * => we expect the newents chain to have nnewents entrys on it and
1274 * we expect newents->prev to point to the last entry on the list
1275 * => note newents is allowed to be NULL
1276 */
1277
1278 int
1279 uvm_map_replace(map, start, end, newents, nnewents)
1280 struct vm_map *map;
1281 vaddr_t start, end;
1282 struct vm_map_entry *newents;
1283 int nnewents;
1284 {
1285 struct vm_map_entry *oldent, *last;
1286
1287 /*
1288 * first find the blank map entry at the specified address
1289 */
1290
1291 if (!uvm_map_lookup_entry(map, start, &oldent)) {
1292 return(FALSE);
1293 }
1294
1295 /*
1296 * check to make sure we have a proper blank entry
1297 */
1298
1299 if (oldent->start != start || oldent->end != end ||
1300 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1301 return (FALSE);
1302 }
1303
1304 #ifdef DIAGNOSTIC
1305
1306 /*
1307 * sanity check the newents chain
1308 */
1309
1310 {
1311 struct vm_map_entry *tmpent = newents;
1312 int nent = 0;
1313 vaddr_t cur = start;
1314
1315 while (tmpent) {
1316 nent++;
1317 if (tmpent->start < cur)
1318 panic("uvm_map_replace1");
1319 if (tmpent->start > tmpent->end || tmpent->end > end) {
1320 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1321 tmpent->start, tmpent->end, end);
1322 panic("uvm_map_replace2");
1323 }
1324 cur = tmpent->end;
1325 if (tmpent->next) {
1326 if (tmpent->next->prev != tmpent)
1327 panic("uvm_map_replace3");
1328 } else {
1329 if (newents->prev != tmpent)
1330 panic("uvm_map_replace4");
1331 }
1332 tmpent = tmpent->next;
1333 }
1334 if (nent != nnewents)
1335 panic("uvm_map_replace5");
1336 }
1337 #endif
1338
1339 /*
1340 * map entry is a valid blank! replace it. (this does all the
1341 * work of map entry link/unlink...).
1342 */
1343
1344 if (newents) {
1345 last = newents->prev;
1346
1347 /* critical: flush stale hints out of map */
1348 SAVE_HINT(map, map->hint, newents);
1349 if (map->first_free == oldent)
1350 map->first_free = last;
1351
1352 last->next = oldent->next;
1353 last->next->prev = last;
1354 newents->prev = oldent->prev;
1355 newents->prev->next = newents;
1356 map->nentries = map->nentries + (nnewents - 1);
1357
1358 } else {
1359
1360 /* critical: flush stale hints out of map */
1361 SAVE_HINT(map, map->hint, oldent->prev);
1362 if (map->first_free == oldent)
1363 map->first_free = oldent->prev;
1364
1365 /* NULL list of new entries: just remove the old one */
1366 uvm_map_entry_unlink(map, oldent);
1367 }
1368
1369
1370 /*
1371 * now we can free the old blank entry, unlock the map and return.
1372 */
1373
1374 uvm_mapent_free(oldent);
1375 return(TRUE);
1376 }
1377
1378 /*
1379 * uvm_map_extract: extract a mapping from a map and put it somewhere
1380 * (maybe removing the old mapping)
1381 *
1382 * => maps should be unlocked (we will write lock them)
1383 * => returns 0 on success, error code otherwise
1384 * => start must be page aligned
1385 * => len must be page sized
1386 * => flags:
1387 * UVM_EXTRACT_REMOVE: remove mappings from srcmap
1388 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1389 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1390 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1391 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1392 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1393 * be used from within the kernel in a kernel level map <<<
1394 */
1395
1396 int
1397 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1398 struct vm_map *srcmap, *dstmap;
1399 vaddr_t start, *dstaddrp;
1400 vsize_t len;
1401 int flags;
1402 {
1403 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1404 oldstart;
1405 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
1406 *deadentry, *oldentry;
1407 vsize_t elen;
1408 int nchain, error, copy_ok;
1409 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1410
1411 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1412 len,0);
1413 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1414
1415 /*
1416 * step 0: sanity check: start must be on a page boundary, length
1417 * must be page sized. can't ask for CONTIG/QREF if you asked for
1418 * REMOVE.
1419 */
1420
1421 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
1422 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
1423 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
1424
1425 /*
1426 * step 1: reserve space in the target map for the extracted area
1427 */
1428
1429 dstaddr = vm_map_min(dstmap);
1430 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE)
1431 return(ENOMEM);
1432 *dstaddrp = dstaddr; /* pass address back to caller */
1433 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0);
1434
1435 /*
1436 * step 2: setup for the extraction process loop by init'ing the
1437 * map entry chain, locking src map, and looking up the first useful
1438 * entry in the map.
1439 */
1440
1441 end = start + len;
1442 newend = dstaddr + len;
1443 chain = endchain = NULL;
1444 nchain = 0;
1445 vm_map_lock(srcmap);
1446
1447 if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1448
1449 /* "start" is within an entry */
1450 if (flags & UVM_EXTRACT_QREF) {
1451
1452 /*
1453 * for quick references we don't clip the entry, so
1454 * the entry may map space "before" the starting
1455 * virtual address... this is the "fudge" factor
1456 * (which can be non-zero only the first time
1457 * through the "while" loop in step 3).
1458 */
1459
1460 fudge = start - entry->start;
1461 } else {
1462
1463 /*
1464 * normal reference: we clip the map to fit (thus
1465 * fudge is zero)
1466 */
1467
1468 UVM_MAP_CLIP_START(srcmap, entry, start);
1469 SAVE_HINT(srcmap, srcmap->hint, entry->prev);
1470 fudge = 0;
1471 }
1472 } else {
1473
1474 /* "start" is not within an entry ... skip to next entry */
1475 if (flags & UVM_EXTRACT_CONTIG) {
1476 error = EINVAL;
1477 goto bad; /* definite hole here ... */
1478 }
1479
1480 entry = entry->next;
1481 fudge = 0;
1482 }
1483
1484 /* save values from srcmap for step 6 */
1485 orig_entry = entry;
1486 orig_fudge = fudge;
1487
1488 /*
1489 * step 3: now start looping through the map entries, extracting
1490 * as we go.
1491 */
1492
1493 while (entry->start < end && entry != &srcmap->header) {
1494
1495 /* if we are not doing a quick reference, clip it */
1496 if ((flags & UVM_EXTRACT_QREF) == 0)
1497 UVM_MAP_CLIP_END(srcmap, entry, end);
1498
1499 /* clear needs_copy (allow chunking) */
1500 if (UVM_ET_ISNEEDSCOPY(entry)) {
1501 if (fudge)
1502 oldstart = entry->start;
1503 else
1504 oldstart = 0; /* XXX: gcc */
1505 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1506 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */
1507 error = ENOMEM;
1508 goto bad;
1509 }
1510
1511 /* amap_copy could clip (during chunk)! update fudge */
1512 if (fudge) {
1513 fudge = fudge - (entry->start - oldstart);
1514 orig_fudge = fudge;
1515 }
1516 }
1517
1518 /* calculate the offset of this from "start" */
1519 oldoffset = (entry->start + fudge) - start;
1520
1521 /* allocate a new map entry */
1522 newentry = uvm_mapent_alloc(dstmap);
1523 if (newentry == NULL) {
1524 error = ENOMEM;
1525 goto bad;
1526 }
1527
1528 /* set up new map entry */
1529 newentry->next = NULL;
1530 newentry->prev = endchain;
1531 newentry->start = dstaddr + oldoffset;
1532 newentry->end =
1533 newentry->start + (entry->end - (entry->start + fudge));
1534 if (newentry->end > newend || newentry->end < newentry->start)
1535 newentry->end = newend;
1536 newentry->object.uvm_obj = entry->object.uvm_obj;
1537 if (newentry->object.uvm_obj) {
1538 if (newentry->object.uvm_obj->pgops->pgo_reference)
1539 newentry->object.uvm_obj->pgops->
1540 pgo_reference(newentry->object.uvm_obj);
1541 newentry->offset = entry->offset + fudge;
1542 } else {
1543 newentry->offset = 0;
1544 }
1545 newentry->etype = entry->etype;
1546 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1547 entry->max_protection : entry->protection;
1548 newentry->max_protection = entry->max_protection;
1549 newentry->inheritance = entry->inheritance;
1550 newentry->wired_count = 0;
1551 newentry->aref.ar_amap = entry->aref.ar_amap;
1552 if (newentry->aref.ar_amap) {
1553 newentry->aref.ar_pageoff =
1554 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1555 uvm_map_reference_amap(newentry, AMAP_SHARED |
1556 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1557 } else {
1558 newentry->aref.ar_pageoff = 0;
1559 }
1560 newentry->advice = entry->advice;
1561
1562 /* now link it on the chain */
1563 nchain++;
1564 if (endchain == NULL) {
1565 chain = endchain = newentry;
1566 } else {
1567 endchain->next = newentry;
1568 endchain = newentry;
1569 }
1570
1571 /* end of 'while' loop! */
1572 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1573 (entry->next == &srcmap->header ||
1574 entry->next->start != entry->end)) {
1575 error = EINVAL;
1576 goto bad;
1577 }
1578 entry = entry->next;
1579 fudge = 0;
1580 }
1581
1582 /*
1583 * step 4: close off chain (in format expected by uvm_map_replace)
1584 */
1585
1586 if (chain)
1587 chain->prev = endchain;
1588
1589 /*
1590 * step 5: attempt to lock the dest map so we can pmap_copy.
1591 * note usage of copy_ok:
1592 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1593 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1594 */
1595
1596 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1597 copy_ok = 1;
1598 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1599 nchain)) {
1600 if (srcmap != dstmap)
1601 vm_map_unlock(dstmap);
1602 error = EIO;
1603 goto bad;
1604 }
1605 } else {
1606 copy_ok = 0;
1607 /* replace defered until step 7 */
1608 }
1609
1610 /*
1611 * step 6: traverse the srcmap a second time to do the following:
1612 * - if we got a lock on the dstmap do pmap_copy
1613 * - if UVM_EXTRACT_REMOVE remove the entries
1614 * we make use of orig_entry and orig_fudge (saved in step 2)
1615 */
1616
1617 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1618
1619 /* purge possible stale hints from srcmap */
1620 if (flags & UVM_EXTRACT_REMOVE) {
1621 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
1622 if (srcmap->first_free->start >= start)
1623 srcmap->first_free = orig_entry->prev;
1624 }
1625
1626 entry = orig_entry;
1627 fudge = orig_fudge;
1628 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */
1629
1630 while (entry->start < end && entry != &srcmap->header) {
1631 if (copy_ok) {
1632 oldoffset = (entry->start + fudge) - start;
1633 elen = MIN(end, entry->end) -
1634 (entry->start + fudge);
1635 pmap_copy(dstmap->pmap, srcmap->pmap,
1636 dstaddr + oldoffset, elen,
1637 entry->start + fudge);
1638 }
1639
1640 /* we advance "entry" in the following if statement */
1641 if (flags & UVM_EXTRACT_REMOVE) {
1642 pmap_remove(srcmap->pmap, entry->start,
1643 entry->end);
1644 oldentry = entry; /* save entry */
1645 entry = entry->next; /* advance */
1646 uvm_map_entry_unlink(srcmap, oldentry);
1647 /* add to dead list */
1648 oldentry->next = deadentry;
1649 deadentry = oldentry;
1650 } else {
1651 entry = entry->next; /* advance */
1652 }
1653
1654 /* end of 'while' loop */
1655 fudge = 0;
1656 }
1657 pmap_update(srcmap->pmap);
1658
1659 /*
1660 * unlock dstmap. we will dispose of deadentry in
1661 * step 7 if needed
1662 */
1663
1664 if (copy_ok && srcmap != dstmap)
1665 vm_map_unlock(dstmap);
1666
1667 } else {
1668 deadentry = NULL;
1669 }
1670
1671 /*
1672 * step 7: we are done with the source map, unlock. if copy_ok
1673 * is 0 then we have not replaced the dummy mapping in dstmap yet
1674 * and we need to do so now.
1675 */
1676
1677 vm_map_unlock(srcmap);
1678 if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1679 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */
1680
1681 /* now do the replacement if we didn't do it in step 5 */
1682 if (copy_ok == 0) {
1683 vm_map_lock(dstmap);
1684 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1685 nchain);
1686 vm_map_unlock(dstmap);
1687
1688 if (error == FALSE) {
1689 error = EIO;
1690 goto bad2;
1691 }
1692 }
1693 return(0);
1694
1695 /*
1696 * bad: failure recovery
1697 */
1698 bad:
1699 vm_map_unlock(srcmap);
1700 bad2: /* src already unlocked */
1701 if (chain)
1702 uvm_unmap_detach(chain,
1703 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1704 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */
1705 return(error);
1706 }
1707
1708 /* end of extraction functions */
1709
1710 /*
1711 * uvm_map_submap: punch down part of a map into a submap
1712 *
1713 * => only the kernel_map is allowed to be submapped
1714 * => the purpose of submapping is to break up the locking granularity
1715 * of a larger map
1716 * => the range specified must have been mapped previously with a uvm_map()
1717 * call [with uobj==NULL] to create a blank map entry in the main map.
1718 * [And it had better still be blank!]
1719 * => maps which contain submaps should never be copied or forked.
1720 * => to remove a submap, use uvm_unmap() on the main map
1721 * and then uvm_map_deallocate() the submap.
1722 * => main map must be unlocked.
1723 * => submap must have been init'd and have a zero reference count.
1724 * [need not be locked as we don't actually reference it]
1725 */
1726
1727 int
1728 uvm_map_submap(map, start, end, submap)
1729 struct vm_map *map, *submap;
1730 vaddr_t start, end;
1731 {
1732 struct vm_map_entry *entry;
1733 int error;
1734
1735 vm_map_lock(map);
1736 VM_MAP_RANGE_CHECK(map, start, end);
1737
1738 if (uvm_map_lookup_entry(map, start, &entry)) {
1739 UVM_MAP_CLIP_START(map, entry, start);
1740 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */
1741 } else {
1742 entry = NULL;
1743 }
1744
1745 if (entry != NULL &&
1746 entry->start == start && entry->end == end &&
1747 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1748 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1749 entry->etype |= UVM_ET_SUBMAP;
1750 entry->object.sub_map = submap;
1751 entry->offset = 0;
1752 uvm_map_reference(submap);
1753 error = 0;
1754 } else {
1755 error = EINVAL;
1756 }
1757 vm_map_unlock(map);
1758 return error;
1759 }
1760
1761
1762 /*
1763 * uvm_map_protect: change map protection
1764 *
1765 * => set_max means set max_protection.
1766 * => map must be unlocked.
1767 */
1768
1769 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \
1770 ~VM_PROT_WRITE : VM_PROT_ALL)
1771
1772 int
1773 uvm_map_protect(map, start, end, new_prot, set_max)
1774 struct vm_map *map;
1775 vaddr_t start, end;
1776 vm_prot_t new_prot;
1777 boolean_t set_max;
1778 {
1779 struct vm_map_entry *current, *entry;
1780 int error = 0;
1781 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1782 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1783 map, start, end, new_prot);
1784
1785 vm_map_lock(map);
1786 VM_MAP_RANGE_CHECK(map, start, end);
1787 if (uvm_map_lookup_entry(map, start, &entry)) {
1788 UVM_MAP_CLIP_START(map, entry, start);
1789 } else {
1790 entry = entry->next;
1791 }
1792
1793 /*
1794 * make a first pass to check for protection violations.
1795 */
1796
1797 current = entry;
1798 while ((current != &map->header) && (current->start < end)) {
1799 if (UVM_ET_ISSUBMAP(current)) {
1800 error = EINVAL;
1801 goto out;
1802 }
1803 if ((new_prot & current->max_protection) != new_prot) {
1804 error = EACCES;
1805 goto out;
1806 }
1807 /*
1808 * Don't allow VM_PROT_EXECUTE to be set on entries that
1809 * point to vnodes that are associated with a NOEXEC file
1810 * system.
1811 */
1812 if (UVM_ET_ISOBJ(current) &&
1813 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
1814 struct vnode *vp =
1815 (struct vnode *) current->object.uvm_obj;
1816
1817 if ((new_prot & VM_PROT_EXECUTE) != 0 &&
1818 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
1819 error = EACCES;
1820 goto out;
1821 }
1822 }
1823 current = current->next;
1824 }
1825
1826 /* go back and fix up protections (no need to clip this time). */
1827
1828 current = entry;
1829 while ((current != &map->header) && (current->start < end)) {
1830 vm_prot_t old_prot;
1831
1832 UVM_MAP_CLIP_END(map, current, end);
1833 old_prot = current->protection;
1834 if (set_max)
1835 current->protection =
1836 (current->max_protection = new_prot) & old_prot;
1837 else
1838 current->protection = new_prot;
1839
1840 /*
1841 * update physical map if necessary. worry about copy-on-write
1842 * here -- CHECK THIS XXX
1843 */
1844
1845 if (current->protection != old_prot) {
1846 /* update pmap! */
1847 pmap_protect(map->pmap, current->start, current->end,
1848 current->protection & MASK(entry));
1849
1850 /*
1851 * If this entry points at a vnode, and the
1852 * protection includes VM_PROT_EXECUTE, mark
1853 * the vnode as VEXECMAP.
1854 */
1855 if (UVM_ET_ISOBJ(current)) {
1856 struct uvm_object *uobj =
1857 current->object.uvm_obj;
1858
1859 if (UVM_OBJ_IS_VNODE(uobj) &&
1860 (current->protection & VM_PROT_EXECUTE))
1861 vn_markexec((struct vnode *) uobj);
1862 }
1863 }
1864
1865 /*
1866 * If the map is configured to lock any future mappings,
1867 * wire this entry now if the old protection was VM_PROT_NONE
1868 * and the new protection is not VM_PROT_NONE.
1869 */
1870
1871 if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
1872 VM_MAPENT_ISWIRED(entry) == 0 &&
1873 old_prot == VM_PROT_NONE &&
1874 new_prot != VM_PROT_NONE) {
1875 if (uvm_map_pageable(map, entry->start,
1876 entry->end, FALSE,
1877 UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
1878
1879 /*
1880 * If locking the entry fails, remember the
1881 * error if it's the first one. Note we
1882 * still continue setting the protection in
1883 * the map, but will return the error
1884 * condition regardless.
1885 *
1886 * XXX Ignore what the actual error is,
1887 * XXX just call it a resource shortage
1888 * XXX so that it doesn't get confused
1889 * XXX what uvm_map_protect() itself would
1890 * XXX normally return.
1891 */
1892
1893 error = ENOMEM;
1894 }
1895 }
1896 current = current->next;
1897 }
1898 pmap_update(map->pmap);
1899
1900 out:
1901 vm_map_unlock(map);
1902 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
1903 return error;
1904 }
1905
1906 #undef MASK
1907
1908 /*
1909 * uvm_map_inherit: set inheritance code for range of addrs in map.
1910 *
1911 * => map must be unlocked
1912 * => note that the inherit code is used during a "fork". see fork
1913 * code for details.
1914 */
1915
1916 int
1917 uvm_map_inherit(map, start, end, new_inheritance)
1918 struct vm_map *map;
1919 vaddr_t start;
1920 vaddr_t end;
1921 vm_inherit_t new_inheritance;
1922 {
1923 struct vm_map_entry *entry, *temp_entry;
1924 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1925 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1926 map, start, end, new_inheritance);
1927
1928 switch (new_inheritance) {
1929 case MAP_INHERIT_NONE:
1930 case MAP_INHERIT_COPY:
1931 case MAP_INHERIT_SHARE:
1932 break;
1933 default:
1934 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1935 return EINVAL;
1936 }
1937
1938 vm_map_lock(map);
1939 VM_MAP_RANGE_CHECK(map, start, end);
1940 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1941 entry = temp_entry;
1942 UVM_MAP_CLIP_START(map, entry, start);
1943 } else {
1944 entry = temp_entry->next;
1945 }
1946 while ((entry != &map->header) && (entry->start < end)) {
1947 UVM_MAP_CLIP_END(map, entry, end);
1948 entry->inheritance = new_inheritance;
1949 entry = entry->next;
1950 }
1951 vm_map_unlock(map);
1952 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1953 return 0;
1954 }
1955
1956 /*
1957 * uvm_map_advice: set advice code for range of addrs in map.
1958 *
1959 * => map must be unlocked
1960 */
1961
1962 int
1963 uvm_map_advice(map, start, end, new_advice)
1964 struct vm_map *map;
1965 vaddr_t start;
1966 vaddr_t end;
1967 int new_advice;
1968 {
1969 struct vm_map_entry *entry, *temp_entry;
1970 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
1971 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
1972 map, start, end, new_advice);
1973
1974 vm_map_lock(map);
1975 VM_MAP_RANGE_CHECK(map, start, end);
1976 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1977 entry = temp_entry;
1978 UVM_MAP_CLIP_START(map, entry, start);
1979 } else {
1980 entry = temp_entry->next;
1981 }
1982
1983 /*
1984 * XXXJRT: disallow holes?
1985 */
1986
1987 while ((entry != &map->header) && (entry->start < end)) {
1988 UVM_MAP_CLIP_END(map, entry, end);
1989
1990 switch (new_advice) {
1991 case MADV_NORMAL:
1992 case MADV_RANDOM:
1993 case MADV_SEQUENTIAL:
1994 /* nothing special here */
1995 break;
1996
1997 default:
1998 vm_map_unlock(map);
1999 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2000 return EINVAL;
2001 }
2002 entry->advice = new_advice;
2003 entry = entry->next;
2004 }
2005
2006 vm_map_unlock(map);
2007 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2008 return 0;
2009 }
2010
2011 /*
2012 * uvm_map_pageable: sets the pageability of a range in a map.
2013 *
2014 * => wires map entries. should not be used for transient page locking.
2015 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
2016 * => regions sepcified as not pageable require lock-down (wired) memory
2017 * and page tables.
2018 * => map must never be read-locked
2019 * => if islocked is TRUE, map is already write-locked
2020 * => we always unlock the map, since we must downgrade to a read-lock
2021 * to call uvm_fault_wire()
2022 * => XXXCDC: check this and try and clean it up.
2023 */
2024
2025 int
2026 uvm_map_pageable(map, start, end, new_pageable, lockflags)
2027 struct vm_map *map;
2028 vaddr_t start, end;
2029 boolean_t new_pageable;
2030 int lockflags;
2031 {
2032 struct vm_map_entry *entry, *start_entry, *failed_entry;
2033 int rv;
2034 #ifdef DIAGNOSTIC
2035 u_int timestamp_save;
2036 #endif
2037 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
2038 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
2039 map, start, end, new_pageable);
2040 KASSERT(map->flags & VM_MAP_PAGEABLE);
2041
2042 if ((lockflags & UVM_LK_ENTER) == 0)
2043 vm_map_lock(map);
2044 VM_MAP_RANGE_CHECK(map, start, end);
2045
2046 /*
2047 * only one pageability change may take place at one time, since
2048 * uvm_fault_wire assumes it will be called only once for each
2049 * wiring/unwiring. therefore, we have to make sure we're actually
2050 * changing the pageability for the entire region. we do so before
2051 * making any changes.
2052 */
2053
2054 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
2055 if ((lockflags & UVM_LK_EXIT) == 0)
2056 vm_map_unlock(map);
2057
2058 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
2059 return EFAULT;
2060 }
2061 entry = start_entry;
2062
2063 /*
2064 * handle wiring and unwiring separately.
2065 */
2066
2067 if (new_pageable) { /* unwire */
2068 UVM_MAP_CLIP_START(map, entry, start);
2069
2070 /*
2071 * unwiring. first ensure that the range to be unwired is
2072 * really wired down and that there are no holes.
2073 */
2074
2075 while ((entry != &map->header) && (entry->start < end)) {
2076 if (entry->wired_count == 0 ||
2077 (entry->end < end &&
2078 (entry->next == &map->header ||
2079 entry->next->start > entry->end))) {
2080 if ((lockflags & UVM_LK_EXIT) == 0)
2081 vm_map_unlock(map);
2082 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
2083 return EINVAL;
2084 }
2085 entry = entry->next;
2086 }
2087
2088 /*
2089 * POSIX 1003.1b - a single munlock call unlocks a region,
2090 * regardless of the number of mlock calls made on that
2091 * region.
2092 */
2093
2094 entry = start_entry;
2095 while ((entry != &map->header) && (entry->start < end)) {
2096 UVM_MAP_CLIP_END(map, entry, end);
2097 if (VM_MAPENT_ISWIRED(entry))
2098 uvm_map_entry_unwire(map, entry);
2099 entry = entry->next;
2100 }
2101 if ((lockflags & UVM_LK_EXIT) == 0)
2102 vm_map_unlock(map);
2103 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2104 return 0;
2105 }
2106
2107 /*
2108 * wire case: in two passes [XXXCDC: ugly block of code here]
2109 *
2110 * 1: holding the write lock, we create any anonymous maps that need
2111 * to be created. then we clip each map entry to the region to
2112 * be wired and increment its wiring count.
2113 *
2114 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2115 * in the pages for any newly wired area (wired_count == 1).
2116 *
2117 * downgrading to a read lock for uvm_fault_wire avoids a possible
2118 * deadlock with another thread that may have faulted on one of
2119 * the pages to be wired (it would mark the page busy, blocking
2120 * us, then in turn block on the map lock that we hold). because
2121 * of problems in the recursive lock package, we cannot upgrade
2122 * to a write lock in vm_map_lookup. thus, any actions that
2123 * require the write lock must be done beforehand. because we
2124 * keep the read lock on the map, the copy-on-write status of the
2125 * entries we modify here cannot change.
2126 */
2127
2128 while ((entry != &map->header) && (entry->start < end)) {
2129 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2130
2131 /*
2132 * perform actions of vm_map_lookup that need the
2133 * write lock on the map: create an anonymous map
2134 * for a copy-on-write region, or an anonymous map
2135 * for a zero-fill region. (XXXCDC: submap case
2136 * ok?)
2137 */
2138
2139 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2140 if (UVM_ET_ISNEEDSCOPY(entry) &&
2141 ((entry->max_protection & VM_PROT_WRITE) ||
2142 (entry->object.uvm_obj == NULL))) {
2143 amap_copy(map, entry, M_WAITOK, TRUE,
2144 start, end);
2145 /* XXXCDC: wait OK? */
2146 }
2147 }
2148 }
2149 UVM_MAP_CLIP_START(map, entry, start);
2150 UVM_MAP_CLIP_END(map, entry, end);
2151 entry->wired_count++;
2152
2153 /*
2154 * Check for holes
2155 */
2156
2157 if (entry->protection == VM_PROT_NONE ||
2158 (entry->end < end &&
2159 (entry->next == &map->header ||
2160 entry->next->start > entry->end))) {
2161
2162 /*
2163 * found one. amap creation actions do not need to
2164 * be undone, but the wired counts need to be restored.
2165 */
2166
2167 while (entry != &map->header && entry->end > start) {
2168 entry->wired_count--;
2169 entry = entry->prev;
2170 }
2171 if ((lockflags & UVM_LK_EXIT) == 0)
2172 vm_map_unlock(map);
2173 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2174 return EINVAL;
2175 }
2176 entry = entry->next;
2177 }
2178
2179 /*
2180 * Pass 2.
2181 */
2182
2183 #ifdef DIAGNOSTIC
2184 timestamp_save = map->timestamp;
2185 #endif
2186 vm_map_busy(map);
2187 vm_map_downgrade(map);
2188
2189 rv = 0;
2190 entry = start_entry;
2191 while (entry != &map->header && entry->start < end) {
2192 if (entry->wired_count == 1) {
2193 rv = uvm_fault_wire(map, entry->start, entry->end,
2194 VM_FAULT_WIREMAX, entry->max_protection);
2195 if (rv) {
2196
2197 /*
2198 * wiring failed. break out of the loop.
2199 * we'll clean up the map below, once we
2200 * have a write lock again.
2201 */
2202
2203 break;
2204 }
2205 }
2206 entry = entry->next;
2207 }
2208
2209 if (rv) { /* failed? */
2210
2211 /*
2212 * Get back to an exclusive (write) lock.
2213 */
2214
2215 vm_map_upgrade(map);
2216 vm_map_unbusy(map);
2217
2218 #ifdef DIAGNOSTIC
2219 if (timestamp_save != map->timestamp)
2220 panic("uvm_map_pageable: stale map");
2221 #endif
2222
2223 /*
2224 * first drop the wiring count on all the entries
2225 * which haven't actually been wired yet.
2226 */
2227
2228 failed_entry = entry;
2229 while (entry != &map->header && entry->start < end) {
2230 entry->wired_count--;
2231 entry = entry->next;
2232 }
2233
2234 /*
2235 * now, unwire all the entries that were successfully
2236 * wired above.
2237 */
2238
2239 entry = start_entry;
2240 while (entry != failed_entry) {
2241 entry->wired_count--;
2242 if (VM_MAPENT_ISWIRED(entry) == 0)
2243 uvm_map_entry_unwire(map, entry);
2244 entry = entry->next;
2245 }
2246 if ((lockflags & UVM_LK_EXIT) == 0)
2247 vm_map_unlock(map);
2248 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2249 return(rv);
2250 }
2251
2252 /* We are holding a read lock here. */
2253 if ((lockflags & UVM_LK_EXIT) == 0) {
2254 vm_map_unbusy(map);
2255 vm_map_unlock_read(map);
2256 } else {
2257
2258 /*
2259 * Get back to an exclusive (write) lock.
2260 */
2261
2262 vm_map_upgrade(map);
2263 vm_map_unbusy(map);
2264 }
2265
2266 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2267 return 0;
2268 }
2269
2270 /*
2271 * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2272 * all mapped regions.
2273 *
2274 * => map must not be locked.
2275 * => if no flags are specified, all regions are unwired.
2276 * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2277 */
2278
2279 int
2280 uvm_map_pageable_all(map, flags, limit)
2281 struct vm_map *map;
2282 int flags;
2283 vsize_t limit;
2284 {
2285 struct vm_map_entry *entry, *failed_entry;
2286 vsize_t size;
2287 int rv;
2288 #ifdef DIAGNOSTIC
2289 u_int timestamp_save;
2290 #endif
2291 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2292 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2293
2294 KASSERT(map->flags & VM_MAP_PAGEABLE);
2295
2296 vm_map_lock(map);
2297
2298 /*
2299 * handle wiring and unwiring separately.
2300 */
2301
2302 if (flags == 0) { /* unwire */
2303
2304 /*
2305 * POSIX 1003.1b -- munlockall unlocks all regions,
2306 * regardless of how many times mlockall has been called.
2307 */
2308
2309 for (entry = map->header.next; entry != &map->header;
2310 entry = entry->next) {
2311 if (VM_MAPENT_ISWIRED(entry))
2312 uvm_map_entry_unwire(map, entry);
2313 }
2314 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2315 vm_map_unlock(map);
2316 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2317 return 0;
2318 }
2319
2320 if (flags & MCL_FUTURE) {
2321
2322 /*
2323 * must wire all future mappings; remember this.
2324 */
2325
2326 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2327 }
2328
2329 if ((flags & MCL_CURRENT) == 0) {
2330
2331 /*
2332 * no more work to do!
2333 */
2334
2335 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2336 vm_map_unlock(map);
2337 return 0;
2338 }
2339
2340 /*
2341 * wire case: in three passes [XXXCDC: ugly block of code here]
2342 *
2343 * 1: holding the write lock, count all pages mapped by non-wired
2344 * entries. if this would cause us to go over our limit, we fail.
2345 *
2346 * 2: still holding the write lock, we create any anonymous maps that
2347 * need to be created. then we increment its wiring count.
2348 *
2349 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2350 * in the pages for any newly wired area (wired_count == 1).
2351 *
2352 * downgrading to a read lock for uvm_fault_wire avoids a possible
2353 * deadlock with another thread that may have faulted on one of
2354 * the pages to be wired (it would mark the page busy, blocking
2355 * us, then in turn block on the map lock that we hold). because
2356 * of problems in the recursive lock package, we cannot upgrade
2357 * to a write lock in vm_map_lookup. thus, any actions that
2358 * require the write lock must be done beforehand. because we
2359 * keep the read lock on the map, the copy-on-write status of the
2360 * entries we modify here cannot change.
2361 */
2362
2363 for (size = 0, entry = map->header.next; entry != &map->header;
2364 entry = entry->next) {
2365 if (entry->protection != VM_PROT_NONE &&
2366 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2367 size += entry->end - entry->start;
2368 }
2369 }
2370
2371 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2372 vm_map_unlock(map);
2373 return ENOMEM;
2374 }
2375
2376 /* XXX non-pmap_wired_count case must be handled by caller */
2377 #ifdef pmap_wired_count
2378 if (limit != 0 &&
2379 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2380 vm_map_unlock(map);
2381 return ENOMEM;
2382 }
2383 #endif
2384
2385 /*
2386 * Pass 2.
2387 */
2388
2389 for (entry = map->header.next; entry != &map->header;
2390 entry = entry->next) {
2391 if (entry->protection == VM_PROT_NONE)
2392 continue;
2393 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2394
2395 /*
2396 * perform actions of vm_map_lookup that need the
2397 * write lock on the map: create an anonymous map
2398 * for a copy-on-write region, or an anonymous map
2399 * for a zero-fill region. (XXXCDC: submap case
2400 * ok?)
2401 */
2402
2403 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2404 if (UVM_ET_ISNEEDSCOPY(entry) &&
2405 ((entry->max_protection & VM_PROT_WRITE) ||
2406 (entry->object.uvm_obj == NULL))) {
2407 amap_copy(map, entry, M_WAITOK, TRUE,
2408 entry->start, entry->end);
2409 /* XXXCDC: wait OK? */
2410 }
2411 }
2412 }
2413 entry->wired_count++;
2414 }
2415
2416 /*
2417 * Pass 3.
2418 */
2419
2420 #ifdef DIAGNOSTIC
2421 timestamp_save = map->timestamp;
2422 #endif
2423 vm_map_busy(map);
2424 vm_map_downgrade(map);
2425
2426 rv = 0;
2427 for (entry = map->header.next; entry != &map->header;
2428 entry = entry->next) {
2429 if (entry->wired_count == 1) {
2430 rv = uvm_fault_wire(map, entry->start, entry->end,
2431 VM_FAULT_WIREMAX, entry->max_protection);
2432 if (rv) {
2433
2434 /*
2435 * wiring failed. break out of the loop.
2436 * we'll clean up the map below, once we
2437 * have a write lock again.
2438 */
2439
2440 break;
2441 }
2442 }
2443 }
2444
2445 if (rv) {
2446
2447 /*
2448 * Get back an exclusive (write) lock.
2449 */
2450
2451 vm_map_upgrade(map);
2452 vm_map_unbusy(map);
2453
2454 #ifdef DIAGNOSTIC
2455 if (timestamp_save != map->timestamp)
2456 panic("uvm_map_pageable_all: stale map");
2457 #endif
2458
2459 /*
2460 * first drop the wiring count on all the entries
2461 * which haven't actually been wired yet.
2462 *
2463 * Skip VM_PROT_NONE entries like we did above.
2464 */
2465
2466 failed_entry = entry;
2467 for (/* nothing */; entry != &map->header;
2468 entry = entry->next) {
2469 if (entry->protection == VM_PROT_NONE)
2470 continue;
2471 entry->wired_count--;
2472 }
2473
2474 /*
2475 * now, unwire all the entries that were successfully
2476 * wired above.
2477 *
2478 * Skip VM_PROT_NONE entries like we did above.
2479 */
2480
2481 for (entry = map->header.next; entry != failed_entry;
2482 entry = entry->next) {
2483 if (entry->protection == VM_PROT_NONE)
2484 continue;
2485 entry->wired_count--;
2486 if (VM_MAPENT_ISWIRED(entry))
2487 uvm_map_entry_unwire(map, entry);
2488 }
2489 vm_map_unlock(map);
2490 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2491 return (rv);
2492 }
2493
2494 /* We are holding a read lock here. */
2495 vm_map_unbusy(map);
2496 vm_map_unlock_read(map);
2497
2498 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2499 return 0;
2500 }
2501
2502 /*
2503 * uvm_map_clean: clean out a map range
2504 *
2505 * => valid flags:
2506 * if (flags & PGO_CLEANIT): dirty pages are cleaned first
2507 * if (flags & PGO_SYNCIO): dirty pages are written synchronously
2508 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2509 * if (flags & PGO_FREE): any cached pages are freed after clean
2510 * => returns an error if any part of the specified range isn't mapped
2511 * => never a need to flush amap layer since the anonymous memory has
2512 * no permanent home, but may deactivate pages there
2513 * => called from sys_msync() and sys_madvise()
2514 * => caller must not write-lock map (read OK).
2515 * => we may sleep while cleaning if SYNCIO [with map read-locked]
2516 */
2517
2518 int
2519 uvm_map_clean(map, start, end, flags)
2520 struct vm_map *map;
2521 vaddr_t start, end;
2522 int flags;
2523 {
2524 struct vm_map_entry *current, *entry;
2525 struct uvm_object *uobj;
2526 struct vm_amap *amap;
2527 struct vm_anon *anon;
2528 struct vm_page *pg;
2529 vaddr_t offset;
2530 vsize_t size;
2531 int error, refs;
2532 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2533
2534 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2535 map, start, end, flags);
2536 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
2537 (PGO_FREE|PGO_DEACTIVATE));
2538
2539 vm_map_lock_read(map);
2540 VM_MAP_RANGE_CHECK(map, start, end);
2541 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2542 vm_map_unlock_read(map);
2543 return EFAULT;
2544 }
2545
2546 /*
2547 * Make a first pass to check for holes.
2548 */
2549
2550 for (current = entry; current->start < end; current = current->next) {
2551 if (UVM_ET_ISSUBMAP(current)) {
2552 vm_map_unlock_read(map);
2553 return EINVAL;
2554 }
2555 if (end <= current->end) {
2556 break;
2557 }
2558 if (current->end != current->next->start) {
2559 vm_map_unlock_read(map);
2560 return EFAULT;
2561 }
2562 }
2563
2564 error = 0;
2565 for (current = entry; start < end; current = current->next) {
2566 amap = current->aref.ar_amap; /* top layer */
2567 uobj = current->object.uvm_obj; /* bottom layer */
2568 KASSERT(start >= current->start);
2569
2570 /*
2571 * No amap cleaning necessary if:
2572 *
2573 * (1) There's no amap.
2574 *
2575 * (2) We're not deactivating or freeing pages.
2576 */
2577
2578 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2579 goto flush_object;
2580
2581 amap_lock(amap);
2582 offset = start - current->start;
2583 size = MIN(end, current->end) - start;
2584 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
2585 anon = amap_lookup(¤t->aref, offset);
2586 if (anon == NULL)
2587 continue;
2588
2589 simple_lock(&anon->an_lock);
2590 pg = anon->u.an_page;
2591 if (pg == NULL) {
2592 simple_unlock(&anon->an_lock);
2593 continue;
2594 }
2595
2596 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2597
2598 /*
2599 * In these first 3 cases, we just deactivate the page.
2600 */
2601
2602 case PGO_CLEANIT|PGO_FREE:
2603 case PGO_CLEANIT|PGO_DEACTIVATE:
2604 case PGO_DEACTIVATE:
2605 deactivate_it:
2606 /*
2607 * skip the page if it's loaned or wired,
2608 * since it shouldn't be on a paging queue
2609 * at all in these cases.
2610 */
2611
2612 uvm_lock_pageq();
2613 if (pg->loan_count != 0 ||
2614 pg->wire_count != 0) {
2615 uvm_unlock_pageq();
2616 simple_unlock(&anon->an_lock);
2617 continue;
2618 }
2619 KASSERT(pg->uanon == anon);
2620 pmap_clear_reference(pg);
2621 uvm_pagedeactivate(pg);
2622 uvm_unlock_pageq();
2623 simple_unlock(&anon->an_lock);
2624 continue;
2625
2626 case PGO_FREE:
2627
2628 /*
2629 * If there are multiple references to
2630 * the amap, just deactivate the page.
2631 */
2632
2633 if (amap_refs(amap) > 1)
2634 goto deactivate_it;
2635
2636 /* skip the page if it's wired */
2637 if (pg->wire_count != 0) {
2638 simple_unlock(&anon->an_lock);
2639 continue;
2640 }
2641 amap_unadd(¤t->aref, offset);
2642 refs = --anon->an_ref;
2643 simple_unlock(&anon->an_lock);
2644 if (refs == 0)
2645 uvm_anfree(anon);
2646 continue;
2647 }
2648 }
2649 amap_unlock(amap);
2650
2651 flush_object:
2652 /*
2653 * flush pages if we've got a valid backing object.
2654 * note that we must always clean object pages before
2655 * freeing them since otherwise we could reveal stale
2656 * data from files.
2657 */
2658
2659 offset = current->offset + (start - current->start);
2660 size = MIN(end, current->end) - start;
2661 if (uobj != NULL) {
2662 simple_lock(&uobj->vmobjlock);
2663 error = (uobj->pgops->pgo_put)(uobj, offset,
2664 offset + size, flags | PGO_CLEANIT);
2665 }
2666 start += size;
2667 }
2668 vm_map_unlock_read(map);
2669 return (error);
2670 }
2671
2672
2673 /*
2674 * uvm_map_checkprot: check protection in map
2675 *
2676 * => must allow specified protection in a fully allocated region.
2677 * => map must be read or write locked by caller.
2678 */
2679
2680 boolean_t
2681 uvm_map_checkprot(map, start, end, protection)
2682 struct vm_map * map;
2683 vaddr_t start, end;
2684 vm_prot_t protection;
2685 {
2686 struct vm_map_entry *entry;
2687 struct vm_map_entry *tmp_entry;
2688
2689 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2690 return(FALSE);
2691 }
2692 entry = tmp_entry;
2693 while (start < end) {
2694 if (entry == &map->header) {
2695 return(FALSE);
2696 }
2697
2698 /*
2699 * no holes allowed
2700 */
2701
2702 if (start < entry->start) {
2703 return(FALSE);
2704 }
2705
2706 /*
2707 * check protection associated with entry
2708 */
2709
2710 if ((entry->protection & protection) != protection) {
2711 return(FALSE);
2712 }
2713 start = entry->end;
2714 entry = entry->next;
2715 }
2716 return(TRUE);
2717 }
2718
2719 /*
2720 * uvmspace_alloc: allocate a vmspace structure.
2721 *
2722 * - structure includes vm_map and pmap
2723 * - XXX: no locking on this structure
2724 * - refcnt set to 1, rest must be init'd by caller
2725 */
2726 struct vmspace *
2727 uvmspace_alloc(min, max)
2728 vaddr_t min, max;
2729 {
2730 struct vmspace *vm;
2731 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2732
2733 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2734 uvmspace_init(vm, NULL, min, max);
2735 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2736 return (vm);
2737 }
2738
2739 /*
2740 * uvmspace_init: initialize a vmspace structure.
2741 *
2742 * - XXX: no locking on this structure
2743 * - refcnt set to 1, rest must me init'd by caller
2744 */
2745 void
2746 uvmspace_init(vm, pmap, min, max)
2747 struct vmspace *vm;
2748 struct pmap *pmap;
2749 vaddr_t min, max;
2750 {
2751 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2752
2753 memset(vm, 0, sizeof(*vm));
2754 uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE);
2755 if (pmap)
2756 pmap_reference(pmap);
2757 else
2758 pmap = pmap_create();
2759 vm->vm_map.pmap = pmap;
2760 vm->vm_refcnt = 1;
2761 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2762 }
2763
2764 /*
2765 * uvmspace_share: share a vmspace between two proceses
2766 *
2767 * - XXX: no locking on vmspace
2768 * - used for vfork, threads(?)
2769 */
2770
2771 void
2772 uvmspace_share(p1, p2)
2773 struct proc *p1, *p2;
2774 {
2775 p2->p_vmspace = p1->p_vmspace;
2776 p1->p_vmspace->vm_refcnt++;
2777 }
2778
2779 /*
2780 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2781 *
2782 * - XXX: no locking on vmspace
2783 */
2784
2785 void
2786 uvmspace_unshare(p)
2787 struct proc *p;
2788 {
2789 struct vmspace *nvm, *ovm = p->p_vmspace;
2790
2791 if (ovm->vm_refcnt == 1)
2792 /* nothing to do: vmspace isn't shared in the first place */
2793 return;
2794
2795 /* make a new vmspace, still holding old one */
2796 nvm = uvmspace_fork(ovm);
2797
2798 pmap_deactivate(p); /* unbind old vmspace */
2799 p->p_vmspace = nvm;
2800 pmap_activate(p); /* switch to new vmspace */
2801
2802 uvmspace_free(ovm); /* drop reference to old vmspace */
2803 }
2804
2805 /*
2806 * uvmspace_exec: the process wants to exec a new program
2807 *
2808 * - XXX: no locking on vmspace
2809 */
2810
2811 void
2812 uvmspace_exec(p, start, end)
2813 struct proc *p;
2814 vaddr_t start, end;
2815 {
2816 struct vmspace *nvm, *ovm = p->p_vmspace;
2817 struct vm_map *map = &ovm->vm_map;
2818
2819 #ifdef __sparc__
2820 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2821 kill_user_windows(p); /* before stack addresses go away */
2822 #endif
2823
2824 /*
2825 * see if more than one process is using this vmspace...
2826 */
2827
2828 if (ovm->vm_refcnt == 1) {
2829
2830 /*
2831 * if p is the only process using its vmspace then we can safely
2832 * recycle that vmspace for the program that is being exec'd.
2833 */
2834
2835 #ifdef SYSVSHM
2836 /*
2837 * SYSV SHM semantics require us to kill all segments on an exec
2838 */
2839
2840 if (ovm->vm_shm)
2841 shmexit(ovm);
2842 #endif
2843
2844 /*
2845 * POSIX 1003.1b -- "lock future mappings" is revoked
2846 * when a process execs another program image.
2847 */
2848
2849 vm_map_lock(map);
2850 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2851 vm_map_unlock(map);
2852
2853 /*
2854 * now unmap the old program
2855 */
2856
2857 uvm_unmap(map, map->min_offset, map->max_offset);
2858
2859 /*
2860 * resize the map
2861 */
2862
2863 vm_map_lock(map);
2864 map->min_offset = start;
2865 map->max_offset = end;
2866 vm_map_unlock(map);
2867 } else {
2868
2869 /*
2870 * p's vmspace is being shared, so we can't reuse it for p since
2871 * it is still being used for others. allocate a new vmspace
2872 * for p
2873 */
2874
2875 nvm = uvmspace_alloc(start, end);
2876
2877 /*
2878 * install new vmspace and drop our ref to the old one.
2879 */
2880
2881 pmap_deactivate(p);
2882 p->p_vmspace = nvm;
2883 pmap_activate(p);
2884
2885 uvmspace_free(ovm);
2886 }
2887 }
2888
2889 /*
2890 * uvmspace_free: free a vmspace data structure
2891 *
2892 * - XXX: no locking on vmspace
2893 */
2894
2895 void
2896 uvmspace_free(vm)
2897 struct vmspace *vm;
2898 {
2899 struct vm_map_entry *dead_entries;
2900 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2901
2902 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2903 if (--vm->vm_refcnt == 0) {
2904
2905 /*
2906 * lock the map, to wait out all other references to it. delete
2907 * all of the mappings and pages they hold, then call the pmap
2908 * module to reclaim anything left.
2909 */
2910
2911 #ifdef SYSVSHM
2912 /* Get rid of any SYSV shared memory segments. */
2913 if (vm->vm_shm != NULL)
2914 shmexit(vm);
2915 #endif
2916 vm_map_lock(&vm->vm_map);
2917 if (vm->vm_map.nentries) {
2918 uvm_unmap_remove(&vm->vm_map,
2919 vm->vm_map.min_offset, vm->vm_map.max_offset,
2920 &dead_entries);
2921 if (dead_entries != NULL)
2922 uvm_unmap_detach(dead_entries, 0);
2923 }
2924 pmap_destroy(vm->vm_map.pmap);
2925 vm->vm_map.pmap = NULL;
2926 pool_put(&uvm_vmspace_pool, vm);
2927 }
2928 UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2929 }
2930
2931 /*
2932 * F O R K - m a i n e n t r y p o i n t
2933 */
2934 /*
2935 * uvmspace_fork: fork a process' main map
2936 *
2937 * => create a new vmspace for child process from parent.
2938 * => parent's map must not be locked.
2939 */
2940
2941 struct vmspace *
2942 uvmspace_fork(vm1)
2943 struct vmspace *vm1;
2944 {
2945 struct vmspace *vm2;
2946 struct vm_map *old_map = &vm1->vm_map;
2947 struct vm_map *new_map;
2948 struct vm_map_entry *old_entry;
2949 struct vm_map_entry *new_entry;
2950 pmap_t new_pmap;
2951 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2952
2953 vm_map_lock(old_map);
2954
2955 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset);
2956 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2957 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2958 new_map = &vm2->vm_map; /* XXX */
2959 new_pmap = new_map->pmap;
2960
2961 old_entry = old_map->header.next;
2962
2963 /*
2964 * go entry-by-entry
2965 */
2966
2967 while (old_entry != &old_map->header) {
2968
2969 /*
2970 * first, some sanity checks on the old entry
2971 */
2972
2973 KASSERT(!UVM_ET_ISSUBMAP(old_entry));
2974 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
2975 !UVM_ET_ISNEEDSCOPY(old_entry));
2976
2977 switch (old_entry->inheritance) {
2978 case MAP_INHERIT_NONE:
2979
2980 /*
2981 * drop the mapping
2982 */
2983
2984 break;
2985
2986 case MAP_INHERIT_SHARE:
2987
2988 /*
2989 * share the mapping: this means we want the old and
2990 * new entries to share amaps and backing objects.
2991 */
2992 /*
2993 * if the old_entry needs a new amap (due to prev fork)
2994 * then we need to allocate it now so that we have
2995 * something we own to share with the new_entry. [in
2996 * other words, we need to clear needs_copy]
2997 */
2998
2999 if (UVM_ET_ISNEEDSCOPY(old_entry)) {
3000 /* get our own amap, clears needs_copy */
3001 amap_copy(old_map, old_entry, M_WAITOK, FALSE,
3002 0, 0);
3003 /* XXXCDC: WAITOK??? */
3004 }
3005
3006 new_entry = uvm_mapent_alloc(new_map);
3007 /* old_entry -> new_entry */
3008 uvm_mapent_copy(old_entry, new_entry);
3009
3010 /* new pmap has nothing wired in it */
3011 new_entry->wired_count = 0;
3012
3013 /*
3014 * gain reference to object backing the map (can't
3015 * be a submap, already checked this case).
3016 */
3017
3018 if (new_entry->aref.ar_amap)
3019 uvm_map_reference_amap(new_entry, AMAP_SHARED);
3020
3021 if (new_entry->object.uvm_obj &&
3022 new_entry->object.uvm_obj->pgops->pgo_reference)
3023 new_entry->object.uvm_obj->
3024 pgops->pgo_reference(
3025 new_entry->object.uvm_obj);
3026
3027 /* insert entry at end of new_map's entry list */
3028 uvm_map_entry_link(new_map, new_map->header.prev,
3029 new_entry);
3030
3031 break;
3032
3033 case MAP_INHERIT_COPY:
3034
3035 /*
3036 * copy-on-write the mapping (using mmap's
3037 * MAP_PRIVATE semantics)
3038 *
3039 * allocate new_entry, adjust reference counts.
3040 * (note that new references are read-only).
3041 */
3042
3043 new_entry = uvm_mapent_alloc(new_map);
3044 /* old_entry -> new_entry */
3045 uvm_mapent_copy(old_entry, new_entry);
3046
3047 if (new_entry->aref.ar_amap)
3048 uvm_map_reference_amap(new_entry, 0);
3049
3050 if (new_entry->object.uvm_obj &&
3051 new_entry->object.uvm_obj->pgops->pgo_reference)
3052 new_entry->object.uvm_obj->pgops->pgo_reference
3053 (new_entry->object.uvm_obj);
3054
3055 /* new pmap has nothing wired in it */
3056 new_entry->wired_count = 0;
3057
3058 new_entry->etype |=
3059 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3060 uvm_map_entry_link(new_map, new_map->header.prev,
3061 new_entry);
3062
3063 /*
3064 * the new entry will need an amap. it will either
3065 * need to be copied from the old entry or created
3066 * from scratch (if the old entry does not have an
3067 * amap). can we defer this process until later
3068 * (by setting "needs_copy") or do we need to copy
3069 * the amap now?
3070 *
3071 * we must copy the amap now if any of the following
3072 * conditions hold:
3073 * 1. the old entry has an amap and that amap is
3074 * being shared. this means that the old (parent)
3075 * process is sharing the amap with another
3076 * process. if we do not clear needs_copy here
3077 * we will end up in a situation where both the
3078 * parent and child process are refering to the
3079 * same amap with "needs_copy" set. if the
3080 * parent write-faults, the fault routine will
3081 * clear "needs_copy" in the parent by allocating
3082 * a new amap. this is wrong because the
3083 * parent is supposed to be sharing the old amap
3084 * and the new amap will break that.
3085 *
3086 * 2. if the old entry has an amap and a non-zero
3087 * wire count then we are going to have to call
3088 * amap_cow_now to avoid page faults in the
3089 * parent process. since amap_cow_now requires
3090 * "needs_copy" to be clear we might as well
3091 * clear it here as well.
3092 *
3093 */
3094
3095 if (old_entry->aref.ar_amap != NULL) {
3096 if ((amap_flags(old_entry->aref.ar_amap) &
3097 AMAP_SHARED) != 0 ||
3098 VM_MAPENT_ISWIRED(old_entry)) {
3099
3100 amap_copy(new_map, new_entry, M_WAITOK,
3101 FALSE, 0, 0);
3102 /* XXXCDC: M_WAITOK ... ok? */
3103 }
3104 }
3105
3106 /*
3107 * if the parent's entry is wired down, then the
3108 * parent process does not want page faults on
3109 * access to that memory. this means that we
3110 * cannot do copy-on-write because we can't write
3111 * protect the old entry. in this case we
3112 * resolve all copy-on-write faults now, using
3113 * amap_cow_now. note that we have already
3114 * allocated any needed amap (above).
3115 */
3116
3117 if (VM_MAPENT_ISWIRED(old_entry)) {
3118
3119 /*
3120 * resolve all copy-on-write faults now
3121 * (note that there is nothing to do if
3122 * the old mapping does not have an amap).
3123 */
3124 if (old_entry->aref.ar_amap)
3125 amap_cow_now(new_map, new_entry);
3126
3127 } else {
3128
3129 /*
3130 * setup mappings to trigger copy-on-write faults
3131 * we must write-protect the parent if it has
3132 * an amap and it is not already "needs_copy"...
3133 * if it is already "needs_copy" then the parent
3134 * has already been write-protected by a previous
3135 * fork operation.
3136 */
3137
3138 if (old_entry->aref.ar_amap &&
3139 !UVM_ET_ISNEEDSCOPY(old_entry)) {
3140 if (old_entry->max_protection & VM_PROT_WRITE) {
3141 pmap_protect(old_map->pmap,
3142 old_entry->start,
3143 old_entry->end,
3144 old_entry->protection &
3145 ~VM_PROT_WRITE);
3146 pmap_update(old_map->pmap);
3147 }
3148 old_entry->etype |= UVM_ET_NEEDSCOPY;
3149 }
3150 }
3151 break;
3152 } /* end of switch statement */
3153 old_entry = old_entry->next;
3154 }
3155
3156 new_map->size = old_map->size;
3157 vm_map_unlock(old_map);
3158
3159 #ifdef SYSVSHM
3160 if (vm1->vm_shm)
3161 shmfork(vm1, vm2);
3162 #endif
3163
3164 #ifdef PMAP_FORK
3165 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3166 #endif
3167
3168 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3169 return(vm2);
3170 }
3171
3172
3173 #if defined(DDB)
3174
3175 /*
3176 * DDB hooks
3177 */
3178
3179 /*
3180 * uvm_map_printit: actually prints the map
3181 */
3182
3183 void
3184 uvm_map_printit(map, full, pr)
3185 struct vm_map *map;
3186 boolean_t full;
3187 void (*pr) __P((const char *, ...));
3188 {
3189 struct vm_map_entry *entry;
3190
3191 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3192 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3193 map->nentries, map->size, map->ref_count, map->timestamp,
3194 map->flags);
3195 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3196 pmap_resident_count(map->pmap));
3197 if (!full)
3198 return;
3199 for (entry = map->header.next; entry != &map->header;
3200 entry = entry->next) {
3201 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
3202 entry, entry->start, entry->end, entry->object.uvm_obj,
3203 (long long)entry->offset, entry->aref.ar_amap,
3204 entry->aref.ar_pageoff);
3205 (*pr)(
3206 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
3207 "wc=%d, adv=%d\n",
3208 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3209 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3210 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3211 entry->protection, entry->max_protection,
3212 entry->inheritance, entry->wired_count, entry->advice);
3213 }
3214 }
3215
3216 /*
3217 * uvm_object_printit: actually prints the object
3218 */
3219
3220 void
3221 uvm_object_printit(uobj, full, pr)
3222 struct uvm_object *uobj;
3223 boolean_t full;
3224 void (*pr) __P((const char *, ...));
3225 {
3226 struct vm_page *pg;
3227 int cnt = 0;
3228
3229 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
3230 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
3231 if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3232 (*pr)("refs=<SYSTEM>\n");
3233 else
3234 (*pr)("refs=%d\n", uobj->uo_refs);
3235
3236 if (!full) {
3237 return;
3238 }
3239 (*pr)(" PAGES <pg,offset>:\n ");
3240 TAILQ_FOREACH(pg, &uobj->memq, listq) {
3241 cnt++;
3242 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
3243 if ((cnt % 3) == 0) {
3244 (*pr)("\n ");
3245 }
3246 }
3247 if ((cnt % 3) != 0) {
3248 (*pr)("\n");
3249 }
3250 }
3251
3252 /*
3253 * uvm_page_printit: actually print the page
3254 */
3255
3256 static const char page_flagbits[] =
3257 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY"
3258 "\11ZERO\15PAGER1";
3259 static const char page_pqflagbits[] =
3260 "\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ";
3261
3262 void
3263 uvm_page_printit(pg, full, pr)
3264 struct vm_page *pg;
3265 boolean_t full;
3266 void (*pr) __P((const char *, ...));
3267 {
3268 struct vm_page *tpg;
3269 struct uvm_object *uobj;
3270 struct pglist *pgl;
3271 char pgbuf[128];
3272 char pqbuf[128];
3273
3274 (*pr)("PAGE %p:\n", pg);
3275 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
3276 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
3277 (*pr)(" flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
3278 pgbuf, pqbuf, pg->wire_count, (long)pg->phys_addr);
3279 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
3280 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
3281 #if defined(UVM_PAGE_TRKOWN)
3282 if (pg->flags & PG_BUSY)
3283 (*pr)(" owning process = %d, tag=%s\n",
3284 pg->owner, pg->owner_tag);
3285 else
3286 (*pr)(" page not busy, no owner\n");
3287 #else
3288 (*pr)(" [page ownership tracking disabled]\n");
3289 #endif
3290
3291 if (!full)
3292 return;
3293
3294 /* cross-verify object/anon */
3295 if ((pg->pqflags & PQ_FREE) == 0) {
3296 if (pg->pqflags & PQ_ANON) {
3297 if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3298 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3299 (pg->uanon) ? pg->uanon->u.an_page : NULL);
3300 else
3301 (*pr)(" anon backpointer is OK\n");
3302 } else {
3303 uobj = pg->uobject;
3304 if (uobj) {
3305 (*pr)(" checking object list\n");
3306 TAILQ_FOREACH(tpg, &uobj->memq, listq) {
3307 if (tpg == pg) {
3308 break;
3309 }
3310 }
3311 if (tpg)
3312 (*pr)(" page found on object list\n");
3313 else
3314 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3315 }
3316 }
3317 }
3318
3319 /* cross-verify page queue */
3320 if (pg->pqflags & PQ_FREE) {
3321 int fl = uvm_page_lookup_freelist(pg);
3322 int color = VM_PGCOLOR_BUCKET(pg);
3323 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
3324 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
3325 } else if (pg->pqflags & PQ_INACTIVE) {
3326 pgl = &uvm.page_inactive;
3327 } else if (pg->pqflags & PQ_ACTIVE) {
3328 pgl = &uvm.page_active;
3329 } else {
3330 pgl = NULL;
3331 }
3332
3333 if (pgl) {
3334 (*pr)(" checking pageq list\n");
3335 TAILQ_FOREACH(tpg, pgl, pageq) {
3336 if (tpg == pg) {
3337 break;
3338 }
3339 }
3340 if (tpg)
3341 (*pr)(" page found on pageq list\n");
3342 else
3343 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3344 }
3345 }
3346 #endif
3347