uvm_map.c revision 1.63 1 /* $NetBSD: uvm_map.c,v 1.63 1999/07/07 21:51:35 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 #include "opt_ddb.h"
70 #include "opt_uvmhist.h"
71 #include "opt_pmap_new.h"
72 #include "opt_sysv.h"
73
74 /*
75 * uvm_map.c: uvm map operations
76 */
77
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/mman.h>
81 #include <sys/proc.h>
82 #include <sys/malloc.h>
83 #include <sys/pool.h>
84
85 #ifdef SYSVSHM
86 #include <sys/shm.h>
87 #endif
88
89 #include <vm/vm.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_kern.h>
92
93 #define UVM_MAP
94 #include <uvm/uvm.h>
95
96 #ifdef DDB
97 #include <uvm/uvm_ddb.h>
98 #endif
99
100
101 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
102 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
103
104 /*
105 * pool for vmspace structures.
106 */
107
108 struct pool uvm_vmspace_pool;
109
110 /*
111 * pool for dynamically-allocated map entries.
112 */
113
114 struct pool uvm_map_entry_pool;
115
116 #ifdef PMAP_GROWKERNEL
117 /*
118 * This global represents the end of the kernel virtual address
119 * space. If we want to exceed this, we must grow the kernel
120 * virtual address space dynamically.
121 *
122 * Note, this variable is locked by kernel_map's lock.
123 */
124 vaddr_t uvm_maxkaddr;
125 #endif
126
127 /*
128 * macros
129 */
130
131 /*
132 * uvm_map_entry_link: insert entry into a map
133 *
134 * => map must be locked
135 */
136 #define uvm_map_entry_link(map, after_where, entry) do { \
137 (map)->nentries++; \
138 (entry)->prev = (after_where); \
139 (entry)->next = (after_where)->next; \
140 (entry)->prev->next = (entry); \
141 (entry)->next->prev = (entry); \
142 } while (0)
143
144 /*
145 * uvm_map_entry_unlink: remove entry from a map
146 *
147 * => map must be locked
148 */
149 #define uvm_map_entry_unlink(map, entry) do { \
150 (map)->nentries--; \
151 (entry)->next->prev = (entry)->prev; \
152 (entry)->prev->next = (entry)->next; \
153 } while (0)
154
155 /*
156 * SAVE_HINT: saves the specified entry as the hint for future lookups.
157 *
158 * => map need not be locked (protected by hint_lock).
159 */
160 #define SAVE_HINT(map,value) do { \
161 simple_lock(&(map)->hint_lock); \
162 (map)->hint = (value); \
163 simple_unlock(&(map)->hint_lock); \
164 } while (0)
165
166 /*
167 * VM_MAP_RANGE_CHECK: check and correct range
168 *
169 * => map must at least be read locked
170 */
171
172 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
173 if (start < vm_map_min(map)) \
174 start = vm_map_min(map); \
175 if (end > vm_map_max(map)) \
176 end = vm_map_max(map); \
177 if (start > end) \
178 start = end; \
179 } while (0)
180
181 /*
182 * local prototypes
183 */
184
185 static vm_map_entry_t uvm_mapent_alloc __P((vm_map_t));
186 static void uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t));
187 static void uvm_mapent_free __P((vm_map_entry_t));
188 static void uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
189
190 /*
191 * local inlines
192 */
193
194 /*
195 * uvm_mapent_alloc: allocate a map entry
196 *
197 * => XXX: static pool for kernel map?
198 */
199
200 static __inline vm_map_entry_t
201 uvm_mapent_alloc(map)
202 vm_map_t map;
203 {
204 vm_map_entry_t me;
205 int s;
206 UVMHIST_FUNC("uvm_mapent_alloc");
207 UVMHIST_CALLED(maphist);
208
209 if ((map->flags & VM_MAP_INTRSAFE) == 0 &&
210 map != kernel_map && kernel_map != NULL /* XXX */) {
211 me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
212 me->flags = 0;
213 /* me can't be null, wait ok */
214 } else {
215 s = splimp(); /* protect kentry_free list with splimp */
216 simple_lock(&uvm.kentry_lock);
217 me = uvm.kentry_free;
218 if (me) uvm.kentry_free = me->next;
219 simple_unlock(&uvm.kentry_lock);
220 splx(s);
221 if (!me)
222 panic("mapent_alloc: out of static map entries, check MAX_KMAPENT");
223 me->flags = UVM_MAP_STATIC;
224 }
225
226 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]",
227 me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map)
228 ? TRUE : FALSE, 0, 0);
229 return(me);
230 }
231
232 /*
233 * uvm_mapent_free: free map entry
234 *
235 * => XXX: static pool for kernel map?
236 */
237
238 static __inline void
239 uvm_mapent_free(me)
240 vm_map_entry_t me;
241 {
242 int s;
243 UVMHIST_FUNC("uvm_mapent_free");
244 UVMHIST_CALLED(maphist);
245 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
246 me, me->flags, 0, 0);
247 if ((me->flags & UVM_MAP_STATIC) == 0) {
248 pool_put(&uvm_map_entry_pool, me);
249 } else {
250 s = splimp(); /* protect kentry_free list with splimp */
251 simple_lock(&uvm.kentry_lock);
252 me->next = uvm.kentry_free;
253 uvm.kentry_free = me;
254 simple_unlock(&uvm.kentry_lock);
255 splx(s);
256 }
257 }
258
259 /*
260 * uvm_mapent_copy: copy a map entry, preserving flags
261 */
262
263 static __inline void
264 uvm_mapent_copy(src, dst)
265 vm_map_entry_t src;
266 vm_map_entry_t dst;
267 {
268
269 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src));
270 }
271
272 /*
273 * uvm_map_entry_unwire: unwire a map entry
274 *
275 * => map should be locked by caller
276 */
277
278 static __inline void
279 uvm_map_entry_unwire(map, entry)
280 vm_map_t map;
281 vm_map_entry_t entry;
282 {
283
284 entry->wired_count = 0;
285 uvm_fault_unwire_locked(map, entry->start, entry->end);
286 }
287
288 /*
289 * uvm_map_init: init mapping system at boot time. note that we allocate
290 * and init the static pool of vm_map_entry_t's for the kernel here.
291 */
292
293 void
294 uvm_map_init()
295 {
296 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
297 #if defined(UVMHIST)
298 static struct uvm_history_ent maphistbuf[100];
299 static struct uvm_history_ent pdhistbuf[100];
300 #endif
301 int lcv;
302
303 /*
304 * first, init logging system.
305 */
306
307 UVMHIST_FUNC("uvm_map_init");
308 UVMHIST_INIT_STATIC(maphist, maphistbuf);
309 UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
310 UVMHIST_CALLED(maphist);
311 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
312 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0,
313 "# uvm_map() successful calls", 0);
314 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
315 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
316 0);
317 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0);
318 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0);
319
320 /*
321 * now set up static pool of kernel map entrys ...
322 */
323
324 simple_lock_init(&uvm.kentry_lock);
325 uvm.kentry_free = NULL;
326 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
327 kernel_map_entry[lcv].next = uvm.kentry_free;
328 uvm.kentry_free = &kernel_map_entry[lcv];
329 }
330
331 /*
332 * initialize the map-related pools.
333 */
334 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
335 0, 0, 0, "vmsppl", 0,
336 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
337 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
338 0, 0, 0, "vmmpepl", 0,
339 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
340 }
341
342 /*
343 * clippers
344 */
345
346 /*
347 * uvm_map_clip_start: ensure that the entry begins at or after
348 * the starting address, if it doesn't we split the entry.
349 *
350 * => caller should use UVM_MAP_CLIP_START macro rather than calling
351 * this directly
352 * => map must be locked by caller
353 */
354
355 void uvm_map_clip_start(map, entry, start)
356 vm_map_t map;
357 vm_map_entry_t entry;
358 vaddr_t start;
359 {
360 vm_map_entry_t new_entry;
361 vaddr_t new_adj;
362
363 /* uvm_map_simplify_entry(map, entry); */ /* XXX */
364
365 /*
366 * Split off the front portion. note that we must insert the new
367 * entry BEFORE this one, so that this entry has the specified
368 * starting address.
369 */
370
371 new_entry = uvm_mapent_alloc(map);
372 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
373
374 new_entry->end = start;
375 new_adj = start - new_entry->start;
376 if (entry->object.uvm_obj)
377 entry->offset += new_adj; /* shift start over */
378 entry->start = start;
379
380 if (new_entry->aref.ar_amap) {
381 amap_splitref(&new_entry->aref, &entry->aref, new_adj);
382 }
383
384 uvm_map_entry_link(map, entry->prev, new_entry);
385
386 if (UVM_ET_ISSUBMAP(entry)) {
387 /* ... unlikely to happen, but play it safe */
388 uvm_map_reference(new_entry->object.sub_map);
389 } else {
390 if (UVM_ET_ISOBJ(entry) &&
391 entry->object.uvm_obj->pgops &&
392 entry->object.uvm_obj->pgops->pgo_reference)
393 entry->object.uvm_obj->pgops->pgo_reference(
394 entry->object.uvm_obj);
395 }
396 }
397
398 /*
399 * uvm_map_clip_end: ensure that the entry ends at or before
400 * the ending address, if it does't we split the reference
401 *
402 * => caller should use UVM_MAP_CLIP_END macro rather than calling
403 * this directly
404 * => map must be locked by caller
405 */
406
407 void
408 uvm_map_clip_end(map, entry, end)
409 vm_map_t map;
410 vm_map_entry_t entry;
411 vaddr_t end;
412 {
413 vm_map_entry_t new_entry;
414 vaddr_t new_adj; /* #bytes we move start forward */
415
416 /*
417 * Create a new entry and insert it
418 * AFTER the specified entry
419 */
420
421 new_entry = uvm_mapent_alloc(map);
422 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
423
424 new_entry->start = entry->end = end;
425 new_adj = end - entry->start;
426 if (new_entry->object.uvm_obj)
427 new_entry->offset += new_adj;
428
429 if (entry->aref.ar_amap)
430 amap_splitref(&entry->aref, &new_entry->aref, new_adj);
431
432 uvm_map_entry_link(map, entry, new_entry);
433
434 if (UVM_ET_ISSUBMAP(entry)) {
435 /* ... unlikely to happen, but play it safe */
436 uvm_map_reference(new_entry->object.sub_map);
437 } else {
438 if (UVM_ET_ISOBJ(entry) &&
439 entry->object.uvm_obj->pgops &&
440 entry->object.uvm_obj->pgops->pgo_reference)
441 entry->object.uvm_obj->pgops->pgo_reference(
442 entry->object.uvm_obj);
443 }
444 }
445
446
447 /*
448 * M A P - m a i n e n t r y p o i n t
449 */
450 /*
451 * uvm_map: establish a valid mapping in a map
452 *
453 * => assume startp is page aligned.
454 * => assume size is a multiple of PAGE_SIZE.
455 * => assume sys_mmap provides enough of a "hint" to have us skip
456 * over text/data/bss area.
457 * => map must be unlocked (we will lock it)
458 * => <uobj,uoffset> value meanings (4 cases):
459 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER
460 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER
461 * [3] <uobj,uoffset> == normal mapping
462 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA
463 *
464 * case [4] is for kernel mappings where we don't know the offset until
465 * we've found a virtual address. note that kernel object offsets are
466 * always relative to vm_map_min(kernel_map).
467 * => XXXCDC: need way to map in external amap?
468 */
469
470 int
471 uvm_map(map, startp, size, uobj, uoffset, flags)
472 vm_map_t map;
473 vaddr_t *startp; /* IN/OUT */
474 vsize_t size;
475 struct uvm_object *uobj;
476 vaddr_t uoffset;
477 uvm_flag_t flags;
478 {
479 vm_map_entry_t prev_entry, new_entry;
480 vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
481 UVM_MAXPROTECTION(flags);
482 vm_inherit_t inherit = UVM_INHERIT(flags);
483 int advice = UVM_ADVICE(flags);
484 UVMHIST_FUNC("uvm_map");
485 UVMHIST_CALLED(maphist);
486
487 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
488 map, *startp, size, flags);
489 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0);
490
491 /*
492 * step 0: sanity check of protection code
493 */
494
495 if ((prot & maxprot) != prot) {
496 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x",
497 prot, maxprot,0,0);
498 return(KERN_PROTECTION_FAILURE);
499 }
500
501 /*
502 * step 1: figure out where to put new VM range
503 */
504
505 if (vm_map_lock_try(map) == FALSE) {
506 if (flags & UVM_FLAG_TRYLOCK)
507 return(KERN_FAILURE);
508 vm_map_lock(map); /* could sleep here */
509 }
510 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
511 uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) {
512 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
513 vm_map_unlock(map);
514 return (KERN_NO_SPACE);
515 }
516
517 #ifdef PMAP_GROWKERNEL
518 {
519 /*
520 * If the kernel pmap can't map the requested space,
521 * then allocate more resources for it.
522 */
523 if (map == kernel_map && uvm_maxkaddr < (*startp + size))
524 uvm_maxkaddr = pmap_growkernel(*startp + size);
525 }
526 #endif
527
528 UVMCNT_INCR(uvm_map_call);
529
530 /*
531 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
532 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in
533 * either case we want to zero it before storing it in the map entry
534 * (because it looks strange and confusing when debugging...)
535 *
536 * if uobj is not null
537 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
538 * and we do not need to change uoffset.
539 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
540 * now (based on the starting address of the map). this case is
541 * for kernel object mappings where we don't know the offset until
542 * the virtual address is found (with uvm_map_findspace). the
543 * offset is the distance we are from the start of the map.
544 */
545
546 if (uobj == NULL) {
547 uoffset = 0;
548 } else {
549 if (uoffset == UVM_UNKNOWN_OFFSET) {
550 #ifdef DIAGNOSTIC
551 if (UVM_OBJ_IS_KERN_OBJECT(uobj) == 0)
552 panic("uvm_map: unknown offset with "
553 "non-kernel object");
554 #endif
555 uoffset = *startp - vm_map_min(kernel_map);
556 }
557 }
558
559 /*
560 * step 2: try and insert in map by extending previous entry, if
561 * possible
562 * XXX: we don't try and pull back the next entry. might be useful
563 * for a stack, but we are currently allocating our stack in advance.
564 */
565
566 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
567 prev_entry->end == *startp && prev_entry != &map->header &&
568 prev_entry->object.uvm_obj == uobj) {
569
570 if (uobj && prev_entry->offset +
571 (prev_entry->end - prev_entry->start) != uoffset)
572 goto step3;
573
574 if (UVM_ET_ISSUBMAP(prev_entry))
575 goto step3;
576
577 if (prev_entry->protection != prot ||
578 prev_entry->max_protection != maxprot)
579 goto step3;
580
581 if (prev_entry->inheritance != inherit ||
582 prev_entry->advice != advice)
583 goto step3;
584
585 /* wiring status must match (new area is unwired) */
586 if (VM_MAPENT_ISWIRED(prev_entry))
587 goto step3;
588
589 /*
590 * can't extend a shared amap. note: no need to lock amap to
591 * look at refs since we don't care about its exact value.
592 * if it is one (i.e. we have only reference) it will stay there
593 */
594
595 if (prev_entry->aref.ar_amap &&
596 amap_refs(prev_entry->aref.ar_amap) != 1) {
597 goto step3;
598 }
599
600 /* got it! */
601
602 UVMCNT_INCR(map_backmerge);
603 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0);
604
605 /*
606 * drop our reference to uobj since we are extending a reference
607 * that we already have (the ref count can not drop to zero).
608 */
609 if (uobj && uobj->pgops->pgo_detach)
610 uobj->pgops->pgo_detach(uobj);
611
612 if (prev_entry->aref.ar_amap) {
613 amap_extend(prev_entry, size);
614 }
615
616 prev_entry->end += size;
617 map->size += size;
618
619 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
620 vm_map_unlock(map);
621 return (KERN_SUCCESS);
622
623 }
624 step3:
625 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0);
626
627 /*
628 * check for possible forward merge (which we don't do) and count
629 * the number of times we missed a *possible* chance to merge more
630 */
631
632 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
633 prev_entry->next != &map->header &&
634 prev_entry->next->start == (*startp + size))
635 UVMCNT_INCR(map_forwmerge);
636
637 /*
638 * step 3: allocate new entry and link it in
639 */
640
641 new_entry = uvm_mapent_alloc(map);
642 new_entry->start = *startp;
643 new_entry->end = new_entry->start + size;
644 new_entry->object.uvm_obj = uobj;
645 new_entry->offset = uoffset;
646
647 if (uobj)
648 new_entry->etype = UVM_ET_OBJ;
649 else
650 new_entry->etype = 0;
651
652 if (flags & UVM_FLAG_COPYONW) {
653 new_entry->etype |= UVM_ET_COPYONWRITE;
654 if ((flags & UVM_FLAG_OVERLAY) == 0)
655 new_entry->etype |= UVM_ET_NEEDSCOPY;
656 }
657
658 new_entry->protection = prot;
659 new_entry->max_protection = maxprot;
660 new_entry->inheritance = inherit;
661 new_entry->wired_count = 0;
662 new_entry->advice = advice;
663 if (flags & UVM_FLAG_OVERLAY) {
664 /*
665 * to_add: for BSS we overallocate a little since we
666 * are likely to extend
667 */
668 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
669 UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
670 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK);
671 new_entry->aref.ar_pageoff = 0;
672 new_entry->aref.ar_amap = amap;
673 } else {
674 new_entry->aref.ar_amap = NULL;
675 }
676
677 uvm_map_entry_link(map, prev_entry, new_entry);
678
679 map->size += size;
680
681 /*
682 * Update the free space hint
683 */
684
685 if ((map->first_free == prev_entry) &&
686 (prev_entry->end >= new_entry->start))
687 map->first_free = new_entry;
688
689 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
690 vm_map_unlock(map);
691 return(KERN_SUCCESS);
692 }
693
694 /*
695 * uvm_map_lookup_entry: find map entry at or before an address
696 *
697 * => map must at least be read-locked by caller
698 * => entry is returned in "entry"
699 * => return value is true if address is in the returned entry
700 */
701
702 boolean_t
703 uvm_map_lookup_entry(map, address, entry)
704 vm_map_t map;
705 vaddr_t address;
706 vm_map_entry_t *entry; /* OUT */
707 {
708 vm_map_entry_t cur;
709 vm_map_entry_t last;
710 UVMHIST_FUNC("uvm_map_lookup_entry");
711 UVMHIST_CALLED(maphist);
712
713 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
714 map, address, entry, 0);
715
716 /*
717 * start looking either from the head of the
718 * list, or from the hint.
719 */
720
721 simple_lock(&map->hint_lock);
722 cur = map->hint;
723 simple_unlock(&map->hint_lock);
724
725 if (cur == &map->header)
726 cur = cur->next;
727
728 UVMCNT_INCR(uvm_mlk_call);
729 if (address >= cur->start) {
730 /*
731 * go from hint to end of list.
732 *
733 * but first, make a quick check to see if
734 * we are already looking at the entry we
735 * want (which is usually the case).
736 * note also that we don't need to save the hint
737 * here... it is the same hint (unless we are
738 * at the header, in which case the hint didn't
739 * buy us anything anyway).
740 */
741 last = &map->header;
742 if ((cur != last) && (cur->end > address)) {
743 UVMCNT_INCR(uvm_mlk_hint);
744 *entry = cur;
745 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
746 cur, 0, 0, 0);
747 return (TRUE);
748 }
749 } else {
750 /*
751 * go from start to hint, *inclusively*
752 */
753 last = cur->next;
754 cur = map->header.next;
755 }
756
757 /*
758 * search linearly
759 */
760
761 while (cur != last) {
762 if (cur->end > address) {
763 if (address >= cur->start) {
764 /*
765 * save this lookup for future
766 * hints, and return
767 */
768
769 *entry = cur;
770 SAVE_HINT(map, cur);
771 UVMHIST_LOG(maphist,"<- search got it (0x%x)",
772 cur, 0, 0, 0);
773 return (TRUE);
774 }
775 break;
776 }
777 cur = cur->next;
778 }
779 *entry = cur->prev;
780 SAVE_HINT(map, *entry);
781 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
782 return (FALSE);
783 }
784
785
786 /*
787 * uvm_map_findspace: find "length" sized space in "map".
788 *
789 * => "hint" is a hint about where we want it, unless fixed is true
790 * (in which case we insist on using "hint").
791 * => "result" is VA returned
792 * => uobj/uoffset are to be used to handle VAC alignment, if required
793 * => caller must at least have read-locked map
794 * => returns NULL on failure, or pointer to prev. map entry if success
795 * => note this is a cross between the old vm_map_findspace and vm_map_find
796 */
797
798 vm_map_entry_t
799 uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed)
800 vm_map_t map;
801 vaddr_t hint;
802 vsize_t length;
803 vaddr_t *result; /* OUT */
804 struct uvm_object *uobj;
805 vaddr_t uoffset;
806 boolean_t fixed;
807 {
808 vm_map_entry_t entry, next, tmp;
809 vaddr_t end;
810 UVMHIST_FUNC("uvm_map_findspace");
811 UVMHIST_CALLED(maphist);
812
813 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)",
814 map, hint, length, fixed);
815
816 if (hint < map->min_offset) { /* check ranges ... */
817 if (fixed) {
818 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
819 return(NULL);
820 }
821 hint = map->min_offset;
822 }
823 if (hint > map->max_offset) {
824 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
825 hint, map->min_offset, map->max_offset, 0);
826 return(NULL);
827 }
828
829 /*
830 * Look for the first possible address; if there's already
831 * something at this address, we have to start after it.
832 */
833
834 if (!fixed && hint == map->min_offset) {
835 if ((entry = map->first_free) != &map->header)
836 hint = entry->end;
837 } else {
838 if (uvm_map_lookup_entry(map, hint, &tmp)) {
839 /* "hint" address already in use ... */
840 if (fixed) {
841 UVMHIST_LOG(maphist,"<- fixed & VA in use",
842 0, 0, 0, 0);
843 return(NULL);
844 }
845 hint = tmp->end;
846 }
847 entry = tmp;
848 }
849
850 /*
851 * Look through the rest of the map, trying to fit a new region in
852 * the gap between existing regions, or after the very last region.
853 * note: entry->end = base VA of current gap,
854 * next->start = VA of end of current gap
855 */
856 for (;; hint = (entry = next)->end) {
857 /*
858 * Find the end of the proposed new region. Be sure we didn't
859 * go beyond the end of the map, or wrap around the address;
860 * if so, we lose. Otherwise, if this is the last entry, or
861 * if the proposed new region fits before the next entry, we
862 * win.
863 */
864
865 #ifdef PMAP_PREFER
866 /*
867 * push hint forward as needed to avoid VAC alias problems.
868 * we only do this if a valid offset is specified.
869 */
870 if (!fixed && uoffset != UVM_UNKNOWN_OFFSET)
871 PMAP_PREFER(uoffset, &hint);
872 #endif
873 end = hint + length;
874 if (end > map->max_offset || end < hint) {
875 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
876 return (NULL);
877 }
878 next = entry->next;
879 if (next == &map->header || next->start >= end)
880 break;
881 if (fixed) {
882 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
883 return(NULL); /* only one shot at it ... */
884 }
885 }
886 SAVE_HINT(map, entry);
887 *result = hint;
888 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0);
889 return (entry);
890 }
891
892 /*
893 * U N M A P - m a i n h e l p e r f u n c t i o n s
894 */
895
896 /*
897 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
898 *
899 * => caller must check alignment and size
900 * => map must be locked by caller
901 * => we return a list of map entries that we've remove from the map
902 * in "entry_list"
903 */
904
905 int
906 uvm_unmap_remove(map, start, end, entry_list)
907 vm_map_t map;
908 vaddr_t start,end;
909 vm_map_entry_t *entry_list; /* OUT */
910 {
911 vm_map_entry_t entry, first_entry, next;
912 vaddr_t len;
913 UVMHIST_FUNC("uvm_unmap_remove");
914 UVMHIST_CALLED(maphist);
915
916 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
917 map, start, end, 0);
918
919 VM_MAP_RANGE_CHECK(map, start, end);
920
921 /*
922 * find first entry
923 */
924 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
925 /* clip and go... */
926 entry = first_entry;
927 UVM_MAP_CLIP_START(map, entry, start);
928 /* critical! prevents stale hint */
929 SAVE_HINT(map, entry->prev);
930
931 } else {
932 entry = first_entry->next;
933 }
934
935 /*
936 * Save the free space hint
937 */
938
939 if (map->first_free->start >= start)
940 map->first_free = entry->prev;
941
942 /*
943 * note: we now re-use first_entry for a different task. we remove
944 * a number of map entries from the map and save them in a linked
945 * list headed by "first_entry". once we remove them from the map
946 * the caller should unlock the map and drop the references to the
947 * backing objects [c.f. uvm_unmap_detach]. the object is to
948 * seperate unmapping from reference dropping. why?
949 * [1] the map has to be locked for unmapping
950 * [2] the map need not be locked for reference dropping
951 * [3] dropping references may trigger pager I/O, and if we hit
952 * a pager that does synchronous I/O we may have to wait for it.
953 * [4] we would like all waiting for I/O to occur with maps unlocked
954 * so that we don't block other threads.
955 */
956 first_entry = NULL;
957 *entry_list = NULL; /* to be safe */
958
959 /*
960 * break up the area into map entry sized regions and unmap. note
961 * that all mappings have to be removed before we can even consider
962 * dropping references to amaps or VM objects (otherwise we could end
963 * up with a mapping to a page on the free list which would be very bad)
964 */
965
966 while ((entry != &map->header) && (entry->start < end)) {
967
968 UVM_MAP_CLIP_END(map, entry, end);
969 next = entry->next;
970 len = entry->end - entry->start;
971
972 /*
973 * unwire before removing addresses from the pmap; otherwise
974 * unwiring will put the entries back into the pmap (XXX).
975 */
976
977 if (VM_MAPENT_ISWIRED(entry))
978 uvm_map_entry_unwire(map, entry);
979
980 /*
981 * special case: handle mappings to anonymous kernel objects.
982 * we want to free these pages right away...
983 */
984 if (UVM_ET_ISOBJ(entry) &&
985 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
986 #ifdef DIAGNOSTIC
987 if (vm_map_pmap(map) != pmap_kernel())
988 panic("uvm_unmap_remove: kernel object "
989 "mapped by non-kernel map");
990 #endif
991
992 /*
993 * note: kernel object mappings are currently used in
994 * two ways:
995 * [1] "normal" mappings of pages in the kernel object
996 * [2] uvm_km_valloc'd allocations in which we
997 * pmap_enter in some non-kernel-object page
998 * (e.g. vmapbuf).
999 *
1000 * for case [1], we need to remove the mapping from
1001 * the pmap and then remove the page from the kernel
1002 * object (because, once pages in a kernel object are
1003 * unmapped they are no longer needed, unlike, say,
1004 * a vnode where you might want the data to persist
1005 * until flushed out of a queue).
1006 *
1007 * for case [2], we need to remove the mapping from
1008 * the pmap. there shouldn't be any pages at the
1009 * specified offset in the kernel object [but it
1010 * doesn't hurt to call uvm_km_pgremove just to be
1011 * safe?]
1012 *
1013 * uvm_km_pgremove currently does the following:
1014 * for pages in the kernel object in range:
1015 * - drops the swap slot
1016 * - uvm_pagefree the page
1017 *
1018 * note there is version of uvm_km_pgremove() that
1019 * is used for "intrsafe" objects.
1020 */
1021
1022 /*
1023 * remove mappings from pmap and drop the pages
1024 * from the object. offsets are always relative
1025 * to vm_map_min(kernel_map).
1026 */
1027 if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) {
1028 #if defined(PMAP_NEW)
1029 pmap_kremove(entry->start, len);
1030 #else
1031 pmap_remove(pmap_kernel(), entry->start,
1032 entry->start + len);
1033 #endif
1034 uvm_km_pgremove_intrsafe(entry->object.uvm_obj,
1035 entry->start - vm_map_min(kernel_map),
1036 entry->end - vm_map_min(kernel_map));
1037 } else {
1038 pmap_remove(pmap_kernel(), entry->start,
1039 entry->start + len);
1040 uvm_km_pgremove(entry->object.uvm_obj,
1041 entry->start - vm_map_min(kernel_map),
1042 entry->end - vm_map_min(kernel_map));
1043 }
1044
1045 /*
1046 * null out kernel_object reference, we've just
1047 * dropped it
1048 */
1049 entry->etype &= ~UVM_ET_OBJ;
1050 entry->object.uvm_obj = NULL; /* to be safe */
1051
1052 } else {
1053 /*
1054 * remove mappings the standard way.
1055 */
1056 pmap_remove(map->pmap, entry->start, entry->end);
1057 }
1058
1059 /*
1060 * remove entry from map and put it on our list of entries
1061 * that we've nuked. then go do next entry.
1062 */
1063 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0);
1064 uvm_map_entry_unlink(map, entry);
1065 map->size -= len;
1066 entry->next = first_entry;
1067 first_entry = entry;
1068 entry = next; /* next entry, please */
1069 }
1070
1071 /*
1072 * now we've cleaned up the map and are ready for the caller to drop
1073 * references to the mapped objects.
1074 */
1075
1076 *entry_list = first_entry;
1077 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1078 return(KERN_SUCCESS);
1079 }
1080
1081 /*
1082 * uvm_unmap_detach: drop references in a chain of map entries
1083 *
1084 * => we will free the map entries as we traverse the list.
1085 */
1086
1087 void
1088 uvm_unmap_detach(first_entry, amap_unref_flags)
1089 vm_map_entry_t first_entry;
1090 int amap_unref_flags;
1091 {
1092 vm_map_entry_t next_entry;
1093 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1094
1095 while (first_entry) {
1096
1097 #ifdef DIAGNOSTIC
1098 /*
1099 * sanity check
1100 */
1101 /* was part of vm_map_entry_delete() */
1102 if (VM_MAPENT_ISWIRED(first_entry))
1103 panic("unmap: still wired!");
1104 #endif
1105
1106 UVMHIST_LOG(maphist,
1107 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1108 first_entry, first_entry->aref.ar_amap,
1109 first_entry->object.uvm_obj,
1110 UVM_ET_ISSUBMAP(first_entry));
1111
1112 /*
1113 * drop reference to amap, if we've got one
1114 */
1115
1116 if (first_entry->aref.ar_amap)
1117 amap_unref(first_entry, amap_unref_flags);
1118
1119 /*
1120 * drop reference to our backing object, if we've got one
1121 */
1122
1123 if (UVM_ET_ISSUBMAP(first_entry)) {
1124 /* ... unlikely to happen, but play it safe */
1125 uvm_map_deallocate(first_entry->object.sub_map);
1126 } else {
1127 if (UVM_ET_ISOBJ(first_entry) &&
1128 first_entry->object.uvm_obj->pgops->pgo_detach)
1129 first_entry->object.uvm_obj->pgops->
1130 pgo_detach(first_entry->object.uvm_obj);
1131 }
1132
1133 /*
1134 * next entry
1135 */
1136 next_entry = first_entry->next;
1137 uvm_mapent_free(first_entry);
1138 first_entry = next_entry;
1139 }
1140
1141 /*
1142 * done!
1143 */
1144 UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1145 return;
1146 }
1147
1148 /*
1149 * E X T R A C T I O N F U N C T I O N S
1150 */
1151
1152 /*
1153 * uvm_map_reserve: reserve space in a vm_map for future use.
1154 *
1155 * => we reserve space in a map by putting a dummy map entry in the
1156 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1157 * => map should be unlocked (we will write lock it)
1158 * => we return true if we were able to reserve space
1159 * => XXXCDC: should be inline?
1160 */
1161
1162 int
1163 uvm_map_reserve(map, size, offset, raddr)
1164 vm_map_t map;
1165 vsize_t size;
1166 vaddr_t offset; /* hint for pmap_prefer */
1167 vaddr_t *raddr; /* OUT: reserved VA */
1168 {
1169 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1170
1171 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1172 map,size,offset,raddr);
1173
1174 size = round_page(size);
1175 if (*raddr < vm_map_min(map))
1176 *raddr = vm_map_min(map); /* hint */
1177
1178 /*
1179 * reserve some virtual space.
1180 */
1181
1182 if (uvm_map(map, raddr, size, NULL, offset,
1183 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1184 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) {
1185 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1186 return (FALSE);
1187 }
1188
1189 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1190 return (TRUE);
1191 }
1192
1193 /*
1194 * uvm_map_replace: replace a reserved (blank) area of memory with
1195 * real mappings.
1196 *
1197 * => caller must WRITE-LOCK the map
1198 * => we return TRUE if replacement was a success
1199 * => we expect the newents chain to have nnewents entrys on it and
1200 * we expect newents->prev to point to the last entry on the list
1201 * => note newents is allowed to be NULL
1202 */
1203
1204 int
1205 uvm_map_replace(map, start, end, newents, nnewents)
1206 struct vm_map *map;
1207 vaddr_t start, end;
1208 vm_map_entry_t newents;
1209 int nnewents;
1210 {
1211 vm_map_entry_t oldent, last;
1212 UVMHIST_FUNC("uvm_map_replace");
1213 UVMHIST_CALLED(maphist);
1214
1215 /*
1216 * first find the blank map entry at the specified address
1217 */
1218
1219 if (!uvm_map_lookup_entry(map, start, &oldent)) {
1220 return(FALSE);
1221 }
1222
1223 /*
1224 * check to make sure we have a proper blank entry
1225 */
1226
1227 if (oldent->start != start || oldent->end != end ||
1228 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1229 return (FALSE);
1230 }
1231
1232 #ifdef DIAGNOSTIC
1233 /*
1234 * sanity check the newents chain
1235 */
1236 {
1237 vm_map_entry_t tmpent = newents;
1238 int nent = 0;
1239 vaddr_t cur = start;
1240
1241 while (tmpent) {
1242 nent++;
1243 if (tmpent->start < cur)
1244 panic("uvm_map_replace1");
1245 if (tmpent->start > tmpent->end || tmpent->end > end) {
1246 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1247 tmpent->start, tmpent->end, end);
1248 panic("uvm_map_replace2");
1249 }
1250 cur = tmpent->end;
1251 if (tmpent->next) {
1252 if (tmpent->next->prev != tmpent)
1253 panic("uvm_map_replace3");
1254 } else {
1255 if (newents->prev != tmpent)
1256 panic("uvm_map_replace4");
1257 }
1258 tmpent = tmpent->next;
1259 }
1260 if (nent != nnewents)
1261 panic("uvm_map_replace5");
1262 }
1263 #endif
1264
1265 /*
1266 * map entry is a valid blank! replace it. (this does all the
1267 * work of map entry link/unlink...).
1268 */
1269
1270 if (newents) {
1271
1272 last = newents->prev; /* we expect this */
1273
1274 /* critical: flush stale hints out of map */
1275 SAVE_HINT(map, newents);
1276 if (map->first_free == oldent)
1277 map->first_free = last;
1278
1279 last->next = oldent->next;
1280 last->next->prev = last;
1281 newents->prev = oldent->prev;
1282 newents->prev->next = newents;
1283 map->nentries = map->nentries + (nnewents - 1);
1284
1285 } else {
1286
1287 /* critical: flush stale hints out of map */
1288 SAVE_HINT(map, oldent->prev);
1289 if (map->first_free == oldent)
1290 map->first_free = oldent->prev;
1291
1292 /* NULL list of new entries: just remove the old one */
1293 uvm_map_entry_unlink(map, oldent);
1294 }
1295
1296
1297 /*
1298 * now we can free the old blank entry, unlock the map and return.
1299 */
1300
1301 uvm_mapent_free(oldent);
1302 return(TRUE);
1303 }
1304
1305 /*
1306 * uvm_map_extract: extract a mapping from a map and put it somewhere
1307 * (maybe removing the old mapping)
1308 *
1309 * => maps should be unlocked (we will write lock them)
1310 * => returns 0 on success, error code otherwise
1311 * => start must be page aligned
1312 * => len must be page sized
1313 * => flags:
1314 * UVM_EXTRACT_REMOVE: remove mappings from srcmap
1315 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1316 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1317 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1318 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1319 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1320 * be used from within the kernel in a kernel level map <<<
1321 */
1322
1323 int
1324 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1325 vm_map_t srcmap, dstmap;
1326 vaddr_t start, *dstaddrp;
1327 vsize_t len;
1328 int flags;
1329 {
1330 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1331 oldstart;
1332 vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry;
1333 vm_map_entry_t oldentry;
1334 vsize_t elen;
1335 int nchain, error, copy_ok;
1336 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1337 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1338 len,0);
1339 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1340
1341 #ifdef DIAGNOSTIC
1342 /*
1343 * step 0: sanity check: start must be on a page boundary, length
1344 * must be page sized. can't ask for CONTIG/QREF if you asked for
1345 * REMOVE.
1346 */
1347 if ((start & PAGE_MASK) || (len & PAGE_MASK))
1348 panic("uvm_map_extract1");
1349 if (flags & UVM_EXTRACT_REMOVE)
1350 if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF))
1351 panic("uvm_map_extract2");
1352 #endif
1353
1354
1355 /*
1356 * step 1: reserve space in the target map for the extracted area
1357 */
1358
1359 dstaddr = *dstaddrp;
1360 if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE)
1361 return(ENOMEM);
1362 *dstaddrp = dstaddr; /* pass address back to caller */
1363 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0);
1364
1365
1366 /*
1367 * step 2: setup for the extraction process loop by init'ing the
1368 * map entry chain, locking src map, and looking up the first useful
1369 * entry in the map.
1370 */
1371
1372 end = start + len;
1373 newend = dstaddr + len;
1374 chain = endchain = NULL;
1375 nchain = 0;
1376 vm_map_lock(srcmap);
1377
1378 if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1379
1380 /* "start" is within an entry */
1381 if (flags & UVM_EXTRACT_QREF) {
1382 /*
1383 * for quick references we don't clip the entry, so
1384 * the entry may map space "before" the starting
1385 * virtual address... this is the "fudge" factor
1386 * (which can be non-zero only the first time
1387 * through the "while" loop in step 3).
1388 */
1389 fudge = start - entry->start;
1390 } else {
1391 /*
1392 * normal reference: we clip the map to fit (thus
1393 * fudge is zero)
1394 */
1395 UVM_MAP_CLIP_START(srcmap, entry, start);
1396 SAVE_HINT(srcmap, entry->prev);
1397 fudge = 0;
1398 }
1399
1400 } else {
1401
1402 /* "start" is not within an entry ... skip to next entry */
1403 if (flags & UVM_EXTRACT_CONTIG) {
1404 error = EINVAL;
1405 goto bad; /* definite hole here ... */
1406 }
1407
1408 entry = entry->next;
1409 fudge = 0;
1410 }
1411 /* save values from srcmap for step 6 */
1412 orig_entry = entry;
1413 orig_fudge = fudge;
1414
1415
1416 /*
1417 * step 3: now start looping through the map entries, extracting
1418 * as we go.
1419 */
1420
1421 while (entry->start < end && entry != &srcmap->header) {
1422
1423 /* if we are not doing a quick reference, clip it */
1424 if ((flags & UVM_EXTRACT_QREF) == 0)
1425 UVM_MAP_CLIP_END(srcmap, entry, end);
1426
1427 /* clear needs_copy (allow chunking) */
1428 if (UVM_ET_ISNEEDSCOPY(entry)) {
1429 if (fudge)
1430 oldstart = entry->start;
1431 else
1432 oldstart = 0; /* XXX: gcc */
1433 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1434 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */
1435 error = ENOMEM;
1436 goto bad;
1437 }
1438 /* amap_copy could clip (during chunk)! update fudge */
1439 if (fudge) {
1440 fudge = fudge - (entry->start - oldstart);
1441 orig_fudge = fudge;
1442 }
1443 }
1444
1445 /* calculate the offset of this from "start" */
1446 oldoffset = (entry->start + fudge) - start;
1447
1448 /* allocate a new map entry */
1449 newentry = uvm_mapent_alloc(dstmap);
1450 if (newentry == NULL) {
1451 error = ENOMEM;
1452 goto bad;
1453 }
1454
1455 /* set up new map entry */
1456 newentry->next = NULL;
1457 newentry->prev = endchain;
1458 newentry->start = dstaddr + oldoffset;
1459 newentry->end =
1460 newentry->start + (entry->end - (entry->start + fudge));
1461 if (newentry->end > newend || newentry->end < newentry->start)
1462 newentry->end = newend;
1463 newentry->object.uvm_obj = entry->object.uvm_obj;
1464 if (newentry->object.uvm_obj) {
1465 if (newentry->object.uvm_obj->pgops->pgo_reference)
1466 newentry->object.uvm_obj->pgops->
1467 pgo_reference(newentry->object.uvm_obj);
1468 newentry->offset = entry->offset + fudge;
1469 } else {
1470 newentry->offset = 0;
1471 }
1472 newentry->etype = entry->etype;
1473 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1474 entry->max_protection : entry->protection;
1475 newentry->max_protection = entry->max_protection;
1476 newentry->inheritance = entry->inheritance;
1477 newentry->wired_count = 0;
1478 newentry->aref.ar_amap = entry->aref.ar_amap;
1479 if (newentry->aref.ar_amap) {
1480 newentry->aref.ar_pageoff =
1481 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1482 amap_ref(newentry, AMAP_SHARED |
1483 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1484 } else {
1485 newentry->aref.ar_pageoff = 0;
1486 }
1487 newentry->advice = entry->advice;
1488
1489 /* now link it on the chain */
1490 nchain++;
1491 if (endchain == NULL) {
1492 chain = endchain = newentry;
1493 } else {
1494 endchain->next = newentry;
1495 endchain = newentry;
1496 }
1497
1498 /* end of 'while' loop! */
1499 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1500 (entry->next == &srcmap->header ||
1501 entry->next->start != entry->end)) {
1502 error = EINVAL;
1503 goto bad;
1504 }
1505 entry = entry->next;
1506 fudge = 0;
1507 }
1508
1509
1510 /*
1511 * step 4: close off chain (in format expected by uvm_map_replace)
1512 */
1513
1514 if (chain)
1515 chain->prev = endchain;
1516
1517
1518 /*
1519 * step 5: attempt to lock the dest map so we can pmap_copy.
1520 * note usage of copy_ok:
1521 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1522 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1523 */
1524
1525 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1526
1527 copy_ok = 1;
1528 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1529 nchain)) {
1530 if (srcmap != dstmap)
1531 vm_map_unlock(dstmap);
1532 error = EIO;
1533 goto bad;
1534 }
1535
1536 } else {
1537
1538 copy_ok = 0;
1539 /* replace defered until step 7 */
1540
1541 }
1542
1543
1544 /*
1545 * step 6: traverse the srcmap a second time to do the following:
1546 * - if we got a lock on the dstmap do pmap_copy
1547 * - if UVM_EXTRACT_REMOVE remove the entries
1548 * we make use of orig_entry and orig_fudge (saved in step 2)
1549 */
1550
1551 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1552
1553 /* purge possible stale hints from srcmap */
1554 if (flags & UVM_EXTRACT_REMOVE) {
1555 SAVE_HINT(srcmap, orig_entry->prev);
1556 if (srcmap->first_free->start >= start)
1557 srcmap->first_free = orig_entry->prev;
1558 }
1559
1560 entry = orig_entry;
1561 fudge = orig_fudge;
1562 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */
1563
1564 while (entry->start < end && entry != &srcmap->header) {
1565
1566 if (copy_ok) {
1567 oldoffset = (entry->start + fudge) - start;
1568 elen = min(end, entry->end) - (entry->start + fudge);
1569 pmap_copy(dstmap->pmap, srcmap->pmap, dstaddr + oldoffset,
1570 elen, entry->start + fudge);
1571 }
1572
1573 /* we advance "entry" in the following if statement */
1574 if (flags & UVM_EXTRACT_REMOVE) {
1575 pmap_remove(srcmap->pmap, entry->start,
1576 entry->end);
1577 oldentry = entry; /* save entry */
1578 entry = entry->next; /* advance */
1579 uvm_map_entry_unlink(srcmap, oldentry);
1580 /* add to dead list */
1581 oldentry->next = deadentry;
1582 deadentry = oldentry;
1583 } else {
1584 entry = entry->next; /* advance */
1585 }
1586
1587 /* end of 'while' loop */
1588 fudge = 0;
1589 }
1590
1591 /*
1592 * unlock dstmap. we will dispose of deadentry in
1593 * step 7 if needed
1594 */
1595 if (copy_ok && srcmap != dstmap)
1596 vm_map_unlock(dstmap);
1597
1598 }
1599 else
1600 deadentry = NULL; /* XXX: gcc */
1601
1602 /*
1603 * step 7: we are done with the source map, unlock. if copy_ok
1604 * is 0 then we have not replaced the dummy mapping in dstmap yet
1605 * and we need to do so now.
1606 */
1607
1608 vm_map_unlock(srcmap);
1609 if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1610 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */
1611
1612 /* now do the replacement if we didn't do it in step 5 */
1613 if (copy_ok == 0) {
1614 vm_map_lock(dstmap);
1615 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1616 nchain);
1617 vm_map_unlock(dstmap);
1618
1619 if (error == FALSE) {
1620 error = EIO;
1621 goto bad2;
1622 }
1623 }
1624
1625 /*
1626 * done!
1627 */
1628 return(0);
1629
1630 /*
1631 * bad: failure recovery
1632 */
1633 bad:
1634 vm_map_unlock(srcmap);
1635 bad2: /* src already unlocked */
1636 if (chain)
1637 uvm_unmap_detach(chain,
1638 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1639 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */
1640 return(error);
1641 }
1642
1643 /* end of extraction functions */
1644
1645 /*
1646 * uvm_map_submap: punch down part of a map into a submap
1647 *
1648 * => only the kernel_map is allowed to be submapped
1649 * => the purpose of submapping is to break up the locking granularity
1650 * of a larger map
1651 * => the range specified must have been mapped previously with a uvm_map()
1652 * call [with uobj==NULL] to create a blank map entry in the main map.
1653 * [And it had better still be blank!]
1654 * => maps which contain submaps should never be copied or forked.
1655 * => to remove a submap, use uvm_unmap() on the main map
1656 * and then uvm_map_deallocate() the submap.
1657 * => main map must be unlocked.
1658 * => submap must have been init'd and have a zero reference count.
1659 * [need not be locked as we don't actually reference it]
1660 */
1661
1662 int
1663 uvm_map_submap(map, start, end, submap)
1664 vm_map_t map, submap;
1665 vaddr_t start, end;
1666 {
1667 vm_map_entry_t entry;
1668 int result;
1669 UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist);
1670
1671 vm_map_lock(map);
1672
1673 VM_MAP_RANGE_CHECK(map, start, end);
1674
1675 if (uvm_map_lookup_entry(map, start, &entry)) {
1676 UVM_MAP_CLIP_START(map, entry, start);
1677 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */
1678 }
1679 else {
1680 entry = NULL;
1681 }
1682
1683 if (entry != NULL &&
1684 entry->start == start && entry->end == end &&
1685 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1686 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1687
1688 /*
1689 * doit!
1690 */
1691 entry->etype |= UVM_ET_SUBMAP;
1692 entry->object.sub_map = submap;
1693 entry->offset = 0;
1694 uvm_map_reference(submap);
1695 result = KERN_SUCCESS;
1696 } else {
1697 result = KERN_INVALID_ARGUMENT;
1698 }
1699 vm_map_unlock(map);
1700
1701 return(result);
1702 }
1703
1704
1705 /*
1706 * uvm_map_protect: change map protection
1707 *
1708 * => set_max means set max_protection.
1709 * => map must be unlocked.
1710 */
1711
1712 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \
1713 ~VM_PROT_WRITE : VM_PROT_ALL)
1714 #define max(a,b) ((a) > (b) ? (a) : (b))
1715
1716 int
1717 uvm_map_protect(map, start, end, new_prot, set_max)
1718 vm_map_t map;
1719 vaddr_t start, end;
1720 vm_prot_t new_prot;
1721 boolean_t set_max;
1722 {
1723 vm_map_entry_t current, entry;
1724 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1725 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1726 map, start, end, new_prot);
1727
1728 vm_map_lock(map);
1729
1730 VM_MAP_RANGE_CHECK(map, start, end);
1731
1732 if (uvm_map_lookup_entry(map, start, &entry)) {
1733 UVM_MAP_CLIP_START(map, entry, start);
1734 } else {
1735 entry = entry->next;
1736 }
1737
1738 /*
1739 * make a first pass to check for protection violations.
1740 */
1741
1742 current = entry;
1743 while ((current != &map->header) && (current->start < end)) {
1744 if (UVM_ET_ISSUBMAP(current))
1745 return (KERN_INVALID_ARGUMENT);
1746 if ((new_prot & current->max_protection) != new_prot) {
1747 vm_map_unlock(map);
1748 return (KERN_PROTECTION_FAILURE);
1749 }
1750 current = current->next;
1751 }
1752
1753 /* go back and fix up protections (no need to clip this time). */
1754
1755 current = entry;
1756
1757 while ((current != &map->header) && (current->start < end)) {
1758 vm_prot_t old_prot;
1759
1760 UVM_MAP_CLIP_END(map, current, end);
1761
1762 old_prot = current->protection;
1763 if (set_max)
1764 current->protection =
1765 (current->max_protection = new_prot) & old_prot;
1766 else
1767 current->protection = new_prot;
1768
1769 /*
1770 * update physical map if necessary. worry about copy-on-write
1771 * here -- CHECK THIS XXX
1772 */
1773
1774 if (current->protection != old_prot) {
1775
1776 /* update pmap! */
1777 pmap_protect(map->pmap, current->start, current->end,
1778 current->protection & MASK(entry));
1779
1780 }
1781 current = current->next;
1782 }
1783
1784 vm_map_unlock(map);
1785 UVMHIST_LOG(maphist, "<- done",0,0,0,0);
1786 return(KERN_SUCCESS);
1787 }
1788
1789 #undef max
1790 #undef MASK
1791
1792 /*
1793 * uvm_map_inherit: set inheritance code for range of addrs in map.
1794 *
1795 * => map must be unlocked
1796 * => note that the inherit code is used during a "fork". see fork
1797 * code for details.
1798 */
1799
1800 int
1801 uvm_map_inherit(map, start, end, new_inheritance)
1802 vm_map_t map;
1803 vaddr_t start;
1804 vaddr_t end;
1805 vm_inherit_t new_inheritance;
1806 {
1807 vm_map_entry_t entry, temp_entry;
1808 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1809 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1810 map, start, end, new_inheritance);
1811
1812 switch (new_inheritance) {
1813 case VM_INHERIT_NONE:
1814 case VM_INHERIT_COPY:
1815 case VM_INHERIT_SHARE:
1816 break;
1817 default:
1818 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1819 return (KERN_INVALID_ARGUMENT);
1820 }
1821
1822 vm_map_lock(map);
1823
1824 VM_MAP_RANGE_CHECK(map, start, end);
1825
1826 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1827 entry = temp_entry;
1828 UVM_MAP_CLIP_START(map, entry, start);
1829 } else {
1830 entry = temp_entry->next;
1831 }
1832
1833 while ((entry != &map->header) && (entry->start < end)) {
1834 UVM_MAP_CLIP_END(map, entry, end);
1835
1836 entry->inheritance = new_inheritance;
1837
1838 entry = entry->next;
1839 }
1840
1841 vm_map_unlock(map);
1842 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1843 return(KERN_SUCCESS);
1844 }
1845
1846 /*
1847 * uvm_map_advice: set advice code for range of addrs in map.
1848 *
1849 * => map must be unlocked
1850 */
1851
1852 int
1853 uvm_map_advice(map, start, end, new_advice)
1854 vm_map_t map;
1855 vaddr_t start;
1856 vaddr_t end;
1857 int new_advice;
1858 {
1859 vm_map_entry_t entry, temp_entry;
1860 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
1861 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
1862 map, start, end, new_advice);
1863
1864 vm_map_lock(map);
1865
1866 VM_MAP_RANGE_CHECK(map, start, end);
1867
1868 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1869 entry = temp_entry;
1870 UVM_MAP_CLIP_START(map, entry, start);
1871 } else {
1872 entry = temp_entry->next;
1873 }
1874
1875 /*
1876 * XXXJRT: disallow holes?
1877 */
1878
1879 while ((entry != &map->header) && (entry->start < end)) {
1880 UVM_MAP_CLIP_END(map, entry, end);
1881
1882 switch (new_advice) {
1883 case MADV_NORMAL:
1884 case MADV_RANDOM:
1885 case MADV_SEQUENTIAL:
1886 /* nothing special here */
1887 break;
1888
1889 default:
1890 vm_map_unlock(map);
1891 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1892 return (KERN_INVALID_ARGUMENT);
1893 }
1894
1895
1896 entry->advice = new_advice;
1897
1898 entry = entry->next;
1899 }
1900
1901 vm_map_unlock(map);
1902 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1903 return (KERN_SUCCESS);
1904 }
1905
1906 /*
1907 * uvm_map_pageable: sets the pageability of a range in a map.
1908 *
1909 * => wires map entries. should not be used for transient page locking.
1910 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
1911 * => regions sepcified as not pageable require lock-down (wired) memory
1912 * and page tables.
1913 * => map must never be read-locked
1914 * => if islocked is TRUE, map is already write-locked
1915 * => we always unlock the map, since we must downgrade to a read-lock
1916 * to call uvm_fault_wire()
1917 * => XXXCDC: check this and try and clean it up.
1918 */
1919
1920 int
1921 uvm_map_pageable(map, start, end, new_pageable, islocked)
1922 vm_map_t map;
1923 vaddr_t start, end;
1924 boolean_t new_pageable, islocked;
1925 {
1926 vm_map_entry_t entry, start_entry, failed_entry;
1927 int rv;
1928 #ifdef DIAGNOSTIC
1929 u_int timestamp_save;
1930 #endif
1931 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
1932 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1933 map, start, end, new_pageable);
1934
1935 #ifdef DIAGNOSTIC
1936 if ((map->flags & VM_MAP_PAGEABLE) == 0)
1937 panic("uvm_map_pageable: map %p not pageable", map);
1938 #endif
1939
1940 if (islocked == FALSE)
1941 vm_map_lock(map);
1942 VM_MAP_RANGE_CHECK(map, start, end);
1943
1944 /*
1945 * only one pageability change may take place at one time, since
1946 * uvm_fault_wire assumes it will be called only once for each
1947 * wiring/unwiring. therefore, we have to make sure we're actually
1948 * changing the pageability for the entire region. we do so before
1949 * making any changes.
1950 */
1951
1952 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1953 vm_map_unlock(map);
1954
1955 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1956 return (KERN_INVALID_ADDRESS);
1957 }
1958 entry = start_entry;
1959
1960 /*
1961 * handle wiring and unwiring seperately.
1962 */
1963
1964 if (new_pageable) { /* unwire */
1965 UVM_MAP_CLIP_START(map, entry, start);
1966 /*
1967 * unwiring. first ensure that the range to be unwired is
1968 * really wired down and that there are no holes.
1969 */
1970 while ((entry != &map->header) && (entry->start < end)) {
1971 if (entry->wired_count == 0 ||
1972 (entry->end < end &&
1973 (entry->next == &map->header ||
1974 entry->next->start > entry->end))) {
1975 vm_map_unlock(map);
1976 UVMHIST_LOG(maphist,
1977 "<- done (INVALID UNWIRE ARG)",0,0,0,0);
1978 return (KERN_INVALID_ARGUMENT);
1979 }
1980 entry = entry->next;
1981 }
1982
1983 /*
1984 * POSIX 1003.1b - a single munlock call unlocks a region,
1985 * regardless of the number of mlock calls made on that
1986 * region.
1987 */
1988 entry = start_entry;
1989 while ((entry != &map->header) && (entry->start < end)) {
1990 UVM_MAP_CLIP_END(map, entry, end);
1991 if (VM_MAPENT_ISWIRED(entry))
1992 uvm_map_entry_unwire(map, entry);
1993 entry = entry->next;
1994 }
1995 vm_map_unlock(map);
1996 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
1997 return(KERN_SUCCESS);
1998
1999 /*
2000 * end of unwire case!
2001 */
2002 }
2003
2004 /*
2005 * wire case: in two passes [XXXCDC: ugly block of code here]
2006 *
2007 * 1: holding the write lock, we create any anonymous maps that need
2008 * to be created. then we clip each map entry to the region to
2009 * be wired and increment its wiring count.
2010 *
2011 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2012 * in the pages for any newly wired area (wired_count == 1).
2013 *
2014 * downgrading to a read lock for uvm_fault_wire avoids a possible
2015 * deadlock with another thread that may have faulted on one of
2016 * the pages to be wired (it would mark the page busy, blocking
2017 * us, then in turn block on the map lock that we hold). because
2018 * of problems in the recursive lock package, we cannot upgrade
2019 * to a write lock in vm_map_lookup. thus, any actions that
2020 * require the write lock must be done beforehand. because we
2021 * keep the read lock on the map, the copy-on-write status of the
2022 * entries we modify here cannot change.
2023 */
2024
2025 while ((entry != &map->header) && (entry->start < end)) {
2026 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2027 /*
2028 * perform actions of vm_map_lookup that need the
2029 * write lock on the map: create an anonymous map
2030 * for a copy-on-write region, or an anonymous map
2031 * for a zero-fill region. (XXXCDC: submap case
2032 * ok?)
2033 */
2034 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2035 if (UVM_ET_ISNEEDSCOPY(entry) &&
2036 ((entry->protection & VM_PROT_WRITE) ||
2037 (entry->object.uvm_obj == NULL))) {
2038 amap_copy(map, entry, M_WAITOK, TRUE,
2039 start, end);
2040 /* XXXCDC: wait OK? */
2041 }
2042 }
2043 }
2044 UVM_MAP_CLIP_START(map, entry, start);
2045 UVM_MAP_CLIP_END(map, entry, end);
2046 entry->wired_count++;
2047
2048 /*
2049 * Check for holes
2050 */
2051 if (entry->protection == VM_PROT_NONE ||
2052 (entry->end < end &&
2053 (entry->next == &map->header ||
2054 entry->next->start > entry->end))) {
2055 /*
2056 * found one. amap creation actions do not need to
2057 * be undone, but the wired counts need to be restored.
2058 */
2059 while (entry != &map->header && entry->end > start) {
2060 entry->wired_count--;
2061 entry = entry->prev;
2062 }
2063 vm_map_unlock(map);
2064 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2065 return (KERN_INVALID_ARGUMENT);
2066 }
2067 entry = entry->next;
2068 }
2069
2070 /*
2071 * Pass 2.
2072 */
2073
2074 #ifdef DIAGNOSTIC
2075 timestamp_save = map->timestamp;
2076 #endif
2077 vm_map_busy(map);
2078 vm_map_downgrade(map);
2079
2080 rv = 0;
2081 entry = start_entry;
2082 while (entry != &map->header && entry->start < end) {
2083 if (entry->wired_count == 1) {
2084 rv = uvm_fault_wire(map, entry->start, entry->end,
2085 entry->protection);
2086 if (rv) {
2087 /*
2088 * wiring failed. break out of the loop.
2089 * we'll clean up the map below, once we
2090 * have a write lock again.
2091 */
2092 break;
2093 }
2094 }
2095 entry = entry->next;
2096 }
2097
2098 if (rv) { /* failed? */
2099 /*
2100 * Get back to an exclusive (write) lock.
2101 */
2102 vm_map_upgrade(map);
2103 vm_map_unbusy(map);
2104
2105 #ifdef DIAGNOSTIC
2106 if (timestamp_save != map->timestamp)
2107 panic("uvm_map_pageable: stale map");
2108 #endif
2109
2110 /*
2111 * first drop the wiring count on all the entries
2112 * which haven't actually been wired yet.
2113 */
2114 failed_entry = entry;
2115 while (entry != &map->header && entry->start < end) {
2116 entry->wired_count--;
2117 entry = entry->next;
2118 }
2119
2120 /*
2121 * now, unwire all the entries that were successfully
2122 * wired above.
2123 */
2124 entry = start_entry;
2125 while (entry != failed_entry) {
2126 entry->wired_count--;
2127 if (VM_MAPENT_ISWIRED(entry) == 0)
2128 uvm_map_entry_unwire(map, entry);
2129 entry = entry->next;
2130 }
2131 vm_map_unlock(map);
2132 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2133 return(rv);
2134 }
2135
2136 /* We are holding a read lock here. */
2137 vm_map_unbusy(map);
2138 vm_map_unlock_read(map);
2139
2140 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2141 return(KERN_SUCCESS);
2142 }
2143
2144 /*
2145 * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2146 * all mapped regions.
2147 *
2148 * => map must not be locked.
2149 * => if no flags are specified, all regions are unwired.
2150 * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2151 */
2152
2153 int
2154 uvm_map_pageable_all(map, flags, limit)
2155 vm_map_t map;
2156 int flags;
2157 vsize_t limit;
2158 {
2159 vm_map_entry_t entry, failed_entry;
2160 vsize_t size;
2161 int rv;
2162 #ifdef DIAGNOSTIC
2163 u_int timestamp_save;
2164 #endif
2165 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2166 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2167
2168 #ifdef DIAGNOSTIC
2169 if ((map->flags & VM_MAP_PAGEABLE) == 0)
2170 panic("uvm_map_pageable_all: map %p not pageable", map);
2171 #endif
2172
2173 vm_map_lock(map);
2174
2175 /*
2176 * handle wiring and unwiring separately.
2177 */
2178
2179 if (flags == 0) { /* unwire */
2180 /*
2181 * POSIX 1003.1b -- munlockall unlocks all regions,
2182 * regardless of how many times mlockall has been called.
2183 */
2184 for (entry = map->header.next; entry != &map->header;
2185 entry = entry->next) {
2186 if (VM_MAPENT_ISWIRED(entry))
2187 uvm_map_entry_unwire(map, entry);
2188 }
2189 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2190 vm_map_unlock(map);
2191 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2192 return (KERN_SUCCESS);
2193
2194 /*
2195 * end of unwire case!
2196 */
2197 }
2198
2199 if (flags & MCL_FUTURE) {
2200 /*
2201 * must wire all future mappings; remember this.
2202 */
2203 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2204 }
2205
2206 if ((flags & MCL_CURRENT) == 0) {
2207 /*
2208 * no more work to do!
2209 */
2210 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2211 vm_map_unlock(map);
2212 return (KERN_SUCCESS);
2213 }
2214
2215 /*
2216 * wire case: in three passes [XXXCDC: ugly block of code here]
2217 *
2218 * 1: holding the write lock, count all pages mapped by non-wired
2219 * entries. if this would cause us to go over our limit, we fail.
2220 *
2221 * 2: still holding the write lock, we create any anonymous maps that
2222 * need to be created. then we increment its wiring count.
2223 *
2224 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2225 * in the pages for any newly wired area (wired_count == 1).
2226 *
2227 * downgrading to a read lock for uvm_fault_wire avoids a possible
2228 * deadlock with another thread that may have faulted on one of
2229 * the pages to be wired (it would mark the page busy, blocking
2230 * us, then in turn block on the map lock that we hold). because
2231 * of problems in the recursive lock package, we cannot upgrade
2232 * to a write lock in vm_map_lookup. thus, any actions that
2233 * require the write lock must be done beforehand. because we
2234 * keep the read lock on the map, the copy-on-write status of the
2235 * entries we modify here cannot change.
2236 */
2237
2238 for (size = 0, entry = map->header.next; entry != &map->header;
2239 entry = entry->next) {
2240 if (entry->protection != VM_PROT_NONE &&
2241 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2242 size += entry->end - entry->start;
2243 }
2244 }
2245
2246 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2247 vm_map_unlock(map);
2248 return (KERN_NO_SPACE); /* XXX overloaded */
2249 }
2250
2251 /* XXX non-pmap_wired_count case must be handled by caller */
2252 #ifdef pmap_wired_count
2253 if (limit != 0 &&
2254 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2255 vm_map_unlock(map);
2256 return (KERN_NO_SPACE); /* XXX overloaded */
2257 }
2258 #endif
2259
2260 /*
2261 * Pass 2.
2262 */
2263
2264 for (entry = map->header.next; entry != &map->header;
2265 entry = entry->next) {
2266 if (entry->protection == VM_PROT_NONE)
2267 continue;
2268 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2269 /*
2270 * perform actions of vm_map_lookup that need the
2271 * write lock on the map: create an anonymous map
2272 * for a copy-on-write region, or an anonymous map
2273 * for a zero-fill region. (XXXCDC: submap case
2274 * ok?)
2275 */
2276 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2277 if (UVM_ET_ISNEEDSCOPY(entry) &&
2278 ((entry->protection & VM_PROT_WRITE) ||
2279 (entry->object.uvm_obj == NULL))) {
2280 amap_copy(map, entry, M_WAITOK, TRUE,
2281 entry->start, entry->end);
2282 /* XXXCDC: wait OK? */
2283 }
2284 }
2285 }
2286 entry->wired_count++;
2287 }
2288
2289 /*
2290 * Pass 3.
2291 */
2292
2293 #ifdef DIAGNOSTIC
2294 timestamp_save = map->timestamp;
2295 #endif
2296 vm_map_busy(map);
2297 vm_map_downgrade(map);
2298
2299 rv = KERN_SUCCESS;
2300 for (entry = map->header.next; entry != &map->header;
2301 entry = entry->next) {
2302 if (entry->wired_count == 1) {
2303 rv = uvm_fault_wire(map, entry->start, entry->end,
2304 entry->protection);
2305 if (rv) {
2306 /*
2307 * wiring failed. break out of the loop.
2308 * we'll clean up the map below, once we
2309 * have a write lock again.
2310 */
2311 break;
2312 }
2313 }
2314 }
2315
2316 if (rv) { /* failed? */
2317 /*
2318 * Get back an exclusive (write) lock.
2319 */
2320 vm_map_upgrade(map);
2321 vm_map_unbusy(map);
2322
2323 #ifdef DIAGNOSTIC
2324 if (timestamp_save != map->timestamp)
2325 panic("uvm_map_pageable_all: stale map");
2326 #endif
2327
2328 /*
2329 * first drop the wiring count on all the entries
2330 * which haven't actually been wired yet.
2331 */
2332 failed_entry = entry;
2333 for (/* nothing */; entry != &map->header;
2334 entry = entry->next)
2335 entry->wired_count--;
2336
2337 /*
2338 * now, unwire all the entries that were successfully
2339 * wired above.
2340 */
2341 for (entry = map->header.next; entry != failed_entry;
2342 entry = entry->next) {
2343 entry->wired_count--;
2344 if (VM_MAPENT_ISWIRED(entry) == 0)
2345 uvm_map_entry_unwire(map, entry);
2346 }
2347 vm_map_unlock(map);
2348 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2349 return (rv);
2350 }
2351
2352 /* We are holding a read lock here. */
2353 vm_map_unbusy(map);
2354 vm_map_unlock_read(map);
2355
2356 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2357 return (KERN_SUCCESS);
2358 }
2359
2360 /*
2361 * uvm_map_clean: clean out a map range
2362 *
2363 * => valid flags:
2364 * if (flags & PGO_CLEANIT): dirty pages are cleaned first
2365 * if (flags & PGO_SYNCIO): dirty pages are written synchronously
2366 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2367 * if (flags & PGO_FREE): any cached pages are freed after clean
2368 * => returns an error if any part of the specified range isn't mapped
2369 * => never a need to flush amap layer since the anonymous memory has
2370 * no permanent home, but may deactivate pages there
2371 * => called from sys_msync() and sys_madvise()
2372 * => caller must not write-lock map (read OK).
2373 * => we may sleep while cleaning if SYNCIO [with map read-locked]
2374 */
2375
2376 int amap_clean_works = 1; /* XXX for now, just in case... */
2377
2378 int
2379 uvm_map_clean(map, start, end, flags)
2380 vm_map_t map;
2381 vaddr_t start, end;
2382 int flags;
2383 {
2384 vm_map_entry_t current, entry;
2385 struct uvm_object *uobj;
2386 struct vm_amap *amap;
2387 struct vm_anon *anon;
2388 struct vm_page *pg;
2389 vaddr_t offset;
2390 vsize_t size;
2391 int rv, error, refs;
2392 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2393 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2394 map, start, end, flags);
2395
2396 #ifdef DIAGNOSTIC
2397 if ((flags & (PGO_FREE|PGO_DEACTIVATE)) == (PGO_FREE|PGO_DEACTIVATE))
2398 panic("uvm_map_clean: FREE and DEACTIVATE");
2399 #endif
2400
2401 vm_map_lock_read(map);
2402 VM_MAP_RANGE_CHECK(map, start, end);
2403 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2404 vm_map_unlock_read(map);
2405 return(KERN_INVALID_ADDRESS);
2406 }
2407
2408 /*
2409 * Make a first pass to check for holes.
2410 */
2411 for (current = entry; current->start < end; current = current->next) {
2412 if (UVM_ET_ISSUBMAP(current)) {
2413 vm_map_unlock_read(map);
2414 return (KERN_INVALID_ARGUMENT);
2415 }
2416 if (end > current->end && (current->next == &map->header ||
2417 current->end != current->next->start)) {
2418 vm_map_unlock_read(map);
2419 return (KERN_INVALID_ADDRESS);
2420 }
2421 }
2422
2423 error = KERN_SUCCESS;
2424
2425 for (current = entry; current->start < end; current = current->next) {
2426 amap = current->aref.ar_amap; /* top layer */
2427 uobj = current->object.uvm_obj; /* bottom layer */
2428
2429 #ifdef DIAGNOSTIC
2430 if (start < current->start)
2431 panic("uvm_map_clean: hole");
2432 #endif
2433
2434 /*
2435 * No amap cleaning necessary if:
2436 *
2437 * (1) There's no amap.
2438 *
2439 * (2) We're not deactivating or freeing pages.
2440 */
2441 if (amap == NULL ||
2442 (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2443 goto flush_object;
2444
2445 /* XXX for now, just in case... */
2446 if (amap_clean_works == 0)
2447 goto flush_object;
2448
2449 amap_lock(amap);
2450
2451 offset = start - current->start;
2452 size = (end <= current->end ? end : current->end) -
2453 start;
2454
2455 for (/* nothing */; size != 0; size -= PAGE_SIZE,
2456 offset += PAGE_SIZE) {
2457 anon = amap_lookup(¤t->aref, offset);
2458 if (anon == NULL)
2459 continue;
2460
2461 simple_lock(&anon->an_lock);
2462
2463 pg = anon->u.an_page;
2464 if (pg == NULL) {
2465 simple_unlock(&anon->an_lock);
2466 continue;
2467 }
2468
2469 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2470 /*
2471 * XXX In these first 3 cases, we always just
2472 * XXX deactivate the page. We may want to
2473 * XXX handle the different cases more
2474 * XXX specifically, in the future.
2475 */
2476 case PGO_CLEANIT|PGO_FREE:
2477 case PGO_CLEANIT|PGO_DEACTIVATE:
2478 case PGO_DEACTIVATE:
2479 /* skip the page if it's loaned or wired */
2480 if (pg->loan_count != 0 ||
2481 pg->wire_count != 0) {
2482 simple_unlock(&anon->an_lock);
2483 continue;
2484 }
2485
2486 uvm_lock_pageq();
2487
2488 /*
2489 * skip the page if it's not actually owned
2490 * by the anon (may simply be loaned to the
2491 * anon).
2492 */
2493 if ((pg->pqflags & PQ_ANON) == 0) {
2494 #ifdef DIAGNOSTIC
2495 if (pg->uobject != NULL)
2496 panic("uvm_map_clean: "
2497 "page anon vs. object "
2498 "inconsistency");
2499 #endif
2500 uvm_unlock_pageq();
2501 simple_unlock(&anon->an_lock);
2502 continue;
2503 }
2504
2505 #ifdef DIAGNOSTIC
2506 if (pg->uanon != anon)
2507 panic("uvm_map_clean: anon "
2508 "inconsistency");
2509 #endif
2510
2511 /* zap all mappings for the page. */
2512 pmap_page_protect(PMAP_PGARG(pg),
2513 VM_PROT_NONE);
2514
2515 /* ...and deactivate the page. */
2516 uvm_pagedeactivate(pg);
2517
2518 uvm_unlock_pageq();
2519 simple_unlock(&anon->an_lock);
2520 continue;
2521
2522 case PGO_FREE:
2523 /* XXX skip the page if it's wired */
2524 if (pg->wire_count != 0) {
2525 simple_unlock(&anon->an_lock);
2526 continue;
2527 }
2528 amap_unadd(&entry->aref, offset);
2529 refs = --anon->an_ref;
2530 simple_unlock(&anon->an_lock);
2531 if (refs == 0)
2532 uvm_anfree(anon);
2533 continue;
2534
2535 default:
2536 panic("uvm_map_clean: wierd flags");
2537 }
2538 #ifdef DIAGNOSTIC
2539 panic("uvm_map_clean: unreachable code");
2540 #endif
2541 }
2542
2543 amap_unlock(amap);
2544
2545 flush_object:
2546 /*
2547 * flush pages if we've got a valid backing object.
2548 */
2549
2550 offset = current->offset + (start - current->start);
2551 size = (end <= current->end ? end : current->end) - start;
2552
2553 if (uobj != NULL) {
2554 simple_lock(&uobj->vmobjlock);
2555 rv = uobj->pgops->pgo_flush(uobj, offset,
2556 offset + size, flags);
2557 simple_unlock(&uobj->vmobjlock);
2558
2559 if (rv == FALSE)
2560 error = KERN_FAILURE;
2561 }
2562 start += size;
2563 }
2564
2565 vm_map_unlock_read(map);
2566 return (error);
2567 }
2568
2569
2570 /*
2571 * uvm_map_checkprot: check protection in map
2572 *
2573 * => must allow specified protection in a fully allocated region.
2574 * => map must be read or write locked by caller.
2575 */
2576
2577 boolean_t
2578 uvm_map_checkprot(map, start, end, protection)
2579 vm_map_t map;
2580 vaddr_t start, end;
2581 vm_prot_t protection;
2582 {
2583 vm_map_entry_t entry;
2584 vm_map_entry_t tmp_entry;
2585
2586 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2587 return(FALSE);
2588 }
2589
2590 entry = tmp_entry;
2591
2592 while (start < end) {
2593 if (entry == &map->header) {
2594 return(FALSE);
2595 }
2596
2597 /*
2598 * no holes allowed
2599 */
2600
2601 if (start < entry->start) {
2602 return(FALSE);
2603 }
2604
2605 /*
2606 * check protection associated with entry
2607 */
2608
2609 if ((entry->protection & protection) != protection) {
2610 return(FALSE);
2611 }
2612
2613 /* go to next entry */
2614
2615 start = entry->end;
2616 entry = entry->next;
2617 }
2618 return(TRUE);
2619 }
2620
2621 /*
2622 * uvmspace_alloc: allocate a vmspace structure.
2623 *
2624 * - structure includes vm_map and pmap
2625 * - XXX: no locking on this structure
2626 * - refcnt set to 1, rest must be init'd by caller
2627 */
2628 struct vmspace *
2629 uvmspace_alloc(min, max, pageable)
2630 vaddr_t min, max;
2631 int pageable;
2632 {
2633 struct vmspace *vm;
2634 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2635
2636 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2637 uvmspace_init(vm, NULL, min, max, pageable);
2638 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2639 return (vm);
2640 }
2641
2642 /*
2643 * uvmspace_init: initialize a vmspace structure.
2644 *
2645 * - XXX: no locking on this structure
2646 * - refcnt set to 1, rest must me init'd by caller
2647 */
2648 void
2649 uvmspace_init(vm, pmap, min, max, pageable)
2650 struct vmspace *vm;
2651 struct pmap *pmap;
2652 vaddr_t min, max;
2653 boolean_t pageable;
2654 {
2655 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2656
2657 memset(vm, 0, sizeof(*vm));
2658
2659 uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0);
2660
2661 if (pmap)
2662 pmap_reference(pmap);
2663 else
2664 #if defined(PMAP_NEW)
2665 pmap = pmap_create();
2666 #else
2667 pmap = pmap_create(0);
2668 #endif
2669 vm->vm_map.pmap = pmap;
2670
2671 vm->vm_refcnt = 1;
2672 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2673 }
2674
2675 /*
2676 * uvmspace_share: share a vmspace between two proceses
2677 *
2678 * - XXX: no locking on vmspace
2679 * - used for vfork, threads(?)
2680 */
2681
2682 void
2683 uvmspace_share(p1, p2)
2684 struct proc *p1, *p2;
2685 {
2686 p2->p_vmspace = p1->p_vmspace;
2687 p1->p_vmspace->vm_refcnt++;
2688 }
2689
2690 /*
2691 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2692 *
2693 * - XXX: no locking on vmspace
2694 */
2695
2696 void
2697 uvmspace_unshare(p)
2698 struct proc *p;
2699 {
2700 struct vmspace *nvm, *ovm = p->p_vmspace;
2701 int s;
2702
2703 if (ovm->vm_refcnt == 1)
2704 /* nothing to do: vmspace isn't shared in the first place */
2705 return;
2706
2707 /* make a new vmspace, still holding old one */
2708 nvm = uvmspace_fork(ovm);
2709
2710 s = splhigh(); /* make this `atomic' */
2711 pmap_deactivate(p); /* unbind old vmspace */
2712 p->p_vmspace = nvm;
2713 pmap_activate(p); /* switch to new vmspace */
2714 splx(s); /* end of critical section */
2715
2716 uvmspace_free(ovm); /* drop reference to old vmspace */
2717 }
2718
2719 /*
2720 * uvmspace_exec: the process wants to exec a new program
2721 *
2722 * - XXX: no locking on vmspace
2723 */
2724
2725 void
2726 uvmspace_exec(p)
2727 struct proc *p;
2728 {
2729 struct vmspace *nvm, *ovm = p->p_vmspace;
2730 vm_map_t map = &ovm->vm_map;
2731 int s;
2732
2733 #ifdef sparc
2734 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2735 kill_user_windows(p); /* before stack addresses go away */
2736 #endif
2737
2738 /*
2739 * see if more than one process is using this vmspace...
2740 */
2741
2742 if (ovm->vm_refcnt == 1) {
2743
2744 /*
2745 * if p is the only process using its vmspace then we can safely
2746 * recycle that vmspace for the program that is being exec'd.
2747 */
2748
2749 #ifdef SYSVSHM
2750 /*
2751 * SYSV SHM semantics require us to kill all segments on an exec
2752 */
2753 if (ovm->vm_shm)
2754 shmexit(ovm);
2755 #endif
2756
2757 /*
2758 * POSIX 1003.1b -- "lock future mappings" is revoked
2759 * when a process execs another program image.
2760 */
2761 vm_map_lock(map);
2762 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2763 vm_map_unlock(map);
2764
2765 /*
2766 * now unmap the old program
2767 */
2768 uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
2769
2770 } else {
2771
2772 /*
2773 * p's vmspace is being shared, so we can't reuse it for p since
2774 * it is still being used for others. allocate a new vmspace
2775 * for p
2776 */
2777 nvm = uvmspace_alloc(map->min_offset, map->max_offset,
2778 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2779
2780 /*
2781 * install new vmspace and drop our ref to the old one.
2782 */
2783
2784 s = splhigh();
2785 pmap_deactivate(p);
2786 p->p_vmspace = nvm;
2787 pmap_activate(p);
2788 splx(s);
2789
2790 uvmspace_free(ovm);
2791 }
2792 }
2793
2794 /*
2795 * uvmspace_free: free a vmspace data structure
2796 *
2797 * - XXX: no locking on vmspace
2798 */
2799
2800 void
2801 uvmspace_free(vm)
2802 struct vmspace *vm;
2803 {
2804 vm_map_entry_t dead_entries;
2805 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2806
2807 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2808 if (--vm->vm_refcnt == 0) {
2809 /*
2810 * lock the map, to wait out all other references to it. delete
2811 * all of the mappings and pages they hold, then call the pmap
2812 * module to reclaim anything left.
2813 */
2814 vm_map_lock(&vm->vm_map);
2815 if (vm->vm_map.nentries) {
2816 (void)uvm_unmap_remove(&vm->vm_map,
2817 vm->vm_map.min_offset, vm->vm_map.max_offset,
2818 &dead_entries);
2819 if (dead_entries != NULL)
2820 uvm_unmap_detach(dead_entries, 0);
2821 }
2822 pmap_destroy(vm->vm_map.pmap);
2823 vm->vm_map.pmap = NULL;
2824 pool_put(&uvm_vmspace_pool, vm);
2825 }
2826 UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2827 }
2828
2829 /*
2830 * F O R K - m a i n e n t r y p o i n t
2831 */
2832 /*
2833 * uvmspace_fork: fork a process' main map
2834 *
2835 * => create a new vmspace for child process from parent.
2836 * => parent's map must not be locked.
2837 */
2838
2839 struct vmspace *
2840 uvmspace_fork(vm1)
2841 struct vmspace *vm1;
2842 {
2843 struct vmspace *vm2;
2844 vm_map_t old_map = &vm1->vm_map;
2845 vm_map_t new_map;
2846 vm_map_entry_t old_entry;
2847 vm_map_entry_t new_entry;
2848 pmap_t new_pmap;
2849 boolean_t protect_child;
2850 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2851
2852 vm_map_lock(old_map);
2853
2854 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset,
2855 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2856 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2857 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2858 new_map = &vm2->vm_map; /* XXX */
2859 new_pmap = new_map->pmap;
2860
2861 old_entry = old_map->header.next;
2862
2863 /*
2864 * go entry-by-entry
2865 */
2866
2867 while (old_entry != &old_map->header) {
2868
2869 /*
2870 * first, some sanity checks on the old entry
2871 */
2872 if (UVM_ET_ISSUBMAP(old_entry))
2873 panic("fork: encountered a submap during fork (illegal)");
2874
2875 if (!UVM_ET_ISCOPYONWRITE(old_entry) &&
2876 UVM_ET_ISNEEDSCOPY(old_entry))
2877 panic("fork: non-copy_on_write map entry marked needs_copy (illegal)");
2878
2879
2880 switch (old_entry->inheritance) {
2881 case VM_INHERIT_NONE:
2882 /*
2883 * drop the mapping
2884 */
2885 break;
2886
2887 case VM_INHERIT_SHARE:
2888 /*
2889 * share the mapping: this means we want the old and
2890 * new entries to share amaps and backing objects.
2891 */
2892
2893 /*
2894 * if the old_entry needs a new amap (due to prev fork)
2895 * then we need to allocate it now so that we have
2896 * something we own to share with the new_entry. [in
2897 * other words, we need to clear needs_copy]
2898 */
2899
2900 if (UVM_ET_ISNEEDSCOPY(old_entry)) {
2901 /* get our own amap, clears needs_copy */
2902 amap_copy(old_map, old_entry, M_WAITOK, FALSE,
2903 0, 0);
2904 /* XXXCDC: WAITOK??? */
2905 }
2906
2907 new_entry = uvm_mapent_alloc(new_map);
2908 /* old_entry -> new_entry */
2909 uvm_mapent_copy(old_entry, new_entry);
2910
2911 /* new pmap has nothing wired in it */
2912 new_entry->wired_count = 0;
2913
2914 /*
2915 * gain reference to object backing the map (can't
2916 * be a submap, already checked this case).
2917 */
2918 if (new_entry->aref.ar_amap)
2919 /* share reference */
2920 amap_ref(new_entry, AMAP_SHARED);
2921
2922 if (new_entry->object.uvm_obj &&
2923 new_entry->object.uvm_obj->pgops->pgo_reference)
2924 new_entry->object.uvm_obj->
2925 pgops->pgo_reference(
2926 new_entry->object.uvm_obj);
2927
2928 /* insert entry at end of new_map's entry list */
2929 uvm_map_entry_link(new_map, new_map->header.prev,
2930 new_entry);
2931
2932 /*
2933 * pmap_copy the mappings: this routine is optional
2934 * but if it is there it will reduce the number of
2935 * page faults in the new proc.
2936 */
2937
2938 pmap_copy(new_pmap, old_map->pmap, new_entry->start,
2939 (old_entry->end - old_entry->start),
2940 old_entry->start);
2941
2942 break;
2943
2944 case VM_INHERIT_COPY:
2945
2946 /*
2947 * copy-on-write the mapping (using mmap's
2948 * MAP_PRIVATE semantics)
2949 *
2950 * allocate new_entry, adjust reference counts.
2951 * (note that new references are read-only).
2952 */
2953
2954 new_entry = uvm_mapent_alloc(new_map);
2955 /* old_entry -> new_entry */
2956 uvm_mapent_copy(old_entry, new_entry);
2957
2958 if (new_entry->aref.ar_amap)
2959 amap_ref(new_entry, 0);
2960
2961 if (new_entry->object.uvm_obj &&
2962 new_entry->object.uvm_obj->pgops->pgo_reference)
2963 new_entry->object.uvm_obj->pgops->pgo_reference
2964 (new_entry->object.uvm_obj);
2965
2966 /* new pmap has nothing wired in it */
2967 new_entry->wired_count = 0;
2968
2969 new_entry->etype |=
2970 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
2971 uvm_map_entry_link(new_map, new_map->header.prev,
2972 new_entry);
2973
2974 /*
2975 * the new entry will need an amap. it will either
2976 * need to be copied from the old entry or created
2977 * from scratch (if the old entry does not have an
2978 * amap). can we defer this process until later
2979 * (by setting "needs_copy") or do we need to copy
2980 * the amap now?
2981 *
2982 * we must copy the amap now if any of the following
2983 * conditions hold:
2984 * 1. the old entry has an amap and that amap is
2985 * being shared. this means that the old (parent)
2986 * process is sharing the amap with another
2987 * process. if we do not clear needs_copy here
2988 * we will end up in a situation where both the
2989 * parent and child process are refering to the
2990 * same amap with "needs_copy" set. if the
2991 * parent write-faults, the fault routine will
2992 * clear "needs_copy" in the parent by allocating
2993 * a new amap. this is wrong because the
2994 * parent is supposed to be sharing the old amap
2995 * and the new amap will break that.
2996 *
2997 * 2. if the old entry has an amap and a non-zero
2998 * wire count then we are going to have to call
2999 * amap_cow_now to avoid page faults in the
3000 * parent process. since amap_cow_now requires
3001 * "needs_copy" to be clear we might as well
3002 * clear it here as well.
3003 *
3004 */
3005
3006 if (old_entry->aref.ar_amap != NULL) {
3007
3008 if ((amap_flags(old_entry->aref.ar_amap) &
3009 AMAP_SHARED) != 0 ||
3010 VM_MAPENT_ISWIRED(old_entry)) {
3011
3012 amap_copy(new_map, new_entry, M_WAITOK, FALSE,
3013 0, 0);
3014 /* XXXCDC: M_WAITOK ... ok? */
3015 }
3016 }
3017
3018 /*
3019 * if the parent's entry is wired down, then the
3020 * parent process does not want page faults on
3021 * access to that memory. this means that we
3022 * cannot do copy-on-write because we can't write
3023 * protect the old entry. in this case we
3024 * resolve all copy-on-write faults now, using
3025 * amap_cow_now. note that we have already
3026 * allocated any needed amap (above).
3027 */
3028
3029 if (VM_MAPENT_ISWIRED(old_entry)) {
3030
3031 /*
3032 * resolve all copy-on-write faults now
3033 * (note that there is nothing to do if
3034 * the old mapping does not have an amap).
3035 * XXX: is it worthwhile to bother with pmap_copy
3036 * in this case?
3037 */
3038 if (old_entry->aref.ar_amap)
3039 amap_cow_now(new_map, new_entry);
3040
3041 } else {
3042
3043 /*
3044 * setup mappings to trigger copy-on-write faults
3045 * we must write-protect the parent if it has
3046 * an amap and it is not already "needs_copy"...
3047 * if it is already "needs_copy" then the parent
3048 * has already been write-protected by a previous
3049 * fork operation.
3050 *
3051 * if we do not write-protect the parent, then
3052 * we must be sure to write-protect the child
3053 * after the pmap_copy() operation.
3054 *
3055 * XXX: pmap_copy should have some way of telling
3056 * us that it didn't do anything so we can avoid
3057 * calling pmap_protect needlessly.
3058 */
3059
3060 if (old_entry->aref.ar_amap) {
3061
3062 if (!UVM_ET_ISNEEDSCOPY(old_entry)) {
3063 if (old_entry->max_protection & VM_PROT_WRITE) {
3064 pmap_protect(old_map->pmap,
3065 old_entry->start,
3066 old_entry->end,
3067 old_entry->protection &
3068 ~VM_PROT_WRITE);
3069 }
3070 old_entry->etype |= UVM_ET_NEEDSCOPY;
3071 }
3072
3073 /*
3074 * parent must now be write-protected
3075 */
3076 protect_child = FALSE;
3077 } else {
3078
3079 /*
3080 * we only need to protect the child if the
3081 * parent has write access.
3082 */
3083 if (old_entry->max_protection & VM_PROT_WRITE)
3084 protect_child = TRUE;
3085 else
3086 protect_child = FALSE;
3087
3088 }
3089
3090 /*
3091 * copy the mappings
3092 * XXX: need a way to tell if this does anything
3093 */
3094
3095 pmap_copy(new_pmap, old_map->pmap,
3096 new_entry->start,
3097 (old_entry->end - old_entry->start),
3098 old_entry->start);
3099
3100 /*
3101 * protect the child's mappings if necessary
3102 */
3103 if (protect_child) {
3104 pmap_protect(new_pmap, new_entry->start,
3105 new_entry->end,
3106 new_entry->protection &
3107 ~VM_PROT_WRITE);
3108 }
3109
3110 }
3111 break;
3112 } /* end of switch statement */
3113 old_entry = old_entry->next;
3114 }
3115
3116 new_map->size = old_map->size;
3117 vm_map_unlock(old_map);
3118
3119 #ifdef SYSVSHM
3120 if (vm1->vm_shm)
3121 shmfork(vm1, vm2);
3122 #endif
3123
3124 #ifdef PMAP_FORK
3125 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3126 #endif
3127
3128 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3129 return(vm2);
3130 }
3131
3132
3133 #if defined(DDB)
3134
3135 /*
3136 * DDB hooks
3137 */
3138
3139 /*
3140 * uvm_map_print: print out a map
3141 */
3142
3143 void
3144 uvm_map_print(map, full)
3145 vm_map_t map;
3146 boolean_t full;
3147 {
3148
3149 uvm_map_printit(map, full, printf);
3150 }
3151
3152 /*
3153 * uvm_map_printit: actually prints the map
3154 */
3155
3156 void
3157 uvm_map_printit(map, full, pr)
3158 vm_map_t map;
3159 boolean_t full;
3160 void (*pr) __P((const char *, ...));
3161 {
3162 vm_map_entry_t entry;
3163
3164 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3165 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3166 map->nentries, map->size, map->ref_count, map->timestamp,
3167 map->flags);
3168 #ifdef pmap_resident_count
3169 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3170 pmap_resident_count(map->pmap));
3171 #else
3172 /* XXXCDC: this should be required ... */
3173 (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap);
3174 #endif
3175 if (!full)
3176 return;
3177 for (entry = map->header.next; entry != &map->header;
3178 entry = entry->next) {
3179 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%x, amap=%p/%d\n",
3180 entry, entry->start, entry->end, entry->object.uvm_obj,
3181 entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff);
3182 (*pr)(
3183 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n",
3184 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3185 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3186 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3187 entry->protection, entry->max_protection,
3188 entry->inheritance, entry->wired_count, entry->advice);
3189 }
3190 }
3191
3192 /*
3193 * uvm_object_print: print out an object
3194 */
3195
3196 void
3197 uvm_object_print(uobj, full)
3198 struct uvm_object *uobj;
3199 boolean_t full;
3200 {
3201
3202 uvm_object_printit(uobj, full, printf);
3203 }
3204
3205 /*
3206 * uvm_object_printit: actually prints the object
3207 */
3208
3209 void
3210 uvm_object_printit(uobj, full, pr)
3211 struct uvm_object *uobj;
3212 boolean_t full;
3213 void (*pr) __P((const char *, ...));
3214 {
3215 struct vm_page *pg;
3216 int cnt = 0;
3217
3218 (*pr)("OBJECT %p: pgops=%p, npages=%d, ", uobj, uobj->pgops,
3219 uobj->uo_npages);
3220 if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3221 (*pr)("refs=<SYSTEM>\n");
3222 else
3223 (*pr)("refs=%d\n", uobj->uo_refs);
3224
3225 if (!full) return;
3226 (*pr)(" PAGES <pg,offset>:\n ");
3227 for (pg = uobj->memq.tqh_first ; pg ; pg = pg->listq.tqe_next, cnt++) {
3228 (*pr)("<%p,0x%lx> ", pg, pg->offset);
3229 if ((cnt % 3) == 2) (*pr)("\n ");
3230 }
3231 if ((cnt % 3) != 2) (*pr)("\n");
3232 }
3233
3234 /*
3235 * uvm_page_print: print out a page
3236 */
3237
3238 void
3239 uvm_page_print(pg, full)
3240 struct vm_page *pg;
3241 boolean_t full;
3242 {
3243
3244 uvm_page_printit(pg, full, printf);
3245 }
3246
3247 /*
3248 * uvm_page_printit: actually print the page
3249 */
3250
3251 void
3252 uvm_page_printit(pg, full, pr)
3253 struct vm_page *pg;
3254 boolean_t full;
3255 void (*pr) __P((const char *, ...));
3256 {
3257 struct vm_page *lcv;
3258 struct uvm_object *uobj;
3259 struct pglist *pgl;
3260
3261 (*pr)("PAGE %p:\n", pg);
3262 (*pr)(" flags=0x%x, pqflags=0x%x, vers=%d, wire_count=%d, pa=0x%lx\n",
3263 pg->flags, pg->pqflags, pg->version, pg->wire_count, (long)pg->phys_addr);
3264 (*pr)(" uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
3265 pg->uobject, pg->uanon, pg->offset, pg->loan_count);
3266 #if defined(UVM_PAGE_TRKOWN)
3267 if (pg->flags & PG_BUSY)
3268 (*pr)(" owning process = %d, tag=%s\n",
3269 pg->owner, pg->owner_tag);
3270 else
3271 (*pr)(" page not busy, no owner\n");
3272 #else
3273 (*pr)(" [page ownership tracking disabled]\n");
3274 #endif
3275
3276 if (!full)
3277 return;
3278
3279 /* cross-verify object/anon */
3280 if ((pg->pqflags & PQ_FREE) == 0) {
3281 if (pg->pqflags & PQ_ANON) {
3282 if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3283 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3284 (pg->uanon) ? pg->uanon->u.an_page : NULL);
3285 else
3286 (*pr)(" anon backpointer is OK\n");
3287 } else {
3288 uobj = pg->uobject;
3289 if (uobj) {
3290 (*pr)(" checking object list\n");
3291 for (lcv = uobj->memq.tqh_first ; lcv ;
3292 lcv = lcv->listq.tqe_next) {
3293 if (lcv == pg) break;
3294 }
3295 if (lcv)
3296 (*pr)(" page found on object list\n");
3297 else
3298 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3299 }
3300 }
3301 }
3302
3303 /* cross-verify page queue */
3304 if (pg->pqflags & PQ_FREE)
3305 pgl = &uvm.page_free[uvm_page_lookup_freelist(pg)];
3306 else if (pg->pqflags & PQ_INACTIVE)
3307 pgl = (pg->pqflags & PQ_SWAPBACKED) ?
3308 &uvm.page_inactive_swp : &uvm.page_inactive_obj;
3309 else if (pg->pqflags & PQ_ACTIVE)
3310 pgl = &uvm.page_active;
3311 else
3312 pgl = NULL;
3313
3314 if (pgl) {
3315 (*pr)(" checking pageq list\n");
3316 for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) {
3317 if (lcv == pg) break;
3318 }
3319 if (lcv)
3320 (*pr)(" page found on pageq list\n");
3321 else
3322 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3323 }
3324 }
3325 #endif
3326