uvm_map.c revision 1.74 1 /* $NetBSD: uvm_map.c,v 1.74 2000/05/19 17:43:55 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_sysv.h"
72
73 /*
74 * uvm_map.c: uvm map operations
75 */
76
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/mman.h>
80 #include <sys/proc.h>
81 #include <sys/malloc.h>
82 #include <sys/pool.h>
83
84 #ifdef SYSVSHM
85 #include <sys/shm.h>
86 #endif
87
88 #include <vm/vm.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_kern.h>
91
92 #define UVM_MAP
93 #include <uvm/uvm.h>
94
95 #ifdef DDB
96 #include <uvm/uvm_ddb.h>
97 #endif
98
99
100 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
101 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
102
103 /*
104 * pool for vmspace structures.
105 */
106
107 struct pool uvm_vmspace_pool;
108
109 /*
110 * pool for dynamically-allocated map entries.
111 */
112
113 struct pool uvm_map_entry_pool;
114
115 #ifdef PMAP_GROWKERNEL
116 /*
117 * This global represents the end of the kernel virtual address
118 * space. If we want to exceed this, we must grow the kernel
119 * virtual address space dynamically.
120 *
121 * Note, this variable is locked by kernel_map's lock.
122 */
123 vaddr_t uvm_maxkaddr;
124 #endif
125
126 /*
127 * macros
128 */
129
130 /*
131 * uvm_map_entry_link: insert entry into a map
132 *
133 * => map must be locked
134 */
135 #define uvm_map_entry_link(map, after_where, entry) do { \
136 (map)->nentries++; \
137 (entry)->prev = (after_where); \
138 (entry)->next = (after_where)->next; \
139 (entry)->prev->next = (entry); \
140 (entry)->next->prev = (entry); \
141 } while (0)
142
143 /*
144 * uvm_map_entry_unlink: remove entry from a map
145 *
146 * => map must be locked
147 */
148 #define uvm_map_entry_unlink(map, entry) do { \
149 (map)->nentries--; \
150 (entry)->next->prev = (entry)->prev; \
151 (entry)->prev->next = (entry)->next; \
152 } while (0)
153
154 /*
155 * SAVE_HINT: saves the specified entry as the hint for future lookups.
156 *
157 * => map need not be locked (protected by hint_lock).
158 */
159 #define SAVE_HINT(map,value) do { \
160 simple_lock(&(map)->hint_lock); \
161 (map)->hint = (value); \
162 simple_unlock(&(map)->hint_lock); \
163 } while (0)
164
165 /*
166 * VM_MAP_RANGE_CHECK: check and correct range
167 *
168 * => map must at least be read locked
169 */
170
171 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
172 if (start < vm_map_min(map)) \
173 start = vm_map_min(map); \
174 if (end > vm_map_max(map)) \
175 end = vm_map_max(map); \
176 if (start > end) \
177 start = end; \
178 } while (0)
179
180 /*
181 * local prototypes
182 */
183
184 static vm_map_entry_t uvm_mapent_alloc __P((vm_map_t));
185 static void uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t));
186 static void uvm_mapent_free __P((vm_map_entry_t));
187 static void uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
188
189 /*
190 * local inlines
191 */
192
193 /*
194 * uvm_mapent_alloc: allocate a map entry
195 *
196 * => XXX: static pool for kernel map?
197 */
198
199 static __inline vm_map_entry_t
200 uvm_mapent_alloc(map)
201 vm_map_t map;
202 {
203 vm_map_entry_t me;
204 int s;
205 UVMHIST_FUNC("uvm_mapent_alloc");
206 UVMHIST_CALLED(maphist);
207
208 if ((map->flags & VM_MAP_INTRSAFE) == 0 &&
209 map != kernel_map && kernel_map != NULL /* XXX */) {
210 me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
211 me->flags = 0;
212 /* me can't be null, wait ok */
213 } else {
214 s = splimp(); /* protect kentry_free list with splimp */
215 simple_lock(&uvm.kentry_lock);
216 me = uvm.kentry_free;
217 if (me) uvm.kentry_free = me->next;
218 simple_unlock(&uvm.kentry_lock);
219 splx(s);
220 if (!me)
221 panic("mapent_alloc: out of static map entries, check MAX_KMAPENT");
222 me->flags = UVM_MAP_STATIC;
223 }
224
225 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]",
226 me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map)
227 ? TRUE : FALSE, 0, 0);
228 return(me);
229 }
230
231 /*
232 * uvm_mapent_free: free map entry
233 *
234 * => XXX: static pool for kernel map?
235 */
236
237 static __inline void
238 uvm_mapent_free(me)
239 vm_map_entry_t me;
240 {
241 int s;
242 UVMHIST_FUNC("uvm_mapent_free");
243 UVMHIST_CALLED(maphist);
244 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
245 me, me->flags, 0, 0);
246 if ((me->flags & UVM_MAP_STATIC) == 0) {
247 pool_put(&uvm_map_entry_pool, me);
248 } else {
249 s = splimp(); /* protect kentry_free list with splimp */
250 simple_lock(&uvm.kentry_lock);
251 me->next = uvm.kentry_free;
252 uvm.kentry_free = me;
253 simple_unlock(&uvm.kentry_lock);
254 splx(s);
255 }
256 }
257
258 /*
259 * uvm_mapent_copy: copy a map entry, preserving flags
260 */
261
262 static __inline void
263 uvm_mapent_copy(src, dst)
264 vm_map_entry_t src;
265 vm_map_entry_t dst;
266 {
267
268 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src));
269 }
270
271 /*
272 * uvm_map_entry_unwire: unwire a map entry
273 *
274 * => map should be locked by caller
275 */
276
277 static __inline void
278 uvm_map_entry_unwire(map, entry)
279 vm_map_t map;
280 vm_map_entry_t entry;
281 {
282
283 entry->wired_count = 0;
284 uvm_fault_unwire_locked(map, entry->start, entry->end);
285 }
286
287 /*
288 * uvm_map_init: init mapping system at boot time. note that we allocate
289 * and init the static pool of vm_map_entry_t's for the kernel here.
290 */
291
292 void
293 uvm_map_init()
294 {
295 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
296 #if defined(UVMHIST)
297 static struct uvm_history_ent maphistbuf[100];
298 static struct uvm_history_ent pdhistbuf[100];
299 #endif
300 int lcv;
301
302 /*
303 * first, init logging system.
304 */
305
306 UVMHIST_FUNC("uvm_map_init");
307 UVMHIST_INIT_STATIC(maphist, maphistbuf);
308 UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
309 UVMHIST_CALLED(maphist);
310 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
311 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0,
312 "# uvm_map() successful calls", 0);
313 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
314 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
315 0);
316 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0);
317 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0);
318
319 /*
320 * now set up static pool of kernel map entrys ...
321 */
322
323 simple_lock_init(&uvm.kentry_lock);
324 uvm.kentry_free = NULL;
325 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
326 kernel_map_entry[lcv].next = uvm.kentry_free;
327 uvm.kentry_free = &kernel_map_entry[lcv];
328 }
329
330 /*
331 * initialize the map-related pools.
332 */
333 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
334 0, 0, 0, "vmsppl", 0,
335 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
336 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
337 0, 0, 0, "vmmpepl", 0,
338 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
339 }
340
341 /*
342 * clippers
343 */
344
345 /*
346 * uvm_map_clip_start: ensure that the entry begins at or after
347 * the starting address, if it doesn't we split the entry.
348 *
349 * => caller should use UVM_MAP_CLIP_START macro rather than calling
350 * this directly
351 * => map must be locked by caller
352 */
353
354 void uvm_map_clip_start(map, entry, start)
355 vm_map_t map;
356 vm_map_entry_t entry;
357 vaddr_t start;
358 {
359 vm_map_entry_t new_entry;
360 vaddr_t new_adj;
361
362 /* uvm_map_simplify_entry(map, entry); */ /* XXX */
363
364 /*
365 * Split off the front portion. note that we must insert the new
366 * entry BEFORE this one, so that this entry has the specified
367 * starting address.
368 */
369
370 new_entry = uvm_mapent_alloc(map);
371 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
372
373 new_entry->end = start;
374 new_adj = start - new_entry->start;
375 if (entry->object.uvm_obj)
376 entry->offset += new_adj; /* shift start over */
377 entry->start = start;
378
379 if (new_entry->aref.ar_amap) {
380 amap_splitref(&new_entry->aref, &entry->aref, new_adj);
381 }
382
383 uvm_map_entry_link(map, entry->prev, new_entry);
384
385 if (UVM_ET_ISSUBMAP(entry)) {
386 /* ... unlikely to happen, but play it safe */
387 uvm_map_reference(new_entry->object.sub_map);
388 } else {
389 if (UVM_ET_ISOBJ(entry) &&
390 entry->object.uvm_obj->pgops &&
391 entry->object.uvm_obj->pgops->pgo_reference)
392 entry->object.uvm_obj->pgops->pgo_reference(
393 entry->object.uvm_obj);
394 }
395 }
396
397 /*
398 * uvm_map_clip_end: ensure that the entry ends at or before
399 * the ending address, if it does't we split the reference
400 *
401 * => caller should use UVM_MAP_CLIP_END macro rather than calling
402 * this directly
403 * => map must be locked by caller
404 */
405
406 void
407 uvm_map_clip_end(map, entry, end)
408 vm_map_t map;
409 vm_map_entry_t entry;
410 vaddr_t end;
411 {
412 vm_map_entry_t new_entry;
413 vaddr_t new_adj; /* #bytes we move start forward */
414
415 /*
416 * Create a new entry and insert it
417 * AFTER the specified entry
418 */
419
420 new_entry = uvm_mapent_alloc(map);
421 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
422
423 new_entry->start = entry->end = end;
424 new_adj = end - entry->start;
425 if (new_entry->object.uvm_obj)
426 new_entry->offset += new_adj;
427
428 if (entry->aref.ar_amap)
429 amap_splitref(&entry->aref, &new_entry->aref, new_adj);
430
431 uvm_map_entry_link(map, entry, new_entry);
432
433 if (UVM_ET_ISSUBMAP(entry)) {
434 /* ... unlikely to happen, but play it safe */
435 uvm_map_reference(new_entry->object.sub_map);
436 } else {
437 if (UVM_ET_ISOBJ(entry) &&
438 entry->object.uvm_obj->pgops &&
439 entry->object.uvm_obj->pgops->pgo_reference)
440 entry->object.uvm_obj->pgops->pgo_reference(
441 entry->object.uvm_obj);
442 }
443 }
444
445
446 /*
447 * M A P - m a i n e n t r y p o i n t
448 */
449 /*
450 * uvm_map: establish a valid mapping in a map
451 *
452 * => assume startp is page aligned.
453 * => assume size is a multiple of PAGE_SIZE.
454 * => assume sys_mmap provides enough of a "hint" to have us skip
455 * over text/data/bss area.
456 * => map must be unlocked (we will lock it)
457 * => <uobj,uoffset> value meanings (4 cases):
458 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER
459 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER
460 * [3] <uobj,uoffset> == normal mapping
461 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA
462 *
463 * case [4] is for kernel mappings where we don't know the offset until
464 * we've found a virtual address. note that kernel object offsets are
465 * always relative to vm_map_min(kernel_map).
466 * => XXXCDC: need way to map in external amap?
467 */
468
469 int
470 uvm_map(map, startp, size, uobj, uoffset, flags)
471 vm_map_t map;
472 vaddr_t *startp; /* IN/OUT */
473 vsize_t size;
474 struct uvm_object *uobj;
475 voff_t uoffset;
476 uvm_flag_t flags;
477 {
478 vm_map_entry_t prev_entry, new_entry;
479 vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
480 UVM_MAXPROTECTION(flags);
481 vm_inherit_t inherit = UVM_INHERIT(flags);
482 int advice = UVM_ADVICE(flags);
483 UVMHIST_FUNC("uvm_map");
484 UVMHIST_CALLED(maphist);
485
486 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
487 map, *startp, size, flags);
488 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0);
489
490 /*
491 * step 0: sanity check of protection code
492 */
493
494 if ((prot & maxprot) != prot) {
495 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x",
496 prot, maxprot,0,0);
497 return(KERN_PROTECTION_FAILURE);
498 }
499
500 /*
501 * step 1: figure out where to put new VM range
502 */
503
504 if (vm_map_lock_try(map) == FALSE) {
505 if (flags & UVM_FLAG_TRYLOCK)
506 return(KERN_FAILURE);
507 vm_map_lock(map); /* could sleep here */
508 }
509 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
510 uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) {
511 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
512 vm_map_unlock(map);
513 return (KERN_NO_SPACE);
514 }
515
516 #ifdef PMAP_GROWKERNEL
517 {
518 /*
519 * If the kernel pmap can't map the requested space,
520 * then allocate more resources for it.
521 */
522 if (map == kernel_map && uvm_maxkaddr < (*startp + size))
523 uvm_maxkaddr = pmap_growkernel(*startp + size);
524 }
525 #endif
526
527 UVMCNT_INCR(uvm_map_call);
528
529 /*
530 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
531 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in
532 * either case we want to zero it before storing it in the map entry
533 * (because it looks strange and confusing when debugging...)
534 *
535 * if uobj is not null
536 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
537 * and we do not need to change uoffset.
538 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
539 * now (based on the starting address of the map). this case is
540 * for kernel object mappings where we don't know the offset until
541 * the virtual address is found (with uvm_map_findspace). the
542 * offset is the distance we are from the start of the map.
543 */
544
545 if (uobj == NULL) {
546 uoffset = 0;
547 } else {
548 if (uoffset == UVM_UNKNOWN_OFFSET) {
549 #ifdef DIAGNOSTIC
550 if (UVM_OBJ_IS_KERN_OBJECT(uobj) == 0)
551 panic("uvm_map: unknown offset with "
552 "non-kernel object");
553 #endif
554 uoffset = *startp - vm_map_min(kernel_map);
555 }
556 }
557
558 /*
559 * step 2: try and insert in map by extending previous entry, if
560 * possible
561 * XXX: we don't try and pull back the next entry. might be useful
562 * for a stack, but we are currently allocating our stack in advance.
563 */
564
565 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
566 prev_entry->end == *startp && prev_entry != &map->header &&
567 prev_entry->object.uvm_obj == uobj) {
568
569 if (uobj && prev_entry->offset +
570 (prev_entry->end - prev_entry->start) != uoffset)
571 goto step3;
572
573 if (UVM_ET_ISSUBMAP(prev_entry))
574 goto step3;
575
576 if (prev_entry->protection != prot ||
577 prev_entry->max_protection != maxprot)
578 goto step3;
579
580 if (prev_entry->inheritance != inherit ||
581 prev_entry->advice != advice)
582 goto step3;
583
584 /* wiring status must match (new area is unwired) */
585 if (VM_MAPENT_ISWIRED(prev_entry))
586 goto step3;
587
588 /*
589 * can't extend a shared amap. note: no need to lock amap to
590 * look at refs since we don't care about its exact value.
591 * if it is one (i.e. we have only reference) it will stay there
592 */
593
594 if (prev_entry->aref.ar_amap &&
595 amap_refs(prev_entry->aref.ar_amap) != 1) {
596 goto step3;
597 }
598
599 /* got it! */
600
601 UVMCNT_INCR(map_backmerge);
602 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0);
603
604 /*
605 * drop our reference to uobj since we are extending a reference
606 * that we already have (the ref count can not drop to zero).
607 */
608 if (uobj && uobj->pgops->pgo_detach)
609 uobj->pgops->pgo_detach(uobj);
610
611 if (prev_entry->aref.ar_amap) {
612 amap_extend(prev_entry, size);
613 }
614
615 prev_entry->end += size;
616 map->size += size;
617
618 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
619 vm_map_unlock(map);
620 return (KERN_SUCCESS);
621
622 }
623 step3:
624 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0);
625
626 /*
627 * check for possible forward merge (which we don't do) and count
628 * the number of times we missed a *possible* chance to merge more
629 */
630
631 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
632 prev_entry->next != &map->header &&
633 prev_entry->next->start == (*startp + size))
634 UVMCNT_INCR(map_forwmerge);
635
636 /*
637 * step 3: allocate new entry and link it in
638 */
639
640 new_entry = uvm_mapent_alloc(map);
641 new_entry->start = *startp;
642 new_entry->end = new_entry->start + size;
643 new_entry->object.uvm_obj = uobj;
644 new_entry->offset = uoffset;
645
646 if (uobj)
647 new_entry->etype = UVM_ET_OBJ;
648 else
649 new_entry->etype = 0;
650
651 if (flags & UVM_FLAG_COPYONW) {
652 new_entry->etype |= UVM_ET_COPYONWRITE;
653 if ((flags & UVM_FLAG_OVERLAY) == 0)
654 new_entry->etype |= UVM_ET_NEEDSCOPY;
655 }
656
657 new_entry->protection = prot;
658 new_entry->max_protection = maxprot;
659 new_entry->inheritance = inherit;
660 new_entry->wired_count = 0;
661 new_entry->advice = advice;
662 if (flags & UVM_FLAG_OVERLAY) {
663 /*
664 * to_add: for BSS we overallocate a little since we
665 * are likely to extend
666 */
667 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
668 UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
669 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK);
670 new_entry->aref.ar_pageoff = 0;
671 new_entry->aref.ar_amap = amap;
672 } else {
673 new_entry->aref.ar_amap = NULL;
674 }
675
676 uvm_map_entry_link(map, prev_entry, new_entry);
677
678 map->size += size;
679
680 /*
681 * Update the free space hint
682 */
683
684 if ((map->first_free == prev_entry) &&
685 (prev_entry->end >= new_entry->start))
686 map->first_free = new_entry;
687
688 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
689 vm_map_unlock(map);
690 return(KERN_SUCCESS);
691 }
692
693 /*
694 * uvm_map_lookup_entry: find map entry at or before an address
695 *
696 * => map must at least be read-locked by caller
697 * => entry is returned in "entry"
698 * => return value is true if address is in the returned entry
699 */
700
701 boolean_t
702 uvm_map_lookup_entry(map, address, entry)
703 vm_map_t map;
704 vaddr_t address;
705 vm_map_entry_t *entry; /* OUT */
706 {
707 vm_map_entry_t cur;
708 vm_map_entry_t last;
709 UVMHIST_FUNC("uvm_map_lookup_entry");
710 UVMHIST_CALLED(maphist);
711
712 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
713 map, address, entry, 0);
714
715 /*
716 * start looking either from the head of the
717 * list, or from the hint.
718 */
719
720 simple_lock(&map->hint_lock);
721 cur = map->hint;
722 simple_unlock(&map->hint_lock);
723
724 if (cur == &map->header)
725 cur = cur->next;
726
727 UVMCNT_INCR(uvm_mlk_call);
728 if (address >= cur->start) {
729 /*
730 * go from hint to end of list.
731 *
732 * but first, make a quick check to see if
733 * we are already looking at the entry we
734 * want (which is usually the case).
735 * note also that we don't need to save the hint
736 * here... it is the same hint (unless we are
737 * at the header, in which case the hint didn't
738 * buy us anything anyway).
739 */
740 last = &map->header;
741 if ((cur != last) && (cur->end > address)) {
742 UVMCNT_INCR(uvm_mlk_hint);
743 *entry = cur;
744 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
745 cur, 0, 0, 0);
746 return (TRUE);
747 }
748 } else {
749 /*
750 * go from start to hint, *inclusively*
751 */
752 last = cur->next;
753 cur = map->header.next;
754 }
755
756 /*
757 * search linearly
758 */
759
760 while (cur != last) {
761 if (cur->end > address) {
762 if (address >= cur->start) {
763 /*
764 * save this lookup for future
765 * hints, and return
766 */
767
768 *entry = cur;
769 SAVE_HINT(map, cur);
770 UVMHIST_LOG(maphist,"<- search got it (0x%x)",
771 cur, 0, 0, 0);
772 return (TRUE);
773 }
774 break;
775 }
776 cur = cur->next;
777 }
778 *entry = cur->prev;
779 SAVE_HINT(map, *entry);
780 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
781 return (FALSE);
782 }
783
784
785 /*
786 * uvm_map_findspace: find "length" sized space in "map".
787 *
788 * => "hint" is a hint about where we want it, unless fixed is true
789 * (in which case we insist on using "hint").
790 * => "result" is VA returned
791 * => uobj/uoffset are to be used to handle VAC alignment, if required
792 * => caller must at least have read-locked map
793 * => returns NULL on failure, or pointer to prev. map entry if success
794 * => note this is a cross between the old vm_map_findspace and vm_map_find
795 */
796
797 vm_map_entry_t
798 uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed)
799 vm_map_t map;
800 vaddr_t hint;
801 vsize_t length;
802 vaddr_t *result; /* OUT */
803 struct uvm_object *uobj;
804 voff_t uoffset;
805 boolean_t fixed;
806 {
807 vm_map_entry_t entry, next, tmp;
808 vaddr_t end;
809 UVMHIST_FUNC("uvm_map_findspace");
810 UVMHIST_CALLED(maphist);
811
812 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)",
813 map, hint, length, fixed);
814
815 if (hint < map->min_offset) { /* check ranges ... */
816 if (fixed) {
817 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
818 return(NULL);
819 }
820 hint = map->min_offset;
821 }
822 if (hint > map->max_offset) {
823 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
824 hint, map->min_offset, map->max_offset, 0);
825 return(NULL);
826 }
827
828 /*
829 * Look for the first possible address; if there's already
830 * something at this address, we have to start after it.
831 */
832
833 if (!fixed && hint == map->min_offset) {
834 if ((entry = map->first_free) != &map->header)
835 hint = entry->end;
836 } else {
837 if (uvm_map_lookup_entry(map, hint, &tmp)) {
838 /* "hint" address already in use ... */
839 if (fixed) {
840 UVMHIST_LOG(maphist,"<- fixed & VA in use",
841 0, 0, 0, 0);
842 return(NULL);
843 }
844 hint = tmp->end;
845 }
846 entry = tmp;
847 }
848
849 /*
850 * Look through the rest of the map, trying to fit a new region in
851 * the gap between existing regions, or after the very last region.
852 * note: entry->end = base VA of current gap,
853 * next->start = VA of end of current gap
854 */
855 for (;; hint = (entry = next)->end) {
856 /*
857 * Find the end of the proposed new region. Be sure we didn't
858 * go beyond the end of the map, or wrap around the address;
859 * if so, we lose. Otherwise, if this is the last entry, or
860 * if the proposed new region fits before the next entry, we
861 * win.
862 */
863
864 #ifdef PMAP_PREFER
865 /*
866 * push hint forward as needed to avoid VAC alias problems.
867 * we only do this if a valid offset is specified.
868 */
869 if (!fixed && uoffset != UVM_UNKNOWN_OFFSET)
870 PMAP_PREFER(uoffset, &hint);
871 #endif
872 end = hint + length;
873 if (end > map->max_offset || end < hint) {
874 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
875 return (NULL);
876 }
877 next = entry->next;
878 if (next == &map->header || next->start >= end)
879 break;
880 if (fixed) {
881 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
882 return(NULL); /* only one shot at it ... */
883 }
884 }
885 SAVE_HINT(map, entry);
886 *result = hint;
887 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0);
888 return (entry);
889 }
890
891 /*
892 * U N M A P - m a i n h e l p e r f u n c t i o n s
893 */
894
895 /*
896 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
897 *
898 * => caller must check alignment and size
899 * => map must be locked by caller
900 * => we return a list of map entries that we've remove from the map
901 * in "entry_list"
902 */
903
904 int
905 uvm_unmap_remove(map, start, end, entry_list)
906 vm_map_t map;
907 vaddr_t start,end;
908 vm_map_entry_t *entry_list; /* OUT */
909 {
910 vm_map_entry_t entry, first_entry, next;
911 vaddr_t len;
912 UVMHIST_FUNC("uvm_unmap_remove");
913 UVMHIST_CALLED(maphist);
914
915 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
916 map, start, end, 0);
917
918 VM_MAP_RANGE_CHECK(map, start, end);
919
920 /*
921 * find first entry
922 */
923 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
924 /* clip and go... */
925 entry = first_entry;
926 UVM_MAP_CLIP_START(map, entry, start);
927 /* critical! prevents stale hint */
928 SAVE_HINT(map, entry->prev);
929
930 } else {
931 entry = first_entry->next;
932 }
933
934 /*
935 * Save the free space hint
936 */
937
938 if (map->first_free->start >= start)
939 map->first_free = entry->prev;
940
941 /*
942 * note: we now re-use first_entry for a different task. we remove
943 * a number of map entries from the map and save them in a linked
944 * list headed by "first_entry". once we remove them from the map
945 * the caller should unlock the map and drop the references to the
946 * backing objects [c.f. uvm_unmap_detach]. the object is to
947 * seperate unmapping from reference dropping. why?
948 * [1] the map has to be locked for unmapping
949 * [2] the map need not be locked for reference dropping
950 * [3] dropping references may trigger pager I/O, and if we hit
951 * a pager that does synchronous I/O we may have to wait for it.
952 * [4] we would like all waiting for I/O to occur with maps unlocked
953 * so that we don't block other threads.
954 */
955 first_entry = NULL;
956 *entry_list = NULL; /* to be safe */
957
958 /*
959 * break up the area into map entry sized regions and unmap. note
960 * that all mappings have to be removed before we can even consider
961 * dropping references to amaps or VM objects (otherwise we could end
962 * up with a mapping to a page on the free list which would be very bad)
963 */
964
965 while ((entry != &map->header) && (entry->start < end)) {
966
967 UVM_MAP_CLIP_END(map, entry, end);
968 next = entry->next;
969 len = entry->end - entry->start;
970
971 /*
972 * unwire before removing addresses from the pmap; otherwise
973 * unwiring will put the entries back into the pmap (XXX).
974 */
975
976 if (VM_MAPENT_ISWIRED(entry))
977 uvm_map_entry_unwire(map, entry);
978
979 /*
980 * special case: handle mappings to anonymous kernel objects.
981 * we want to free these pages right away...
982 */
983 if (UVM_ET_ISOBJ(entry) &&
984 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
985 #ifdef DIAGNOSTIC
986 if (vm_map_pmap(map) != pmap_kernel())
987 panic("uvm_unmap_remove: kernel object "
988 "mapped by non-kernel map");
989 #endif
990
991 /*
992 * note: kernel object mappings are currently used in
993 * two ways:
994 * [1] "normal" mappings of pages in the kernel object
995 * [2] uvm_km_valloc'd allocations in which we
996 * pmap_enter in some non-kernel-object page
997 * (e.g. vmapbuf).
998 *
999 * for case [1], we need to remove the mapping from
1000 * the pmap and then remove the page from the kernel
1001 * object (because, once pages in a kernel object are
1002 * unmapped they are no longer needed, unlike, say,
1003 * a vnode where you might want the data to persist
1004 * until flushed out of a queue).
1005 *
1006 * for case [2], we need to remove the mapping from
1007 * the pmap. there shouldn't be any pages at the
1008 * specified offset in the kernel object [but it
1009 * doesn't hurt to call uvm_km_pgremove just to be
1010 * safe?]
1011 *
1012 * uvm_km_pgremove currently does the following:
1013 * for pages in the kernel object in range:
1014 * - drops the swap slot
1015 * - uvm_pagefree the page
1016 *
1017 * note there is version of uvm_km_pgremove() that
1018 * is used for "intrsafe" objects.
1019 */
1020
1021 /*
1022 * remove mappings from pmap and drop the pages
1023 * from the object. offsets are always relative
1024 * to vm_map_min(kernel_map).
1025 */
1026 if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) {
1027 pmap_kremove(entry->start, len);
1028 uvm_km_pgremove_intrsafe(entry->object.uvm_obj,
1029 entry->start - vm_map_min(kernel_map),
1030 entry->end - vm_map_min(kernel_map));
1031 } else {
1032 pmap_remove(pmap_kernel(), entry->start,
1033 entry->start + len);
1034 uvm_km_pgremove(entry->object.uvm_obj,
1035 entry->start - vm_map_min(kernel_map),
1036 entry->end - vm_map_min(kernel_map));
1037 }
1038
1039 /*
1040 * null out kernel_object reference, we've just
1041 * dropped it
1042 */
1043 entry->etype &= ~UVM_ET_OBJ;
1044 entry->object.uvm_obj = NULL; /* to be safe */
1045
1046 } else {
1047 /*
1048 * remove mappings the standard way.
1049 */
1050 pmap_remove(map->pmap, entry->start, entry->end);
1051 }
1052
1053 /*
1054 * remove entry from map and put it on our list of entries
1055 * that we've nuked. then go do next entry.
1056 */
1057 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0);
1058 uvm_map_entry_unlink(map, entry);
1059 map->size -= len;
1060 entry->next = first_entry;
1061 first_entry = entry;
1062 entry = next; /* next entry, please */
1063 }
1064
1065 /*
1066 * now we've cleaned up the map and are ready for the caller to drop
1067 * references to the mapped objects.
1068 */
1069
1070 *entry_list = first_entry;
1071 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1072 return(KERN_SUCCESS);
1073 }
1074
1075 /*
1076 * uvm_unmap_detach: drop references in a chain of map entries
1077 *
1078 * => we will free the map entries as we traverse the list.
1079 */
1080
1081 void
1082 uvm_unmap_detach(first_entry, amap_unref_flags)
1083 vm_map_entry_t first_entry;
1084 int amap_unref_flags;
1085 {
1086 vm_map_entry_t next_entry;
1087 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1088
1089 while (first_entry) {
1090
1091 #ifdef DIAGNOSTIC
1092 /*
1093 * sanity check
1094 */
1095 /* was part of vm_map_entry_delete() */
1096 if (VM_MAPENT_ISWIRED(first_entry))
1097 panic("unmap: still wired!");
1098 #endif
1099
1100 UVMHIST_LOG(maphist,
1101 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1102 first_entry, first_entry->aref.ar_amap,
1103 first_entry->object.uvm_obj,
1104 UVM_ET_ISSUBMAP(first_entry));
1105
1106 /*
1107 * drop reference to amap, if we've got one
1108 */
1109
1110 if (first_entry->aref.ar_amap)
1111 amap_unref(first_entry, amap_unref_flags);
1112
1113 /*
1114 * drop reference to our backing object, if we've got one
1115 */
1116
1117 if (UVM_ET_ISSUBMAP(first_entry)) {
1118 /* ... unlikely to happen, but play it safe */
1119 uvm_map_deallocate(first_entry->object.sub_map);
1120 } else {
1121 if (UVM_ET_ISOBJ(first_entry) &&
1122 first_entry->object.uvm_obj->pgops->pgo_detach)
1123 first_entry->object.uvm_obj->pgops->
1124 pgo_detach(first_entry->object.uvm_obj);
1125 }
1126
1127 /*
1128 * next entry
1129 */
1130 next_entry = first_entry->next;
1131 uvm_mapent_free(first_entry);
1132 first_entry = next_entry;
1133 }
1134
1135 /*
1136 * done!
1137 */
1138 UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1139 return;
1140 }
1141
1142 /*
1143 * E X T R A C T I O N F U N C T I O N S
1144 */
1145
1146 /*
1147 * uvm_map_reserve: reserve space in a vm_map for future use.
1148 *
1149 * => we reserve space in a map by putting a dummy map entry in the
1150 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1151 * => map should be unlocked (we will write lock it)
1152 * => we return true if we were able to reserve space
1153 * => XXXCDC: should be inline?
1154 */
1155
1156 int
1157 uvm_map_reserve(map, size, offset, raddr)
1158 vm_map_t map;
1159 vsize_t size;
1160 vaddr_t offset; /* hint for pmap_prefer */
1161 vaddr_t *raddr; /* OUT: reserved VA */
1162 {
1163 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1164
1165 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1166 map,size,offset,raddr);
1167
1168 size = round_page(size);
1169 if (*raddr < vm_map_min(map))
1170 *raddr = vm_map_min(map); /* hint */
1171
1172 /*
1173 * reserve some virtual space.
1174 */
1175
1176 if (uvm_map(map, raddr, size, NULL, offset,
1177 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1178 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) {
1179 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1180 return (FALSE);
1181 }
1182
1183 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1184 return (TRUE);
1185 }
1186
1187 /*
1188 * uvm_map_replace: replace a reserved (blank) area of memory with
1189 * real mappings.
1190 *
1191 * => caller must WRITE-LOCK the map
1192 * => we return TRUE if replacement was a success
1193 * => we expect the newents chain to have nnewents entrys on it and
1194 * we expect newents->prev to point to the last entry on the list
1195 * => note newents is allowed to be NULL
1196 */
1197
1198 int
1199 uvm_map_replace(map, start, end, newents, nnewents)
1200 struct vm_map *map;
1201 vaddr_t start, end;
1202 vm_map_entry_t newents;
1203 int nnewents;
1204 {
1205 vm_map_entry_t oldent, last;
1206 UVMHIST_FUNC("uvm_map_replace");
1207 UVMHIST_CALLED(maphist);
1208
1209 /*
1210 * first find the blank map entry at the specified address
1211 */
1212
1213 if (!uvm_map_lookup_entry(map, start, &oldent)) {
1214 return(FALSE);
1215 }
1216
1217 /*
1218 * check to make sure we have a proper blank entry
1219 */
1220
1221 if (oldent->start != start || oldent->end != end ||
1222 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1223 return (FALSE);
1224 }
1225
1226 #ifdef DIAGNOSTIC
1227 /*
1228 * sanity check the newents chain
1229 */
1230 {
1231 vm_map_entry_t tmpent = newents;
1232 int nent = 0;
1233 vaddr_t cur = start;
1234
1235 while (tmpent) {
1236 nent++;
1237 if (tmpent->start < cur)
1238 panic("uvm_map_replace1");
1239 if (tmpent->start > tmpent->end || tmpent->end > end) {
1240 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1241 tmpent->start, tmpent->end, end);
1242 panic("uvm_map_replace2");
1243 }
1244 cur = tmpent->end;
1245 if (tmpent->next) {
1246 if (tmpent->next->prev != tmpent)
1247 panic("uvm_map_replace3");
1248 } else {
1249 if (newents->prev != tmpent)
1250 panic("uvm_map_replace4");
1251 }
1252 tmpent = tmpent->next;
1253 }
1254 if (nent != nnewents)
1255 panic("uvm_map_replace5");
1256 }
1257 #endif
1258
1259 /*
1260 * map entry is a valid blank! replace it. (this does all the
1261 * work of map entry link/unlink...).
1262 */
1263
1264 if (newents) {
1265
1266 last = newents->prev; /* we expect this */
1267
1268 /* critical: flush stale hints out of map */
1269 SAVE_HINT(map, newents);
1270 if (map->first_free == oldent)
1271 map->first_free = last;
1272
1273 last->next = oldent->next;
1274 last->next->prev = last;
1275 newents->prev = oldent->prev;
1276 newents->prev->next = newents;
1277 map->nentries = map->nentries + (nnewents - 1);
1278
1279 } else {
1280
1281 /* critical: flush stale hints out of map */
1282 SAVE_HINT(map, oldent->prev);
1283 if (map->first_free == oldent)
1284 map->first_free = oldent->prev;
1285
1286 /* NULL list of new entries: just remove the old one */
1287 uvm_map_entry_unlink(map, oldent);
1288 }
1289
1290
1291 /*
1292 * now we can free the old blank entry, unlock the map and return.
1293 */
1294
1295 uvm_mapent_free(oldent);
1296 return(TRUE);
1297 }
1298
1299 /*
1300 * uvm_map_extract: extract a mapping from a map and put it somewhere
1301 * (maybe removing the old mapping)
1302 *
1303 * => maps should be unlocked (we will write lock them)
1304 * => returns 0 on success, error code otherwise
1305 * => start must be page aligned
1306 * => len must be page sized
1307 * => flags:
1308 * UVM_EXTRACT_REMOVE: remove mappings from srcmap
1309 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1310 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1311 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1312 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1313 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1314 * be used from within the kernel in a kernel level map <<<
1315 */
1316
1317 int
1318 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1319 vm_map_t srcmap, dstmap;
1320 vaddr_t start, *dstaddrp;
1321 vsize_t len;
1322 int flags;
1323 {
1324 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1325 oldstart;
1326 vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry;
1327 vm_map_entry_t oldentry;
1328 vsize_t elen;
1329 int nchain, error, copy_ok;
1330 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1331 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1332 len,0);
1333 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1334
1335 #ifdef DIAGNOSTIC
1336 /*
1337 * step 0: sanity check: start must be on a page boundary, length
1338 * must be page sized. can't ask for CONTIG/QREF if you asked for
1339 * REMOVE.
1340 */
1341 if ((start & PAGE_MASK) || (len & PAGE_MASK))
1342 panic("uvm_map_extract1");
1343 if (flags & UVM_EXTRACT_REMOVE)
1344 if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF))
1345 panic("uvm_map_extract2");
1346 #endif
1347
1348
1349 /*
1350 * step 1: reserve space in the target map for the extracted area
1351 */
1352
1353 dstaddr = *dstaddrp;
1354 if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE)
1355 return(ENOMEM);
1356 *dstaddrp = dstaddr; /* pass address back to caller */
1357 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0);
1358
1359
1360 /*
1361 * step 2: setup for the extraction process loop by init'ing the
1362 * map entry chain, locking src map, and looking up the first useful
1363 * entry in the map.
1364 */
1365
1366 end = start + len;
1367 newend = dstaddr + len;
1368 chain = endchain = NULL;
1369 nchain = 0;
1370 vm_map_lock(srcmap);
1371
1372 if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1373
1374 /* "start" is within an entry */
1375 if (flags & UVM_EXTRACT_QREF) {
1376 /*
1377 * for quick references we don't clip the entry, so
1378 * the entry may map space "before" the starting
1379 * virtual address... this is the "fudge" factor
1380 * (which can be non-zero only the first time
1381 * through the "while" loop in step 3).
1382 */
1383 fudge = start - entry->start;
1384 } else {
1385 /*
1386 * normal reference: we clip the map to fit (thus
1387 * fudge is zero)
1388 */
1389 UVM_MAP_CLIP_START(srcmap, entry, start);
1390 SAVE_HINT(srcmap, entry->prev);
1391 fudge = 0;
1392 }
1393
1394 } else {
1395
1396 /* "start" is not within an entry ... skip to next entry */
1397 if (flags & UVM_EXTRACT_CONTIG) {
1398 error = EINVAL;
1399 goto bad; /* definite hole here ... */
1400 }
1401
1402 entry = entry->next;
1403 fudge = 0;
1404 }
1405 /* save values from srcmap for step 6 */
1406 orig_entry = entry;
1407 orig_fudge = fudge;
1408
1409
1410 /*
1411 * step 3: now start looping through the map entries, extracting
1412 * as we go.
1413 */
1414
1415 while (entry->start < end && entry != &srcmap->header) {
1416
1417 /* if we are not doing a quick reference, clip it */
1418 if ((flags & UVM_EXTRACT_QREF) == 0)
1419 UVM_MAP_CLIP_END(srcmap, entry, end);
1420
1421 /* clear needs_copy (allow chunking) */
1422 if (UVM_ET_ISNEEDSCOPY(entry)) {
1423 if (fudge)
1424 oldstart = entry->start;
1425 else
1426 oldstart = 0; /* XXX: gcc */
1427 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1428 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */
1429 error = ENOMEM;
1430 goto bad;
1431 }
1432 /* amap_copy could clip (during chunk)! update fudge */
1433 if (fudge) {
1434 fudge = fudge - (entry->start - oldstart);
1435 orig_fudge = fudge;
1436 }
1437 }
1438
1439 /* calculate the offset of this from "start" */
1440 oldoffset = (entry->start + fudge) - start;
1441
1442 /* allocate a new map entry */
1443 newentry = uvm_mapent_alloc(dstmap);
1444 if (newentry == NULL) {
1445 error = ENOMEM;
1446 goto bad;
1447 }
1448
1449 /* set up new map entry */
1450 newentry->next = NULL;
1451 newentry->prev = endchain;
1452 newentry->start = dstaddr + oldoffset;
1453 newentry->end =
1454 newentry->start + (entry->end - (entry->start + fudge));
1455 if (newentry->end > newend || newentry->end < newentry->start)
1456 newentry->end = newend;
1457 newentry->object.uvm_obj = entry->object.uvm_obj;
1458 if (newentry->object.uvm_obj) {
1459 if (newentry->object.uvm_obj->pgops->pgo_reference)
1460 newentry->object.uvm_obj->pgops->
1461 pgo_reference(newentry->object.uvm_obj);
1462 newentry->offset = entry->offset + fudge;
1463 } else {
1464 newentry->offset = 0;
1465 }
1466 newentry->etype = entry->etype;
1467 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1468 entry->max_protection : entry->protection;
1469 newentry->max_protection = entry->max_protection;
1470 newentry->inheritance = entry->inheritance;
1471 newentry->wired_count = 0;
1472 newentry->aref.ar_amap = entry->aref.ar_amap;
1473 if (newentry->aref.ar_amap) {
1474 newentry->aref.ar_pageoff =
1475 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1476 amap_ref(newentry, AMAP_SHARED |
1477 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1478 } else {
1479 newentry->aref.ar_pageoff = 0;
1480 }
1481 newentry->advice = entry->advice;
1482
1483 /* now link it on the chain */
1484 nchain++;
1485 if (endchain == NULL) {
1486 chain = endchain = newentry;
1487 } else {
1488 endchain->next = newentry;
1489 endchain = newentry;
1490 }
1491
1492 /* end of 'while' loop! */
1493 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1494 (entry->next == &srcmap->header ||
1495 entry->next->start != entry->end)) {
1496 error = EINVAL;
1497 goto bad;
1498 }
1499 entry = entry->next;
1500 fudge = 0;
1501 }
1502
1503
1504 /*
1505 * step 4: close off chain (in format expected by uvm_map_replace)
1506 */
1507
1508 if (chain)
1509 chain->prev = endchain;
1510
1511
1512 /*
1513 * step 5: attempt to lock the dest map so we can pmap_copy.
1514 * note usage of copy_ok:
1515 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1516 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1517 */
1518
1519 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1520
1521 copy_ok = 1;
1522 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1523 nchain)) {
1524 if (srcmap != dstmap)
1525 vm_map_unlock(dstmap);
1526 error = EIO;
1527 goto bad;
1528 }
1529
1530 } else {
1531
1532 copy_ok = 0;
1533 /* replace defered until step 7 */
1534
1535 }
1536
1537
1538 /*
1539 * step 6: traverse the srcmap a second time to do the following:
1540 * - if we got a lock on the dstmap do pmap_copy
1541 * - if UVM_EXTRACT_REMOVE remove the entries
1542 * we make use of orig_entry and orig_fudge (saved in step 2)
1543 */
1544
1545 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1546
1547 /* purge possible stale hints from srcmap */
1548 if (flags & UVM_EXTRACT_REMOVE) {
1549 SAVE_HINT(srcmap, orig_entry->prev);
1550 if (srcmap->first_free->start >= start)
1551 srcmap->first_free = orig_entry->prev;
1552 }
1553
1554 entry = orig_entry;
1555 fudge = orig_fudge;
1556 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */
1557
1558 while (entry->start < end && entry != &srcmap->header) {
1559
1560 if (copy_ok) {
1561 oldoffset = (entry->start + fudge) - start;
1562 elen = min(end, entry->end) -
1563 (entry->start + fudge);
1564 pmap_copy(dstmap->pmap, srcmap->pmap,
1565 dstaddr + oldoffset, elen,
1566 entry->start + fudge);
1567 }
1568
1569 /* we advance "entry" in the following if statement */
1570 if (flags & UVM_EXTRACT_REMOVE) {
1571 pmap_remove(srcmap->pmap, entry->start,
1572 entry->end);
1573 oldentry = entry; /* save entry */
1574 entry = entry->next; /* advance */
1575 uvm_map_entry_unlink(srcmap, oldentry);
1576 /* add to dead list */
1577 oldentry->next = deadentry;
1578 deadentry = oldentry;
1579 } else {
1580 entry = entry->next; /* advance */
1581 }
1582
1583 /* end of 'while' loop */
1584 fudge = 0;
1585 }
1586
1587 /*
1588 * unlock dstmap. we will dispose of deadentry in
1589 * step 7 if needed
1590 */
1591 if (copy_ok && srcmap != dstmap)
1592 vm_map_unlock(dstmap);
1593
1594 }
1595 else
1596 deadentry = NULL; /* XXX: gcc */
1597
1598 /*
1599 * step 7: we are done with the source map, unlock. if copy_ok
1600 * is 0 then we have not replaced the dummy mapping in dstmap yet
1601 * and we need to do so now.
1602 */
1603
1604 vm_map_unlock(srcmap);
1605 if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1606 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */
1607
1608 /* now do the replacement if we didn't do it in step 5 */
1609 if (copy_ok == 0) {
1610 vm_map_lock(dstmap);
1611 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1612 nchain);
1613 vm_map_unlock(dstmap);
1614
1615 if (error == FALSE) {
1616 error = EIO;
1617 goto bad2;
1618 }
1619 }
1620
1621 /*
1622 * done!
1623 */
1624 return(0);
1625
1626 /*
1627 * bad: failure recovery
1628 */
1629 bad:
1630 vm_map_unlock(srcmap);
1631 bad2: /* src already unlocked */
1632 if (chain)
1633 uvm_unmap_detach(chain,
1634 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1635 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */
1636 return(error);
1637 }
1638
1639 /* end of extraction functions */
1640
1641 /*
1642 * uvm_map_submap: punch down part of a map into a submap
1643 *
1644 * => only the kernel_map is allowed to be submapped
1645 * => the purpose of submapping is to break up the locking granularity
1646 * of a larger map
1647 * => the range specified must have been mapped previously with a uvm_map()
1648 * call [with uobj==NULL] to create a blank map entry in the main map.
1649 * [And it had better still be blank!]
1650 * => maps which contain submaps should never be copied or forked.
1651 * => to remove a submap, use uvm_unmap() on the main map
1652 * and then uvm_map_deallocate() the submap.
1653 * => main map must be unlocked.
1654 * => submap must have been init'd and have a zero reference count.
1655 * [need not be locked as we don't actually reference it]
1656 */
1657
1658 int
1659 uvm_map_submap(map, start, end, submap)
1660 vm_map_t map, submap;
1661 vaddr_t start, end;
1662 {
1663 vm_map_entry_t entry;
1664 int result;
1665 UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist);
1666
1667 vm_map_lock(map);
1668
1669 VM_MAP_RANGE_CHECK(map, start, end);
1670
1671 if (uvm_map_lookup_entry(map, start, &entry)) {
1672 UVM_MAP_CLIP_START(map, entry, start);
1673 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */
1674 }
1675 else {
1676 entry = NULL;
1677 }
1678
1679 if (entry != NULL &&
1680 entry->start == start && entry->end == end &&
1681 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1682 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1683
1684 /*
1685 * doit!
1686 */
1687 entry->etype |= UVM_ET_SUBMAP;
1688 entry->object.sub_map = submap;
1689 entry->offset = 0;
1690 uvm_map_reference(submap);
1691 result = KERN_SUCCESS;
1692 } else {
1693 result = KERN_INVALID_ARGUMENT;
1694 }
1695 vm_map_unlock(map);
1696
1697 return(result);
1698 }
1699
1700
1701 /*
1702 * uvm_map_protect: change map protection
1703 *
1704 * => set_max means set max_protection.
1705 * => map must be unlocked.
1706 */
1707
1708 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \
1709 ~VM_PROT_WRITE : VM_PROT_ALL)
1710 #define max(a,b) ((a) > (b) ? (a) : (b))
1711
1712 int
1713 uvm_map_protect(map, start, end, new_prot, set_max)
1714 vm_map_t map;
1715 vaddr_t start, end;
1716 vm_prot_t new_prot;
1717 boolean_t set_max;
1718 {
1719 vm_map_entry_t current, entry;
1720 int rv = KERN_SUCCESS;
1721 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1722 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1723 map, start, end, new_prot);
1724
1725 vm_map_lock(map);
1726
1727 VM_MAP_RANGE_CHECK(map, start, end);
1728
1729 if (uvm_map_lookup_entry(map, start, &entry)) {
1730 UVM_MAP_CLIP_START(map, entry, start);
1731 } else {
1732 entry = entry->next;
1733 }
1734
1735 /*
1736 * make a first pass to check for protection violations.
1737 */
1738
1739 current = entry;
1740 while ((current != &map->header) && (current->start < end)) {
1741 if (UVM_ET_ISSUBMAP(current)) {
1742 rv = KERN_INVALID_ARGUMENT;
1743 goto out;
1744 }
1745 if ((new_prot & current->max_protection) != new_prot) {
1746 rv = KERN_PROTECTION_FAILURE;
1747 goto out;
1748 }
1749 current = current->next;
1750 }
1751
1752 /* go back and fix up protections (no need to clip this time). */
1753
1754 current = entry;
1755
1756 while ((current != &map->header) && (current->start < end)) {
1757 vm_prot_t old_prot;
1758
1759 UVM_MAP_CLIP_END(map, current, end);
1760
1761 old_prot = current->protection;
1762 if (set_max)
1763 current->protection =
1764 (current->max_protection = new_prot) & old_prot;
1765 else
1766 current->protection = new_prot;
1767
1768 /*
1769 * update physical map if necessary. worry about copy-on-write
1770 * here -- CHECK THIS XXX
1771 */
1772
1773 if (current->protection != old_prot) {
1774 /* update pmap! */
1775 pmap_protect(map->pmap, current->start, current->end,
1776 current->protection & MASK(entry));
1777 }
1778
1779 /*
1780 * If the map is configured to lock any future mappings,
1781 * wire this entry now if the old protection was VM_PROT_NONE
1782 * and the new protection is not VM_PROT_NONE.
1783 */
1784
1785 if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
1786 VM_MAPENT_ISWIRED(entry) == 0 &&
1787 old_prot == VM_PROT_NONE &&
1788 new_prot != VM_PROT_NONE) {
1789 if (uvm_map_pageable(map, entry->start,
1790 entry->end, FALSE,
1791 UVM_LK_ENTER|UVM_LK_EXIT) != KERN_SUCCESS) {
1792 /*
1793 * If locking the entry fails, remember the
1794 * error if it's the first one. Note we
1795 * still continue setting the protection in
1796 * the map, but will return the resource
1797 * shortage condition regardless.
1798 *
1799 * XXX Ignore what the actual error is,
1800 * XXX just call it a resource shortage
1801 * XXX so that it doesn't get confused
1802 * XXX what uvm_map_protect() itself would
1803 * XXX normally return.
1804 */
1805 rv = KERN_RESOURCE_SHORTAGE;
1806 }
1807 }
1808
1809 current = current->next;
1810 }
1811
1812 out:
1813 vm_map_unlock(map);
1814 UVMHIST_LOG(maphist, "<- done, rv=%d",rv,0,0,0);
1815 return (rv);
1816 }
1817
1818 #undef max
1819 #undef MASK
1820
1821 /*
1822 * uvm_map_inherit: set inheritance code for range of addrs in map.
1823 *
1824 * => map must be unlocked
1825 * => note that the inherit code is used during a "fork". see fork
1826 * code for details.
1827 */
1828
1829 int
1830 uvm_map_inherit(map, start, end, new_inheritance)
1831 vm_map_t map;
1832 vaddr_t start;
1833 vaddr_t end;
1834 vm_inherit_t new_inheritance;
1835 {
1836 vm_map_entry_t entry, temp_entry;
1837 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1838 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1839 map, start, end, new_inheritance);
1840
1841 switch (new_inheritance) {
1842 case VM_INHERIT_NONE:
1843 case VM_INHERIT_COPY:
1844 case VM_INHERIT_SHARE:
1845 break;
1846 default:
1847 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1848 return (KERN_INVALID_ARGUMENT);
1849 }
1850
1851 vm_map_lock(map);
1852
1853 VM_MAP_RANGE_CHECK(map, start, end);
1854
1855 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1856 entry = temp_entry;
1857 UVM_MAP_CLIP_START(map, entry, start);
1858 } else {
1859 entry = temp_entry->next;
1860 }
1861
1862 while ((entry != &map->header) && (entry->start < end)) {
1863 UVM_MAP_CLIP_END(map, entry, end);
1864
1865 entry->inheritance = new_inheritance;
1866
1867 entry = entry->next;
1868 }
1869
1870 vm_map_unlock(map);
1871 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1872 return(KERN_SUCCESS);
1873 }
1874
1875 /*
1876 * uvm_map_advice: set advice code for range of addrs in map.
1877 *
1878 * => map must be unlocked
1879 */
1880
1881 int
1882 uvm_map_advice(map, start, end, new_advice)
1883 vm_map_t map;
1884 vaddr_t start;
1885 vaddr_t end;
1886 int new_advice;
1887 {
1888 vm_map_entry_t entry, temp_entry;
1889 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
1890 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
1891 map, start, end, new_advice);
1892
1893 vm_map_lock(map);
1894
1895 VM_MAP_RANGE_CHECK(map, start, end);
1896
1897 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1898 entry = temp_entry;
1899 UVM_MAP_CLIP_START(map, entry, start);
1900 } else {
1901 entry = temp_entry->next;
1902 }
1903
1904 /*
1905 * XXXJRT: disallow holes?
1906 */
1907
1908 while ((entry != &map->header) && (entry->start < end)) {
1909 UVM_MAP_CLIP_END(map, entry, end);
1910
1911 switch (new_advice) {
1912 case MADV_NORMAL:
1913 case MADV_RANDOM:
1914 case MADV_SEQUENTIAL:
1915 /* nothing special here */
1916 break;
1917
1918 default:
1919 vm_map_unlock(map);
1920 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1921 return (KERN_INVALID_ARGUMENT);
1922 }
1923
1924
1925 entry->advice = new_advice;
1926
1927 entry = entry->next;
1928 }
1929
1930 vm_map_unlock(map);
1931 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1932 return (KERN_SUCCESS);
1933 }
1934
1935 /*
1936 * uvm_map_pageable: sets the pageability of a range in a map.
1937 *
1938 * => wires map entries. should not be used for transient page locking.
1939 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
1940 * => regions sepcified as not pageable require lock-down (wired) memory
1941 * and page tables.
1942 * => map must never be read-locked
1943 * => if islocked is TRUE, map is already write-locked
1944 * => we always unlock the map, since we must downgrade to a read-lock
1945 * to call uvm_fault_wire()
1946 * => XXXCDC: check this and try and clean it up.
1947 */
1948
1949 int
1950 uvm_map_pageable(map, start, end, new_pageable, lockflags)
1951 vm_map_t map;
1952 vaddr_t start, end;
1953 boolean_t new_pageable;
1954 int lockflags;
1955 {
1956 vm_map_entry_t entry, start_entry, failed_entry;
1957 int rv;
1958 #ifdef DIAGNOSTIC
1959 u_int timestamp_save;
1960 #endif
1961 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
1962 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1963 map, start, end, new_pageable);
1964
1965 #ifdef DIAGNOSTIC
1966 if ((map->flags & VM_MAP_PAGEABLE) == 0)
1967 panic("uvm_map_pageable: map %p not pageable", map);
1968 #endif
1969
1970 if ((lockflags & UVM_LK_ENTER) == 0)
1971 vm_map_lock(map);
1972
1973 VM_MAP_RANGE_CHECK(map, start, end);
1974
1975 /*
1976 * only one pageability change may take place at one time, since
1977 * uvm_fault_wire assumes it will be called only once for each
1978 * wiring/unwiring. therefore, we have to make sure we're actually
1979 * changing the pageability for the entire region. we do so before
1980 * making any changes.
1981 */
1982
1983 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1984 if ((lockflags & UVM_LK_EXIT) == 0)
1985 vm_map_unlock(map);
1986
1987 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1988 return (KERN_INVALID_ADDRESS);
1989 }
1990 entry = start_entry;
1991
1992 /*
1993 * handle wiring and unwiring seperately.
1994 */
1995
1996 if (new_pageable) { /* unwire */
1997 UVM_MAP_CLIP_START(map, entry, start);
1998 /*
1999 * unwiring. first ensure that the range to be unwired is
2000 * really wired down and that there are no holes.
2001 */
2002 while ((entry != &map->header) && (entry->start < end)) {
2003 if (entry->wired_count == 0 ||
2004 (entry->end < end &&
2005 (entry->next == &map->header ||
2006 entry->next->start > entry->end))) {
2007 if ((lockflags & UVM_LK_EXIT) == 0)
2008 vm_map_unlock(map);
2009 UVMHIST_LOG(maphist,
2010 "<- done (INVALID UNWIRE ARG)",0,0,0,0);
2011 return (KERN_INVALID_ARGUMENT);
2012 }
2013 entry = entry->next;
2014 }
2015
2016 /*
2017 * POSIX 1003.1b - a single munlock call unlocks a region,
2018 * regardless of the number of mlock calls made on that
2019 * region.
2020 */
2021 entry = start_entry;
2022 while ((entry != &map->header) && (entry->start < end)) {
2023 UVM_MAP_CLIP_END(map, entry, end);
2024 if (VM_MAPENT_ISWIRED(entry))
2025 uvm_map_entry_unwire(map, entry);
2026 entry = entry->next;
2027 }
2028 if ((lockflags & UVM_LK_EXIT) == 0)
2029 vm_map_unlock(map);
2030 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2031 return(KERN_SUCCESS);
2032
2033 /*
2034 * end of unwire case!
2035 */
2036 }
2037
2038 /*
2039 * wire case: in two passes [XXXCDC: ugly block of code here]
2040 *
2041 * 1: holding the write lock, we create any anonymous maps that need
2042 * to be created. then we clip each map entry to the region to
2043 * be wired and increment its wiring count.
2044 *
2045 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2046 * in the pages for any newly wired area (wired_count == 1).
2047 *
2048 * downgrading to a read lock for uvm_fault_wire avoids a possible
2049 * deadlock with another thread that may have faulted on one of
2050 * the pages to be wired (it would mark the page busy, blocking
2051 * us, then in turn block on the map lock that we hold). because
2052 * of problems in the recursive lock package, we cannot upgrade
2053 * to a write lock in vm_map_lookup. thus, any actions that
2054 * require the write lock must be done beforehand. because we
2055 * keep the read lock on the map, the copy-on-write status of the
2056 * entries we modify here cannot change.
2057 */
2058
2059 while ((entry != &map->header) && (entry->start < end)) {
2060 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2061 /*
2062 * perform actions of vm_map_lookup that need the
2063 * write lock on the map: create an anonymous map
2064 * for a copy-on-write region, or an anonymous map
2065 * for a zero-fill region. (XXXCDC: submap case
2066 * ok?)
2067 */
2068 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2069 if (UVM_ET_ISNEEDSCOPY(entry) &&
2070 ((entry->protection & VM_PROT_WRITE) ||
2071 (entry->object.uvm_obj == NULL))) {
2072 amap_copy(map, entry, M_WAITOK, TRUE,
2073 start, end);
2074 /* XXXCDC: wait OK? */
2075 }
2076 }
2077 }
2078 UVM_MAP_CLIP_START(map, entry, start);
2079 UVM_MAP_CLIP_END(map, entry, end);
2080 entry->wired_count++;
2081
2082 /*
2083 * Check for holes
2084 */
2085 if (entry->protection == VM_PROT_NONE ||
2086 (entry->end < end &&
2087 (entry->next == &map->header ||
2088 entry->next->start > entry->end))) {
2089 /*
2090 * found one. amap creation actions do not need to
2091 * be undone, but the wired counts need to be restored.
2092 */
2093 while (entry != &map->header && entry->end > start) {
2094 entry->wired_count--;
2095 entry = entry->prev;
2096 }
2097 if ((lockflags & UVM_LK_EXIT) == 0)
2098 vm_map_unlock(map);
2099 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2100 return (KERN_INVALID_ARGUMENT);
2101 }
2102 entry = entry->next;
2103 }
2104
2105 /*
2106 * Pass 2.
2107 */
2108
2109 #ifdef DIAGNOSTIC
2110 timestamp_save = map->timestamp;
2111 #endif
2112 vm_map_busy(map);
2113 vm_map_downgrade(map);
2114
2115 rv = 0;
2116 entry = start_entry;
2117 while (entry != &map->header && entry->start < end) {
2118 if (entry->wired_count == 1) {
2119 rv = uvm_fault_wire(map, entry->start, entry->end,
2120 entry->protection);
2121 if (rv) {
2122 /*
2123 * wiring failed. break out of the loop.
2124 * we'll clean up the map below, once we
2125 * have a write lock again.
2126 */
2127 break;
2128 }
2129 }
2130 entry = entry->next;
2131 }
2132
2133 if (rv) { /* failed? */
2134 /*
2135 * Get back to an exclusive (write) lock.
2136 */
2137 vm_map_upgrade(map);
2138 vm_map_unbusy(map);
2139
2140 #ifdef DIAGNOSTIC
2141 if (timestamp_save != map->timestamp)
2142 panic("uvm_map_pageable: stale map");
2143 #endif
2144
2145 /*
2146 * first drop the wiring count on all the entries
2147 * which haven't actually been wired yet.
2148 */
2149 failed_entry = entry;
2150 while (entry != &map->header && entry->start < end) {
2151 entry->wired_count--;
2152 entry = entry->next;
2153 }
2154
2155 /*
2156 * now, unwire all the entries that were successfully
2157 * wired above.
2158 */
2159 entry = start_entry;
2160 while (entry != failed_entry) {
2161 entry->wired_count--;
2162 if (VM_MAPENT_ISWIRED(entry) == 0)
2163 uvm_map_entry_unwire(map, entry);
2164 entry = entry->next;
2165 }
2166 if ((lockflags & UVM_LK_EXIT) == 0)
2167 vm_map_unlock(map);
2168 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2169 return(rv);
2170 }
2171
2172 /* We are holding a read lock here. */
2173 if ((lockflags & UVM_LK_EXIT) == 0) {
2174 vm_map_unbusy(map);
2175 vm_map_unlock_read(map);
2176 } else {
2177 /*
2178 * Get back to an exclusive (write) lock.
2179 */
2180 vm_map_upgrade(map);
2181 vm_map_unbusy(map);
2182 }
2183
2184 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2185 return(KERN_SUCCESS);
2186 }
2187
2188 /*
2189 * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2190 * all mapped regions.
2191 *
2192 * => map must not be locked.
2193 * => if no flags are specified, all regions are unwired.
2194 * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2195 */
2196
2197 int
2198 uvm_map_pageable_all(map, flags, limit)
2199 vm_map_t map;
2200 int flags;
2201 vsize_t limit;
2202 {
2203 vm_map_entry_t entry, failed_entry;
2204 vsize_t size;
2205 int rv;
2206 #ifdef DIAGNOSTIC
2207 u_int timestamp_save;
2208 #endif
2209 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2210 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2211
2212 #ifdef DIAGNOSTIC
2213 if ((map->flags & VM_MAP_PAGEABLE) == 0)
2214 panic("uvm_map_pageable_all: map %p not pageable", map);
2215 #endif
2216
2217 vm_map_lock(map);
2218
2219 /*
2220 * handle wiring and unwiring separately.
2221 */
2222
2223 if (flags == 0) { /* unwire */
2224 /*
2225 * POSIX 1003.1b -- munlockall unlocks all regions,
2226 * regardless of how many times mlockall has been called.
2227 */
2228 for (entry = map->header.next; entry != &map->header;
2229 entry = entry->next) {
2230 if (VM_MAPENT_ISWIRED(entry))
2231 uvm_map_entry_unwire(map, entry);
2232 }
2233 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2234 vm_map_unlock(map);
2235 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2236 return (KERN_SUCCESS);
2237
2238 /*
2239 * end of unwire case!
2240 */
2241 }
2242
2243 if (flags & MCL_FUTURE) {
2244 /*
2245 * must wire all future mappings; remember this.
2246 */
2247 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2248 }
2249
2250 if ((flags & MCL_CURRENT) == 0) {
2251 /*
2252 * no more work to do!
2253 */
2254 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2255 vm_map_unlock(map);
2256 return (KERN_SUCCESS);
2257 }
2258
2259 /*
2260 * wire case: in three passes [XXXCDC: ugly block of code here]
2261 *
2262 * 1: holding the write lock, count all pages mapped by non-wired
2263 * entries. if this would cause us to go over our limit, we fail.
2264 *
2265 * 2: still holding the write lock, we create any anonymous maps that
2266 * need to be created. then we increment its wiring count.
2267 *
2268 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2269 * in the pages for any newly wired area (wired_count == 1).
2270 *
2271 * downgrading to a read lock for uvm_fault_wire avoids a possible
2272 * deadlock with another thread that may have faulted on one of
2273 * the pages to be wired (it would mark the page busy, blocking
2274 * us, then in turn block on the map lock that we hold). because
2275 * of problems in the recursive lock package, we cannot upgrade
2276 * to a write lock in vm_map_lookup. thus, any actions that
2277 * require the write lock must be done beforehand. because we
2278 * keep the read lock on the map, the copy-on-write status of the
2279 * entries we modify here cannot change.
2280 */
2281
2282 for (size = 0, entry = map->header.next; entry != &map->header;
2283 entry = entry->next) {
2284 if (entry->protection != VM_PROT_NONE &&
2285 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2286 size += entry->end - entry->start;
2287 }
2288 }
2289
2290 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2291 vm_map_unlock(map);
2292 return (KERN_NO_SPACE); /* XXX overloaded */
2293 }
2294
2295 /* XXX non-pmap_wired_count case must be handled by caller */
2296 #ifdef pmap_wired_count
2297 if (limit != 0 &&
2298 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2299 vm_map_unlock(map);
2300 return (KERN_NO_SPACE); /* XXX overloaded */
2301 }
2302 #endif
2303
2304 /*
2305 * Pass 2.
2306 */
2307
2308 for (entry = map->header.next; entry != &map->header;
2309 entry = entry->next) {
2310 if (entry->protection == VM_PROT_NONE)
2311 continue;
2312 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2313 /*
2314 * perform actions of vm_map_lookup that need the
2315 * write lock on the map: create an anonymous map
2316 * for a copy-on-write region, or an anonymous map
2317 * for a zero-fill region. (XXXCDC: submap case
2318 * ok?)
2319 */
2320 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2321 if (UVM_ET_ISNEEDSCOPY(entry) &&
2322 ((entry->protection & VM_PROT_WRITE) ||
2323 (entry->object.uvm_obj == NULL))) {
2324 amap_copy(map, entry, M_WAITOK, TRUE,
2325 entry->start, entry->end);
2326 /* XXXCDC: wait OK? */
2327 }
2328 }
2329 }
2330 entry->wired_count++;
2331 }
2332
2333 /*
2334 * Pass 3.
2335 */
2336
2337 #ifdef DIAGNOSTIC
2338 timestamp_save = map->timestamp;
2339 #endif
2340 vm_map_busy(map);
2341 vm_map_downgrade(map);
2342
2343 rv = KERN_SUCCESS;
2344 for (entry = map->header.next; entry != &map->header;
2345 entry = entry->next) {
2346 if (entry->wired_count == 1) {
2347 rv = uvm_fault_wire(map, entry->start, entry->end,
2348 entry->protection);
2349 if (rv) {
2350 /*
2351 * wiring failed. break out of the loop.
2352 * we'll clean up the map below, once we
2353 * have a write lock again.
2354 */
2355 break;
2356 }
2357 }
2358 }
2359
2360 if (rv) { /* failed? */
2361 /*
2362 * Get back an exclusive (write) lock.
2363 */
2364 vm_map_upgrade(map);
2365 vm_map_unbusy(map);
2366
2367 #ifdef DIAGNOSTIC
2368 if (timestamp_save != map->timestamp)
2369 panic("uvm_map_pageable_all: stale map");
2370 #endif
2371
2372 /*
2373 * first drop the wiring count on all the entries
2374 * which haven't actually been wired yet.
2375 *
2376 * Skip VM_PROT_NONE entries like we did above.
2377 */
2378 failed_entry = entry;
2379 for (/* nothing */; entry != &map->header;
2380 entry = entry->next) {
2381 if (entry->protection == VM_PROT_NONE)
2382 continue;
2383 entry->wired_count--;
2384 }
2385
2386 /*
2387 * now, unwire all the entries that were successfully
2388 * wired above.
2389 *
2390 * Skip VM_PROT_NONE entries like we did above.
2391 */
2392 for (entry = map->header.next; entry != failed_entry;
2393 entry = entry->next) {
2394 if (entry->protection == VM_PROT_NONE)
2395 continue;
2396 entry->wired_count--;
2397 if (VM_MAPENT_ISWIRED(entry))
2398 uvm_map_entry_unwire(map, entry);
2399 }
2400 vm_map_unlock(map);
2401 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2402 return (rv);
2403 }
2404
2405 /* We are holding a read lock here. */
2406 vm_map_unbusy(map);
2407 vm_map_unlock_read(map);
2408
2409 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2410 return (KERN_SUCCESS);
2411 }
2412
2413 /*
2414 * uvm_map_clean: clean out a map range
2415 *
2416 * => valid flags:
2417 * if (flags & PGO_CLEANIT): dirty pages are cleaned first
2418 * if (flags & PGO_SYNCIO): dirty pages are written synchronously
2419 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2420 * if (flags & PGO_FREE): any cached pages are freed after clean
2421 * => returns an error if any part of the specified range isn't mapped
2422 * => never a need to flush amap layer since the anonymous memory has
2423 * no permanent home, but may deactivate pages there
2424 * => called from sys_msync() and sys_madvise()
2425 * => caller must not write-lock map (read OK).
2426 * => we may sleep while cleaning if SYNCIO [with map read-locked]
2427 */
2428
2429 int amap_clean_works = 1; /* XXX for now, just in case... */
2430
2431 int
2432 uvm_map_clean(map, start, end, flags)
2433 vm_map_t map;
2434 vaddr_t start, end;
2435 int flags;
2436 {
2437 vm_map_entry_t current, entry;
2438 struct uvm_object *uobj;
2439 struct vm_amap *amap;
2440 struct vm_anon *anon;
2441 struct vm_page *pg;
2442 vaddr_t offset;
2443 vsize_t size;
2444 int rv, error, refs;
2445 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2446 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2447 map, start, end, flags);
2448
2449 #ifdef DIAGNOSTIC
2450 if ((flags & (PGO_FREE|PGO_DEACTIVATE)) == (PGO_FREE|PGO_DEACTIVATE))
2451 panic("uvm_map_clean: FREE and DEACTIVATE");
2452 #endif
2453
2454 vm_map_lock_read(map);
2455 VM_MAP_RANGE_CHECK(map, start, end);
2456 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2457 vm_map_unlock_read(map);
2458 return(KERN_INVALID_ADDRESS);
2459 }
2460
2461 /*
2462 * Make a first pass to check for holes.
2463 */
2464 for (current = entry; current->start < end; current = current->next) {
2465 if (UVM_ET_ISSUBMAP(current)) {
2466 vm_map_unlock_read(map);
2467 return (KERN_INVALID_ARGUMENT);
2468 }
2469 if (end > current->end && (current->next == &map->header ||
2470 current->end != current->next->start)) {
2471 vm_map_unlock_read(map);
2472 return (KERN_INVALID_ADDRESS);
2473 }
2474 }
2475
2476 error = KERN_SUCCESS;
2477
2478 for (current = entry; current->start < end; current = current->next) {
2479 amap = current->aref.ar_amap; /* top layer */
2480 uobj = current->object.uvm_obj; /* bottom layer */
2481
2482 #ifdef DIAGNOSTIC
2483 if (start < current->start)
2484 panic("uvm_map_clean: hole");
2485 #endif
2486
2487 /*
2488 * No amap cleaning necessary if:
2489 *
2490 * (1) There's no amap.
2491 *
2492 * (2) We're not deactivating or freeing pages.
2493 */
2494 if (amap == NULL ||
2495 (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2496 goto flush_object;
2497
2498 /* XXX for now, just in case... */
2499 if (amap_clean_works == 0)
2500 goto flush_object;
2501
2502 amap_lock(amap);
2503
2504 offset = start - current->start;
2505 size = (end <= current->end ? end : current->end) -
2506 start;
2507
2508 for (/* nothing */; size != 0; size -= PAGE_SIZE,
2509 offset += PAGE_SIZE) {
2510 anon = amap_lookup(¤t->aref, offset);
2511 if (anon == NULL)
2512 continue;
2513
2514 simple_lock(&anon->an_lock);
2515
2516 pg = anon->u.an_page;
2517 if (pg == NULL) {
2518 simple_unlock(&anon->an_lock);
2519 continue;
2520 }
2521
2522 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2523 /*
2524 * XXX In these first 3 cases, we always just
2525 * XXX deactivate the page. We may want to
2526 * XXX handle the different cases more
2527 * XXX specifically, in the future.
2528 */
2529 case PGO_CLEANIT|PGO_FREE:
2530 case PGO_CLEANIT|PGO_DEACTIVATE:
2531 case PGO_DEACTIVATE:
2532 deactivate_it:
2533 /* skip the page if it's loaned or wired */
2534 if (pg->loan_count != 0 ||
2535 pg->wire_count != 0) {
2536 simple_unlock(&anon->an_lock);
2537 continue;
2538 }
2539
2540 uvm_lock_pageq();
2541
2542 /*
2543 * skip the page if it's not actually owned
2544 * by the anon (may simply be loaned to the
2545 * anon).
2546 */
2547 if ((pg->pqflags & PQ_ANON) == 0) {
2548 #ifdef DIAGNOSTIC
2549 if (pg->uobject != NULL)
2550 panic("uvm_map_clean: "
2551 "page anon vs. object "
2552 "inconsistency");
2553 #endif
2554 uvm_unlock_pageq();
2555 simple_unlock(&anon->an_lock);
2556 continue;
2557 }
2558
2559 #ifdef DIAGNOSTIC
2560 if (pg->uanon != anon)
2561 panic("uvm_map_clean: anon "
2562 "inconsistency");
2563 #endif
2564
2565 /* zap all mappings for the page. */
2566 pmap_page_protect(pg, VM_PROT_NONE);
2567
2568 /* ...and deactivate the page. */
2569 uvm_pagedeactivate(pg);
2570
2571 uvm_unlock_pageq();
2572 simple_unlock(&anon->an_lock);
2573 continue;
2574
2575 case PGO_FREE:
2576 /*
2577 * If there are multiple references to
2578 * the amap, just deactivate the page.
2579 */
2580 if (amap_refs(amap) > 1)
2581 goto deactivate_it;
2582
2583 /* XXX skip the page if it's wired */
2584 if (pg->wire_count != 0) {
2585 simple_unlock(&anon->an_lock);
2586 continue;
2587 }
2588 amap_unadd(¤t->aref, offset);
2589 refs = --anon->an_ref;
2590 simple_unlock(&anon->an_lock);
2591 if (refs == 0)
2592 uvm_anfree(anon);
2593 continue;
2594
2595 default:
2596 panic("uvm_map_clean: wierd flags");
2597 }
2598 #ifdef DIAGNOSTIC
2599 panic("uvm_map_clean: unreachable code");
2600 #endif
2601 }
2602
2603 amap_unlock(amap);
2604
2605 flush_object:
2606 /*
2607 * flush pages if we've got a valid backing object.
2608 */
2609
2610 offset = current->offset + (start - current->start);
2611 size = (end <= current->end ? end : current->end) - start;
2612
2613 if (uobj != NULL) {
2614 simple_lock(&uobj->vmobjlock);
2615 rv = uobj->pgops->pgo_flush(uobj, offset,
2616 offset + size, flags);
2617 simple_unlock(&uobj->vmobjlock);
2618
2619 if (rv == FALSE)
2620 error = KERN_FAILURE;
2621 }
2622 start += size;
2623 }
2624
2625 vm_map_unlock_read(map);
2626 return (error);
2627 }
2628
2629
2630 /*
2631 * uvm_map_checkprot: check protection in map
2632 *
2633 * => must allow specified protection in a fully allocated region.
2634 * => map must be read or write locked by caller.
2635 */
2636
2637 boolean_t
2638 uvm_map_checkprot(map, start, end, protection)
2639 vm_map_t map;
2640 vaddr_t start, end;
2641 vm_prot_t protection;
2642 {
2643 vm_map_entry_t entry;
2644 vm_map_entry_t tmp_entry;
2645
2646 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2647 return(FALSE);
2648 }
2649
2650 entry = tmp_entry;
2651
2652 while (start < end) {
2653 if (entry == &map->header) {
2654 return(FALSE);
2655 }
2656
2657 /*
2658 * no holes allowed
2659 */
2660
2661 if (start < entry->start) {
2662 return(FALSE);
2663 }
2664
2665 /*
2666 * check protection associated with entry
2667 */
2668
2669 if ((entry->protection & protection) != protection) {
2670 return(FALSE);
2671 }
2672
2673 /* go to next entry */
2674
2675 start = entry->end;
2676 entry = entry->next;
2677 }
2678 return(TRUE);
2679 }
2680
2681 /*
2682 * uvmspace_alloc: allocate a vmspace structure.
2683 *
2684 * - structure includes vm_map and pmap
2685 * - XXX: no locking on this structure
2686 * - refcnt set to 1, rest must be init'd by caller
2687 */
2688 struct vmspace *
2689 uvmspace_alloc(min, max, pageable)
2690 vaddr_t min, max;
2691 int pageable;
2692 {
2693 struct vmspace *vm;
2694 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2695
2696 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2697 uvmspace_init(vm, NULL, min, max, pageable);
2698 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2699 return (vm);
2700 }
2701
2702 /*
2703 * uvmspace_init: initialize a vmspace structure.
2704 *
2705 * - XXX: no locking on this structure
2706 * - refcnt set to 1, rest must me init'd by caller
2707 */
2708 void
2709 uvmspace_init(vm, pmap, min, max, pageable)
2710 struct vmspace *vm;
2711 struct pmap *pmap;
2712 vaddr_t min, max;
2713 boolean_t pageable;
2714 {
2715 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2716
2717 memset(vm, 0, sizeof(*vm));
2718
2719 uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0);
2720
2721 if (pmap)
2722 pmap_reference(pmap);
2723 else
2724 pmap = pmap_create();
2725 vm->vm_map.pmap = pmap;
2726
2727 vm->vm_refcnt = 1;
2728 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2729 }
2730
2731 /*
2732 * uvmspace_share: share a vmspace between two proceses
2733 *
2734 * - XXX: no locking on vmspace
2735 * - used for vfork, threads(?)
2736 */
2737
2738 void
2739 uvmspace_share(p1, p2)
2740 struct proc *p1, *p2;
2741 {
2742 p2->p_vmspace = p1->p_vmspace;
2743 p1->p_vmspace->vm_refcnt++;
2744 }
2745
2746 /*
2747 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2748 *
2749 * - XXX: no locking on vmspace
2750 */
2751
2752 void
2753 uvmspace_unshare(p)
2754 struct proc *p;
2755 {
2756 struct vmspace *nvm, *ovm = p->p_vmspace;
2757
2758 if (ovm->vm_refcnt == 1)
2759 /* nothing to do: vmspace isn't shared in the first place */
2760 return;
2761
2762 /* make a new vmspace, still holding old one */
2763 nvm = uvmspace_fork(ovm);
2764
2765 pmap_deactivate(p); /* unbind old vmspace */
2766 p->p_vmspace = nvm;
2767 pmap_activate(p); /* switch to new vmspace */
2768
2769 uvmspace_free(ovm); /* drop reference to old vmspace */
2770 }
2771
2772 /*
2773 * uvmspace_exec: the process wants to exec a new program
2774 *
2775 * - XXX: no locking on vmspace
2776 */
2777
2778 void
2779 uvmspace_exec(p)
2780 struct proc *p;
2781 {
2782 struct vmspace *nvm, *ovm = p->p_vmspace;
2783 vm_map_t map = &ovm->vm_map;
2784
2785 #ifdef __sparc__
2786 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2787 kill_user_windows(p); /* before stack addresses go away */
2788 #endif
2789
2790 /*
2791 * see if more than one process is using this vmspace...
2792 */
2793
2794 if (ovm->vm_refcnt == 1) {
2795
2796 /*
2797 * if p is the only process using its vmspace then we can safely
2798 * recycle that vmspace for the program that is being exec'd.
2799 */
2800
2801 #ifdef SYSVSHM
2802 /*
2803 * SYSV SHM semantics require us to kill all segments on an exec
2804 */
2805 if (ovm->vm_shm)
2806 shmexit(ovm);
2807 #endif
2808
2809 /*
2810 * POSIX 1003.1b -- "lock future mappings" is revoked
2811 * when a process execs another program image.
2812 */
2813 vm_map_lock(map);
2814 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2815 vm_map_unlock(map);
2816
2817 /*
2818 * now unmap the old program
2819 */
2820 uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
2821
2822 } else {
2823
2824 /*
2825 * p's vmspace is being shared, so we can't reuse it for p since
2826 * it is still being used for others. allocate a new vmspace
2827 * for p
2828 */
2829 nvm = uvmspace_alloc(map->min_offset, map->max_offset,
2830 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2831
2832 /*
2833 * install new vmspace and drop our ref to the old one.
2834 */
2835
2836 pmap_deactivate(p);
2837 p->p_vmspace = nvm;
2838 pmap_activate(p);
2839
2840 uvmspace_free(ovm);
2841 }
2842 }
2843
2844 /*
2845 * uvmspace_free: free a vmspace data structure
2846 *
2847 * - XXX: no locking on vmspace
2848 */
2849
2850 void
2851 uvmspace_free(vm)
2852 struct vmspace *vm;
2853 {
2854 vm_map_entry_t dead_entries;
2855 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2856
2857 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2858 if (--vm->vm_refcnt == 0) {
2859 /*
2860 * lock the map, to wait out all other references to it. delete
2861 * all of the mappings and pages they hold, then call the pmap
2862 * module to reclaim anything left.
2863 */
2864 vm_map_lock(&vm->vm_map);
2865 if (vm->vm_map.nentries) {
2866 (void)uvm_unmap_remove(&vm->vm_map,
2867 vm->vm_map.min_offset, vm->vm_map.max_offset,
2868 &dead_entries);
2869 if (dead_entries != NULL)
2870 uvm_unmap_detach(dead_entries, 0);
2871 }
2872 pmap_destroy(vm->vm_map.pmap);
2873 vm->vm_map.pmap = NULL;
2874 pool_put(&uvm_vmspace_pool, vm);
2875 }
2876 UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2877 }
2878
2879 /*
2880 * F O R K - m a i n e n t r y p o i n t
2881 */
2882 /*
2883 * uvmspace_fork: fork a process' main map
2884 *
2885 * => create a new vmspace for child process from parent.
2886 * => parent's map must not be locked.
2887 */
2888
2889 struct vmspace *
2890 uvmspace_fork(vm1)
2891 struct vmspace *vm1;
2892 {
2893 struct vmspace *vm2;
2894 vm_map_t old_map = &vm1->vm_map;
2895 vm_map_t new_map;
2896 vm_map_entry_t old_entry;
2897 vm_map_entry_t new_entry;
2898 pmap_t new_pmap;
2899 boolean_t protect_child;
2900 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2901
2902 vm_map_lock(old_map);
2903
2904 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset,
2905 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2906 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2907 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2908 new_map = &vm2->vm_map; /* XXX */
2909 new_pmap = new_map->pmap;
2910
2911 old_entry = old_map->header.next;
2912
2913 /*
2914 * go entry-by-entry
2915 */
2916
2917 while (old_entry != &old_map->header) {
2918
2919 /*
2920 * first, some sanity checks on the old entry
2921 */
2922 if (UVM_ET_ISSUBMAP(old_entry))
2923 panic("fork: encountered a submap during fork (illegal)");
2924
2925 if (!UVM_ET_ISCOPYONWRITE(old_entry) &&
2926 UVM_ET_ISNEEDSCOPY(old_entry))
2927 panic("fork: non-copy_on_write map entry marked needs_copy (illegal)");
2928
2929
2930 switch (old_entry->inheritance) {
2931 case VM_INHERIT_NONE:
2932 /*
2933 * drop the mapping
2934 */
2935 break;
2936
2937 case VM_INHERIT_SHARE:
2938 /*
2939 * share the mapping: this means we want the old and
2940 * new entries to share amaps and backing objects.
2941 */
2942
2943 /*
2944 * if the old_entry needs a new amap (due to prev fork)
2945 * then we need to allocate it now so that we have
2946 * something we own to share with the new_entry. [in
2947 * other words, we need to clear needs_copy]
2948 */
2949
2950 if (UVM_ET_ISNEEDSCOPY(old_entry)) {
2951 /* get our own amap, clears needs_copy */
2952 amap_copy(old_map, old_entry, M_WAITOK, FALSE,
2953 0, 0);
2954 /* XXXCDC: WAITOK??? */
2955 }
2956
2957 new_entry = uvm_mapent_alloc(new_map);
2958 /* old_entry -> new_entry */
2959 uvm_mapent_copy(old_entry, new_entry);
2960
2961 /* new pmap has nothing wired in it */
2962 new_entry->wired_count = 0;
2963
2964 /*
2965 * gain reference to object backing the map (can't
2966 * be a submap, already checked this case).
2967 */
2968 if (new_entry->aref.ar_amap)
2969 /* share reference */
2970 amap_ref(new_entry, AMAP_SHARED);
2971
2972 if (new_entry->object.uvm_obj &&
2973 new_entry->object.uvm_obj->pgops->pgo_reference)
2974 new_entry->object.uvm_obj->
2975 pgops->pgo_reference(
2976 new_entry->object.uvm_obj);
2977
2978 /* insert entry at end of new_map's entry list */
2979 uvm_map_entry_link(new_map, new_map->header.prev,
2980 new_entry);
2981
2982 /*
2983 * pmap_copy the mappings: this routine is optional
2984 * but if it is there it will reduce the number of
2985 * page faults in the new proc.
2986 */
2987
2988 pmap_copy(new_pmap, old_map->pmap, new_entry->start,
2989 (old_entry->end - old_entry->start),
2990 old_entry->start);
2991
2992 break;
2993
2994 case VM_INHERIT_COPY:
2995
2996 /*
2997 * copy-on-write the mapping (using mmap's
2998 * MAP_PRIVATE semantics)
2999 *
3000 * allocate new_entry, adjust reference counts.
3001 * (note that new references are read-only).
3002 */
3003
3004 new_entry = uvm_mapent_alloc(new_map);
3005 /* old_entry -> new_entry */
3006 uvm_mapent_copy(old_entry, new_entry);
3007
3008 if (new_entry->aref.ar_amap)
3009 amap_ref(new_entry, 0);
3010
3011 if (new_entry->object.uvm_obj &&
3012 new_entry->object.uvm_obj->pgops->pgo_reference)
3013 new_entry->object.uvm_obj->pgops->pgo_reference
3014 (new_entry->object.uvm_obj);
3015
3016 /* new pmap has nothing wired in it */
3017 new_entry->wired_count = 0;
3018
3019 new_entry->etype |=
3020 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3021 uvm_map_entry_link(new_map, new_map->header.prev,
3022 new_entry);
3023
3024 /*
3025 * the new entry will need an amap. it will either
3026 * need to be copied from the old entry or created
3027 * from scratch (if the old entry does not have an
3028 * amap). can we defer this process until later
3029 * (by setting "needs_copy") or do we need to copy
3030 * the amap now?
3031 *
3032 * we must copy the amap now if any of the following
3033 * conditions hold:
3034 * 1. the old entry has an amap and that amap is
3035 * being shared. this means that the old (parent)
3036 * process is sharing the amap with another
3037 * process. if we do not clear needs_copy here
3038 * we will end up in a situation where both the
3039 * parent and child process are refering to the
3040 * same amap with "needs_copy" set. if the
3041 * parent write-faults, the fault routine will
3042 * clear "needs_copy" in the parent by allocating
3043 * a new amap. this is wrong because the
3044 * parent is supposed to be sharing the old amap
3045 * and the new amap will break that.
3046 *
3047 * 2. if the old entry has an amap and a non-zero
3048 * wire count then we are going to have to call
3049 * amap_cow_now to avoid page faults in the
3050 * parent process. since amap_cow_now requires
3051 * "needs_copy" to be clear we might as well
3052 * clear it here as well.
3053 *
3054 */
3055
3056 if (old_entry->aref.ar_amap != NULL) {
3057
3058 if ((amap_flags(old_entry->aref.ar_amap) &
3059 AMAP_SHARED) != 0 ||
3060 VM_MAPENT_ISWIRED(old_entry)) {
3061
3062 amap_copy(new_map, new_entry, M_WAITOK, FALSE,
3063 0, 0);
3064 /* XXXCDC: M_WAITOK ... ok? */
3065 }
3066 }
3067
3068 /*
3069 * if the parent's entry is wired down, then the
3070 * parent process does not want page faults on
3071 * access to that memory. this means that we
3072 * cannot do copy-on-write because we can't write
3073 * protect the old entry. in this case we
3074 * resolve all copy-on-write faults now, using
3075 * amap_cow_now. note that we have already
3076 * allocated any needed amap (above).
3077 */
3078
3079 if (VM_MAPENT_ISWIRED(old_entry)) {
3080
3081 /*
3082 * resolve all copy-on-write faults now
3083 * (note that there is nothing to do if
3084 * the old mapping does not have an amap).
3085 * XXX: is it worthwhile to bother with pmap_copy
3086 * in this case?
3087 */
3088 if (old_entry->aref.ar_amap)
3089 amap_cow_now(new_map, new_entry);
3090
3091 } else {
3092
3093 /*
3094 * setup mappings to trigger copy-on-write faults
3095 * we must write-protect the parent if it has
3096 * an amap and it is not already "needs_copy"...
3097 * if it is already "needs_copy" then the parent
3098 * has already been write-protected by a previous
3099 * fork operation.
3100 *
3101 * if we do not write-protect the parent, then
3102 * we must be sure to write-protect the child
3103 * after the pmap_copy() operation.
3104 *
3105 * XXX: pmap_copy should have some way of telling
3106 * us that it didn't do anything so we can avoid
3107 * calling pmap_protect needlessly.
3108 */
3109
3110 if (old_entry->aref.ar_amap) {
3111
3112 if (!UVM_ET_ISNEEDSCOPY(old_entry)) {
3113 if (old_entry->max_protection & VM_PROT_WRITE) {
3114 pmap_protect(old_map->pmap,
3115 old_entry->start,
3116 old_entry->end,
3117 old_entry->protection &
3118 ~VM_PROT_WRITE);
3119 }
3120 old_entry->etype |= UVM_ET_NEEDSCOPY;
3121 }
3122
3123 /*
3124 * parent must now be write-protected
3125 */
3126 protect_child = FALSE;
3127 } else {
3128
3129 /*
3130 * we only need to protect the child if the
3131 * parent has write access.
3132 */
3133 if (old_entry->max_protection & VM_PROT_WRITE)
3134 protect_child = TRUE;
3135 else
3136 protect_child = FALSE;
3137
3138 }
3139
3140 /*
3141 * copy the mappings
3142 * XXX: need a way to tell if this does anything
3143 */
3144
3145 pmap_copy(new_pmap, old_map->pmap,
3146 new_entry->start,
3147 (old_entry->end - old_entry->start),
3148 old_entry->start);
3149
3150 /*
3151 * protect the child's mappings if necessary
3152 */
3153 if (protect_child) {
3154 pmap_protect(new_pmap, new_entry->start,
3155 new_entry->end,
3156 new_entry->protection &
3157 ~VM_PROT_WRITE);
3158 }
3159
3160 }
3161 break;
3162 } /* end of switch statement */
3163 old_entry = old_entry->next;
3164 }
3165
3166 new_map->size = old_map->size;
3167 vm_map_unlock(old_map);
3168
3169 #ifdef SYSVSHM
3170 if (vm1->vm_shm)
3171 shmfork(vm1, vm2);
3172 #endif
3173
3174 #ifdef PMAP_FORK
3175 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3176 #endif
3177
3178 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3179 return(vm2);
3180 }
3181
3182
3183 #if defined(DDB)
3184
3185 /*
3186 * DDB hooks
3187 */
3188
3189 /*
3190 * uvm_map_print: print out a map
3191 */
3192
3193 void
3194 uvm_map_print(map, full)
3195 vm_map_t map;
3196 boolean_t full;
3197 {
3198
3199 uvm_map_printit(map, full, printf);
3200 }
3201
3202 /*
3203 * uvm_map_printit: actually prints the map
3204 */
3205
3206 void
3207 uvm_map_printit(map, full, pr)
3208 vm_map_t map;
3209 boolean_t full;
3210 void (*pr) __P((const char *, ...));
3211 {
3212 vm_map_entry_t entry;
3213
3214 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3215 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3216 map->nentries, map->size, map->ref_count, map->timestamp,
3217 map->flags);
3218 #ifdef pmap_resident_count
3219 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3220 pmap_resident_count(map->pmap));
3221 #else
3222 /* XXXCDC: this should be required ... */
3223 (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap);
3224 #endif
3225 if (!full)
3226 return;
3227 for (entry = map->header.next; entry != &map->header;
3228 entry = entry->next) {
3229 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
3230 entry, entry->start, entry->end, entry->object.uvm_obj,
3231 (long long)entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff);
3232 (*pr)(
3233 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n",
3234 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3235 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3236 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3237 entry->protection, entry->max_protection,
3238 entry->inheritance, entry->wired_count, entry->advice);
3239 }
3240 }
3241
3242 /*
3243 * uvm_object_print: print out an object
3244 */
3245
3246 void
3247 uvm_object_print(uobj, full)
3248 struct uvm_object *uobj;
3249 boolean_t full;
3250 {
3251
3252 uvm_object_printit(uobj, full, printf);
3253 }
3254
3255 /*
3256 * uvm_object_printit: actually prints the object
3257 */
3258
3259 void
3260 uvm_object_printit(uobj, full, pr)
3261 struct uvm_object *uobj;
3262 boolean_t full;
3263 void (*pr) __P((const char *, ...));
3264 {
3265 struct vm_page *pg;
3266 int cnt = 0;
3267
3268 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
3269 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
3270 if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3271 (*pr)("refs=<SYSTEM>\n");
3272 else
3273 (*pr)("refs=%d\n", uobj->uo_refs);
3274
3275 if (!full) {
3276 return;
3277 }
3278 (*pr)(" PAGES <pg,offset>:\n ");
3279 for (pg = TAILQ_FIRST(&uobj->memq);
3280 pg != NULL;
3281 pg = TAILQ_NEXT(pg, listq), cnt++) {
3282 (*pr)("<%p,0x%lx> ", pg, pg->offset);
3283 if ((cnt % 3) == 2) {
3284 (*pr)("\n ");
3285 }
3286 }
3287 if ((cnt % 3) != 2) {
3288 (*pr)("\n");
3289 }
3290 }
3291
3292 const char page_flagbits[] =
3293 "\20\4CLEAN\5BUSY\6WANTED\7TABLED\12FAKE\13FILLED\14DIRTY\15RELEASED"
3294 "\16FAULTING\17CLEANCHK";
3295 const char page_pqflagbits[] =
3296 "\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ";
3297
3298 /*
3299 * uvm_page_print: print out a page
3300 */
3301
3302 void
3303 uvm_page_print(pg, full)
3304 struct vm_page *pg;
3305 boolean_t full;
3306 {
3307
3308 uvm_page_printit(pg, full, printf);
3309 }
3310
3311 /*
3312 * uvm_page_printit: actually print the page
3313 */
3314
3315 void
3316 uvm_page_printit(pg, full, pr)
3317 struct vm_page *pg;
3318 boolean_t full;
3319 void (*pr) __P((const char *, ...));
3320 {
3321 struct vm_page *lcv;
3322 struct uvm_object *uobj;
3323 struct pglist *pgl;
3324 char pgbuf[128];
3325 char pqbuf[128];
3326
3327 (*pr)("PAGE %p:\n", pg);
3328 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
3329 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
3330 (*pr)(" flags=%s, pqflags=%s, vers=%d, wire_count=%d, pa=0x%lx\n",
3331 pgbuf, pqbuf, pg->version, pg->wire_count, (long)pg->phys_addr);
3332 (*pr)(" uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
3333 pg->uobject, pg->uanon, pg->offset, pg->loan_count);
3334 #if defined(UVM_PAGE_TRKOWN)
3335 if (pg->flags & PG_BUSY)
3336 (*pr)(" owning process = %d, tag=%s\n",
3337 pg->owner, pg->owner_tag);
3338 else
3339 (*pr)(" page not busy, no owner\n");
3340 #else
3341 (*pr)(" [page ownership tracking disabled]\n");
3342 #endif
3343
3344 if (!full)
3345 return;
3346
3347 /* cross-verify object/anon */
3348 if ((pg->pqflags & PQ_FREE) == 0) {
3349 if (pg->pqflags & PQ_ANON) {
3350 if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3351 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3352 (pg->uanon) ? pg->uanon->u.an_page : NULL);
3353 else
3354 (*pr)(" anon backpointer is OK\n");
3355 } else {
3356 uobj = pg->uobject;
3357 if (uobj) {
3358 (*pr)(" checking object list\n");
3359 for (lcv = uobj->memq.tqh_first ; lcv ;
3360 lcv = lcv->listq.tqe_next) {
3361 if (lcv == pg) break;
3362 }
3363 if (lcv)
3364 (*pr)(" page found on object list\n");
3365 else
3366 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3367 }
3368 }
3369 }
3370
3371 /* cross-verify page queue */
3372 if (pg->pqflags & PQ_FREE) {
3373 int fl = uvm_page_lookup_freelist(pg);
3374 pgl = &uvm.page_free[fl].pgfl_queues[((pg)->flags & PG_ZERO) ?
3375 PGFL_ZEROS : PGFL_UNKNOWN];
3376 }
3377 else if (pg->pqflags & PQ_INACTIVE)
3378 pgl = (pg->pqflags & PQ_SWAPBACKED) ?
3379 &uvm.page_inactive_swp : &uvm.page_inactive_obj;
3380 else if (pg->pqflags & PQ_ACTIVE)
3381 pgl = &uvm.page_active;
3382 else
3383 pgl = NULL;
3384
3385 if (pgl) {
3386 (*pr)(" checking pageq list\n");
3387 for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) {
3388 if (lcv == pg) break;
3389 }
3390 if (lcv)
3391 (*pr)(" page found on pageq list\n");
3392 else
3393 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3394 }
3395 }
3396 #endif
3397