uvm_map.c revision 1.54 1 /* $NetBSD: uvm_map.c,v 1.54 1999/06/15 23:27:47 thorpej Exp $ */
2
3 /*
4 * Copyright (c) 1997 Charles D. Cranor and Washington University.
5 * Copyright (c) 1991, 1993, The Regents of the University of California.
6 *
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * The Mach Operating System project at Carnegie-Mellon University.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by Charles D. Cranor,
23 * Washington University, the University of California, Berkeley and
24 * its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * SUCH DAMAGE.
40 *
41 * @(#)vm_map.c 8.3 (Berkeley) 1/12/94
42 * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
43 *
44 *
45 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46 * All rights reserved.
47 *
48 * Permission to use, copy, modify and distribute this software and
49 * its documentation is hereby granted, provided that both the copyright
50 * notice and this permission notice appear in all copies of the
51 * software, derivative works or modified versions, and any portions
52 * thereof, and that both notices appear in supporting documentation.
53 *
54 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57 *
58 * Carnegie Mellon requests users of this software to return to
59 *
60 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
61 * School of Computer Science
62 * Carnegie Mellon University
63 * Pittsburgh PA 15213-3890
64 *
65 * any improvements or extensions that they make and grant Carnegie the
66 * rights to redistribute these changes.
67 */
68
69 #include "opt_ddb.h"
70 #include "opt_uvmhist.h"
71 #include "opt_pmap_new.h"
72 #include "opt_sysv.h"
73
74 /*
75 * uvm_map.c: uvm map operations
76 */
77
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/mman.h>
81 #include <sys/proc.h>
82 #include <sys/malloc.h>
83 #include <sys/pool.h>
84
85 #ifdef SYSVSHM
86 #include <sys/shm.h>
87 #endif
88
89 #include <vm/vm.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_kern.h>
92
93 #define UVM_MAP
94 #include <uvm/uvm.h>
95
96 #ifdef DDB
97 #include <uvm/uvm_ddb.h>
98 #endif
99
100
101 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
102 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
103
104 /*
105 * pool for vmspace structures.
106 */
107
108 struct pool uvm_vmspace_pool;
109
110 /*
111 * pool for dynamically-allocated map entries.
112 */
113
114 struct pool uvm_map_entry_pool;
115
116 #ifdef PMAP_GROWKERNEL
117 /*
118 * This global represents the end of the kernel virtual address
119 * space. If we want to exceed this, we must grow the kernel
120 * virtual address space dynamically.
121 *
122 * Note, this variable is locked by kernel_map's lock.
123 */
124 vaddr_t uvm_maxkaddr;
125 #endif
126
127 /*
128 * macros
129 */
130
131 /*
132 * uvm_map_entry_link: insert entry into a map
133 *
134 * => map must be locked
135 */
136 #define uvm_map_entry_link(map, after_where, entry) do { \
137 (map)->nentries++; \
138 (entry)->prev = (after_where); \
139 (entry)->next = (after_where)->next; \
140 (entry)->prev->next = (entry); \
141 (entry)->next->prev = (entry); \
142 } while (0)
143
144 /*
145 * uvm_map_entry_unlink: remove entry from a map
146 *
147 * => map must be locked
148 */
149 #define uvm_map_entry_unlink(map, entry) do { \
150 (map)->nentries--; \
151 (entry)->next->prev = (entry)->prev; \
152 (entry)->prev->next = (entry)->next; \
153 } while (0)
154
155 /*
156 * SAVE_HINT: saves the specified entry as the hint for future lookups.
157 *
158 * => map need not be locked (protected by hint_lock).
159 */
160 #define SAVE_HINT(map,value) do { \
161 simple_lock(&(map)->hint_lock); \
162 (map)->hint = (value); \
163 simple_unlock(&(map)->hint_lock); \
164 } while (0)
165
166 /*
167 * VM_MAP_RANGE_CHECK: check and correct range
168 *
169 * => map must at least be read locked
170 */
171
172 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
173 if (start < vm_map_min(map)) \
174 start = vm_map_min(map); \
175 if (end > vm_map_max(map)) \
176 end = vm_map_max(map); \
177 if (start > end) \
178 start = end; \
179 } while (0)
180
181 /*
182 * local prototypes
183 */
184
185 static vm_map_entry_t uvm_mapent_alloc __P((vm_map_t));
186 static void uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t));
187 static void uvm_mapent_free __P((vm_map_entry_t));
188 static void uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
189
190 /*
191 * local inlines
192 */
193
194 /* XXX Should not exist! */
195 #define vm_map_downgrade(map) \
196 (void) lockmgr(&(map)->lock, LK_DOWNGRADE, NULL)
197
198 /* XXX Should not exist! */
199 #ifdef DIAGNOSTIC
200 #define vm_map_upgrade(map) \
201 do { \
202 if (lockmgr(&(map)->lock, LK_UPGRADE, NULL) != 0) \
203 panic("vm_map_upgrade: failed to upgrade lock"); \
204 } while (0)
205 #else
206 #define vm_map_upgrade(map) \
207 (void) lockmgr(&(map)->lock, LK_UPGRADE, NULL)
208 #endif /* DIAGNOSTIC */
209
210
211 /*
212 * uvm_mapent_alloc: allocate a map entry
213 *
214 * => XXX: static pool for kernel map?
215 */
216
217 static __inline vm_map_entry_t
218 uvm_mapent_alloc(map)
219 vm_map_t map;
220 {
221 vm_map_entry_t me;
222 int s;
223 UVMHIST_FUNC("uvm_mapent_alloc");
224 UVMHIST_CALLED(maphist);
225
226 if ((map->flags & VM_MAP_INTRSAFE) == 0 &&
227 map != kernel_map && kernel_map != NULL /* XXX */) {
228 me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
229 me->flags = 0;
230 /* me can't be null, wait ok */
231 } else {
232 s = splimp(); /* protect kentry_free list with splimp */
233 simple_lock(&uvm.kentry_lock);
234 me = uvm.kentry_free;
235 if (me) uvm.kentry_free = me->next;
236 simple_unlock(&uvm.kentry_lock);
237 splx(s);
238 if (!me)
239 panic("mapent_alloc: out of static map entries, check MAX_KMAPENT");
240 me->flags = UVM_MAP_STATIC;
241 }
242
243 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]",
244 me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map)
245 ? TRUE : FALSE, 0, 0);
246 return(me);
247 }
248
249 /*
250 * uvm_mapent_free: free map entry
251 *
252 * => XXX: static pool for kernel map?
253 */
254
255 static __inline void
256 uvm_mapent_free(me)
257 vm_map_entry_t me;
258 {
259 int s;
260 UVMHIST_FUNC("uvm_mapent_free");
261 UVMHIST_CALLED(maphist);
262 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
263 me, me->flags, 0, 0);
264 if ((me->flags & UVM_MAP_STATIC) == 0) {
265 pool_put(&uvm_map_entry_pool, me);
266 } else {
267 s = splimp(); /* protect kentry_free list with splimp */
268 simple_lock(&uvm.kentry_lock);
269 me->next = uvm.kentry_free;
270 uvm.kentry_free = me;
271 simple_unlock(&uvm.kentry_lock);
272 splx(s);
273 }
274 }
275
276 /*
277 * uvm_mapent_copy: copy a map entry, preserving flags
278 */
279
280 static __inline void
281 uvm_mapent_copy(src, dst)
282 vm_map_entry_t src;
283 vm_map_entry_t dst;
284 {
285
286 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src));
287 }
288
289 /*
290 * uvm_map_entry_unwire: unwire a map entry
291 *
292 * => map should be locked by caller
293 */
294
295 static __inline void
296 uvm_map_entry_unwire(map, entry)
297 vm_map_t map;
298 vm_map_entry_t entry;
299 {
300
301 uvm_fault_unwire(map, entry->start, entry->end);
302 entry->wired_count = 0;
303 }
304
305 /*
306 * uvm_map_init: init mapping system at boot time. note that we allocate
307 * and init the static pool of vm_map_entry_t's for the kernel here.
308 */
309
310 void
311 uvm_map_init()
312 {
313 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
314 #if defined(UVMHIST)
315 static struct uvm_history_ent maphistbuf[100];
316 static struct uvm_history_ent pdhistbuf[100];
317 #endif
318 int lcv;
319
320 /*
321 * first, init logging system.
322 */
323
324 UVMHIST_FUNC("uvm_map_init");
325 UVMHIST_INIT_STATIC(maphist, maphistbuf);
326 UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
327 UVMHIST_CALLED(maphist);
328 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
329 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0,
330 "# uvm_map() successful calls", 0);
331 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
332 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
333 0);
334 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0);
335 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0);
336
337 /*
338 * now set up static pool of kernel map entrys ...
339 */
340
341 simple_lock_init(&uvm.kentry_lock);
342 uvm.kentry_free = NULL;
343 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
344 kernel_map_entry[lcv].next = uvm.kentry_free;
345 uvm.kentry_free = &kernel_map_entry[lcv];
346 }
347
348 /*
349 * initialize the map-related pools.
350 */
351 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
352 0, 0, 0, "vmsppl", 0,
353 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
354 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
355 0, 0, 0, "vmmpepl", 0,
356 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
357 }
358
359 /*
360 * clippers
361 */
362
363 /*
364 * uvm_map_clip_start: ensure that the entry begins at or after
365 * the starting address, if it doesn't we split the entry.
366 *
367 * => caller should use UVM_MAP_CLIP_START macro rather than calling
368 * this directly
369 * => map must be locked by caller
370 */
371
372 void uvm_map_clip_start(map, entry, start)
373 vm_map_t map;
374 vm_map_entry_t entry;
375 vaddr_t start;
376 {
377 vm_map_entry_t new_entry;
378 vaddr_t new_adj;
379
380 /* uvm_map_simplify_entry(map, entry); */ /* XXX */
381
382 /*
383 * Split off the front portion. note that we must insert the new
384 * entry BEFORE this one, so that this entry has the specified
385 * starting address.
386 */
387
388 new_entry = uvm_mapent_alloc(map);
389 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
390
391 new_entry->end = start;
392 new_adj = start - new_entry->start;
393 if (entry->object.uvm_obj)
394 entry->offset += new_adj; /* shift start over */
395 entry->start = start;
396
397 if (new_entry->aref.ar_amap) {
398 amap_splitref(&new_entry->aref, &entry->aref, new_adj);
399 }
400
401 uvm_map_entry_link(map, entry->prev, new_entry);
402
403 if (UVM_ET_ISSUBMAP(entry)) {
404 /* ... unlikely to happen, but play it safe */
405 uvm_map_reference(new_entry->object.sub_map);
406 } else {
407 if (UVM_ET_ISOBJ(entry) &&
408 entry->object.uvm_obj->pgops &&
409 entry->object.uvm_obj->pgops->pgo_reference)
410 entry->object.uvm_obj->pgops->pgo_reference(
411 entry->object.uvm_obj);
412 }
413 }
414
415 /*
416 * uvm_map_clip_end: ensure that the entry ends at or before
417 * the ending address, if it does't we split the reference
418 *
419 * => caller should use UVM_MAP_CLIP_END macro rather than calling
420 * this directly
421 * => map must be locked by caller
422 */
423
424 void
425 uvm_map_clip_end(map, entry, end)
426 vm_map_t map;
427 vm_map_entry_t entry;
428 vaddr_t end;
429 {
430 vm_map_entry_t new_entry;
431 vaddr_t new_adj; /* #bytes we move start forward */
432
433 /*
434 * Create a new entry and insert it
435 * AFTER the specified entry
436 */
437
438 new_entry = uvm_mapent_alloc(map);
439 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
440
441 new_entry->start = entry->end = end;
442 new_adj = end - entry->start;
443 if (new_entry->object.uvm_obj)
444 new_entry->offset += new_adj;
445
446 if (entry->aref.ar_amap)
447 amap_splitref(&entry->aref, &new_entry->aref, new_adj);
448
449 uvm_map_entry_link(map, entry, new_entry);
450
451 if (UVM_ET_ISSUBMAP(entry)) {
452 /* ... unlikely to happen, but play it safe */
453 uvm_map_reference(new_entry->object.sub_map);
454 } else {
455 if (UVM_ET_ISOBJ(entry) &&
456 entry->object.uvm_obj->pgops &&
457 entry->object.uvm_obj->pgops->pgo_reference)
458 entry->object.uvm_obj->pgops->pgo_reference(
459 entry->object.uvm_obj);
460 }
461 }
462
463
464 /*
465 * M A P - m a i n e n t r y p o i n t
466 */
467 /*
468 * uvm_map: establish a valid mapping in a map
469 *
470 * => assume startp is page aligned.
471 * => assume size is a multiple of PAGE_SIZE.
472 * => assume sys_mmap provides enough of a "hint" to have us skip
473 * over text/data/bss area.
474 * => map must be unlocked (we will lock it)
475 * => <uobj,uoffset> value meanings (4 cases):
476 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER
477 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER
478 * [3] <uobj,uoffset> == normal mapping
479 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA
480 *
481 * case [4] is for kernel mappings where we don't know the offset until
482 * we've found a virtual address. note that kernel object offsets are
483 * always relative to vm_map_min(kernel_map).
484 * => XXXCDC: need way to map in external amap?
485 */
486
487 int
488 uvm_map(map, startp, size, uobj, uoffset, flags)
489 vm_map_t map;
490 vaddr_t *startp; /* IN/OUT */
491 vsize_t size;
492 struct uvm_object *uobj;
493 vaddr_t uoffset;
494 uvm_flag_t flags;
495 {
496 vm_map_entry_t prev_entry, new_entry;
497 vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
498 UVM_MAXPROTECTION(flags);
499 vm_inherit_t inherit = UVM_INHERIT(flags);
500 int advice = UVM_ADVICE(flags);
501 UVMHIST_FUNC("uvm_map");
502 UVMHIST_CALLED(maphist);
503
504 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
505 map, *startp, size, flags);
506 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0);
507
508 /*
509 * step 0: sanity check of protection code
510 */
511
512 if ((prot & maxprot) != prot) {
513 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x",
514 prot, maxprot,0,0);
515 return(KERN_PROTECTION_FAILURE);
516 }
517
518 /*
519 * step 1: figure out where to put new VM range
520 */
521
522 if (vm_map_lock_try(map) == FALSE) {
523 if (flags & UVM_FLAG_TRYLOCK)
524 return(KERN_FAILURE);
525 vm_map_lock(map); /* could sleep here */
526 }
527 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
528 uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) {
529 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
530 vm_map_unlock(map);
531 return (KERN_NO_SPACE);
532 }
533
534 #ifdef PMAP_GROWKERNEL
535 {
536 /*
537 * If the kernel pmap can't map the requested space,
538 * then allocate more resources for it.
539 */
540 if (map == kernel_map && uvm_maxkaddr < (*startp + size))
541 uvm_maxkaddr = pmap_growkernel(*startp + size);
542 }
543 #endif
544
545 UVMCNT_INCR(uvm_map_call);
546
547 /*
548 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
549 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in
550 * either case we want to zero it before storing it in the map entry
551 * (because it looks strange and confusing when debugging...)
552 *
553 * if uobj is not null
554 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
555 * and we do not need to change uoffset.
556 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
557 * now (based on the starting address of the map). this case is
558 * for kernel object mappings where we don't know the offset until
559 * the virtual address is found (with uvm_map_findspace). the
560 * offset is the distance we are from the start of the map.
561 */
562
563 if (uobj == NULL) {
564 uoffset = 0;
565 } else {
566 if (uoffset == UVM_UNKNOWN_OFFSET) {
567 #ifdef DIAGNOSTIC
568 if (UVM_OBJ_IS_KERN_OBJECT(uobj) == 0)
569 panic("uvm_map: unknown offset with "
570 "non-kernel object");
571 #endif
572 uoffset = *startp - vm_map_min(kernel_map);
573 }
574 }
575
576 /*
577 * step 2: try and insert in map by extending previous entry, if
578 * possible
579 * XXX: we don't try and pull back the next entry. might be useful
580 * for a stack, but we are currently allocating our stack in advance.
581 */
582
583 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
584 prev_entry->end == *startp && prev_entry != &map->header &&
585 prev_entry->object.uvm_obj == uobj) {
586
587 if (uobj && prev_entry->offset +
588 (prev_entry->end - prev_entry->start) != uoffset)
589 goto step3;
590
591 if (UVM_ET_ISSUBMAP(prev_entry))
592 goto step3;
593
594 if (prev_entry->protection != prot ||
595 prev_entry->max_protection != maxprot)
596 goto step3;
597
598 if (prev_entry->inheritance != inherit ||
599 prev_entry->advice != advice)
600 goto step3;
601
602 /* wired_count's must match (new area is unwired) */
603 if (prev_entry->wired_count)
604 goto step3;
605
606 /*
607 * can't extend a shared amap. note: no need to lock amap to
608 * look at refs since we don't care about its exact value.
609 * if it is one (i.e. we have only reference) it will stay there
610 */
611
612 if (prev_entry->aref.ar_amap &&
613 amap_refs(prev_entry->aref.ar_amap) != 1) {
614 goto step3;
615 }
616
617 /* got it! */
618
619 UVMCNT_INCR(map_backmerge);
620 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0);
621
622 /*
623 * drop our reference to uobj since we are extending a reference
624 * that we already have (the ref count can not drop to zero).
625 */
626 if (uobj && uobj->pgops->pgo_detach)
627 uobj->pgops->pgo_detach(uobj);
628
629 if (prev_entry->aref.ar_amap) {
630 amap_extend(prev_entry, size);
631 }
632
633 prev_entry->end += size;
634 map->size += size;
635
636 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
637 vm_map_unlock(map);
638 return (KERN_SUCCESS);
639
640 }
641 step3:
642 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0);
643
644 /*
645 * check for possible forward merge (which we don't do) and count
646 * the number of times we missed a *possible* chance to merge more
647 */
648
649 if ((flags & UVM_FLAG_NOMERGE) == 0 &&
650 prev_entry->next != &map->header &&
651 prev_entry->next->start == (*startp + size))
652 UVMCNT_INCR(map_forwmerge);
653
654 /*
655 * step 3: allocate new entry and link it in
656 */
657
658 new_entry = uvm_mapent_alloc(map);
659 new_entry->start = *startp;
660 new_entry->end = new_entry->start + size;
661 new_entry->object.uvm_obj = uobj;
662 new_entry->offset = uoffset;
663
664 if (uobj)
665 new_entry->etype = UVM_ET_OBJ;
666 else
667 new_entry->etype = 0;
668
669 if (flags & UVM_FLAG_COPYONW) {
670 new_entry->etype |= UVM_ET_COPYONWRITE;
671 if ((flags & UVM_FLAG_OVERLAY) == 0)
672 new_entry->etype |= UVM_ET_NEEDSCOPY;
673 }
674
675 new_entry->protection = prot;
676 new_entry->max_protection = maxprot;
677 new_entry->inheritance = inherit;
678 new_entry->wired_count = 0;
679 new_entry->advice = advice;
680 if (flags & UVM_FLAG_OVERLAY) {
681 /*
682 * to_add: for BSS we overallocate a little since we
683 * are likely to extend
684 */
685 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
686 UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
687 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK);
688 new_entry->aref.ar_pageoff = 0;
689 new_entry->aref.ar_amap = amap;
690 } else {
691 new_entry->aref.ar_amap = NULL;
692 }
693
694 uvm_map_entry_link(map, prev_entry, new_entry);
695
696 map->size += size;
697
698 /*
699 * Update the free space hint
700 */
701
702 if ((map->first_free == prev_entry) &&
703 (prev_entry->end >= new_entry->start))
704 map->first_free = new_entry;
705
706 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
707 vm_map_unlock(map);
708 return(KERN_SUCCESS);
709 }
710
711 /*
712 * uvm_map_lookup_entry: find map entry at or before an address
713 *
714 * => map must at least be read-locked by caller
715 * => entry is returned in "entry"
716 * => return value is true if address is in the returned entry
717 */
718
719 boolean_t
720 uvm_map_lookup_entry(map, address, entry)
721 vm_map_t map;
722 vaddr_t address;
723 vm_map_entry_t *entry; /* OUT */
724 {
725 vm_map_entry_t cur;
726 vm_map_entry_t last;
727 UVMHIST_FUNC("uvm_map_lookup_entry");
728 UVMHIST_CALLED(maphist);
729
730 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
731 map, address, entry, 0);
732
733 /*
734 * start looking either from the head of the
735 * list, or from the hint.
736 */
737
738 simple_lock(&map->hint_lock);
739 cur = map->hint;
740 simple_unlock(&map->hint_lock);
741
742 if (cur == &map->header)
743 cur = cur->next;
744
745 UVMCNT_INCR(uvm_mlk_call);
746 if (address >= cur->start) {
747 /*
748 * go from hint to end of list.
749 *
750 * but first, make a quick check to see if
751 * we are already looking at the entry we
752 * want (which is usually the case).
753 * note also that we don't need to save the hint
754 * here... it is the same hint (unless we are
755 * at the header, in which case the hint didn't
756 * buy us anything anyway).
757 */
758 last = &map->header;
759 if ((cur != last) && (cur->end > address)) {
760 UVMCNT_INCR(uvm_mlk_hint);
761 *entry = cur;
762 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
763 cur, 0, 0, 0);
764 return (TRUE);
765 }
766 } else {
767 /*
768 * go from start to hint, *inclusively*
769 */
770 last = cur->next;
771 cur = map->header.next;
772 }
773
774 /*
775 * search linearly
776 */
777
778 while (cur != last) {
779 if (cur->end > address) {
780 if (address >= cur->start) {
781 /*
782 * save this lookup for future
783 * hints, and return
784 */
785
786 *entry = cur;
787 SAVE_HINT(map, cur);
788 UVMHIST_LOG(maphist,"<- search got it (0x%x)",
789 cur, 0, 0, 0);
790 return (TRUE);
791 }
792 break;
793 }
794 cur = cur->next;
795 }
796 *entry = cur->prev;
797 SAVE_HINT(map, *entry);
798 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
799 return (FALSE);
800 }
801
802
803 /*
804 * uvm_map_findspace: find "length" sized space in "map".
805 *
806 * => "hint" is a hint about where we want it, unless fixed is true
807 * (in which case we insist on using "hint").
808 * => "result" is VA returned
809 * => uobj/uoffset are to be used to handle VAC alignment, if required
810 * => caller must at least have read-locked map
811 * => returns NULL on failure, or pointer to prev. map entry if success
812 * => note this is a cross between the old vm_map_findspace and vm_map_find
813 */
814
815 vm_map_entry_t
816 uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed)
817 vm_map_t map;
818 vaddr_t hint;
819 vsize_t length;
820 vaddr_t *result; /* OUT */
821 struct uvm_object *uobj;
822 vaddr_t uoffset;
823 boolean_t fixed;
824 {
825 vm_map_entry_t entry, next, tmp;
826 vaddr_t end;
827 UVMHIST_FUNC("uvm_map_findspace");
828 UVMHIST_CALLED(maphist);
829
830 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)",
831 map, hint, length, fixed);
832
833 if (hint < map->min_offset) { /* check ranges ... */
834 if (fixed) {
835 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
836 return(NULL);
837 }
838 hint = map->min_offset;
839 }
840 if (hint > map->max_offset) {
841 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
842 hint, map->min_offset, map->max_offset, 0);
843 return(NULL);
844 }
845
846 /*
847 * Look for the first possible address; if there's already
848 * something at this address, we have to start after it.
849 */
850
851 if (!fixed && hint == map->min_offset) {
852 if ((entry = map->first_free) != &map->header)
853 hint = entry->end;
854 } else {
855 if (uvm_map_lookup_entry(map, hint, &tmp)) {
856 /* "hint" address already in use ... */
857 if (fixed) {
858 UVMHIST_LOG(maphist,"<- fixed & VA in use",
859 0, 0, 0, 0);
860 return(NULL);
861 }
862 hint = tmp->end;
863 }
864 entry = tmp;
865 }
866
867 /*
868 * Look through the rest of the map, trying to fit a new region in
869 * the gap between existing regions, or after the very last region.
870 * note: entry->end = base VA of current gap,
871 * next->start = VA of end of current gap
872 */
873 for (;; hint = (entry = next)->end) {
874 /*
875 * Find the end of the proposed new region. Be sure we didn't
876 * go beyond the end of the map, or wrap around the address;
877 * if so, we lose. Otherwise, if this is the last entry, or
878 * if the proposed new region fits before the next entry, we
879 * win.
880 */
881
882 #ifdef PMAP_PREFER
883 /*
884 * push hint forward as needed to avoid VAC alias problems.
885 * we only do this if a valid offset is specified.
886 */
887 if (!fixed && uoffset != UVM_UNKNOWN_OFFSET)
888 PMAP_PREFER(uoffset, &hint);
889 #endif
890 end = hint + length;
891 if (end > map->max_offset || end < hint) {
892 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
893 return (NULL);
894 }
895 next = entry->next;
896 if (next == &map->header || next->start >= end)
897 break;
898 if (fixed) {
899 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
900 return(NULL); /* only one shot at it ... */
901 }
902 }
903 SAVE_HINT(map, entry);
904 *result = hint;
905 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0);
906 return (entry);
907 }
908
909 /*
910 * U N M A P - m a i n h e l p e r f u n c t i o n s
911 */
912
913 /*
914 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
915 *
916 * => caller must check alignment and size
917 * => map must be locked by caller
918 * => we return a list of map entries that we've remove from the map
919 * in "entry_list"
920 */
921
922 int
923 uvm_unmap_remove(map, start, end, entry_list)
924 vm_map_t map;
925 vaddr_t start,end;
926 vm_map_entry_t *entry_list; /* OUT */
927 {
928 vm_map_entry_t entry, first_entry, next;
929 vaddr_t len;
930 UVMHIST_FUNC("uvm_unmap_remove");
931 UVMHIST_CALLED(maphist);
932
933 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
934 map, start, end, 0);
935
936 VM_MAP_RANGE_CHECK(map, start, end);
937
938 /*
939 * find first entry
940 */
941 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
942 /* clip and go... */
943 entry = first_entry;
944 UVM_MAP_CLIP_START(map, entry, start);
945 /* critical! prevents stale hint */
946 SAVE_HINT(map, entry->prev);
947
948 } else {
949 entry = first_entry->next;
950 }
951
952 /*
953 * Save the free space hint
954 */
955
956 if (map->first_free->start >= start)
957 map->first_free = entry->prev;
958
959 /*
960 * note: we now re-use first_entry for a different task. we remove
961 * a number of map entries from the map and save them in a linked
962 * list headed by "first_entry". once we remove them from the map
963 * the caller should unlock the map and drop the references to the
964 * backing objects [c.f. uvm_unmap_detach]. the object is to
965 * seperate unmapping from reference dropping. why?
966 * [1] the map has to be locked for unmapping
967 * [2] the map need not be locked for reference dropping
968 * [3] dropping references may trigger pager I/O, and if we hit
969 * a pager that does synchronous I/O we may have to wait for it.
970 * [4] we would like all waiting for I/O to occur with maps unlocked
971 * so that we don't block other threads.
972 */
973 first_entry = NULL;
974 *entry_list = NULL; /* to be safe */
975
976 /*
977 * break up the area into map entry sized regions and unmap. note
978 * that all mappings have to be removed before we can even consider
979 * dropping references to amaps or VM objects (otherwise we could end
980 * up with a mapping to a page on the free list which would be very bad)
981 */
982
983 while ((entry != &map->header) && (entry->start < end)) {
984
985 UVM_MAP_CLIP_END(map, entry, end);
986 next = entry->next;
987 len = entry->end - entry->start;
988
989 /*
990 * unwire before removing addresses from the pmap; otherwise
991 * unwiring will put the entries back into the pmap (XXX).
992 */
993
994 if (entry->wired_count)
995 uvm_map_entry_unwire(map, entry);
996
997 /*
998 * special case: handle mappings to anonymous kernel objects.
999 * we want to free these pages right away...
1000 */
1001 if (UVM_ET_ISOBJ(entry) &&
1002 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
1003 #ifdef DIAGNOSTIC
1004 if (vm_map_pmap(map) != pmap_kernel())
1005 panic("uvm_unmap_remove: kernel object "
1006 "mapped by non-kernel map");
1007 #endif
1008
1009 /*
1010 * note: kernel object mappings are currently used in
1011 * two ways:
1012 * [1] "normal" mappings of pages in the kernel object
1013 * [2] uvm_km_valloc'd allocations in which we
1014 * pmap_enter in some non-kernel-object page
1015 * (e.g. vmapbuf).
1016 *
1017 * for case [1], we need to remove the mapping from
1018 * the pmap and then remove the page from the kernel
1019 * object (because, once pages in a kernel object are
1020 * unmapped they are no longer needed, unlike, say,
1021 * a vnode where you might want the data to persist
1022 * until flushed out of a queue).
1023 *
1024 * for case [2], we need to remove the mapping from
1025 * the pmap. there shouldn't be any pages at the
1026 * specified offset in the kernel object [but it
1027 * doesn't hurt to call uvm_km_pgremove just to be
1028 * safe?]
1029 *
1030 * uvm_km_pgremove currently does the following:
1031 * for pages in the kernel object in range:
1032 * - drops the swap slot
1033 * - uvm_pagefree the page
1034 *
1035 * note there is version of uvm_km_pgremove() that
1036 * is used for "intrsafe" objects.
1037 */
1038
1039 /*
1040 * remove mappings from pmap and drop the pages
1041 * from the object. offsets are always relative
1042 * to vm_map_min(kernel_map).
1043 */
1044 if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) {
1045 #if defined(PMAP_NEW)
1046 pmap_kremove(entry->start, len);
1047 #else
1048 pmap_remove(pmap_kernel(), entry->start,
1049 entry->start + len);
1050 #endif
1051 uvm_km_pgremove_intrsafe(entry->object.uvm_obj,
1052 entry->start - vm_map_min(kernel_map),
1053 entry->end - vm_map_min(kernel_map));
1054 } else {
1055 pmap_remove(pmap_kernel(), entry->start,
1056 entry->start + len);
1057 uvm_km_pgremove(entry->object.uvm_obj,
1058 entry->start - vm_map_min(kernel_map),
1059 entry->end - vm_map_min(kernel_map));
1060 }
1061
1062 /*
1063 * null out kernel_object reference, we've just
1064 * dropped it
1065 */
1066 entry->etype &= ~UVM_ET_OBJ;
1067 entry->object.uvm_obj = NULL; /* to be safe */
1068
1069 } else {
1070 /*
1071 * remove mappings the standard way.
1072 */
1073 pmap_remove(map->pmap, entry->start, entry->end);
1074 }
1075
1076 /*
1077 * remove entry from map and put it on our list of entries
1078 * that we've nuked. then go do next entry.
1079 */
1080 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0);
1081 uvm_map_entry_unlink(map, entry);
1082 map->size -= len;
1083 entry->next = first_entry;
1084 first_entry = entry;
1085 entry = next; /* next entry, please */
1086 }
1087
1088 /*
1089 * now we've cleaned up the map and are ready for the caller to drop
1090 * references to the mapped objects.
1091 */
1092
1093 *entry_list = first_entry;
1094 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1095 return(KERN_SUCCESS);
1096 }
1097
1098 /*
1099 * uvm_unmap_detach: drop references in a chain of map entries
1100 *
1101 * => we will free the map entries as we traverse the list.
1102 */
1103
1104 void
1105 uvm_unmap_detach(first_entry, amap_unref_flags)
1106 vm_map_entry_t first_entry;
1107 int amap_unref_flags;
1108 {
1109 vm_map_entry_t next_entry;
1110 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1111
1112 while (first_entry) {
1113
1114 #ifdef DIAGNOSTIC
1115 /*
1116 * sanity check
1117 */
1118 /* was part of vm_map_entry_delete() */
1119 if (first_entry->wired_count)
1120 panic("unmap: still wired!");
1121 #endif
1122
1123 UVMHIST_LOG(maphist,
1124 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1125 first_entry, first_entry->aref.ar_amap,
1126 first_entry->object.uvm_obj,
1127 UVM_ET_ISSUBMAP(first_entry));
1128
1129 /*
1130 * drop reference to amap, if we've got one
1131 */
1132
1133 if (first_entry->aref.ar_amap)
1134 amap_unref(first_entry, amap_unref_flags);
1135
1136 /*
1137 * drop reference to our backing object, if we've got one
1138 */
1139
1140 if (UVM_ET_ISSUBMAP(first_entry)) {
1141 /* ... unlikely to happen, but play it safe */
1142 uvm_map_deallocate(first_entry->object.sub_map);
1143 } else {
1144 if (UVM_ET_ISOBJ(first_entry) &&
1145 first_entry->object.uvm_obj->pgops->pgo_detach)
1146 first_entry->object.uvm_obj->pgops->
1147 pgo_detach(first_entry->object.uvm_obj);
1148 }
1149
1150 /*
1151 * next entry
1152 */
1153 next_entry = first_entry->next;
1154 uvm_mapent_free(first_entry);
1155 first_entry = next_entry;
1156 }
1157
1158 /*
1159 * done!
1160 */
1161 UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1162 return;
1163 }
1164
1165 /*
1166 * E X T R A C T I O N F U N C T I O N S
1167 */
1168
1169 /*
1170 * uvm_map_reserve: reserve space in a vm_map for future use.
1171 *
1172 * => we reserve space in a map by putting a dummy map entry in the
1173 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1174 * => map should be unlocked (we will write lock it)
1175 * => we return true if we were able to reserve space
1176 * => XXXCDC: should be inline?
1177 */
1178
1179 int
1180 uvm_map_reserve(map, size, offset, raddr)
1181 vm_map_t map;
1182 vsize_t size;
1183 vaddr_t offset; /* hint for pmap_prefer */
1184 vaddr_t *raddr; /* OUT: reserved VA */
1185 {
1186 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1187
1188 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1189 map,size,offset,raddr);
1190
1191 size = round_page(size);
1192 if (*raddr < vm_map_min(map))
1193 *raddr = vm_map_min(map); /* hint */
1194
1195 /*
1196 * reserve some virtual space.
1197 */
1198
1199 if (uvm_map(map, raddr, size, NULL, offset,
1200 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1201 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) {
1202 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1203 return (FALSE);
1204 }
1205
1206 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1207 return (TRUE);
1208 }
1209
1210 /*
1211 * uvm_map_replace: replace a reserved (blank) area of memory with
1212 * real mappings.
1213 *
1214 * => caller must WRITE-LOCK the map
1215 * => we return TRUE if replacement was a success
1216 * => we expect the newents chain to have nnewents entrys on it and
1217 * we expect newents->prev to point to the last entry on the list
1218 * => note newents is allowed to be NULL
1219 */
1220
1221 int
1222 uvm_map_replace(map, start, end, newents, nnewents)
1223 struct vm_map *map;
1224 vaddr_t start, end;
1225 vm_map_entry_t newents;
1226 int nnewents;
1227 {
1228 vm_map_entry_t oldent, last;
1229 UVMHIST_FUNC("uvm_map_replace");
1230 UVMHIST_CALLED(maphist);
1231
1232 /*
1233 * first find the blank map entry at the specified address
1234 */
1235
1236 if (!uvm_map_lookup_entry(map, start, &oldent)) {
1237 return(FALSE);
1238 }
1239
1240 /*
1241 * check to make sure we have a proper blank entry
1242 */
1243
1244 if (oldent->start != start || oldent->end != end ||
1245 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1246 return (FALSE);
1247 }
1248
1249 #ifdef DIAGNOSTIC
1250 /*
1251 * sanity check the newents chain
1252 */
1253 {
1254 vm_map_entry_t tmpent = newents;
1255 int nent = 0;
1256 vaddr_t cur = start;
1257
1258 while (tmpent) {
1259 nent++;
1260 if (tmpent->start < cur)
1261 panic("uvm_map_replace1");
1262 if (tmpent->start > tmpent->end || tmpent->end > end) {
1263 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1264 tmpent->start, tmpent->end, end);
1265 panic("uvm_map_replace2");
1266 }
1267 cur = tmpent->end;
1268 if (tmpent->next) {
1269 if (tmpent->next->prev != tmpent)
1270 panic("uvm_map_replace3");
1271 } else {
1272 if (newents->prev != tmpent)
1273 panic("uvm_map_replace4");
1274 }
1275 tmpent = tmpent->next;
1276 }
1277 if (nent != nnewents)
1278 panic("uvm_map_replace5");
1279 }
1280 #endif
1281
1282 /*
1283 * map entry is a valid blank! replace it. (this does all the
1284 * work of map entry link/unlink...).
1285 */
1286
1287 if (newents) {
1288
1289 last = newents->prev; /* we expect this */
1290
1291 /* critical: flush stale hints out of map */
1292 SAVE_HINT(map, newents);
1293 if (map->first_free == oldent)
1294 map->first_free = last;
1295
1296 last->next = oldent->next;
1297 last->next->prev = last;
1298 newents->prev = oldent->prev;
1299 newents->prev->next = newents;
1300 map->nentries = map->nentries + (nnewents - 1);
1301
1302 } else {
1303
1304 /* critical: flush stale hints out of map */
1305 SAVE_HINT(map, oldent->prev);
1306 if (map->first_free == oldent)
1307 map->first_free = oldent->prev;
1308
1309 /* NULL list of new entries: just remove the old one */
1310 uvm_map_entry_unlink(map, oldent);
1311 }
1312
1313
1314 /*
1315 * now we can free the old blank entry, unlock the map and return.
1316 */
1317
1318 uvm_mapent_free(oldent);
1319 return(TRUE);
1320 }
1321
1322 /*
1323 * uvm_map_extract: extract a mapping from a map and put it somewhere
1324 * (maybe removing the old mapping)
1325 *
1326 * => maps should be unlocked (we will write lock them)
1327 * => returns 0 on success, error code otherwise
1328 * => start must be page aligned
1329 * => len must be page sized
1330 * => flags:
1331 * UVM_EXTRACT_REMOVE: remove mappings from srcmap
1332 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1333 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1334 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1335 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1336 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1337 * be used from within the kernel in a kernel level map <<<
1338 */
1339
1340 int
1341 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1342 vm_map_t srcmap, dstmap;
1343 vaddr_t start, *dstaddrp;
1344 vsize_t len;
1345 int flags;
1346 {
1347 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1348 oldstart;
1349 vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry;
1350 vm_map_entry_t oldentry;
1351 vsize_t elen;
1352 int nchain, error, copy_ok;
1353 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1354 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1355 len,0);
1356 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1357
1358 #ifdef DIAGNOSTIC
1359 /*
1360 * step 0: sanity check: start must be on a page boundary, length
1361 * must be page sized. can't ask for CONTIG/QREF if you asked for
1362 * REMOVE.
1363 */
1364 if ((start & PAGE_MASK) || (len & PAGE_MASK))
1365 panic("uvm_map_extract1");
1366 if (flags & UVM_EXTRACT_REMOVE)
1367 if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF))
1368 panic("uvm_map_extract2");
1369 #endif
1370
1371
1372 /*
1373 * step 1: reserve space in the target map for the extracted area
1374 */
1375
1376 dstaddr = *dstaddrp;
1377 if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE)
1378 return(ENOMEM);
1379 *dstaddrp = dstaddr; /* pass address back to caller */
1380 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0);
1381
1382
1383 /*
1384 * step 2: setup for the extraction process loop by init'ing the
1385 * map entry chain, locking src map, and looking up the first useful
1386 * entry in the map.
1387 */
1388
1389 end = start + len;
1390 newend = dstaddr + len;
1391 chain = endchain = NULL;
1392 nchain = 0;
1393 vm_map_lock(srcmap);
1394
1395 if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1396
1397 /* "start" is within an entry */
1398 if (flags & UVM_EXTRACT_QREF) {
1399 /*
1400 * for quick references we don't clip the entry, so
1401 * the entry may map space "before" the starting
1402 * virtual address... this is the "fudge" factor
1403 * (which can be non-zero only the first time
1404 * through the "while" loop in step 3).
1405 */
1406 fudge = start - entry->start;
1407 } else {
1408 /*
1409 * normal reference: we clip the map to fit (thus
1410 * fudge is zero)
1411 */
1412 UVM_MAP_CLIP_START(srcmap, entry, start);
1413 SAVE_HINT(srcmap, entry->prev);
1414 fudge = 0;
1415 }
1416
1417 } else {
1418
1419 /* "start" is not within an entry ... skip to next entry */
1420 if (flags & UVM_EXTRACT_CONTIG) {
1421 error = EINVAL;
1422 goto bad; /* definite hole here ... */
1423 }
1424
1425 entry = entry->next;
1426 fudge = 0;
1427 }
1428 /* save values from srcmap for step 6 */
1429 orig_entry = entry;
1430 orig_fudge = fudge;
1431
1432
1433 /*
1434 * step 3: now start looping through the map entries, extracting
1435 * as we go.
1436 */
1437
1438 while (entry->start < end && entry != &srcmap->header) {
1439
1440 /* if we are not doing a quick reference, clip it */
1441 if ((flags & UVM_EXTRACT_QREF) == 0)
1442 UVM_MAP_CLIP_END(srcmap, entry, end);
1443
1444 /* clear needs_copy (allow chunking) */
1445 if (UVM_ET_ISNEEDSCOPY(entry)) {
1446 if (fudge)
1447 oldstart = entry->start;
1448 else
1449 oldstart = 0; /* XXX: gcc */
1450 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1451 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */
1452 error = ENOMEM;
1453 goto bad;
1454 }
1455 /* amap_copy could clip (during chunk)! update fudge */
1456 if (fudge) {
1457 fudge = fudge - (entry->start - oldstart);
1458 orig_fudge = fudge;
1459 }
1460 }
1461
1462 /* calculate the offset of this from "start" */
1463 oldoffset = (entry->start + fudge) - start;
1464
1465 /* allocate a new map entry */
1466 newentry = uvm_mapent_alloc(dstmap);
1467 if (newentry == NULL) {
1468 error = ENOMEM;
1469 goto bad;
1470 }
1471
1472 /* set up new map entry */
1473 newentry->next = NULL;
1474 newentry->prev = endchain;
1475 newentry->start = dstaddr + oldoffset;
1476 newentry->end =
1477 newentry->start + (entry->end - (entry->start + fudge));
1478 if (newentry->end > newend || newentry->end < newentry->start)
1479 newentry->end = newend;
1480 newentry->object.uvm_obj = entry->object.uvm_obj;
1481 if (newentry->object.uvm_obj) {
1482 if (newentry->object.uvm_obj->pgops->pgo_reference)
1483 newentry->object.uvm_obj->pgops->
1484 pgo_reference(newentry->object.uvm_obj);
1485 newentry->offset = entry->offset + fudge;
1486 } else {
1487 newentry->offset = 0;
1488 }
1489 newentry->etype = entry->etype;
1490 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1491 entry->max_protection : entry->protection;
1492 newentry->max_protection = entry->max_protection;
1493 newentry->inheritance = entry->inheritance;
1494 newentry->wired_count = 0;
1495 newentry->aref.ar_amap = entry->aref.ar_amap;
1496 if (newentry->aref.ar_amap) {
1497 newentry->aref.ar_pageoff =
1498 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1499 amap_ref(newentry, AMAP_SHARED |
1500 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1501 } else {
1502 newentry->aref.ar_pageoff = 0;
1503 }
1504 newentry->advice = entry->advice;
1505
1506 /* now link it on the chain */
1507 nchain++;
1508 if (endchain == NULL) {
1509 chain = endchain = newentry;
1510 } else {
1511 endchain->next = newentry;
1512 endchain = newentry;
1513 }
1514
1515 /* end of 'while' loop! */
1516 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1517 (entry->next == &srcmap->header ||
1518 entry->next->start != entry->end)) {
1519 error = EINVAL;
1520 goto bad;
1521 }
1522 entry = entry->next;
1523 fudge = 0;
1524 }
1525
1526
1527 /*
1528 * step 4: close off chain (in format expected by uvm_map_replace)
1529 */
1530
1531 if (chain)
1532 chain->prev = endchain;
1533
1534
1535 /*
1536 * step 5: attempt to lock the dest map so we can pmap_copy.
1537 * note usage of copy_ok:
1538 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1539 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1540 */
1541
1542 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1543
1544 copy_ok = 1;
1545 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1546 nchain)) {
1547 if (srcmap != dstmap)
1548 vm_map_unlock(dstmap);
1549 error = EIO;
1550 goto bad;
1551 }
1552
1553 } else {
1554
1555 copy_ok = 0;
1556 /* replace defered until step 7 */
1557
1558 }
1559
1560
1561 /*
1562 * step 6: traverse the srcmap a second time to do the following:
1563 * - if we got a lock on the dstmap do pmap_copy
1564 * - if UVM_EXTRACT_REMOVE remove the entries
1565 * we make use of orig_entry and orig_fudge (saved in step 2)
1566 */
1567
1568 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1569
1570 /* purge possible stale hints from srcmap */
1571 if (flags & UVM_EXTRACT_REMOVE) {
1572 SAVE_HINT(srcmap, orig_entry->prev);
1573 if (srcmap->first_free->start >= start)
1574 srcmap->first_free = orig_entry->prev;
1575 }
1576
1577 entry = orig_entry;
1578 fudge = orig_fudge;
1579 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */
1580
1581 while (entry->start < end && entry != &srcmap->header) {
1582
1583 if (copy_ok) {
1584 oldoffset = (entry->start + fudge) - start;
1585 elen = min(end, entry->end) - (entry->start + fudge);
1586 pmap_copy(dstmap->pmap, srcmap->pmap, dstaddr + oldoffset,
1587 elen, entry->start + fudge);
1588 }
1589
1590 /* we advance "entry" in the following if statement */
1591 if (flags & UVM_EXTRACT_REMOVE) {
1592 pmap_remove(srcmap->pmap, entry->start,
1593 entry->end);
1594 oldentry = entry; /* save entry */
1595 entry = entry->next; /* advance */
1596 uvm_map_entry_unlink(srcmap, oldentry);
1597 /* add to dead list */
1598 oldentry->next = deadentry;
1599 deadentry = oldentry;
1600 } else {
1601 entry = entry->next; /* advance */
1602 }
1603
1604 /* end of 'while' loop */
1605 fudge = 0;
1606 }
1607
1608 /*
1609 * unlock dstmap. we will dispose of deadentry in
1610 * step 7 if needed
1611 */
1612 if (copy_ok && srcmap != dstmap)
1613 vm_map_unlock(dstmap);
1614
1615 }
1616 else
1617 deadentry = NULL; /* XXX: gcc */
1618
1619 /*
1620 * step 7: we are done with the source map, unlock. if copy_ok
1621 * is 0 then we have not replaced the dummy mapping in dstmap yet
1622 * and we need to do so now.
1623 */
1624
1625 vm_map_unlock(srcmap);
1626 if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1627 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */
1628
1629 /* now do the replacement if we didn't do it in step 5 */
1630 if (copy_ok == 0) {
1631 vm_map_lock(dstmap);
1632 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1633 nchain);
1634 vm_map_unlock(dstmap);
1635
1636 if (error == FALSE) {
1637 error = EIO;
1638 goto bad2;
1639 }
1640 }
1641
1642 /*
1643 * done!
1644 */
1645 return(0);
1646
1647 /*
1648 * bad: failure recovery
1649 */
1650 bad:
1651 vm_map_unlock(srcmap);
1652 bad2: /* src already unlocked */
1653 if (chain)
1654 uvm_unmap_detach(chain,
1655 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1656 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */
1657 return(error);
1658 }
1659
1660 /* end of extraction functions */
1661
1662 /*
1663 * uvm_map_submap: punch down part of a map into a submap
1664 *
1665 * => only the kernel_map is allowed to be submapped
1666 * => the purpose of submapping is to break up the locking granularity
1667 * of a larger map
1668 * => the range specified must have been mapped previously with a uvm_map()
1669 * call [with uobj==NULL] to create a blank map entry in the main map.
1670 * [And it had better still be blank!]
1671 * => maps which contain submaps should never be copied or forked.
1672 * => to remove a submap, use uvm_unmap() on the main map
1673 * and then uvm_map_deallocate() the submap.
1674 * => main map must be unlocked.
1675 * => submap must have been init'd and have a zero reference count.
1676 * [need not be locked as we don't actually reference it]
1677 */
1678
1679 int
1680 uvm_map_submap(map, start, end, submap)
1681 vm_map_t map, submap;
1682 vaddr_t start, end;
1683 {
1684 vm_map_entry_t entry;
1685 int result;
1686 UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist);
1687
1688 vm_map_lock(map);
1689
1690 VM_MAP_RANGE_CHECK(map, start, end);
1691
1692 if (uvm_map_lookup_entry(map, start, &entry)) {
1693 UVM_MAP_CLIP_START(map, entry, start);
1694 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */
1695 }
1696 else {
1697 entry = NULL;
1698 }
1699
1700 if (entry != NULL &&
1701 entry->start == start && entry->end == end &&
1702 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1703 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1704
1705 /*
1706 * doit!
1707 */
1708 entry->etype |= UVM_ET_SUBMAP;
1709 entry->object.sub_map = submap;
1710 entry->offset = 0;
1711 uvm_map_reference(submap);
1712 result = KERN_SUCCESS;
1713 } else {
1714 result = KERN_INVALID_ARGUMENT;
1715 }
1716 vm_map_unlock(map);
1717
1718 return(result);
1719 }
1720
1721
1722 /*
1723 * uvm_map_protect: change map protection
1724 *
1725 * => set_max means set max_protection.
1726 * => map must be unlocked.
1727 */
1728
1729 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \
1730 ~VM_PROT_WRITE : VM_PROT_ALL)
1731 #define max(a,b) ((a) > (b) ? (a) : (b))
1732
1733 int
1734 uvm_map_protect(map, start, end, new_prot, set_max)
1735 vm_map_t map;
1736 vaddr_t start, end;
1737 vm_prot_t new_prot;
1738 boolean_t set_max;
1739 {
1740 vm_map_entry_t current, entry;
1741 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1742 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1743 map, start, end, new_prot);
1744
1745 vm_map_lock(map);
1746
1747 VM_MAP_RANGE_CHECK(map, start, end);
1748
1749 if (uvm_map_lookup_entry(map, start, &entry)) {
1750 UVM_MAP_CLIP_START(map, entry, start);
1751 } else {
1752 entry = entry->next;
1753 }
1754
1755 /*
1756 * make a first pass to check for protection violations.
1757 */
1758
1759 current = entry;
1760 while ((current != &map->header) && (current->start < end)) {
1761 if (UVM_ET_ISSUBMAP(current))
1762 return (KERN_INVALID_ARGUMENT);
1763 if ((new_prot & current->max_protection) != new_prot) {
1764 vm_map_unlock(map);
1765 return (KERN_PROTECTION_FAILURE);
1766 }
1767 current = current->next;
1768 }
1769
1770 /* go back and fix up protections (no need to clip this time). */
1771
1772 current = entry;
1773
1774 while ((current != &map->header) && (current->start < end)) {
1775 vm_prot_t old_prot;
1776
1777 UVM_MAP_CLIP_END(map, current, end);
1778
1779 old_prot = current->protection;
1780 if (set_max)
1781 current->protection =
1782 (current->max_protection = new_prot) & old_prot;
1783 else
1784 current->protection = new_prot;
1785
1786 /*
1787 * update physical map if necessary. worry about copy-on-write
1788 * here -- CHECK THIS XXX
1789 */
1790
1791 if (current->protection != old_prot) {
1792
1793 /* update pmap! */
1794 pmap_protect(map->pmap, current->start, current->end,
1795 current->protection & MASK(entry));
1796
1797 }
1798 current = current->next;
1799 }
1800
1801 vm_map_unlock(map);
1802 UVMHIST_LOG(maphist, "<- done",0,0,0,0);
1803 return(KERN_SUCCESS);
1804 }
1805
1806 #undef max
1807 #undef MASK
1808
1809 /*
1810 * uvm_map_inherit: set inheritance code for range of addrs in map.
1811 *
1812 * => map must be unlocked
1813 * => note that the inherit code is used during a "fork". see fork
1814 * code for details.
1815 */
1816
1817 int
1818 uvm_map_inherit(map, start, end, new_inheritance)
1819 vm_map_t map;
1820 vaddr_t start;
1821 vaddr_t end;
1822 vm_inherit_t new_inheritance;
1823 {
1824 vm_map_entry_t entry, temp_entry;
1825 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1826 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1827 map, start, end, new_inheritance);
1828
1829 switch (new_inheritance) {
1830 case VM_INHERIT_NONE:
1831 case VM_INHERIT_COPY:
1832 case VM_INHERIT_SHARE:
1833 break;
1834 default:
1835 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1836 return (KERN_INVALID_ARGUMENT);
1837 }
1838
1839 vm_map_lock(map);
1840
1841 VM_MAP_RANGE_CHECK(map, start, end);
1842
1843 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1844 entry = temp_entry;
1845 UVM_MAP_CLIP_START(map, entry, start);
1846 } else {
1847 entry = temp_entry->next;
1848 }
1849
1850 while ((entry != &map->header) && (entry->start < end)) {
1851 UVM_MAP_CLIP_END(map, entry, end);
1852
1853 entry->inheritance = new_inheritance;
1854
1855 entry = entry->next;
1856 }
1857
1858 vm_map_unlock(map);
1859 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1860 return(KERN_SUCCESS);
1861 }
1862
1863 /*
1864 * uvm_map_advice: set advice code for range of addrs in map.
1865 *
1866 * => map must be unlocked
1867 */
1868
1869 int
1870 uvm_map_advice(map, start, end, new_advice)
1871 vm_map_t map;
1872 vaddr_t start;
1873 vaddr_t end;
1874 int new_advice;
1875 {
1876 vm_map_entry_t entry, temp_entry;
1877 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
1878 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
1879 map, start, end, new_advice);
1880
1881 vm_map_lock(map);
1882
1883 VM_MAP_RANGE_CHECK(map, start, end);
1884
1885 if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1886 entry = temp_entry;
1887 UVM_MAP_CLIP_START(map, entry, start);
1888 } else {
1889 entry = temp_entry->next;
1890 }
1891
1892 while ((entry != &map->header) && (entry->start < end)) {
1893 UVM_MAP_CLIP_END(map, entry, end);
1894
1895 switch (new_advice) {
1896 case MADV_NORMAL:
1897 case MADV_RANDOM:
1898 case MADV_SEQUENTIAL:
1899 /* nothing special here */
1900 break;
1901
1902 #if 0
1903 case MADV_WILLNEED:
1904 /* activate all these pages */
1905 /* XXX */
1906 /*
1907 * should invent a "weak" mode for uvm_fault()
1908 * which would only do the PGO_LOCKED pgo_get().
1909 */
1910 break;
1911
1912 case MADV_DONTNEED:
1913 /* deactivate this page */
1914 /* XXX */
1915 /*
1916 * vm_page_t p;
1917 * uvm_lock_pageq();
1918 * for (p in each page)
1919 * if (not_wired)
1920 * uvm_pagedeactivate(p);
1921 * uvm_unlock_pageq();
1922 */
1923 break;
1924
1925 case MADV_SPACEAVAIL:
1926 /*
1927 * XXXMRG
1928 * what is this? i think: "ensure that we have
1929 * allocated backing-store for these pages". this
1930 * is going to require changes in the page daemon,
1931 * as it will free swap space allocated to pages in
1932 * core. there's also what to do for
1933 * device/file/anonymous memory..
1934 */
1935 break;
1936
1937 case MADV_GARBAGE:
1938 /* pages are `empty' and can be garbage collected */
1939 /* XXX */
1940 /*
1941 * (perhaps MADV_FREE? check freebsd's MADV_FREE).
1942 *
1943 * need to do this:
1944 * - clear all the referenced and modified bits on
1945 * the pages,
1946 * - delete any backing store,
1947 * - mark the page as `recycable'.
1948 *
1949 * So, if you start paging, the pages would be thrown out
1950 * and then zero-filled the next time they're used.
1951 * Otherwise you'd just reuse them directly. Once the
1952 * page has been modified again, it would no longer be
1953 * recyclable. That way, malloc() can just tell the
1954 * system when pages are `empty'; if memory is needed,
1955 * they'll be tossed; if memory is not needed, there
1956 * will be no additional overhead.
1957 */
1958 break;
1959 #endif
1960
1961 default:
1962 vm_map_unlock(map);
1963 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1964 return (KERN_INVALID_ARGUMENT);
1965 }
1966
1967
1968 entry->advice = new_advice;
1969
1970 entry = entry->next;
1971 }
1972
1973 vm_map_unlock(map);
1974 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1975 return (KERN_SUCCESS);
1976 }
1977
1978 /*
1979 * uvm_map_pageable: sets the pageability of a range in a map.
1980 *
1981 * => regions sepcified as not pageable require lock-down (wired) memory
1982 * and page tables.
1983 * => map must not be locked.
1984 * => XXXCDC: check this and try and clean it up.
1985 */
1986
1987 int
1988 uvm_map_pageable(map, start, end, new_pageable)
1989 vm_map_t map;
1990 vaddr_t start, end;
1991 boolean_t new_pageable;
1992 {
1993 vm_map_entry_t entry, start_entry, failed_entry;
1994 int rv;
1995 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
1996 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1997 map, start, end, new_pageable);
1998
1999 #ifdef DIAGNOSTIC
2000 if ((map->flags & VM_MAP_PAGEABLE) == 0)
2001 panic("uvm_map_pageable: map %p not pageable", map);
2002 #endif
2003
2004 vm_map_lock(map);
2005 VM_MAP_RANGE_CHECK(map, start, end);
2006
2007 /*
2008 * only one pageability change may take place at one time, since
2009 * uvm_fault_wire assumes it will be called only once for each
2010 * wiring/unwiring. therefore, we have to make sure we're actually
2011 * changing the pageability for the entire region. we do so before
2012 * making any changes.
2013 */
2014
2015 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
2016 vm_map_unlock(map);
2017
2018 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2019 return (KERN_INVALID_ADDRESS);
2020 }
2021 entry = start_entry;
2022
2023 /*
2024 * handle wiring and unwiring seperately.
2025 */
2026
2027 if (new_pageable) { /* unwire */
2028
2029 UVM_MAP_CLIP_START(map, entry, start);
2030
2031 /*
2032 * unwiring. first ensure that the range to be unwired is
2033 * really wired down and that there are no holes.
2034 */
2035 while ((entry != &map->header) && (entry->start < end)) {
2036
2037 if (entry->wired_count == 0 ||
2038 (entry->end < end &&
2039 (entry->next == &map->header ||
2040 entry->next->start > entry->end))) {
2041 vm_map_unlock(map);
2042 UVMHIST_LOG(maphist,
2043 "<- done (INVALID UNWIRE ARG)",0,0,0,0);
2044 return (KERN_INVALID_ARGUMENT);
2045 }
2046 entry = entry->next;
2047 }
2048
2049 /*
2050 * now decrement the wiring count for each region. if a region
2051 * becomes completely unwired, unwire its physical pages and
2052 * mappings.
2053 *
2054 * Note, uvm_fault_unwire() (called via uvm_map_entry_unwire())
2055 * does not lock the map, so we don't have to do anything
2056 * special regarding locking here.
2057 */
2058
2059 entry = start_entry;
2060 while ((entry != &map->header) && (entry->start < end)) {
2061 UVM_MAP_CLIP_END(map, entry, end);
2062 entry->wired_count--;
2063 if (entry->wired_count == 0)
2064 uvm_map_entry_unwire(map, entry);
2065 entry = entry->next;
2066 }
2067 vm_map_unlock(map);
2068 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2069 return(KERN_SUCCESS);
2070
2071 /*
2072 * end of unwire case!
2073 */
2074 }
2075
2076 /*
2077 * wire case: in two passes [XXXCDC: ugly block of code here]
2078 *
2079 * 1: holding the write lock, we create any anonymous maps that need
2080 * to be created. then we clip each map entry to the region to
2081 * be wired and increment its wiring count.
2082 *
2083 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2084 * in the pages for any newly wired area (wired_count is 1).
2085 *
2086 * downgrading to a read lock for uvm_fault_wire avoids a possible
2087 * deadlock with another thread that may have faulted on one of
2088 * the pages to be wired (it would mark the page busy, blocking
2089 * us, then in turn block on the map lock that we hold). because
2090 * of problems in the recursive lock package, we cannot upgrade
2091 * to a write lock in vm_map_lookup. thus, any actions that
2092 * require the write lock must be done beforehand. because we
2093 * keep the read lock on the map, the copy-on-write status of the
2094 * entries we modify here cannot change.
2095 */
2096
2097 while ((entry != &map->header) && (entry->start < end)) {
2098
2099 if (entry->wired_count == 0) { /* not already wired? */
2100
2101 /*
2102 * perform actions of vm_map_lookup that need the
2103 * write lock on the map: create an anonymous map
2104 * for a copy-on-write region, or an anonymous map
2105 * for a zero-fill region. (XXXCDC: submap case
2106 * ok?)
2107 */
2108
2109 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2110 if (UVM_ET_ISNEEDSCOPY(entry) &&
2111 ((entry->protection & VM_PROT_WRITE) ||
2112 (entry->object.uvm_obj == NULL))) {
2113 amap_copy(map, entry, M_WAITOK, TRUE,
2114 start, end);
2115 /* XXXCDC: wait OK? */
2116 }
2117 }
2118 } /* wired_count == 0 */
2119 UVM_MAP_CLIP_START(map, entry, start);
2120 UVM_MAP_CLIP_END(map, entry, end);
2121 entry->wired_count++;
2122
2123 /*
2124 * Check for holes
2125 */
2126 if (entry->protection == VM_PROT_NONE ||
2127 (entry->end < end &&
2128 (entry->next == &map->header ||
2129 entry->next->start > entry->end))) {
2130 /*
2131 * found one. amap creation actions do not need to
2132 * be undone, but the wired counts need to be restored.
2133 */
2134 while (entry != &map->header && entry->end > start) {
2135 entry->wired_count--;
2136 entry = entry->prev;
2137 }
2138 vm_map_unlock(map);
2139 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2140 return (KERN_INVALID_ARGUMENT);
2141 }
2142 entry = entry->next;
2143 }
2144
2145 /*
2146 * Pass 2.
2147 */
2148
2149 vm_map_downgrade(map);
2150
2151 rv = 0;
2152 entry = start_entry;
2153 while (entry != &map->header && entry->start < end) {
2154 if (entry->wired_count == 1) {
2155 rv = uvm_fault_wire(map, entry->start, entry->end,
2156 entry->protection);
2157 if (rv) {
2158 /*
2159 * wiring failed. break out of the loop.
2160 * we'll clean up the map below, once we
2161 * have a write lock again.
2162 */
2163 break;
2164 }
2165 }
2166 entry = entry->next;
2167 }
2168
2169 if (rv) { /* failed? */
2170 /*
2171 * Get back to an exclusive (write) lock.
2172 */
2173 vm_map_upgrade(map);
2174
2175 /*
2176 * first drop the wiring count on all the entries
2177 * which haven't actually been wired yet.
2178 */
2179 failed_entry = entry;
2180 while (entry != &map->header && entry->start < end) {
2181 entry->wired_count--;
2182 entry = entry->next;
2183 }
2184
2185 /*
2186 * now, unwire all the entries that were successfully
2187 * wired above.
2188 */
2189 entry = start_entry;
2190 while (entry != failed_entry) {
2191 entry->wired_count--;
2192 if (entry->wired_count == 0)
2193 uvm_map_entry_unwire(map, entry);
2194 entry = entry->next;
2195 }
2196 vm_map_unlock(map);
2197 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2198 return(rv);
2199 }
2200
2201 /* We are holding a read lock here. */
2202 vm_map_unlock_read(map);
2203
2204 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2205 return(KERN_SUCCESS);
2206 }
2207
2208 /*
2209 * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2210 * all mapped regions.
2211 *
2212 * => map must not be locked.
2213 * => if no flags are specified, all regions are unwired.
2214 * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2215 */
2216
2217 int
2218 uvm_map_pageable_all(map, flags, limit)
2219 vm_map_t map;
2220 int flags;
2221 vsize_t limit;
2222 {
2223 vm_map_entry_t entry, failed_entry;
2224 vsize_t size;
2225 int rv;
2226 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2227 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2228
2229 #ifdef DIAGNOSTIC
2230 if ((map->flags & VM_MAP_PAGEABLE) == 0)
2231 panic("uvm_map_pageable_all: map %p not pageable", map);
2232 #endif
2233
2234 vm_map_lock(map);
2235
2236 /*
2237 * handle wiring and unwiring separately.
2238 */
2239
2240 if (flags == 0) { /* unwire */
2241 /*
2242 * Decrement the wiring count on the entries. If they
2243 * reach zero, unwire them.
2244 *
2245 * Note, uvm_fault_unwire() (called via uvm_map_entry_unwire())
2246 * does not lock the map, so we don't have to do anything
2247 * special regarding locking here.
2248 */
2249 for (entry = map->header.next; entry != &map->header;
2250 entry = entry->next) {
2251 if (entry->wired_count) {
2252 if (--entry->wired_count == 0)
2253 uvm_map_entry_unwire(map, entry);
2254 }
2255 }
2256 map->flags &= ~VM_MAP_WIREFUTURE;
2257 vm_map_unlock(map);
2258 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2259 return (KERN_SUCCESS);
2260
2261 /*
2262 * end of unwire case!
2263 */
2264 }
2265
2266 if (flags & MCL_FUTURE) {
2267 /*
2268 * must wire all future mappings; remember this.
2269 */
2270 map->flags |= VM_MAP_WIREFUTURE;
2271 }
2272
2273 if ((flags & MCL_CURRENT) == 0) {
2274 /*
2275 * no more work to do!
2276 */
2277 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2278 vm_map_unlock(map);
2279 return (KERN_SUCCESS);
2280 }
2281
2282 /*
2283 * wire case: in three passes [XXXCDC: ugly block of code here]
2284 *
2285 * 1: holding the write lock, count all pages mapped by non-wired
2286 * entries. if this would cause us to go over our limit, we fail.
2287 *
2288 * 2: still holding the write lock, we create any anonymous maps that
2289 * need to be created. then we increment its wiring count.
2290 *
2291 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2292 * in the pages for any newly wired area (wired count is 1).
2293 *
2294 * downgrading to a read lock for uvm_fault_wire avoids a possible
2295 * deadlock with another thread that may have faulted on one of
2296 * the pages to be wired (it would mark the page busy, blocking
2297 * us, then in turn block on the map lock that we hold). because
2298 * of problems in the recursive lock package, we cannot upgrade
2299 * to a write lock in vm_map_lookup. thus, any actions that
2300 * require the write lock must be done beforehand. because we
2301 * keep the read lock on the map, the copy-on-write status of the
2302 * entries we modify here cannot change.
2303 */
2304
2305 for (size = 0, entry = map->header.next; entry != &map->header;
2306 entry = entry->next) {
2307 if (entry->protection != VM_PROT_NONE &&
2308 entry->wired_count == 0) { /* not already wired? */
2309 size += entry->end - entry->start;
2310 }
2311 }
2312
2313 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2314 vm_map_unlock(map);
2315 return (KERN_NO_SPACE); /* XXX overloaded */
2316 }
2317
2318 /* XXX non-pmap_wired_count case must be handled by caller */
2319 #ifdef pmap_wired_count
2320 if (limit != 0 &&
2321 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2322 vm_map_unlock(map);
2323 return (KERN_NO_SPACE); /* XXX overloaded */
2324 }
2325 #endif
2326
2327 /*
2328 * Pass 2.
2329 */
2330
2331 for (entry = map->header.next; entry != &map->header;
2332 entry = entry->next) {
2333 if (entry->protection == VM_PROT_NONE)
2334 continue;
2335 if (entry->wired_count == 0) { /* not already wired? */
2336 /*
2337 * perform actions of vm_map_lookup that need the
2338 * write lock on the map: create an anonymous map
2339 * for a copy-on-write region, or an anonymous map
2340 * for a zero-fill region. (XXXCDC: submap case
2341 * ok?)
2342 */
2343 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2344 if (UVM_ET_ISNEEDSCOPY(entry) &&
2345 ((entry->protection & VM_PROT_WRITE) ||
2346 (entry->object.uvm_obj == NULL))) {
2347 amap_copy(map, entry, M_WAITOK, TRUE,
2348 entry->start, entry->end);
2349 /* XXXCDC: wait OK? */
2350 }
2351 }
2352 } /* wired_count == 0 */
2353 entry->wired_count++;
2354 }
2355
2356 /*
2357 * Pass 3.
2358 */
2359
2360 vm_map_downgrade(map);
2361
2362 rv = KERN_SUCCESS;
2363 for (entry = map->header.next; entry != &map->header;
2364 entry = entry->next) {
2365 if (entry->wired_count == 1) {
2366 rv = uvm_fault_wire(map, entry->start, entry->end,
2367 entry->protection);
2368 if (rv) {
2369 /*
2370 * wiring failed. break out of the loop.
2371 * we'll clean up the map below, once we
2372 * have a write lock again.
2373 */
2374 break;
2375 }
2376 }
2377 }
2378
2379 if (rv) { /* failed? */
2380 /*
2381 * Get back an exclusive (write) lock.
2382 */
2383 vm_map_upgrade(map);
2384
2385 /*
2386 * first drop the wiring count on all the entries
2387 * which haven't actually been wired yet.
2388 */
2389 failed_entry = entry;
2390 for (/* nothing */; entry != &map->header;
2391 entry = entry->next)
2392 entry->wired_count--;
2393
2394 /*
2395 * now, unwire all the entries that were successfully
2396 * wired above.
2397 */
2398 for (entry = map->header.next; entry != failed_entry;
2399 entry = entry->next) {
2400 entry->wired_count--;
2401 if (entry->wired_count == 0)
2402 uvm_map_entry_unwire(map, entry);
2403 }
2404 vm_map_unlock(map);
2405 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2406 return (rv);
2407 }
2408
2409 /* We are holding a read lock here. */
2410 vm_map_unlock_read(map);
2411
2412 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2413 return (KERN_SUCCESS);
2414 }
2415
2416 /*
2417 * uvm_map_clean: push dirty pages off to backing store.
2418 *
2419 * => valid flags:
2420 * if (flags & PGO_SYNCIO): dirty pages are written synchronously
2421 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2422 * if (flags & PGO_FREE): any cached pages are freed after clean
2423 * => returns an error if any part of the specified range isn't mapped
2424 * => never a need to flush amap layer since the anonymous memory has
2425 * no permanent home...
2426 * => called from sys_msync()
2427 * => caller must not write-lock map (read OK).
2428 * => we may sleep while cleaning if SYNCIO [with map read-locked]
2429 */
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;
2438 vm_map_entry_t entry;
2439 vsize_t size;
2440 struct uvm_object *object;
2441 vaddr_t offset;
2442 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2443 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2444 map, start, end, flags);
2445
2446 vm_map_lock_read(map);
2447 VM_MAP_RANGE_CHECK(map, start, end);
2448 if (!uvm_map_lookup_entry(map, start, &entry)) {
2449 vm_map_unlock_read(map);
2450 return(KERN_INVALID_ADDRESS);
2451 }
2452
2453 /*
2454 * Make a first pass to check for holes.
2455 */
2456 for (current = entry; current->start < end; current = current->next) {
2457 if (UVM_ET_ISSUBMAP(current)) {
2458 vm_map_unlock_read(map);
2459 return (KERN_INVALID_ARGUMENT);
2460 }
2461 if (end > current->end && (current->next == &map->header ||
2462 current->end != current->next->start)) {
2463 vm_map_unlock_read(map);
2464 return (KERN_INVALID_ADDRESS);
2465 }
2466 }
2467
2468 /*
2469 * add "cleanit" flag to flags (for generic flush routine).
2470 * then make a second pass, cleaning/uncaching pages from
2471 * the indicated objects as we go.
2472 */
2473 flags = flags | PGO_CLEANIT;
2474 for (current = entry; current->start < end; current = current->next) {
2475 offset = current->offset + (start - current->start);
2476 size = (end <= current->end ? end : current->end) - start;
2477
2478 /*
2479 * get object/offset. can't be submap (checked above).
2480 */
2481 object = current->object.uvm_obj;
2482 simple_lock(&object->vmobjlock);
2483
2484 /*
2485 * flush pages if we've got a valid backing object.
2486 * note that object is locked.
2487 * XXX should we continue on an error?
2488 */
2489
2490 if (object && object->pgops) {
2491 if (!object->pgops->pgo_flush(object, offset,
2492 offset+size, flags)) {
2493 simple_unlock(&object->vmobjlock);
2494 vm_map_unlock_read(map);
2495 return (KERN_FAILURE);
2496 }
2497 }
2498 simple_unlock(&object->vmobjlock);
2499 start += size;
2500 }
2501 vm_map_unlock_read(map);
2502 return(KERN_SUCCESS);
2503 }
2504
2505
2506 /*
2507 * uvm_map_checkprot: check protection in map
2508 *
2509 * => must allow specified protection in a fully allocated region.
2510 * => map must be read or write locked by caller.
2511 */
2512
2513 boolean_t
2514 uvm_map_checkprot(map, start, end, protection)
2515 vm_map_t map;
2516 vaddr_t start, end;
2517 vm_prot_t protection;
2518 {
2519 vm_map_entry_t entry;
2520 vm_map_entry_t tmp_entry;
2521
2522 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2523 return(FALSE);
2524 }
2525
2526 entry = tmp_entry;
2527
2528 while (start < end) {
2529 if (entry == &map->header) {
2530 return(FALSE);
2531 }
2532
2533 /*
2534 * no holes allowed
2535 */
2536
2537 if (start < entry->start) {
2538 return(FALSE);
2539 }
2540
2541 /*
2542 * check protection associated with entry
2543 */
2544
2545 if ((entry->protection & protection) != protection) {
2546 return(FALSE);
2547 }
2548
2549 /* go to next entry */
2550
2551 start = entry->end;
2552 entry = entry->next;
2553 }
2554 return(TRUE);
2555 }
2556
2557 /*
2558 * uvmspace_alloc: allocate a vmspace structure.
2559 *
2560 * - structure includes vm_map and pmap
2561 * - XXX: no locking on this structure
2562 * - refcnt set to 1, rest must be init'd by caller
2563 */
2564 struct vmspace *
2565 uvmspace_alloc(min, max, pageable)
2566 vaddr_t min, max;
2567 int pageable;
2568 {
2569 struct vmspace *vm;
2570 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2571
2572 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2573 uvmspace_init(vm, NULL, min, max, pageable);
2574 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2575 return (vm);
2576 }
2577
2578 /*
2579 * uvmspace_init: initialize a vmspace structure.
2580 *
2581 * - XXX: no locking on this structure
2582 * - refcnt set to 1, rest must me init'd by caller
2583 */
2584 void
2585 uvmspace_init(vm, pmap, min, max, pageable)
2586 struct vmspace *vm;
2587 struct pmap *pmap;
2588 vaddr_t min, max;
2589 boolean_t pageable;
2590 {
2591 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2592
2593 memset(vm, 0, sizeof(*vm));
2594
2595 uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0);
2596
2597 if (pmap)
2598 pmap_reference(pmap);
2599 else
2600 #if defined(PMAP_NEW)
2601 pmap = pmap_create();
2602 #else
2603 pmap = pmap_create(0);
2604 #endif
2605 vm->vm_map.pmap = pmap;
2606
2607 vm->vm_refcnt = 1;
2608 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2609 }
2610
2611 /*
2612 * uvmspace_share: share a vmspace between two proceses
2613 *
2614 * - XXX: no locking on vmspace
2615 * - used for vfork, threads(?)
2616 */
2617
2618 void
2619 uvmspace_share(p1, p2)
2620 struct proc *p1, *p2;
2621 {
2622 p2->p_vmspace = p1->p_vmspace;
2623 p1->p_vmspace->vm_refcnt++;
2624 }
2625
2626 /*
2627 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2628 *
2629 * - XXX: no locking on vmspace
2630 */
2631
2632 void
2633 uvmspace_unshare(p)
2634 struct proc *p;
2635 {
2636 struct vmspace *nvm, *ovm = p->p_vmspace;
2637 int s;
2638
2639 if (ovm->vm_refcnt == 1)
2640 /* nothing to do: vmspace isn't shared in the first place */
2641 return;
2642
2643 /* make a new vmspace, still holding old one */
2644 nvm = uvmspace_fork(ovm);
2645
2646 s = splhigh(); /* make this `atomic' */
2647 pmap_deactivate(p); /* unbind old vmspace */
2648 p->p_vmspace = nvm;
2649 pmap_activate(p); /* switch to new vmspace */
2650 splx(s); /* end of critical section */
2651
2652 uvmspace_free(ovm); /* drop reference to old vmspace */
2653 }
2654
2655 /*
2656 * uvmspace_exec: the process wants to exec a new program
2657 *
2658 * - XXX: no locking on vmspace
2659 */
2660
2661 void
2662 uvmspace_exec(p)
2663 struct proc *p;
2664 {
2665 struct vmspace *nvm, *ovm = p->p_vmspace;
2666 vm_map_t map = &ovm->vm_map;
2667 int s;
2668
2669 #ifdef sparc
2670 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2671 kill_user_windows(p); /* before stack addresses go away */
2672 #endif
2673
2674 /*
2675 * see if more than one process is using this vmspace...
2676 */
2677
2678 if (ovm->vm_refcnt == 1) {
2679
2680 /*
2681 * if p is the only process using its vmspace then we can safely
2682 * recycle that vmspace for the program that is being exec'd.
2683 */
2684
2685 #ifdef SYSVSHM
2686 /*
2687 * SYSV SHM semantics require us to kill all segments on an exec
2688 */
2689 if (ovm->vm_shm)
2690 shmexit(ovm);
2691 #endif
2692
2693 /*
2694 * POSIX 1003.1b -- "lock future mappings" is revoked
2695 * when a process execs another program image.
2696 */
2697 vm_map_lock(map);
2698 map->flags &= ~VM_MAP_WIREFUTURE;
2699 vm_map_unlock(map);
2700
2701 /*
2702 * now unmap the old program
2703 */
2704 uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
2705
2706 } else {
2707
2708 /*
2709 * p's vmspace is being shared, so we can't reuse it for p since
2710 * it is still being used for others. allocate a new vmspace
2711 * for p
2712 */
2713 nvm = uvmspace_alloc(map->min_offset, map->max_offset,
2714 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2715
2716 #if (defined(i386) || defined(pc532)) && !defined(PMAP_NEW)
2717 /*
2718 * allocate zero fill area in the new vmspace's map for user
2719 * page tables for ports that have old style pmaps that keep
2720 * user page tables in the top part of the process' address
2721 * space.
2722 *
2723 * XXXCDC: this should go away once all pmaps are fixed
2724 */
2725 {
2726 vaddr_t addr = VM_MAXUSER_ADDRESS;
2727 if (uvm_map(&nvm->vm_map, &addr, VM_MAX_ADDRESS - addr,
2728 NULL, UVM_UNKNOWN_OFFSET, UVM_MAPFLAG(UVM_PROT_ALL,
2729 UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_NORMAL,
2730 UVM_FLAG_FIXED|UVM_FLAG_COPYONW)) != KERN_SUCCESS)
2731 panic("vm_allocate of PT page area failed");
2732 }
2733 #endif
2734
2735 /*
2736 * install new vmspace and drop our ref to the old one.
2737 */
2738
2739 s = splhigh();
2740 pmap_deactivate(p);
2741 p->p_vmspace = nvm;
2742 pmap_activate(p);
2743 splx(s);
2744
2745 uvmspace_free(ovm);
2746 }
2747 }
2748
2749 /*
2750 * uvmspace_free: free a vmspace data structure
2751 *
2752 * - XXX: no locking on vmspace
2753 */
2754
2755 void
2756 uvmspace_free(vm)
2757 struct vmspace *vm;
2758 {
2759 vm_map_entry_t dead_entries;
2760 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2761
2762 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2763 if (--vm->vm_refcnt == 0) {
2764 /*
2765 * lock the map, to wait out all other references to it. delete
2766 * all of the mappings and pages they hold, then call the pmap
2767 * module to reclaim anything left.
2768 */
2769 vm_map_lock(&vm->vm_map);
2770 if (vm->vm_map.nentries) {
2771 (void)uvm_unmap_remove(&vm->vm_map,
2772 vm->vm_map.min_offset, vm->vm_map.max_offset,
2773 &dead_entries);
2774 if (dead_entries != NULL)
2775 uvm_unmap_detach(dead_entries, 0);
2776 }
2777 pmap_destroy(vm->vm_map.pmap);
2778 vm->vm_map.pmap = NULL;
2779 pool_put(&uvm_vmspace_pool, vm);
2780 }
2781 UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2782 }
2783
2784 /*
2785 * F O R K - m a i n e n t r y p o i n t
2786 */
2787 /*
2788 * uvmspace_fork: fork a process' main map
2789 *
2790 * => create a new vmspace for child process from parent.
2791 * => parent's map must not be locked.
2792 */
2793
2794 struct vmspace *
2795 uvmspace_fork(vm1)
2796 struct vmspace *vm1;
2797 {
2798 struct vmspace *vm2;
2799 vm_map_t old_map = &vm1->vm_map;
2800 vm_map_t new_map;
2801 vm_map_entry_t old_entry;
2802 vm_map_entry_t new_entry;
2803 pmap_t new_pmap;
2804 boolean_t protect_child;
2805 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2806
2807 #if (defined(i386) || defined(pc532)) && !defined(PMAP_NEW)
2808 /*
2809 * avoid copying any of the parent's pagetables or other per-process
2810 * objects that reside in the map by marking all of them non-inheritable
2811 * XXXCDC: should go away
2812 */
2813 (void) uvm_map_inherit(old_map, VM_MAXUSER_ADDRESS, VM_MAX_ADDRESS,
2814 VM_INHERIT_NONE);
2815 #endif
2816
2817 vm_map_lock(old_map);
2818
2819 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset,
2820 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2821 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2822 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2823 new_map = &vm2->vm_map; /* XXX */
2824 new_pmap = new_map->pmap;
2825
2826 old_entry = old_map->header.next;
2827
2828 /*
2829 * go entry-by-entry
2830 */
2831
2832 while (old_entry != &old_map->header) {
2833
2834 /*
2835 * first, some sanity checks on the old entry
2836 */
2837 if (UVM_ET_ISSUBMAP(old_entry))
2838 panic("fork: encountered a submap during fork (illegal)");
2839
2840 if (!UVM_ET_ISCOPYONWRITE(old_entry) &&
2841 UVM_ET_ISNEEDSCOPY(old_entry))
2842 panic("fork: non-copy_on_write map entry marked needs_copy (illegal)");
2843
2844
2845 switch (old_entry->inheritance) {
2846 case VM_INHERIT_NONE:
2847 /*
2848 * drop the mapping
2849 */
2850 break;
2851
2852 case VM_INHERIT_SHARE:
2853 /*
2854 * share the mapping: this means we want the old and
2855 * new entries to share amaps and backing objects.
2856 */
2857
2858 /*
2859 * if the old_entry needs a new amap (due to prev fork)
2860 * then we need to allocate it now so that we have
2861 * something we own to share with the new_entry. [in
2862 * other words, we need to clear needs_copy]
2863 */
2864
2865 if (UVM_ET_ISNEEDSCOPY(old_entry)) {
2866 /* get our own amap, clears needs_copy */
2867 amap_copy(old_map, old_entry, M_WAITOK, FALSE,
2868 0, 0);
2869 /* XXXCDC: WAITOK??? */
2870 }
2871
2872 new_entry = uvm_mapent_alloc(new_map);
2873 /* old_entry -> new_entry */
2874 uvm_mapent_copy(old_entry, new_entry);
2875
2876 /* new pmap has nothing wired in it */
2877 new_entry->wired_count = 0;
2878
2879 /*
2880 * gain reference to object backing the map (can't
2881 * be a submap, already checked this case).
2882 */
2883 if (new_entry->aref.ar_amap)
2884 /* share reference */
2885 amap_ref(new_entry, AMAP_SHARED);
2886
2887 if (new_entry->object.uvm_obj &&
2888 new_entry->object.uvm_obj->pgops->pgo_reference)
2889 new_entry->object.uvm_obj->
2890 pgops->pgo_reference(
2891 new_entry->object.uvm_obj);
2892
2893 /* insert entry at end of new_map's entry list */
2894 uvm_map_entry_link(new_map, new_map->header.prev,
2895 new_entry);
2896
2897 /*
2898 * pmap_copy the mappings: this routine is optional
2899 * but if it is there it will reduce the number of
2900 * page faults in the new proc.
2901 */
2902
2903 pmap_copy(new_pmap, old_map->pmap, new_entry->start,
2904 (old_entry->end - old_entry->start),
2905 old_entry->start);
2906
2907 break;
2908
2909 case VM_INHERIT_COPY:
2910
2911 /*
2912 * copy-on-write the mapping (using mmap's
2913 * MAP_PRIVATE semantics)
2914 *
2915 * allocate new_entry, adjust reference counts.
2916 * (note that new references are read-only).
2917 */
2918
2919 new_entry = uvm_mapent_alloc(new_map);
2920 /* old_entry -> new_entry */
2921 uvm_mapent_copy(old_entry, new_entry);
2922
2923 if (new_entry->aref.ar_amap)
2924 amap_ref(new_entry, 0);
2925
2926 if (new_entry->object.uvm_obj &&
2927 new_entry->object.uvm_obj->pgops->pgo_reference)
2928 new_entry->object.uvm_obj->pgops->pgo_reference
2929 (new_entry->object.uvm_obj);
2930
2931 /* new pmap has nothing wired in it */
2932 new_entry->wired_count = 0;
2933
2934 new_entry->etype |=
2935 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
2936 uvm_map_entry_link(new_map, new_map->header.prev,
2937 new_entry);
2938
2939 /*
2940 * the new entry will need an amap. it will either
2941 * need to be copied from the old entry or created
2942 * from scratch (if the old entry does not have an
2943 * amap). can we defer this process until later
2944 * (by setting "needs_copy") or do we need to copy
2945 * the amap now?
2946 *
2947 * we must copy the amap now if any of the following
2948 * conditions hold:
2949 * 1. the old entry has an amap and that amap is
2950 * being shared. this means that the old (parent)
2951 * process is sharing the amap with another
2952 * process. if we do not clear needs_copy here
2953 * we will end up in a situation where both the
2954 * parent and child process are refering to the
2955 * same amap with "needs_copy" set. if the
2956 * parent write-faults, the fault routine will
2957 * clear "needs_copy" in the parent by allocating
2958 * a new amap. this is wrong because the
2959 * parent is supposed to be sharing the old amap
2960 * and the new amap will break that.
2961 *
2962 * 2. if the old entry has an amap and a non-zero
2963 * wire count then we are going to have to call
2964 * amap_cow_now to avoid page faults in the
2965 * parent process. since amap_cow_now requires
2966 * "needs_copy" to be clear we might as well
2967 * clear it here as well.
2968 *
2969 */
2970
2971 if (old_entry->aref.ar_amap != NULL) {
2972
2973 if ((amap_flags(old_entry->aref.ar_amap) &
2974 AMAP_SHARED) != 0 ||
2975 old_entry->wired_count != 0) {
2976
2977 amap_copy(new_map, new_entry, M_WAITOK, FALSE,
2978 0, 0);
2979 /* XXXCDC: M_WAITOK ... ok? */
2980 }
2981 }
2982
2983 /*
2984 * if the parent's entry is wired down, then the
2985 * parent process does not want page faults on
2986 * access to that memory. this means that we
2987 * cannot do copy-on-write because we can't write
2988 * protect the old entry. in this case we
2989 * resolve all copy-on-write faults now, using
2990 * amap_cow_now. note that we have already
2991 * allocated any needed amap (above).
2992 */
2993
2994 if (old_entry->wired_count != 0) {
2995
2996 /*
2997 * resolve all copy-on-write faults now
2998 * (note that there is nothing to do if
2999 * the old mapping does not have an amap).
3000 * XXX: is it worthwhile to bother with pmap_copy
3001 * in this case?
3002 */
3003 if (old_entry->aref.ar_amap)
3004 amap_cow_now(new_map, new_entry);
3005
3006 } else {
3007
3008 /*
3009 * setup mappings to trigger copy-on-write faults
3010 * we must write-protect the parent if it has
3011 * an amap and it is not already "needs_copy"...
3012 * if it is already "needs_copy" then the parent
3013 * has already been write-protected by a previous
3014 * fork operation.
3015 *
3016 * if we do not write-protect the parent, then
3017 * we must be sure to write-protect the child
3018 * after the pmap_copy() operation.
3019 *
3020 * XXX: pmap_copy should have some way of telling
3021 * us that it didn't do anything so we can avoid
3022 * calling pmap_protect needlessly.
3023 */
3024
3025 if (old_entry->aref.ar_amap) {
3026
3027 if (!UVM_ET_ISNEEDSCOPY(old_entry)) {
3028 if (old_entry->max_protection & VM_PROT_WRITE) {
3029 pmap_protect(old_map->pmap,
3030 old_entry->start,
3031 old_entry->end,
3032 old_entry->protection &
3033 ~VM_PROT_WRITE);
3034 }
3035 old_entry->etype |= UVM_ET_NEEDSCOPY;
3036 }
3037
3038 /*
3039 * parent must now be write-protected
3040 */
3041 protect_child = FALSE;
3042 } else {
3043
3044 /*
3045 * we only need to protect the child if the
3046 * parent has write access.
3047 */
3048 if (old_entry->max_protection & VM_PROT_WRITE)
3049 protect_child = TRUE;
3050 else
3051 protect_child = FALSE;
3052
3053 }
3054
3055 /*
3056 * copy the mappings
3057 * XXX: need a way to tell if this does anything
3058 */
3059
3060 pmap_copy(new_pmap, old_map->pmap,
3061 new_entry->start,
3062 (old_entry->end - old_entry->start),
3063 old_entry->start);
3064
3065 /*
3066 * protect the child's mappings if necessary
3067 */
3068 if (protect_child) {
3069 pmap_protect(new_pmap, new_entry->start,
3070 new_entry->end,
3071 new_entry->protection &
3072 ~VM_PROT_WRITE);
3073 }
3074
3075 }
3076 break;
3077 } /* end of switch statement */
3078 old_entry = old_entry->next;
3079 }
3080
3081 new_map->size = old_map->size;
3082 vm_map_unlock(old_map);
3083
3084 #if (defined(i386) || defined(pc532)) && !defined(PMAP_NEW)
3085 /*
3086 * allocate zero fill area in the new vmspace's map for user
3087 * page tables for ports that have old style pmaps that keep
3088 * user page tables in the top part of the process' address
3089 * space.
3090 *
3091 * XXXCDC: this should go away once all pmaps are fixed
3092 */
3093 {
3094 vaddr_t addr = VM_MAXUSER_ADDRESS;
3095 if (uvm_map(new_map, &addr, VM_MAX_ADDRESS - addr, NULL,
3096 UVM_UNKNOWN_OFFSET, UVM_MAPFLAG(UVM_PROT_ALL,
3097 UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_NORMAL,
3098 UVM_FLAG_FIXED|UVM_FLAG_COPYONW)) != KERN_SUCCESS)
3099 panic("vm_allocate of PT page area failed");
3100 }
3101 #endif
3102
3103 #ifdef SYSVSHM
3104 if (vm1->vm_shm)
3105 shmfork(vm1, vm2);
3106 #endif
3107
3108 #ifdef PMAP_FORK
3109 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3110 #endif
3111
3112 UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3113 return(vm2);
3114 }
3115
3116
3117 #if defined(DDB)
3118
3119 /*
3120 * DDB hooks
3121 */
3122
3123 /*
3124 * uvm_map_print: print out a map
3125 */
3126
3127 void
3128 uvm_map_print(map, full)
3129 vm_map_t map;
3130 boolean_t full;
3131 {
3132
3133 uvm_map_printit(map, full, printf);
3134 }
3135
3136 /*
3137 * uvm_map_printit: actually prints the map
3138 */
3139
3140 void
3141 uvm_map_printit(map, full, pr)
3142 vm_map_t map;
3143 boolean_t full;
3144 void (*pr) __P((const char *, ...));
3145 {
3146 vm_map_entry_t entry;
3147
3148 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3149 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3150 map->nentries, map->size, map->ref_count, map->timestamp,
3151 map->flags);
3152 #ifdef pmap_resident_count
3153 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3154 pmap_resident_count(map->pmap));
3155 #else
3156 /* XXXCDC: this should be required ... */
3157 (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap);
3158 #endif
3159 if (!full)
3160 return;
3161 for (entry = map->header.next; entry != &map->header;
3162 entry = entry->next) {
3163 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%x, amap=%p/%d\n",
3164 entry, entry->start, entry->end, entry->object.uvm_obj,
3165 entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff);
3166 (*pr)(
3167 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n",
3168 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3169 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3170 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3171 entry->protection, entry->max_protection,
3172 entry->inheritance, entry->wired_count, entry->advice);
3173 }
3174 }
3175
3176 /*
3177 * uvm_object_print: print out an object
3178 */
3179
3180 void
3181 uvm_object_print(uobj, full)
3182 struct uvm_object *uobj;
3183 boolean_t full;
3184 {
3185
3186 uvm_object_printit(uobj, full, printf);
3187 }
3188
3189 /*
3190 * uvm_object_printit: actually prints the object
3191 */
3192
3193 void
3194 uvm_object_printit(uobj, full, pr)
3195 struct uvm_object *uobj;
3196 boolean_t full;
3197 void (*pr) __P((const char *, ...));
3198 {
3199 struct vm_page *pg;
3200 int cnt = 0;
3201
3202 (*pr)("OBJECT %p: pgops=%p, npages=%d, ", uobj, uobj->pgops,
3203 uobj->uo_npages);
3204 if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3205 (*pr)("refs=<SYSTEM>\n");
3206 else
3207 (*pr)("refs=%d\n", uobj->uo_refs);
3208
3209 if (!full) return;
3210 (*pr)(" PAGES <pg,offset>:\n ");
3211 for (pg = uobj->memq.tqh_first ; pg ; pg = pg->listq.tqe_next, cnt++) {
3212 (*pr)("<%p,0x%lx> ", pg, pg->offset);
3213 if ((cnt % 3) == 2) (*pr)("\n ");
3214 }
3215 if ((cnt % 3) != 2) (*pr)("\n");
3216 }
3217
3218 /*
3219 * uvm_page_print: print out a page
3220 */
3221
3222 void
3223 uvm_page_print(pg, full)
3224 struct vm_page *pg;
3225 boolean_t full;
3226 {
3227
3228 uvm_page_printit(pg, full, printf);
3229 }
3230
3231 /*
3232 * uvm_page_printit: actually print the page
3233 */
3234
3235 void
3236 uvm_page_printit(pg, full, pr)
3237 struct vm_page *pg;
3238 boolean_t full;
3239 void (*pr) __P((const char *, ...));
3240 {
3241 struct vm_page *lcv;
3242 struct uvm_object *uobj;
3243 struct pglist *pgl;
3244
3245 (*pr)("PAGE %p:\n", pg);
3246 (*pr)(" flags=0x%x, pqflags=0x%x, vers=%d, wire_count=%d, pa=0x%lx\n",
3247 pg->flags, pg->pqflags, pg->version, pg->wire_count, (long)pg->phys_addr);
3248 (*pr)(" uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
3249 pg->uobject, pg->uanon, pg->offset, pg->loan_count);
3250 #if defined(UVM_PAGE_TRKOWN)
3251 if (pg->flags & PG_BUSY)
3252 (*pr)(" owning process = %d, tag=%s\n",
3253 pg->owner, pg->owner_tag);
3254 else
3255 (*pr)(" page not busy, no owner\n");
3256 #else
3257 (*pr)(" [page ownership tracking disabled]\n");
3258 #endif
3259
3260 if (!full)
3261 return;
3262
3263 /* cross-verify object/anon */
3264 if ((pg->pqflags & PQ_FREE) == 0) {
3265 if (pg->pqflags & PQ_ANON) {
3266 if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3267 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3268 (pg->uanon) ? pg->uanon->u.an_page : NULL);
3269 else
3270 (*pr)(" anon backpointer is OK\n");
3271 } else {
3272 uobj = pg->uobject;
3273 if (uobj) {
3274 (*pr)(" checking object list\n");
3275 for (lcv = uobj->memq.tqh_first ; lcv ;
3276 lcv = lcv->listq.tqe_next) {
3277 if (lcv == pg) break;
3278 }
3279 if (lcv)
3280 (*pr)(" page found on object list\n");
3281 else
3282 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3283 }
3284 }
3285 }
3286
3287 /* cross-verify page queue */
3288 if (pg->pqflags & PQ_FREE)
3289 pgl = &uvm.page_free[uvm_page_lookup_freelist(pg)];
3290 else if (pg->pqflags & PQ_INACTIVE)
3291 pgl = (pg->pqflags & PQ_SWAPBACKED) ?
3292 &uvm.page_inactive_swp : &uvm.page_inactive_obj;
3293 else if (pg->pqflags & PQ_ACTIVE)
3294 pgl = &uvm.page_active;
3295 else
3296 pgl = NULL;
3297
3298 if (pgl) {
3299 (*pr)(" checking pageq list\n");
3300 for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) {
3301 if (lcv == pg) break;
3302 }
3303 if (lcv)
3304 (*pr)(" page found on pageq list\n");
3305 else
3306 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3307 }
3308 }
3309 #endif
3310