uvm_map.c revision 1.74 1 1.74 thorpej /* $NetBSD: uvm_map.c,v 1.74 2000/05/19 17:43:55 thorpej Exp $ */
2 1.1 mrg
3 1.1 mrg /*
4 1.1 mrg * Copyright (c) 1997 Charles D. Cranor and Washington University.
5 1.1 mrg * Copyright (c) 1991, 1993, The Regents of the University of California.
6 1.1 mrg *
7 1.1 mrg * All rights reserved.
8 1.1 mrg *
9 1.1 mrg * This code is derived from software contributed to Berkeley by
10 1.1 mrg * The Mach Operating System project at Carnegie-Mellon University.
11 1.1 mrg *
12 1.1 mrg * Redistribution and use in source and binary forms, with or without
13 1.1 mrg * modification, are permitted provided that the following conditions
14 1.1 mrg * are met:
15 1.1 mrg * 1. Redistributions of source code must retain the above copyright
16 1.1 mrg * notice, this list of conditions and the following disclaimer.
17 1.1 mrg * 2. Redistributions in binary form must reproduce the above copyright
18 1.1 mrg * notice, this list of conditions and the following disclaimer in the
19 1.1 mrg * documentation and/or other materials provided with the distribution.
20 1.1 mrg * 3. All advertising materials mentioning features or use of this software
21 1.1 mrg * must display the following acknowledgement:
22 1.1 mrg * This product includes software developed by Charles D. Cranor,
23 1.1 mrg * Washington University, the University of California, Berkeley and
24 1.1 mrg * its contributors.
25 1.1 mrg * 4. Neither the name of the University nor the names of its contributors
26 1.1 mrg * may be used to endorse or promote products derived from this software
27 1.1 mrg * without specific prior written permission.
28 1.1 mrg *
29 1.1 mrg * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 1.1 mrg * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 1.1 mrg * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 1.1 mrg * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 1.1 mrg * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 1.1 mrg * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 1.1 mrg * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 1.1 mrg * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 1.1 mrg * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 1.1 mrg * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 1.1 mrg * SUCH DAMAGE.
40 1.1 mrg *
41 1.1 mrg * @(#)vm_map.c 8.3 (Berkeley) 1/12/94
42 1.3 mrg * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
43 1.1 mrg *
44 1.1 mrg *
45 1.1 mrg * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46 1.1 mrg * All rights reserved.
47 1.1 mrg *
48 1.1 mrg * Permission to use, copy, modify and distribute this software and
49 1.1 mrg * its documentation is hereby granted, provided that both the copyright
50 1.1 mrg * notice and this permission notice appear in all copies of the
51 1.1 mrg * software, derivative works or modified versions, and any portions
52 1.1 mrg * thereof, and that both notices appear in supporting documentation.
53 1.1 mrg *
54 1.1 mrg * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55 1.1 mrg * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56 1.1 mrg * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57 1.1 mrg *
58 1.1 mrg * Carnegie Mellon requests users of this software to return to
59 1.1 mrg *
60 1.1 mrg * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
61 1.1 mrg * School of Computer Science
62 1.1 mrg * Carnegie Mellon University
63 1.1 mrg * Pittsburgh PA 15213-3890
64 1.1 mrg *
65 1.1 mrg * any improvements or extensions that they make and grant Carnegie the
66 1.1 mrg * rights to redistribute these changes.
67 1.1 mrg */
68 1.1 mrg
69 1.21 jonathan #include "opt_ddb.h"
70 1.6 mrg #include "opt_uvmhist.h"
71 1.31 tron #include "opt_sysv.h"
72 1.6 mrg
73 1.1 mrg /*
74 1.1 mrg * uvm_map.c: uvm map operations
75 1.1 mrg */
76 1.1 mrg
77 1.1 mrg #include <sys/param.h>
78 1.1 mrg #include <sys/systm.h>
79 1.1 mrg #include <sys/mman.h>
80 1.1 mrg #include <sys/proc.h>
81 1.1 mrg #include <sys/malloc.h>
82 1.25 thorpej #include <sys/pool.h>
83 1.1 mrg
84 1.1 mrg #ifdef SYSVSHM
85 1.1 mrg #include <sys/shm.h>
86 1.1 mrg #endif
87 1.1 mrg
88 1.1 mrg #include <vm/vm.h>
89 1.1 mrg #include <vm/vm_page.h>
90 1.1 mrg #include <vm/vm_kern.h>
91 1.1 mrg
92 1.1 mrg #define UVM_MAP
93 1.1 mrg #include <uvm/uvm.h>
94 1.21 jonathan
95 1.21 jonathan #ifdef DDB
96 1.21 jonathan #include <uvm/uvm_ddb.h>
97 1.21 jonathan #endif
98 1.21 jonathan
99 1.1 mrg
100 1.1 mrg struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
101 1.1 mrg struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
102 1.1 mrg
103 1.1 mrg /*
104 1.25 thorpej * pool for vmspace structures.
105 1.25 thorpej */
106 1.25 thorpej
107 1.25 thorpej struct pool uvm_vmspace_pool;
108 1.25 thorpej
109 1.26 thorpej /*
110 1.26 thorpej * pool for dynamically-allocated map entries.
111 1.26 thorpej */
112 1.26 thorpej
113 1.26 thorpej struct pool uvm_map_entry_pool;
114 1.25 thorpej
115 1.40 thorpej #ifdef PMAP_GROWKERNEL
116 1.40 thorpej /*
117 1.40 thorpej * This global represents the end of the kernel virtual address
118 1.40 thorpej * space. If we want to exceed this, we must grow the kernel
119 1.40 thorpej * virtual address space dynamically.
120 1.40 thorpej *
121 1.40 thorpej * Note, this variable is locked by kernel_map's lock.
122 1.40 thorpej */
123 1.40 thorpej vaddr_t uvm_maxkaddr;
124 1.40 thorpej #endif
125 1.40 thorpej
126 1.25 thorpej /*
127 1.1 mrg * macros
128 1.1 mrg */
129 1.1 mrg
130 1.1 mrg /*
131 1.1 mrg * uvm_map_entry_link: insert entry into a map
132 1.1 mrg *
133 1.1 mrg * => map must be locked
134 1.1 mrg */
135 1.10 mrg #define uvm_map_entry_link(map, after_where, entry) do { \
136 1.10 mrg (map)->nentries++; \
137 1.10 mrg (entry)->prev = (after_where); \
138 1.10 mrg (entry)->next = (after_where)->next; \
139 1.10 mrg (entry)->prev->next = (entry); \
140 1.10 mrg (entry)->next->prev = (entry); \
141 1.10 mrg } while (0)
142 1.10 mrg
143 1.1 mrg /*
144 1.1 mrg * uvm_map_entry_unlink: remove entry from a map
145 1.1 mrg *
146 1.1 mrg * => map must be locked
147 1.1 mrg */
148 1.10 mrg #define uvm_map_entry_unlink(map, entry) do { \
149 1.10 mrg (map)->nentries--; \
150 1.10 mrg (entry)->next->prev = (entry)->prev; \
151 1.10 mrg (entry)->prev->next = (entry)->next; \
152 1.10 mrg } while (0)
153 1.1 mrg
154 1.1 mrg /*
155 1.1 mrg * SAVE_HINT: saves the specified entry as the hint for future lookups.
156 1.1 mrg *
157 1.1 mrg * => map need not be locked (protected by hint_lock).
158 1.1 mrg */
159 1.10 mrg #define SAVE_HINT(map,value) do { \
160 1.10 mrg simple_lock(&(map)->hint_lock); \
161 1.10 mrg (map)->hint = (value); \
162 1.10 mrg simple_unlock(&(map)->hint_lock); \
163 1.10 mrg } while (0)
164 1.1 mrg
165 1.1 mrg /*
166 1.1 mrg * VM_MAP_RANGE_CHECK: check and correct range
167 1.1 mrg *
168 1.1 mrg * => map must at least be read locked
169 1.1 mrg */
170 1.1 mrg
171 1.10 mrg #define VM_MAP_RANGE_CHECK(map, start, end) do { \
172 1.10 mrg if (start < vm_map_min(map)) \
173 1.10 mrg start = vm_map_min(map); \
174 1.10 mrg if (end > vm_map_max(map)) \
175 1.10 mrg end = vm_map_max(map); \
176 1.10 mrg if (start > end) \
177 1.10 mrg start = end; \
178 1.10 mrg } while (0)
179 1.1 mrg
180 1.1 mrg /*
181 1.1 mrg * local prototypes
182 1.1 mrg */
183 1.1 mrg
184 1.1 mrg static vm_map_entry_t uvm_mapent_alloc __P((vm_map_t));
185 1.1 mrg static void uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t));
186 1.1 mrg static void uvm_mapent_free __P((vm_map_entry_t));
187 1.1 mrg static void uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
188 1.1 mrg
189 1.1 mrg /*
190 1.1 mrg * local inlines
191 1.1 mrg */
192 1.1 mrg
193 1.1 mrg /*
194 1.1 mrg * uvm_mapent_alloc: allocate a map entry
195 1.1 mrg *
196 1.1 mrg * => XXX: static pool for kernel map?
197 1.1 mrg */
198 1.1 mrg
199 1.10 mrg static __inline vm_map_entry_t
200 1.10 mrg uvm_mapent_alloc(map)
201 1.10 mrg vm_map_t map;
202 1.10 mrg {
203 1.10 mrg vm_map_entry_t me;
204 1.10 mrg int s;
205 1.10 mrg UVMHIST_FUNC("uvm_mapent_alloc");
206 1.10 mrg UVMHIST_CALLED(maphist);
207 1.1 mrg
208 1.45 thorpej if ((map->flags & VM_MAP_INTRSAFE) == 0 &&
209 1.45 thorpej map != kernel_map && kernel_map != NULL /* XXX */) {
210 1.26 thorpej me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
211 1.10 mrg me->flags = 0;
212 1.10 mrg /* me can't be null, wait ok */
213 1.10 mrg } else {
214 1.10 mrg s = splimp(); /* protect kentry_free list with splimp */
215 1.10 mrg simple_lock(&uvm.kentry_lock);
216 1.10 mrg me = uvm.kentry_free;
217 1.10 mrg if (me) uvm.kentry_free = me->next;
218 1.10 mrg simple_unlock(&uvm.kentry_lock);
219 1.10 mrg splx(s);
220 1.10 mrg if (!me)
221 1.45 thorpej panic("mapent_alloc: out of static map entries, check MAX_KMAPENT");
222 1.10 mrg me->flags = UVM_MAP_STATIC;
223 1.10 mrg }
224 1.1 mrg
225 1.45 thorpej UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]",
226 1.45 thorpej me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map)
227 1.45 thorpej ? TRUE : FALSE, 0, 0);
228 1.10 mrg return(me);
229 1.1 mrg }
230 1.1 mrg
231 1.1 mrg /*
232 1.1 mrg * uvm_mapent_free: free map entry
233 1.1 mrg *
234 1.1 mrg * => XXX: static pool for kernel map?
235 1.1 mrg */
236 1.1 mrg
237 1.10 mrg static __inline void
238 1.10 mrg uvm_mapent_free(me)
239 1.10 mrg vm_map_entry_t me;
240 1.1 mrg {
241 1.10 mrg int s;
242 1.10 mrg UVMHIST_FUNC("uvm_mapent_free");
243 1.10 mrg UVMHIST_CALLED(maphist);
244 1.10 mrg UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
245 1.1 mrg me, me->flags, 0, 0);
246 1.10 mrg if ((me->flags & UVM_MAP_STATIC) == 0) {
247 1.26 thorpej pool_put(&uvm_map_entry_pool, me);
248 1.10 mrg } else {
249 1.10 mrg s = splimp(); /* protect kentry_free list with splimp */
250 1.10 mrg simple_lock(&uvm.kentry_lock);
251 1.10 mrg me->next = uvm.kentry_free;
252 1.10 mrg uvm.kentry_free = me;
253 1.10 mrg simple_unlock(&uvm.kentry_lock);
254 1.10 mrg splx(s);
255 1.10 mrg }
256 1.1 mrg }
257 1.1 mrg
258 1.1 mrg /*
259 1.1 mrg * uvm_mapent_copy: copy a map entry, preserving flags
260 1.1 mrg */
261 1.1 mrg
262 1.10 mrg static __inline void
263 1.10 mrg uvm_mapent_copy(src, dst)
264 1.10 mrg vm_map_entry_t src;
265 1.10 mrg vm_map_entry_t dst;
266 1.10 mrg {
267 1.1 mrg
268 1.23 perry memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src));
269 1.1 mrg }
270 1.1 mrg
271 1.1 mrg /*
272 1.1 mrg * uvm_map_entry_unwire: unwire a map entry
273 1.1 mrg *
274 1.1 mrg * => map should be locked by caller
275 1.1 mrg */
276 1.1 mrg
277 1.10 mrg static __inline void
278 1.10 mrg uvm_map_entry_unwire(map, entry)
279 1.10 mrg vm_map_t map;
280 1.10 mrg vm_map_entry_t entry;
281 1.10 mrg {
282 1.1 mrg
283 1.10 mrg entry->wired_count = 0;
284 1.57 thorpej uvm_fault_unwire_locked(map, entry->start, entry->end);
285 1.1 mrg }
286 1.1 mrg
287 1.1 mrg /*
288 1.1 mrg * uvm_map_init: init mapping system at boot time. note that we allocate
289 1.1 mrg * and init the static pool of vm_map_entry_t's for the kernel here.
290 1.1 mrg */
291 1.1 mrg
292 1.10 mrg void
293 1.10 mrg uvm_map_init()
294 1.1 mrg {
295 1.10 mrg static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
296 1.1 mrg #if defined(UVMHIST)
297 1.10 mrg static struct uvm_history_ent maphistbuf[100];
298 1.10 mrg static struct uvm_history_ent pdhistbuf[100];
299 1.1 mrg #endif
300 1.10 mrg int lcv;
301 1.10 mrg
302 1.10 mrg /*
303 1.10 mrg * first, init logging system.
304 1.10 mrg */
305 1.1 mrg
306 1.10 mrg UVMHIST_FUNC("uvm_map_init");
307 1.10 mrg UVMHIST_INIT_STATIC(maphist, maphistbuf);
308 1.10 mrg UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
309 1.10 mrg UVMHIST_CALLED(maphist);
310 1.10 mrg UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
311 1.10 mrg UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0,
312 1.10 mrg "# uvm_map() successful calls", 0);
313 1.10 mrg UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
314 1.10 mrg UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
315 1.10 mrg 0);
316 1.10 mrg UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0);
317 1.10 mrg UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0);
318 1.10 mrg
319 1.10 mrg /*
320 1.10 mrg * now set up static pool of kernel map entrys ...
321 1.10 mrg */
322 1.10 mrg
323 1.10 mrg simple_lock_init(&uvm.kentry_lock);
324 1.10 mrg uvm.kentry_free = NULL;
325 1.10 mrg for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
326 1.10 mrg kernel_map_entry[lcv].next = uvm.kentry_free;
327 1.10 mrg uvm.kentry_free = &kernel_map_entry[lcv];
328 1.10 mrg }
329 1.1 mrg
330 1.25 thorpej /*
331 1.25 thorpej * initialize the map-related pools.
332 1.25 thorpej */
333 1.25 thorpej pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
334 1.25 thorpej 0, 0, 0, "vmsppl", 0,
335 1.26 thorpej pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
336 1.26 thorpej pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
337 1.26 thorpej 0, 0, 0, "vmmpepl", 0,
338 1.25 thorpej pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
339 1.1 mrg }
340 1.1 mrg
341 1.1 mrg /*
342 1.1 mrg * clippers
343 1.1 mrg */
344 1.1 mrg
345 1.1 mrg /*
346 1.1 mrg * uvm_map_clip_start: ensure that the entry begins at or after
347 1.1 mrg * the starting address, if it doesn't we split the entry.
348 1.1 mrg *
349 1.1 mrg * => caller should use UVM_MAP_CLIP_START macro rather than calling
350 1.1 mrg * this directly
351 1.1 mrg * => map must be locked by caller
352 1.1 mrg */
353 1.1 mrg
354 1.1 mrg void uvm_map_clip_start(map, entry, start)
355 1.32 mrg vm_map_t map;
356 1.32 mrg vm_map_entry_t entry;
357 1.32 mrg vaddr_t start;
358 1.1 mrg {
359 1.32 mrg vm_map_entry_t new_entry;
360 1.24 eeh vaddr_t new_adj;
361 1.1 mrg
362 1.1 mrg /* uvm_map_simplify_entry(map, entry); */ /* XXX */
363 1.1 mrg
364 1.10 mrg /*
365 1.10 mrg * Split off the front portion. note that we must insert the new
366 1.10 mrg * entry BEFORE this one, so that this entry has the specified
367 1.1 mrg * starting address.
368 1.10 mrg */
369 1.1 mrg
370 1.10 mrg new_entry = uvm_mapent_alloc(map);
371 1.1 mrg uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
372 1.10 mrg
373 1.10 mrg new_entry->end = start;
374 1.1 mrg new_adj = start - new_entry->start;
375 1.1 mrg if (entry->object.uvm_obj)
376 1.10 mrg entry->offset += new_adj; /* shift start over */
377 1.10 mrg entry->start = start;
378 1.1 mrg
379 1.1 mrg if (new_entry->aref.ar_amap) {
380 1.10 mrg amap_splitref(&new_entry->aref, &entry->aref, new_adj);
381 1.1 mrg }
382 1.1 mrg
383 1.10 mrg uvm_map_entry_link(map, entry->prev, new_entry);
384 1.10 mrg
385 1.29 chuck if (UVM_ET_ISSUBMAP(entry)) {
386 1.29 chuck /* ... unlikely to happen, but play it safe */
387 1.29 chuck uvm_map_reference(new_entry->object.sub_map);
388 1.1 mrg } else {
389 1.10 mrg if (UVM_ET_ISOBJ(entry) &&
390 1.10 mrg entry->object.uvm_obj->pgops &&
391 1.10 mrg entry->object.uvm_obj->pgops->pgo_reference)
392 1.10 mrg entry->object.uvm_obj->pgops->pgo_reference(
393 1.10 mrg entry->object.uvm_obj);
394 1.1 mrg }
395 1.1 mrg }
396 1.1 mrg
397 1.1 mrg /*
398 1.1 mrg * uvm_map_clip_end: ensure that the entry ends at or before
399 1.1 mrg * the ending address, if it does't we split the reference
400 1.1 mrg *
401 1.1 mrg * => caller should use UVM_MAP_CLIP_END macro rather than calling
402 1.1 mrg * this directly
403 1.1 mrg * => map must be locked by caller
404 1.1 mrg */
405 1.1 mrg
406 1.10 mrg void
407 1.10 mrg uvm_map_clip_end(map, entry, end)
408 1.10 mrg vm_map_t map;
409 1.10 mrg vm_map_entry_t entry;
410 1.24 eeh vaddr_t end;
411 1.1 mrg {
412 1.10 mrg vm_map_entry_t new_entry;
413 1.24 eeh vaddr_t new_adj; /* #bytes we move start forward */
414 1.1 mrg
415 1.1 mrg /*
416 1.1 mrg * Create a new entry and insert it
417 1.1 mrg * AFTER the specified entry
418 1.1 mrg */
419 1.1 mrg
420 1.1 mrg new_entry = uvm_mapent_alloc(map);
421 1.1 mrg uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
422 1.1 mrg
423 1.1 mrg new_entry->start = entry->end = end;
424 1.1 mrg new_adj = end - entry->start;
425 1.1 mrg if (new_entry->object.uvm_obj)
426 1.1 mrg new_entry->offset += new_adj;
427 1.1 mrg
428 1.10 mrg if (entry->aref.ar_amap)
429 1.10 mrg amap_splitref(&entry->aref, &new_entry->aref, new_adj);
430 1.1 mrg
431 1.1 mrg uvm_map_entry_link(map, entry, new_entry);
432 1.1 mrg
433 1.29 chuck if (UVM_ET_ISSUBMAP(entry)) {
434 1.29 chuck /* ... unlikely to happen, but play it safe */
435 1.29 chuck uvm_map_reference(new_entry->object.sub_map);
436 1.1 mrg } else {
437 1.10 mrg if (UVM_ET_ISOBJ(entry) &&
438 1.10 mrg entry->object.uvm_obj->pgops &&
439 1.10 mrg entry->object.uvm_obj->pgops->pgo_reference)
440 1.10 mrg entry->object.uvm_obj->pgops->pgo_reference(
441 1.10 mrg entry->object.uvm_obj);
442 1.1 mrg }
443 1.1 mrg }
444 1.1 mrg
445 1.1 mrg
446 1.1 mrg /*
447 1.1 mrg * M A P - m a i n e n t r y p o i n t
448 1.1 mrg */
449 1.1 mrg /*
450 1.1 mrg * uvm_map: establish a valid mapping in a map
451 1.1 mrg *
452 1.1 mrg * => assume startp is page aligned.
453 1.1 mrg * => assume size is a multiple of PAGE_SIZE.
454 1.1 mrg * => assume sys_mmap provides enough of a "hint" to have us skip
455 1.1 mrg * over text/data/bss area.
456 1.1 mrg * => map must be unlocked (we will lock it)
457 1.1 mrg * => <uobj,uoffset> value meanings (4 cases):
458 1.1 mrg * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER
459 1.1 mrg * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER
460 1.1 mrg * [3] <uobj,uoffset> == normal mapping
461 1.1 mrg * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA
462 1.1 mrg *
463 1.1 mrg * case [4] is for kernel mappings where we don't know the offset until
464 1.8 chuck * we've found a virtual address. note that kernel object offsets are
465 1.8 chuck * always relative to vm_map_min(kernel_map).
466 1.1 mrg * => XXXCDC: need way to map in external amap?
467 1.1 mrg */
468 1.1 mrg
469 1.10 mrg int
470 1.10 mrg uvm_map(map, startp, size, uobj, uoffset, flags)
471 1.10 mrg vm_map_t map;
472 1.24 eeh vaddr_t *startp; /* IN/OUT */
473 1.24 eeh vsize_t size;
474 1.10 mrg struct uvm_object *uobj;
475 1.70 kleink voff_t uoffset;
476 1.10 mrg uvm_flag_t flags;
477 1.10 mrg {
478 1.10 mrg vm_map_entry_t prev_entry, new_entry;
479 1.10 mrg vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
480 1.10 mrg UVM_MAXPROTECTION(flags);
481 1.10 mrg vm_inherit_t inherit = UVM_INHERIT(flags);
482 1.10 mrg int advice = UVM_ADVICE(flags);
483 1.10 mrg UVMHIST_FUNC("uvm_map");
484 1.10 mrg UVMHIST_CALLED(maphist);
485 1.1 mrg
486 1.10 mrg UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
487 1.10 mrg map, *startp, size, flags);
488 1.10 mrg UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0);
489 1.1 mrg
490 1.10 mrg /*
491 1.10 mrg * step 0: sanity check of protection code
492 1.10 mrg */
493 1.1 mrg
494 1.10 mrg if ((prot & maxprot) != prot) {
495 1.10 mrg UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x",
496 1.10 mrg prot, maxprot,0,0);
497 1.10 mrg return(KERN_PROTECTION_FAILURE);
498 1.10 mrg }
499 1.1 mrg
500 1.10 mrg /*
501 1.10 mrg * step 1: figure out where to put new VM range
502 1.10 mrg */
503 1.1 mrg
504 1.10 mrg if (vm_map_lock_try(map) == FALSE) {
505 1.10 mrg if (flags & UVM_FLAG_TRYLOCK)
506 1.10 mrg return(KERN_FAILURE);
507 1.10 mrg vm_map_lock(map); /* could sleep here */
508 1.10 mrg }
509 1.10 mrg if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
510 1.10 mrg uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) {
511 1.10 mrg UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
512 1.10 mrg vm_map_unlock(map);
513 1.10 mrg return (KERN_NO_SPACE);
514 1.10 mrg }
515 1.1 mrg
516 1.40 thorpej #ifdef PMAP_GROWKERNEL
517 1.10 mrg {
518 1.10 mrg /*
519 1.40 thorpej * If the kernel pmap can't map the requested space,
520 1.40 thorpej * then allocate more resources for it.
521 1.10 mrg */
522 1.40 thorpej if (map == kernel_map && uvm_maxkaddr < (*startp + size))
523 1.40 thorpej uvm_maxkaddr = pmap_growkernel(*startp + size);
524 1.10 mrg }
525 1.10 mrg #endif
526 1.10 mrg
527 1.10 mrg UVMCNT_INCR(uvm_map_call);
528 1.10 mrg
529 1.10 mrg /*
530 1.10 mrg * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
531 1.10 mrg * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in
532 1.10 mrg * either case we want to zero it before storing it in the map entry
533 1.10 mrg * (because it looks strange and confusing when debugging...)
534 1.10 mrg *
535 1.10 mrg * if uobj is not null
536 1.10 mrg * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
537 1.10 mrg * and we do not need to change uoffset.
538 1.10 mrg * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
539 1.10 mrg * now (based on the starting address of the map). this case is
540 1.10 mrg * for kernel object mappings where we don't know the offset until
541 1.10 mrg * the virtual address is found (with uvm_map_findspace). the
542 1.10 mrg * offset is the distance we are from the start of the map.
543 1.10 mrg */
544 1.10 mrg
545 1.10 mrg if (uobj == NULL) {
546 1.10 mrg uoffset = 0;
547 1.10 mrg } else {
548 1.10 mrg if (uoffset == UVM_UNKNOWN_OFFSET) {
549 1.8 chuck #ifdef DIAGNOSTIC
550 1.42 thorpej if (UVM_OBJ_IS_KERN_OBJECT(uobj) == 0)
551 1.42 thorpej panic("uvm_map: unknown offset with "
552 1.42 thorpej "non-kernel object");
553 1.8 chuck #endif
554 1.10 mrg uoffset = *startp - vm_map_min(kernel_map);
555 1.10 mrg }
556 1.10 mrg }
557 1.10 mrg
558 1.10 mrg /*
559 1.10 mrg * step 2: try and insert in map by extending previous entry, if
560 1.10 mrg * possible
561 1.10 mrg * XXX: we don't try and pull back the next entry. might be useful
562 1.10 mrg * for a stack, but we are currently allocating our stack in advance.
563 1.10 mrg */
564 1.10 mrg
565 1.10 mrg if ((flags & UVM_FLAG_NOMERGE) == 0 &&
566 1.10 mrg prev_entry->end == *startp && prev_entry != &map->header &&
567 1.10 mrg prev_entry->object.uvm_obj == uobj) {
568 1.10 mrg
569 1.10 mrg if (uobj && prev_entry->offset +
570 1.10 mrg (prev_entry->end - prev_entry->start) != uoffset)
571 1.10 mrg goto step3;
572 1.10 mrg
573 1.29 chuck if (UVM_ET_ISSUBMAP(prev_entry))
574 1.10 mrg goto step3;
575 1.10 mrg
576 1.10 mrg if (prev_entry->protection != prot ||
577 1.10 mrg prev_entry->max_protection != maxprot)
578 1.10 mrg goto step3;
579 1.10 mrg
580 1.10 mrg if (prev_entry->inheritance != inherit ||
581 1.10 mrg prev_entry->advice != advice)
582 1.10 mrg goto step3;
583 1.10 mrg
584 1.56 thorpej /* wiring status must match (new area is unwired) */
585 1.55 thorpej if (VM_MAPENT_ISWIRED(prev_entry))
586 1.10 mrg goto step3;
587 1.10 mrg
588 1.10 mrg /*
589 1.10 mrg * can't extend a shared amap. note: no need to lock amap to
590 1.34 chuck * look at refs since we don't care about its exact value.
591 1.10 mrg * if it is one (i.e. we have only reference) it will stay there
592 1.10 mrg */
593 1.10 mrg
594 1.10 mrg if (prev_entry->aref.ar_amap &&
595 1.34 chuck amap_refs(prev_entry->aref.ar_amap) != 1) {
596 1.10 mrg goto step3;
597 1.10 mrg }
598 1.10 mrg
599 1.10 mrg /* got it! */
600 1.10 mrg
601 1.10 mrg UVMCNT_INCR(map_backmerge);
602 1.10 mrg UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0);
603 1.10 mrg
604 1.10 mrg /*
605 1.10 mrg * drop our reference to uobj since we are extending a reference
606 1.10 mrg * that we already have (the ref count can not drop to zero).
607 1.10 mrg */
608 1.10 mrg if (uobj && uobj->pgops->pgo_detach)
609 1.10 mrg uobj->pgops->pgo_detach(uobj);
610 1.10 mrg
611 1.10 mrg if (prev_entry->aref.ar_amap) {
612 1.10 mrg amap_extend(prev_entry, size);
613 1.10 mrg }
614 1.10 mrg
615 1.10 mrg prev_entry->end += size;
616 1.10 mrg map->size += size;
617 1.10 mrg
618 1.10 mrg UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
619 1.10 mrg vm_map_unlock(map);
620 1.10 mrg return (KERN_SUCCESS);
621 1.10 mrg
622 1.1 mrg }
623 1.10 mrg step3:
624 1.10 mrg UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0);
625 1.10 mrg
626 1.10 mrg /*
627 1.10 mrg * check for possible forward merge (which we don't do) and count
628 1.10 mrg * the number of times we missed a *possible* chance to merge more
629 1.10 mrg */
630 1.1 mrg
631 1.10 mrg if ((flags & UVM_FLAG_NOMERGE) == 0 &&
632 1.10 mrg prev_entry->next != &map->header &&
633 1.10 mrg prev_entry->next->start == (*startp + size))
634 1.10 mrg UVMCNT_INCR(map_forwmerge);
635 1.1 mrg
636 1.1 mrg /*
637 1.10 mrg * step 3: allocate new entry and link it in
638 1.1 mrg */
639 1.1 mrg
640 1.10 mrg new_entry = uvm_mapent_alloc(map);
641 1.10 mrg new_entry->start = *startp;
642 1.10 mrg new_entry->end = new_entry->start + size;
643 1.10 mrg new_entry->object.uvm_obj = uobj;
644 1.10 mrg new_entry->offset = uoffset;
645 1.10 mrg
646 1.10 mrg if (uobj)
647 1.10 mrg new_entry->etype = UVM_ET_OBJ;
648 1.10 mrg else
649 1.10 mrg new_entry->etype = 0;
650 1.10 mrg
651 1.10 mrg if (flags & UVM_FLAG_COPYONW) {
652 1.10 mrg new_entry->etype |= UVM_ET_COPYONWRITE;
653 1.10 mrg if ((flags & UVM_FLAG_OVERLAY) == 0)
654 1.10 mrg new_entry->etype |= UVM_ET_NEEDSCOPY;
655 1.10 mrg }
656 1.10 mrg
657 1.10 mrg new_entry->protection = prot;
658 1.10 mrg new_entry->max_protection = maxprot;
659 1.10 mrg new_entry->inheritance = inherit;
660 1.10 mrg new_entry->wired_count = 0;
661 1.10 mrg new_entry->advice = advice;
662 1.10 mrg if (flags & UVM_FLAG_OVERLAY) {
663 1.10 mrg /*
664 1.10 mrg * to_add: for BSS we overallocate a little since we
665 1.10 mrg * are likely to extend
666 1.10 mrg */
667 1.24 eeh vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
668 1.30 chs UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
669 1.10 mrg struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK);
670 1.34 chuck new_entry->aref.ar_pageoff = 0;
671 1.10 mrg new_entry->aref.ar_amap = amap;
672 1.10 mrg } else {
673 1.10 mrg new_entry->aref.ar_amap = NULL;
674 1.1 mrg }
675 1.1 mrg
676 1.10 mrg uvm_map_entry_link(map, prev_entry, new_entry);
677 1.10 mrg
678 1.1 mrg map->size += size;
679 1.1 mrg
680 1.10 mrg /*
681 1.10 mrg * Update the free space hint
682 1.10 mrg */
683 1.10 mrg
684 1.10 mrg if ((map->first_free == prev_entry) &&
685 1.10 mrg (prev_entry->end >= new_entry->start))
686 1.10 mrg map->first_free = new_entry;
687 1.10 mrg
688 1.10 mrg UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
689 1.1 mrg vm_map_unlock(map);
690 1.1 mrg return(KERN_SUCCESS);
691 1.1 mrg }
692 1.1 mrg
693 1.1 mrg /*
694 1.1 mrg * uvm_map_lookup_entry: find map entry at or before an address
695 1.1 mrg *
696 1.1 mrg * => map must at least be read-locked by caller
697 1.1 mrg * => entry is returned in "entry"
698 1.1 mrg * => return value is true if address is in the returned entry
699 1.1 mrg */
700 1.1 mrg
701 1.10 mrg boolean_t
702 1.10 mrg uvm_map_lookup_entry(map, address, entry)
703 1.32 mrg vm_map_t map;
704 1.32 mrg vaddr_t address;
705 1.10 mrg vm_map_entry_t *entry; /* OUT */
706 1.1 mrg {
707 1.32 mrg vm_map_entry_t cur;
708 1.32 mrg vm_map_entry_t last;
709 1.1 mrg UVMHIST_FUNC("uvm_map_lookup_entry");
710 1.1 mrg UVMHIST_CALLED(maphist);
711 1.1 mrg
712 1.1 mrg UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
713 1.10 mrg map, address, entry, 0);
714 1.1 mrg
715 1.1 mrg /*
716 1.10 mrg * start looking either from the head of the
717 1.10 mrg * list, or from the hint.
718 1.1 mrg */
719 1.1 mrg
720 1.1 mrg simple_lock(&map->hint_lock);
721 1.1 mrg cur = map->hint;
722 1.1 mrg simple_unlock(&map->hint_lock);
723 1.1 mrg
724 1.1 mrg if (cur == &map->header)
725 1.1 mrg cur = cur->next;
726 1.1 mrg
727 1.1 mrg UVMCNT_INCR(uvm_mlk_call);
728 1.1 mrg if (address >= cur->start) {
729 1.1 mrg /*
730 1.10 mrg * go from hint to end of list.
731 1.1 mrg *
732 1.10 mrg * but first, make a quick check to see if
733 1.10 mrg * we are already looking at the entry we
734 1.10 mrg * want (which is usually the case).
735 1.10 mrg * note also that we don't need to save the hint
736 1.10 mrg * here... it is the same hint (unless we are
737 1.10 mrg * at the header, in which case the hint didn't
738 1.10 mrg * buy us anything anyway).
739 1.1 mrg */
740 1.1 mrg last = &map->header;
741 1.1 mrg if ((cur != last) && (cur->end > address)) {
742 1.1 mrg UVMCNT_INCR(uvm_mlk_hint);
743 1.1 mrg *entry = cur;
744 1.1 mrg UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
745 1.10 mrg cur, 0, 0, 0);
746 1.10 mrg return (TRUE);
747 1.1 mrg }
748 1.10 mrg } else {
749 1.1 mrg /*
750 1.10 mrg * go from start to hint, *inclusively*
751 1.1 mrg */
752 1.1 mrg last = cur->next;
753 1.1 mrg cur = map->header.next;
754 1.1 mrg }
755 1.1 mrg
756 1.1 mrg /*
757 1.10 mrg * search linearly
758 1.1 mrg */
759 1.1 mrg
760 1.1 mrg while (cur != last) {
761 1.1 mrg if (cur->end > address) {
762 1.1 mrg if (address >= cur->start) {
763 1.1 mrg /*
764 1.10 mrg * save this lookup for future
765 1.10 mrg * hints, and return
766 1.1 mrg */
767 1.1 mrg
768 1.1 mrg *entry = cur;
769 1.1 mrg SAVE_HINT(map, cur);
770 1.1 mrg UVMHIST_LOG(maphist,"<- search got it (0x%x)",
771 1.10 mrg cur, 0, 0, 0);
772 1.10 mrg return (TRUE);
773 1.1 mrg }
774 1.1 mrg break;
775 1.1 mrg }
776 1.1 mrg cur = cur->next;
777 1.1 mrg }
778 1.1 mrg *entry = cur->prev;
779 1.1 mrg SAVE_HINT(map, *entry);
780 1.1 mrg UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
781 1.10 mrg return (FALSE);
782 1.1 mrg }
783 1.1 mrg
784 1.1 mrg
785 1.1 mrg /*
786 1.1 mrg * uvm_map_findspace: find "length" sized space in "map".
787 1.1 mrg *
788 1.1 mrg * => "hint" is a hint about where we want it, unless fixed is true
789 1.1 mrg * (in which case we insist on using "hint").
790 1.1 mrg * => "result" is VA returned
791 1.1 mrg * => uobj/uoffset are to be used to handle VAC alignment, if required
792 1.1 mrg * => caller must at least have read-locked map
793 1.1 mrg * => returns NULL on failure, or pointer to prev. map entry if success
794 1.1 mrg * => note this is a cross between the old vm_map_findspace and vm_map_find
795 1.1 mrg */
796 1.1 mrg
797 1.10 mrg vm_map_entry_t
798 1.10 mrg uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed)
799 1.10 mrg vm_map_t map;
800 1.24 eeh vaddr_t hint;
801 1.24 eeh vsize_t length;
802 1.24 eeh vaddr_t *result; /* OUT */
803 1.10 mrg struct uvm_object *uobj;
804 1.70 kleink voff_t uoffset;
805 1.10 mrg boolean_t fixed;
806 1.1 mrg {
807 1.1 mrg vm_map_entry_t entry, next, tmp;
808 1.24 eeh vaddr_t end;
809 1.1 mrg UVMHIST_FUNC("uvm_map_findspace");
810 1.1 mrg UVMHIST_CALLED(maphist);
811 1.1 mrg
812 1.1 mrg UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)",
813 1.1 mrg map, hint, length, fixed);
814 1.1 mrg
815 1.1 mrg if (hint < map->min_offset) { /* check ranges ... */
816 1.1 mrg if (fixed) {
817 1.1 mrg UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
818 1.1 mrg return(NULL);
819 1.1 mrg }
820 1.1 mrg hint = map->min_offset;
821 1.1 mrg }
822 1.1 mrg if (hint > map->max_offset) {
823 1.1 mrg UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
824 1.1 mrg hint, map->min_offset, map->max_offset, 0);
825 1.1 mrg return(NULL);
826 1.1 mrg }
827 1.1 mrg
828 1.1 mrg /*
829 1.1 mrg * Look for the first possible address; if there's already
830 1.1 mrg * something at this address, we have to start after it.
831 1.1 mrg */
832 1.1 mrg
833 1.1 mrg if (!fixed && hint == map->min_offset) {
834 1.1 mrg if ((entry = map->first_free) != &map->header)
835 1.1 mrg hint = entry->end;
836 1.1 mrg } else {
837 1.1 mrg if (uvm_map_lookup_entry(map, hint, &tmp)) {
838 1.1 mrg /* "hint" address already in use ... */
839 1.1 mrg if (fixed) {
840 1.1 mrg UVMHIST_LOG(maphist,"<- fixed & VA in use",
841 1.10 mrg 0, 0, 0, 0);
842 1.1 mrg return(NULL);
843 1.1 mrg }
844 1.1 mrg hint = tmp->end;
845 1.1 mrg }
846 1.1 mrg entry = tmp;
847 1.1 mrg }
848 1.1 mrg
849 1.1 mrg /*
850 1.1 mrg * Look through the rest of the map, trying to fit a new region in
851 1.1 mrg * the gap between existing regions, or after the very last region.
852 1.1 mrg * note: entry->end = base VA of current gap,
853 1.1 mrg * next->start = VA of end of current gap
854 1.1 mrg */
855 1.1 mrg for (;; hint = (entry = next)->end) {
856 1.1 mrg /*
857 1.1 mrg * Find the end of the proposed new region. Be sure we didn't
858 1.1 mrg * go beyond the end of the map, or wrap around the address;
859 1.1 mrg * if so, we lose. Otherwise, if this is the last entry, or
860 1.1 mrg * if the proposed new region fits before the next entry, we
861 1.1 mrg * win.
862 1.1 mrg */
863 1.1 mrg
864 1.1 mrg #ifdef PMAP_PREFER
865 1.1 mrg /*
866 1.1 mrg * push hint forward as needed to avoid VAC alias problems.
867 1.1 mrg * we only do this if a valid offset is specified.
868 1.1 mrg */
869 1.1 mrg if (!fixed && uoffset != UVM_UNKNOWN_OFFSET)
870 1.1 mrg PMAP_PREFER(uoffset, &hint);
871 1.1 mrg #endif
872 1.1 mrg end = hint + length;
873 1.1 mrg if (end > map->max_offset || end < hint) {
874 1.1 mrg UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
875 1.1 mrg return (NULL);
876 1.1 mrg }
877 1.1 mrg next = entry->next;
878 1.1 mrg if (next == &map->header || next->start >= end)
879 1.1 mrg break;
880 1.1 mrg if (fixed) {
881 1.1 mrg UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
882 1.1 mrg return(NULL); /* only one shot at it ... */
883 1.1 mrg }
884 1.1 mrg }
885 1.1 mrg SAVE_HINT(map, entry);
886 1.1 mrg *result = hint;
887 1.1 mrg UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0);
888 1.1 mrg return (entry);
889 1.1 mrg }
890 1.1 mrg
891 1.1 mrg /*
892 1.1 mrg * U N M A P - m a i n h e l p e r f u n c t i o n s
893 1.1 mrg */
894 1.1 mrg
895 1.1 mrg /*
896 1.1 mrg * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
897 1.1 mrg *
898 1.1 mrg * => caller must check alignment and size
899 1.1 mrg * => map must be locked by caller
900 1.1 mrg * => we return a list of map entries that we've remove from the map
901 1.1 mrg * in "entry_list"
902 1.1 mrg */
903 1.1 mrg
904 1.10 mrg int
905 1.29 chuck uvm_unmap_remove(map, start, end, entry_list)
906 1.10 mrg vm_map_t map;
907 1.24 eeh vaddr_t start,end;
908 1.10 mrg vm_map_entry_t *entry_list; /* OUT */
909 1.10 mrg {
910 1.10 mrg vm_map_entry_t entry, first_entry, next;
911 1.24 eeh vaddr_t len;
912 1.10 mrg UVMHIST_FUNC("uvm_unmap_remove");
913 1.10 mrg UVMHIST_CALLED(maphist);
914 1.10 mrg
915 1.10 mrg UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
916 1.10 mrg map, start, end, 0);
917 1.10 mrg
918 1.10 mrg VM_MAP_RANGE_CHECK(map, start, end);
919 1.10 mrg
920 1.10 mrg /*
921 1.10 mrg * find first entry
922 1.10 mrg */
923 1.10 mrg if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
924 1.29 chuck /* clip and go... */
925 1.10 mrg entry = first_entry;
926 1.10 mrg UVM_MAP_CLIP_START(map, entry, start);
927 1.10 mrg /* critical! prevents stale hint */
928 1.10 mrg SAVE_HINT(map, entry->prev);
929 1.10 mrg
930 1.10 mrg } else {
931 1.10 mrg entry = first_entry->next;
932 1.10 mrg }
933 1.10 mrg
934 1.10 mrg /*
935 1.10 mrg * Save the free space hint
936 1.10 mrg */
937 1.10 mrg
938 1.10 mrg if (map->first_free->start >= start)
939 1.10 mrg map->first_free = entry->prev;
940 1.10 mrg
941 1.10 mrg /*
942 1.10 mrg * note: we now re-use first_entry for a different task. we remove
943 1.10 mrg * a number of map entries from the map and save them in a linked
944 1.10 mrg * list headed by "first_entry". once we remove them from the map
945 1.10 mrg * the caller should unlock the map and drop the references to the
946 1.10 mrg * backing objects [c.f. uvm_unmap_detach]. the object is to
947 1.10 mrg * seperate unmapping from reference dropping. why?
948 1.10 mrg * [1] the map has to be locked for unmapping
949 1.10 mrg * [2] the map need not be locked for reference dropping
950 1.10 mrg * [3] dropping references may trigger pager I/O, and if we hit
951 1.10 mrg * a pager that does synchronous I/O we may have to wait for it.
952 1.10 mrg * [4] we would like all waiting for I/O to occur with maps unlocked
953 1.10 mrg * so that we don't block other threads.
954 1.10 mrg */
955 1.10 mrg first_entry = NULL;
956 1.10 mrg *entry_list = NULL; /* to be safe */
957 1.10 mrg
958 1.10 mrg /*
959 1.10 mrg * break up the area into map entry sized regions and unmap. note
960 1.10 mrg * that all mappings have to be removed before we can even consider
961 1.10 mrg * dropping references to amaps or VM objects (otherwise we could end
962 1.10 mrg * up with a mapping to a page on the free list which would be very bad)
963 1.10 mrg */
964 1.10 mrg
965 1.10 mrg while ((entry != &map->header) && (entry->start < end)) {
966 1.10 mrg
967 1.10 mrg UVM_MAP_CLIP_END(map, entry, end);
968 1.10 mrg next = entry->next;
969 1.10 mrg len = entry->end - entry->start;
970 1.10 mrg
971 1.10 mrg /*
972 1.10 mrg * unwire before removing addresses from the pmap; otherwise
973 1.10 mrg * unwiring will put the entries back into the pmap (XXX).
974 1.10 mrg */
975 1.1 mrg
976 1.55 thorpej if (VM_MAPENT_ISWIRED(entry))
977 1.10 mrg uvm_map_entry_unwire(map, entry);
978 1.10 mrg
979 1.10 mrg /*
980 1.10 mrg * special case: handle mappings to anonymous kernel objects.
981 1.10 mrg * we want to free these pages right away...
982 1.10 mrg */
983 1.10 mrg if (UVM_ET_ISOBJ(entry) &&
984 1.42 thorpej UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
985 1.1 mrg #ifdef DIAGNOSTIC
986 1.10 mrg if (vm_map_pmap(map) != pmap_kernel())
987 1.42 thorpej panic("uvm_unmap_remove: kernel object "
988 1.42 thorpej "mapped by non-kernel map");
989 1.1 mrg #endif
990 1.1 mrg
991 1.10 mrg /*
992 1.10 mrg * note: kernel object mappings are currently used in
993 1.10 mrg * two ways:
994 1.10 mrg * [1] "normal" mappings of pages in the kernel object
995 1.10 mrg * [2] uvm_km_valloc'd allocations in which we
996 1.10 mrg * pmap_enter in some non-kernel-object page
997 1.10 mrg * (e.g. vmapbuf).
998 1.10 mrg *
999 1.10 mrg * for case [1], we need to remove the mapping from
1000 1.10 mrg * the pmap and then remove the page from the kernel
1001 1.10 mrg * object (because, once pages in a kernel object are
1002 1.10 mrg * unmapped they are no longer needed, unlike, say,
1003 1.10 mrg * a vnode where you might want the data to persist
1004 1.10 mrg * until flushed out of a queue).
1005 1.10 mrg *
1006 1.10 mrg * for case [2], we need to remove the mapping from
1007 1.10 mrg * the pmap. there shouldn't be any pages at the
1008 1.10 mrg * specified offset in the kernel object [but it
1009 1.10 mrg * doesn't hurt to call uvm_km_pgremove just to be
1010 1.10 mrg * safe?]
1011 1.10 mrg *
1012 1.10 mrg * uvm_km_pgremove currently does the following:
1013 1.10 mrg * for pages in the kernel object in range:
1014 1.43 thorpej * - drops the swap slot
1015 1.10 mrg * - uvm_pagefree the page
1016 1.10 mrg *
1017 1.43 thorpej * note there is version of uvm_km_pgremove() that
1018 1.43 thorpej * is used for "intrsafe" objects.
1019 1.10 mrg */
1020 1.10 mrg
1021 1.10 mrg /*
1022 1.43 thorpej * remove mappings from pmap and drop the pages
1023 1.43 thorpej * from the object. offsets are always relative
1024 1.43 thorpej * to vm_map_min(kernel_map).
1025 1.10 mrg */
1026 1.43 thorpej if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) {
1027 1.43 thorpej pmap_kremove(entry->start, len);
1028 1.43 thorpej uvm_km_pgremove_intrsafe(entry->object.uvm_obj,
1029 1.43 thorpej entry->start - vm_map_min(kernel_map),
1030 1.43 thorpej entry->end - vm_map_min(kernel_map));
1031 1.43 thorpej } else {
1032 1.43 thorpej pmap_remove(pmap_kernel(), entry->start,
1033 1.43 thorpej entry->start + len);
1034 1.43 thorpej uvm_km_pgremove(entry->object.uvm_obj,
1035 1.43 thorpej entry->start - vm_map_min(kernel_map),
1036 1.43 thorpej entry->end - vm_map_min(kernel_map));
1037 1.43 thorpej }
1038 1.10 mrg
1039 1.10 mrg /*
1040 1.10 mrg * null out kernel_object reference, we've just
1041 1.10 mrg * dropped it
1042 1.10 mrg */
1043 1.10 mrg entry->etype &= ~UVM_ET_OBJ;
1044 1.10 mrg entry->object.uvm_obj = NULL; /* to be safe */
1045 1.10 mrg
1046 1.10 mrg } else {
1047 1.29 chuck /*
1048 1.29 chuck * remove mappings the standard way.
1049 1.29 chuck */
1050 1.29 chuck pmap_remove(map->pmap, entry->start, entry->end);
1051 1.10 mrg }
1052 1.10 mrg
1053 1.10 mrg /*
1054 1.29 chuck * remove entry from map and put it on our list of entries
1055 1.29 chuck * that we've nuked. then go do next entry.
1056 1.10 mrg */
1057 1.10 mrg UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0);
1058 1.10 mrg uvm_map_entry_unlink(map, entry);
1059 1.10 mrg map->size -= len;
1060 1.10 mrg entry->next = first_entry;
1061 1.10 mrg first_entry = entry;
1062 1.10 mrg entry = next; /* next entry, please */
1063 1.10 mrg }
1064 1.10 mrg
1065 1.10 mrg /*
1066 1.10 mrg * now we've cleaned up the map and are ready for the caller to drop
1067 1.10 mrg * references to the mapped objects.
1068 1.10 mrg */
1069 1.10 mrg
1070 1.10 mrg *entry_list = first_entry;
1071 1.10 mrg UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1072 1.10 mrg return(KERN_SUCCESS);
1073 1.1 mrg }
1074 1.1 mrg
1075 1.1 mrg /*
1076 1.1 mrg * uvm_unmap_detach: drop references in a chain of map entries
1077 1.1 mrg *
1078 1.1 mrg * => we will free the map entries as we traverse the list.
1079 1.1 mrg */
1080 1.1 mrg
1081 1.10 mrg void
1082 1.10 mrg uvm_unmap_detach(first_entry, amap_unref_flags)
1083 1.10 mrg vm_map_entry_t first_entry;
1084 1.10 mrg int amap_unref_flags;
1085 1.1 mrg {
1086 1.10 mrg vm_map_entry_t next_entry;
1087 1.10 mrg UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1088 1.1 mrg
1089 1.10 mrg while (first_entry) {
1090 1.1 mrg
1091 1.1 mrg #ifdef DIAGNOSTIC
1092 1.10 mrg /*
1093 1.10 mrg * sanity check
1094 1.10 mrg */
1095 1.10 mrg /* was part of vm_map_entry_delete() */
1096 1.55 thorpej if (VM_MAPENT_ISWIRED(first_entry))
1097 1.10 mrg panic("unmap: still wired!");
1098 1.1 mrg #endif
1099 1.1 mrg
1100 1.10 mrg UVMHIST_LOG(maphist,
1101 1.29 chuck " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1102 1.29 chuck first_entry, first_entry->aref.ar_amap,
1103 1.29 chuck first_entry->object.uvm_obj,
1104 1.29 chuck UVM_ET_ISSUBMAP(first_entry));
1105 1.1 mrg
1106 1.10 mrg /*
1107 1.10 mrg * drop reference to amap, if we've got one
1108 1.10 mrg */
1109 1.10 mrg
1110 1.10 mrg if (first_entry->aref.ar_amap)
1111 1.10 mrg amap_unref(first_entry, amap_unref_flags);
1112 1.10 mrg
1113 1.10 mrg /*
1114 1.10 mrg * drop reference to our backing object, if we've got one
1115 1.10 mrg */
1116 1.10 mrg
1117 1.29 chuck if (UVM_ET_ISSUBMAP(first_entry)) {
1118 1.29 chuck /* ... unlikely to happen, but play it safe */
1119 1.29 chuck uvm_map_deallocate(first_entry->object.sub_map);
1120 1.10 mrg } else {
1121 1.10 mrg if (UVM_ET_ISOBJ(first_entry) &&
1122 1.10 mrg first_entry->object.uvm_obj->pgops->pgo_detach)
1123 1.10 mrg first_entry->object.uvm_obj->pgops->
1124 1.10 mrg pgo_detach(first_entry->object.uvm_obj);
1125 1.10 mrg }
1126 1.10 mrg
1127 1.10 mrg /*
1128 1.10 mrg * next entry
1129 1.10 mrg */
1130 1.10 mrg next_entry = first_entry->next;
1131 1.10 mrg uvm_mapent_free(first_entry);
1132 1.10 mrg first_entry = next_entry;
1133 1.10 mrg }
1134 1.10 mrg
1135 1.10 mrg /*
1136 1.10 mrg * done!
1137 1.10 mrg */
1138 1.10 mrg UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1139 1.10 mrg return;
1140 1.1 mrg }
1141 1.1 mrg
1142 1.1 mrg /*
1143 1.1 mrg * E X T R A C T I O N F U N C T I O N S
1144 1.1 mrg */
1145 1.1 mrg
1146 1.1 mrg /*
1147 1.1 mrg * uvm_map_reserve: reserve space in a vm_map for future use.
1148 1.1 mrg *
1149 1.1 mrg * => we reserve space in a map by putting a dummy map entry in the
1150 1.1 mrg * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1151 1.1 mrg * => map should be unlocked (we will write lock it)
1152 1.1 mrg * => we return true if we were able to reserve space
1153 1.1 mrg * => XXXCDC: should be inline?
1154 1.1 mrg */
1155 1.1 mrg
1156 1.10 mrg int
1157 1.10 mrg uvm_map_reserve(map, size, offset, raddr)
1158 1.10 mrg vm_map_t map;
1159 1.24 eeh vsize_t size;
1160 1.24 eeh vaddr_t offset; /* hint for pmap_prefer */
1161 1.24 eeh vaddr_t *raddr; /* OUT: reserved VA */
1162 1.1 mrg {
1163 1.10 mrg UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1164 1.1 mrg
1165 1.10 mrg UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1166 1.1 mrg map,size,offset,raddr);
1167 1.1 mrg
1168 1.10 mrg size = round_page(size);
1169 1.10 mrg if (*raddr < vm_map_min(map))
1170 1.10 mrg *raddr = vm_map_min(map); /* hint */
1171 1.1 mrg
1172 1.10 mrg /*
1173 1.10 mrg * reserve some virtual space.
1174 1.10 mrg */
1175 1.1 mrg
1176 1.10 mrg if (uvm_map(map, raddr, size, NULL, offset,
1177 1.10 mrg UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1178 1.10 mrg UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) {
1179 1.10 mrg UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1180 1.10 mrg return (FALSE);
1181 1.10 mrg }
1182 1.10 mrg
1183 1.10 mrg UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1184 1.10 mrg return (TRUE);
1185 1.1 mrg }
1186 1.1 mrg
1187 1.1 mrg /*
1188 1.1 mrg * uvm_map_replace: replace a reserved (blank) area of memory with
1189 1.1 mrg * real mappings.
1190 1.1 mrg *
1191 1.1 mrg * => caller must WRITE-LOCK the map
1192 1.1 mrg * => we return TRUE if replacement was a success
1193 1.1 mrg * => we expect the newents chain to have nnewents entrys on it and
1194 1.1 mrg * we expect newents->prev to point to the last entry on the list
1195 1.1 mrg * => note newents is allowed to be NULL
1196 1.1 mrg */
1197 1.1 mrg
1198 1.10 mrg int
1199 1.10 mrg uvm_map_replace(map, start, end, newents, nnewents)
1200 1.10 mrg struct vm_map *map;
1201 1.24 eeh vaddr_t start, end;
1202 1.10 mrg vm_map_entry_t newents;
1203 1.10 mrg int nnewents;
1204 1.10 mrg {
1205 1.10 mrg vm_map_entry_t oldent, last;
1206 1.10 mrg UVMHIST_FUNC("uvm_map_replace");
1207 1.10 mrg UVMHIST_CALLED(maphist);
1208 1.1 mrg
1209 1.10 mrg /*
1210 1.10 mrg * first find the blank map entry at the specified address
1211 1.10 mrg */
1212 1.10 mrg
1213 1.10 mrg if (!uvm_map_lookup_entry(map, start, &oldent)) {
1214 1.10 mrg return(FALSE);
1215 1.10 mrg }
1216 1.10 mrg
1217 1.10 mrg /*
1218 1.10 mrg * check to make sure we have a proper blank entry
1219 1.10 mrg */
1220 1.1 mrg
1221 1.10 mrg if (oldent->start != start || oldent->end != end ||
1222 1.10 mrg oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1223 1.10 mrg return (FALSE);
1224 1.10 mrg }
1225 1.1 mrg
1226 1.1 mrg #ifdef DIAGNOSTIC
1227 1.10 mrg /*
1228 1.10 mrg * sanity check the newents chain
1229 1.10 mrg */
1230 1.10 mrg {
1231 1.10 mrg vm_map_entry_t tmpent = newents;
1232 1.10 mrg int nent = 0;
1233 1.24 eeh vaddr_t cur = start;
1234 1.10 mrg
1235 1.10 mrg while (tmpent) {
1236 1.10 mrg nent++;
1237 1.10 mrg if (tmpent->start < cur)
1238 1.10 mrg panic("uvm_map_replace1");
1239 1.10 mrg if (tmpent->start > tmpent->end || tmpent->end > end) {
1240 1.10 mrg printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1241 1.10 mrg tmpent->start, tmpent->end, end);
1242 1.10 mrg panic("uvm_map_replace2");
1243 1.10 mrg }
1244 1.10 mrg cur = tmpent->end;
1245 1.10 mrg if (tmpent->next) {
1246 1.10 mrg if (tmpent->next->prev != tmpent)
1247 1.10 mrg panic("uvm_map_replace3");
1248 1.10 mrg } else {
1249 1.10 mrg if (newents->prev != tmpent)
1250 1.10 mrg panic("uvm_map_replace4");
1251 1.10 mrg }
1252 1.10 mrg tmpent = tmpent->next;
1253 1.10 mrg }
1254 1.10 mrg if (nent != nnewents)
1255 1.10 mrg panic("uvm_map_replace5");
1256 1.10 mrg }
1257 1.10 mrg #endif
1258 1.10 mrg
1259 1.10 mrg /*
1260 1.10 mrg * map entry is a valid blank! replace it. (this does all the
1261 1.10 mrg * work of map entry link/unlink...).
1262 1.10 mrg */
1263 1.10 mrg
1264 1.10 mrg if (newents) {
1265 1.10 mrg
1266 1.10 mrg last = newents->prev; /* we expect this */
1267 1.10 mrg
1268 1.10 mrg /* critical: flush stale hints out of map */
1269 1.10 mrg SAVE_HINT(map, newents);
1270 1.10 mrg if (map->first_free == oldent)
1271 1.10 mrg map->first_free = last;
1272 1.10 mrg
1273 1.10 mrg last->next = oldent->next;
1274 1.10 mrg last->next->prev = last;
1275 1.10 mrg newents->prev = oldent->prev;
1276 1.10 mrg newents->prev->next = newents;
1277 1.10 mrg map->nentries = map->nentries + (nnewents - 1);
1278 1.10 mrg
1279 1.10 mrg } else {
1280 1.10 mrg
1281 1.10 mrg /* critical: flush stale hints out of map */
1282 1.10 mrg SAVE_HINT(map, oldent->prev);
1283 1.10 mrg if (map->first_free == oldent)
1284 1.10 mrg map->first_free = oldent->prev;
1285 1.10 mrg
1286 1.10 mrg /* NULL list of new entries: just remove the old one */
1287 1.10 mrg uvm_map_entry_unlink(map, oldent);
1288 1.10 mrg }
1289 1.10 mrg
1290 1.10 mrg
1291 1.10 mrg /*
1292 1.10 mrg * now we can free the old blank entry, unlock the map and return.
1293 1.10 mrg */
1294 1.1 mrg
1295 1.10 mrg uvm_mapent_free(oldent);
1296 1.10 mrg return(TRUE);
1297 1.1 mrg }
1298 1.1 mrg
1299 1.1 mrg /*
1300 1.1 mrg * uvm_map_extract: extract a mapping from a map and put it somewhere
1301 1.1 mrg * (maybe removing the old mapping)
1302 1.1 mrg *
1303 1.1 mrg * => maps should be unlocked (we will write lock them)
1304 1.1 mrg * => returns 0 on success, error code otherwise
1305 1.1 mrg * => start must be page aligned
1306 1.1 mrg * => len must be page sized
1307 1.1 mrg * => flags:
1308 1.1 mrg * UVM_EXTRACT_REMOVE: remove mappings from srcmap
1309 1.1 mrg * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1310 1.1 mrg * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1311 1.1 mrg * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1312 1.1 mrg * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1313 1.1 mrg * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1314 1.1 mrg * be used from within the kernel in a kernel level map <<<
1315 1.1 mrg */
1316 1.1 mrg
1317 1.10 mrg int
1318 1.10 mrg uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1319 1.10 mrg vm_map_t srcmap, dstmap;
1320 1.24 eeh vaddr_t start, *dstaddrp;
1321 1.24 eeh vsize_t len;
1322 1.10 mrg int flags;
1323 1.10 mrg {
1324 1.24 eeh vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1325 1.10 mrg oldstart;
1326 1.10 mrg vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry;
1327 1.20 chuck vm_map_entry_t oldentry;
1328 1.24 eeh vsize_t elen;
1329 1.10 mrg int nchain, error, copy_ok;
1330 1.10 mrg UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1331 1.10 mrg UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1332 1.10 mrg len,0);
1333 1.10 mrg UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1334 1.10 mrg
1335 1.10 mrg #ifdef DIAGNOSTIC
1336 1.10 mrg /*
1337 1.10 mrg * step 0: sanity check: start must be on a page boundary, length
1338 1.10 mrg * must be page sized. can't ask for CONTIG/QREF if you asked for
1339 1.10 mrg * REMOVE.
1340 1.10 mrg */
1341 1.10 mrg if ((start & PAGE_MASK) || (len & PAGE_MASK))
1342 1.10 mrg panic("uvm_map_extract1");
1343 1.10 mrg if (flags & UVM_EXTRACT_REMOVE)
1344 1.10 mrg if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF))
1345 1.10 mrg panic("uvm_map_extract2");
1346 1.10 mrg #endif
1347 1.10 mrg
1348 1.10 mrg
1349 1.10 mrg /*
1350 1.10 mrg * step 1: reserve space in the target map for the extracted area
1351 1.10 mrg */
1352 1.10 mrg
1353 1.10 mrg dstaddr = *dstaddrp;
1354 1.10 mrg if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE)
1355 1.10 mrg return(ENOMEM);
1356 1.10 mrg *dstaddrp = dstaddr; /* pass address back to caller */
1357 1.10 mrg UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0);
1358 1.10 mrg
1359 1.10 mrg
1360 1.10 mrg /*
1361 1.10 mrg * step 2: setup for the extraction process loop by init'ing the
1362 1.10 mrg * map entry chain, locking src map, and looking up the first useful
1363 1.10 mrg * entry in the map.
1364 1.10 mrg */
1365 1.1 mrg
1366 1.10 mrg end = start + len;
1367 1.10 mrg newend = dstaddr + len;
1368 1.10 mrg chain = endchain = NULL;
1369 1.10 mrg nchain = 0;
1370 1.10 mrg vm_map_lock(srcmap);
1371 1.10 mrg
1372 1.10 mrg if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1373 1.10 mrg
1374 1.10 mrg /* "start" is within an entry */
1375 1.10 mrg if (flags & UVM_EXTRACT_QREF) {
1376 1.10 mrg /*
1377 1.10 mrg * for quick references we don't clip the entry, so
1378 1.10 mrg * the entry may map space "before" the starting
1379 1.10 mrg * virtual address... this is the "fudge" factor
1380 1.10 mrg * (which can be non-zero only the first time
1381 1.10 mrg * through the "while" loop in step 3).
1382 1.10 mrg */
1383 1.10 mrg fudge = start - entry->start;
1384 1.10 mrg } else {
1385 1.10 mrg /*
1386 1.10 mrg * normal reference: we clip the map to fit (thus
1387 1.10 mrg * fudge is zero)
1388 1.10 mrg */
1389 1.10 mrg UVM_MAP_CLIP_START(srcmap, entry, start);
1390 1.10 mrg SAVE_HINT(srcmap, entry->prev);
1391 1.10 mrg fudge = 0;
1392 1.10 mrg }
1393 1.1 mrg
1394 1.10 mrg } else {
1395 1.10 mrg
1396 1.10 mrg /* "start" is not within an entry ... skip to next entry */
1397 1.10 mrg if (flags & UVM_EXTRACT_CONTIG) {
1398 1.10 mrg error = EINVAL;
1399 1.10 mrg goto bad; /* definite hole here ... */
1400 1.10 mrg }
1401 1.1 mrg
1402 1.10 mrg entry = entry->next;
1403 1.10 mrg fudge = 0;
1404 1.10 mrg }
1405 1.10 mrg /* save values from srcmap for step 6 */
1406 1.10 mrg orig_entry = entry;
1407 1.10 mrg orig_fudge = fudge;
1408 1.1 mrg
1409 1.1 mrg
1410 1.10 mrg /*
1411 1.10 mrg * step 3: now start looping through the map entries, extracting
1412 1.10 mrg * as we go.
1413 1.10 mrg */
1414 1.1 mrg
1415 1.10 mrg while (entry->start < end && entry != &srcmap->header) {
1416 1.10 mrg
1417 1.10 mrg /* if we are not doing a quick reference, clip it */
1418 1.10 mrg if ((flags & UVM_EXTRACT_QREF) == 0)
1419 1.10 mrg UVM_MAP_CLIP_END(srcmap, entry, end);
1420 1.10 mrg
1421 1.10 mrg /* clear needs_copy (allow chunking) */
1422 1.10 mrg if (UVM_ET_ISNEEDSCOPY(entry)) {
1423 1.10 mrg if (fudge)
1424 1.10 mrg oldstart = entry->start;
1425 1.10 mrg else
1426 1.10 mrg oldstart = 0; /* XXX: gcc */
1427 1.10 mrg amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1428 1.10 mrg if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */
1429 1.10 mrg error = ENOMEM;
1430 1.10 mrg goto bad;
1431 1.10 mrg }
1432 1.10 mrg /* amap_copy could clip (during chunk)! update fudge */
1433 1.10 mrg if (fudge) {
1434 1.10 mrg fudge = fudge - (entry->start - oldstart);
1435 1.10 mrg orig_fudge = fudge;
1436 1.10 mrg }
1437 1.10 mrg }
1438 1.1 mrg
1439 1.10 mrg /* calculate the offset of this from "start" */
1440 1.10 mrg oldoffset = (entry->start + fudge) - start;
1441 1.1 mrg
1442 1.10 mrg /* allocate a new map entry */
1443 1.10 mrg newentry = uvm_mapent_alloc(dstmap);
1444 1.10 mrg if (newentry == NULL) {
1445 1.10 mrg error = ENOMEM;
1446 1.10 mrg goto bad;
1447 1.10 mrg }
1448 1.10 mrg
1449 1.10 mrg /* set up new map entry */
1450 1.10 mrg newentry->next = NULL;
1451 1.10 mrg newentry->prev = endchain;
1452 1.10 mrg newentry->start = dstaddr + oldoffset;
1453 1.10 mrg newentry->end =
1454 1.10 mrg newentry->start + (entry->end - (entry->start + fudge));
1455 1.37 chs if (newentry->end > newend || newentry->end < newentry->start)
1456 1.10 mrg newentry->end = newend;
1457 1.10 mrg newentry->object.uvm_obj = entry->object.uvm_obj;
1458 1.10 mrg if (newentry->object.uvm_obj) {
1459 1.10 mrg if (newentry->object.uvm_obj->pgops->pgo_reference)
1460 1.10 mrg newentry->object.uvm_obj->pgops->
1461 1.10 mrg pgo_reference(newentry->object.uvm_obj);
1462 1.10 mrg newentry->offset = entry->offset + fudge;
1463 1.10 mrg } else {
1464 1.10 mrg newentry->offset = 0;
1465 1.10 mrg }
1466 1.10 mrg newentry->etype = entry->etype;
1467 1.10 mrg newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1468 1.10 mrg entry->max_protection : entry->protection;
1469 1.10 mrg newentry->max_protection = entry->max_protection;
1470 1.10 mrg newentry->inheritance = entry->inheritance;
1471 1.10 mrg newentry->wired_count = 0;
1472 1.10 mrg newentry->aref.ar_amap = entry->aref.ar_amap;
1473 1.10 mrg if (newentry->aref.ar_amap) {
1474 1.34 chuck newentry->aref.ar_pageoff =
1475 1.34 chuck entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1476 1.10 mrg amap_ref(newentry, AMAP_SHARED |
1477 1.10 mrg ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1478 1.10 mrg } else {
1479 1.34 chuck newentry->aref.ar_pageoff = 0;
1480 1.10 mrg }
1481 1.10 mrg newentry->advice = entry->advice;
1482 1.10 mrg
1483 1.10 mrg /* now link it on the chain */
1484 1.10 mrg nchain++;
1485 1.10 mrg if (endchain == NULL) {
1486 1.10 mrg chain = endchain = newentry;
1487 1.10 mrg } else {
1488 1.10 mrg endchain->next = newentry;
1489 1.10 mrg endchain = newentry;
1490 1.10 mrg }
1491 1.10 mrg
1492 1.10 mrg /* end of 'while' loop! */
1493 1.10 mrg if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1494 1.10 mrg (entry->next == &srcmap->header ||
1495 1.10 mrg entry->next->start != entry->end)) {
1496 1.10 mrg error = EINVAL;
1497 1.10 mrg goto bad;
1498 1.10 mrg }
1499 1.10 mrg entry = entry->next;
1500 1.10 mrg fudge = 0;
1501 1.10 mrg }
1502 1.10 mrg
1503 1.10 mrg
1504 1.10 mrg /*
1505 1.10 mrg * step 4: close off chain (in format expected by uvm_map_replace)
1506 1.10 mrg */
1507 1.10 mrg
1508 1.10 mrg if (chain)
1509 1.10 mrg chain->prev = endchain;
1510 1.10 mrg
1511 1.10 mrg
1512 1.10 mrg /*
1513 1.10 mrg * step 5: attempt to lock the dest map so we can pmap_copy.
1514 1.10 mrg * note usage of copy_ok:
1515 1.10 mrg * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1516 1.10 mrg * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1517 1.10 mrg */
1518 1.10 mrg
1519 1.10 mrg if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1520 1.10 mrg
1521 1.10 mrg copy_ok = 1;
1522 1.10 mrg if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1523 1.10 mrg nchain)) {
1524 1.10 mrg if (srcmap != dstmap)
1525 1.10 mrg vm_map_unlock(dstmap);
1526 1.10 mrg error = EIO;
1527 1.10 mrg goto bad;
1528 1.10 mrg }
1529 1.10 mrg
1530 1.10 mrg } else {
1531 1.10 mrg
1532 1.10 mrg copy_ok = 0;
1533 1.10 mrg /* replace defered until step 7 */
1534 1.10 mrg
1535 1.10 mrg }
1536 1.10 mrg
1537 1.10 mrg
1538 1.10 mrg /*
1539 1.10 mrg * step 6: traverse the srcmap a second time to do the following:
1540 1.10 mrg * - if we got a lock on the dstmap do pmap_copy
1541 1.10 mrg * - if UVM_EXTRACT_REMOVE remove the entries
1542 1.10 mrg * we make use of orig_entry and orig_fudge (saved in step 2)
1543 1.10 mrg */
1544 1.10 mrg
1545 1.10 mrg if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1546 1.10 mrg
1547 1.10 mrg /* purge possible stale hints from srcmap */
1548 1.10 mrg if (flags & UVM_EXTRACT_REMOVE) {
1549 1.10 mrg SAVE_HINT(srcmap, orig_entry->prev);
1550 1.10 mrg if (srcmap->first_free->start >= start)
1551 1.10 mrg srcmap->first_free = orig_entry->prev;
1552 1.10 mrg }
1553 1.10 mrg
1554 1.10 mrg entry = orig_entry;
1555 1.10 mrg fudge = orig_fudge;
1556 1.10 mrg deadentry = NULL; /* for UVM_EXTRACT_REMOVE */
1557 1.10 mrg
1558 1.10 mrg while (entry->start < end && entry != &srcmap->header) {
1559 1.10 mrg
1560 1.10 mrg if (copy_ok) {
1561 1.74 thorpej oldoffset = (entry->start + fudge) - start;
1562 1.74 thorpej elen = min(end, entry->end) -
1563 1.74 thorpej (entry->start + fudge);
1564 1.74 thorpej pmap_copy(dstmap->pmap, srcmap->pmap,
1565 1.74 thorpej dstaddr + oldoffset, elen,
1566 1.74 thorpej entry->start + fudge);
1567 1.10 mrg }
1568 1.10 mrg
1569 1.74 thorpej /* we advance "entry" in the following if statement */
1570 1.10 mrg if (flags & UVM_EXTRACT_REMOVE) {
1571 1.20 chuck pmap_remove(srcmap->pmap, entry->start,
1572 1.20 chuck entry->end);
1573 1.20 chuck oldentry = entry; /* save entry */
1574 1.20 chuck entry = entry->next; /* advance */
1575 1.20 chuck uvm_map_entry_unlink(srcmap, oldentry);
1576 1.20 chuck /* add to dead list */
1577 1.20 chuck oldentry->next = deadentry;
1578 1.20 chuck deadentry = oldentry;
1579 1.20 chuck } else {
1580 1.20 chuck entry = entry->next; /* advance */
1581 1.10 mrg }
1582 1.10 mrg
1583 1.10 mrg /* end of 'while' loop */
1584 1.10 mrg fudge = 0;
1585 1.10 mrg }
1586 1.10 mrg
1587 1.10 mrg /*
1588 1.10 mrg * unlock dstmap. we will dispose of deadentry in
1589 1.10 mrg * step 7 if needed
1590 1.10 mrg */
1591 1.10 mrg if (copy_ok && srcmap != dstmap)
1592 1.10 mrg vm_map_unlock(dstmap);
1593 1.10 mrg
1594 1.10 mrg }
1595 1.10 mrg else
1596 1.10 mrg deadentry = NULL; /* XXX: gcc */
1597 1.10 mrg
1598 1.10 mrg /*
1599 1.10 mrg * step 7: we are done with the source map, unlock. if copy_ok
1600 1.10 mrg * is 0 then we have not replaced the dummy mapping in dstmap yet
1601 1.10 mrg * and we need to do so now.
1602 1.10 mrg */
1603 1.10 mrg
1604 1.10 mrg vm_map_unlock(srcmap);
1605 1.10 mrg if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1606 1.10 mrg uvm_unmap_detach(deadentry, 0); /* dispose of old entries */
1607 1.10 mrg
1608 1.10 mrg /* now do the replacement if we didn't do it in step 5 */
1609 1.10 mrg if (copy_ok == 0) {
1610 1.10 mrg vm_map_lock(dstmap);
1611 1.10 mrg error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1612 1.10 mrg nchain);
1613 1.10 mrg vm_map_unlock(dstmap);
1614 1.10 mrg
1615 1.10 mrg if (error == FALSE) {
1616 1.10 mrg error = EIO;
1617 1.10 mrg goto bad2;
1618 1.10 mrg }
1619 1.10 mrg }
1620 1.10 mrg
1621 1.10 mrg /*
1622 1.10 mrg * done!
1623 1.10 mrg */
1624 1.10 mrg return(0);
1625 1.10 mrg
1626 1.10 mrg /*
1627 1.10 mrg * bad: failure recovery
1628 1.10 mrg */
1629 1.10 mrg bad:
1630 1.10 mrg vm_map_unlock(srcmap);
1631 1.10 mrg bad2: /* src already unlocked */
1632 1.10 mrg if (chain)
1633 1.10 mrg uvm_unmap_detach(chain,
1634 1.10 mrg (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1635 1.29 chuck uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */
1636 1.10 mrg return(error);
1637 1.10 mrg }
1638 1.10 mrg
1639 1.10 mrg /* end of extraction functions */
1640 1.1 mrg
1641 1.1 mrg /*
1642 1.1 mrg * uvm_map_submap: punch down part of a map into a submap
1643 1.1 mrg *
1644 1.1 mrg * => only the kernel_map is allowed to be submapped
1645 1.1 mrg * => the purpose of submapping is to break up the locking granularity
1646 1.1 mrg * of a larger map
1647 1.1 mrg * => the range specified must have been mapped previously with a uvm_map()
1648 1.1 mrg * call [with uobj==NULL] to create a blank map entry in the main map.
1649 1.1 mrg * [And it had better still be blank!]
1650 1.1 mrg * => maps which contain submaps should never be copied or forked.
1651 1.1 mrg * => to remove a submap, use uvm_unmap() on the main map
1652 1.1 mrg * and then uvm_map_deallocate() the submap.
1653 1.1 mrg * => main map must be unlocked.
1654 1.1 mrg * => submap must have been init'd and have a zero reference count.
1655 1.1 mrg * [need not be locked as we don't actually reference it]
1656 1.1 mrg */
1657 1.10 mrg
1658 1.10 mrg int
1659 1.10 mrg uvm_map_submap(map, start, end, submap)
1660 1.10 mrg vm_map_t map, submap;
1661 1.24 eeh vaddr_t start, end;
1662 1.10 mrg {
1663 1.10 mrg vm_map_entry_t entry;
1664 1.10 mrg int result;
1665 1.10 mrg UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist);
1666 1.1 mrg
1667 1.10 mrg vm_map_lock(map);
1668 1.1 mrg
1669 1.10 mrg VM_MAP_RANGE_CHECK(map, start, end);
1670 1.10 mrg
1671 1.10 mrg if (uvm_map_lookup_entry(map, start, &entry)) {
1672 1.10 mrg UVM_MAP_CLIP_START(map, entry, start);
1673 1.10 mrg UVM_MAP_CLIP_END(map, entry, end); /* to be safe */
1674 1.10 mrg }
1675 1.10 mrg else {
1676 1.10 mrg entry = NULL;
1677 1.10 mrg }
1678 1.1 mrg
1679 1.10 mrg if (entry != NULL &&
1680 1.10 mrg entry->start == start && entry->end == end &&
1681 1.10 mrg entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1682 1.10 mrg !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1683 1.10 mrg
1684 1.10 mrg /*
1685 1.10 mrg * doit!
1686 1.10 mrg */
1687 1.29 chuck entry->etype |= UVM_ET_SUBMAP;
1688 1.10 mrg entry->object.sub_map = submap;
1689 1.10 mrg entry->offset = 0;
1690 1.10 mrg uvm_map_reference(submap);
1691 1.10 mrg result = KERN_SUCCESS;
1692 1.10 mrg } else {
1693 1.10 mrg result = KERN_INVALID_ARGUMENT;
1694 1.10 mrg }
1695 1.10 mrg vm_map_unlock(map);
1696 1.1 mrg
1697 1.10 mrg return(result);
1698 1.1 mrg }
1699 1.1 mrg
1700 1.1 mrg
1701 1.1 mrg /*
1702 1.1 mrg * uvm_map_protect: change map protection
1703 1.1 mrg *
1704 1.1 mrg * => set_max means set max_protection.
1705 1.1 mrg * => map must be unlocked.
1706 1.1 mrg */
1707 1.1 mrg
1708 1.36 mycroft #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \
1709 1.36 mycroft ~VM_PROT_WRITE : VM_PROT_ALL)
1710 1.1 mrg #define max(a,b) ((a) > (b) ? (a) : (b))
1711 1.1 mrg
1712 1.10 mrg int
1713 1.10 mrg uvm_map_protect(map, start, end, new_prot, set_max)
1714 1.10 mrg vm_map_t map;
1715 1.24 eeh vaddr_t start, end;
1716 1.10 mrg vm_prot_t new_prot;
1717 1.10 mrg boolean_t set_max;
1718 1.10 mrg {
1719 1.10 mrg vm_map_entry_t current, entry;
1720 1.65 thorpej int rv = KERN_SUCCESS;
1721 1.10 mrg UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1722 1.10 mrg UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1723 1.1 mrg map, start, end, new_prot);
1724 1.10 mrg
1725 1.10 mrg vm_map_lock(map);
1726 1.1 mrg
1727 1.10 mrg VM_MAP_RANGE_CHECK(map, start, end);
1728 1.1 mrg
1729 1.10 mrg if (uvm_map_lookup_entry(map, start, &entry)) {
1730 1.10 mrg UVM_MAP_CLIP_START(map, entry, start);
1731 1.10 mrg } else {
1732 1.10 mrg entry = entry->next;
1733 1.10 mrg }
1734 1.10 mrg
1735 1.1 mrg /*
1736 1.10 mrg * make a first pass to check for protection violations.
1737 1.1 mrg */
1738 1.1 mrg
1739 1.10 mrg current = entry;
1740 1.10 mrg while ((current != &map->header) && (current->start < end)) {
1741 1.65 thorpej if (UVM_ET_ISSUBMAP(current)) {
1742 1.65 thorpej rv = KERN_INVALID_ARGUMENT;
1743 1.65 thorpej goto out;
1744 1.65 thorpej }
1745 1.10 mrg if ((new_prot & current->max_protection) != new_prot) {
1746 1.65 thorpej rv = KERN_PROTECTION_FAILURE;
1747 1.65 thorpej goto out;
1748 1.10 mrg }
1749 1.65 thorpej current = current->next;
1750 1.10 mrg }
1751 1.10 mrg
1752 1.10 mrg /* go back and fix up protections (no need to clip this time). */
1753 1.10 mrg
1754 1.10 mrg current = entry;
1755 1.10 mrg
1756 1.10 mrg while ((current != &map->header) && (current->start < end)) {
1757 1.10 mrg vm_prot_t old_prot;
1758 1.10 mrg
1759 1.10 mrg UVM_MAP_CLIP_END(map, current, end);
1760 1.10 mrg
1761 1.10 mrg old_prot = current->protection;
1762 1.10 mrg if (set_max)
1763 1.10 mrg current->protection =
1764 1.10 mrg (current->max_protection = new_prot) & old_prot;
1765 1.10 mrg else
1766 1.10 mrg current->protection = new_prot;
1767 1.10 mrg
1768 1.10 mrg /*
1769 1.10 mrg * update physical map if necessary. worry about copy-on-write
1770 1.10 mrg * here -- CHECK THIS XXX
1771 1.10 mrg */
1772 1.10 mrg
1773 1.10 mrg if (current->protection != old_prot) {
1774 1.29 chuck /* update pmap! */
1775 1.29 chuck pmap_protect(map->pmap, current->start, current->end,
1776 1.29 chuck current->protection & MASK(entry));
1777 1.65 thorpej }
1778 1.10 mrg
1779 1.65 thorpej /*
1780 1.65 thorpej * If the map is configured to lock any future mappings,
1781 1.65 thorpej * wire this entry now if the old protection was VM_PROT_NONE
1782 1.65 thorpej * and the new protection is not VM_PROT_NONE.
1783 1.65 thorpej */
1784 1.65 thorpej
1785 1.65 thorpej if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
1786 1.65 thorpej VM_MAPENT_ISWIRED(entry) == 0 &&
1787 1.65 thorpej old_prot == VM_PROT_NONE &&
1788 1.65 thorpej new_prot != VM_PROT_NONE) {
1789 1.65 thorpej if (uvm_map_pageable(map, entry->start,
1790 1.65 thorpej entry->end, FALSE,
1791 1.65 thorpej UVM_LK_ENTER|UVM_LK_EXIT) != KERN_SUCCESS) {
1792 1.65 thorpej /*
1793 1.65 thorpej * If locking the entry fails, remember the
1794 1.65 thorpej * error if it's the first one. Note we
1795 1.65 thorpej * still continue setting the protection in
1796 1.65 thorpej * the map, but will return the resource
1797 1.65 thorpej * shortage condition regardless.
1798 1.65 thorpej *
1799 1.65 thorpej * XXX Ignore what the actual error is,
1800 1.65 thorpej * XXX just call it a resource shortage
1801 1.65 thorpej * XXX so that it doesn't get confused
1802 1.65 thorpej * XXX what uvm_map_protect() itself would
1803 1.65 thorpej * XXX normally return.
1804 1.65 thorpej */
1805 1.65 thorpej rv = KERN_RESOURCE_SHORTAGE;
1806 1.65 thorpej }
1807 1.10 mrg }
1808 1.65 thorpej
1809 1.10 mrg current = current->next;
1810 1.10 mrg }
1811 1.10 mrg
1812 1.65 thorpej out:
1813 1.10 mrg vm_map_unlock(map);
1814 1.65 thorpej UVMHIST_LOG(maphist, "<- done, rv=%d",rv,0,0,0);
1815 1.65 thorpej return (rv);
1816 1.1 mrg }
1817 1.1 mrg
1818 1.1 mrg #undef max
1819 1.1 mrg #undef MASK
1820 1.1 mrg
1821 1.1 mrg /*
1822 1.1 mrg * uvm_map_inherit: set inheritance code for range of addrs in map.
1823 1.1 mrg *
1824 1.1 mrg * => map must be unlocked
1825 1.1 mrg * => note that the inherit code is used during a "fork". see fork
1826 1.1 mrg * code for details.
1827 1.1 mrg */
1828 1.1 mrg
1829 1.10 mrg int
1830 1.10 mrg uvm_map_inherit(map, start, end, new_inheritance)
1831 1.10 mrg vm_map_t map;
1832 1.24 eeh vaddr_t start;
1833 1.24 eeh vaddr_t end;
1834 1.10 mrg vm_inherit_t new_inheritance;
1835 1.10 mrg {
1836 1.10 mrg vm_map_entry_t entry, temp_entry;
1837 1.10 mrg UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1838 1.10 mrg UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1839 1.10 mrg map, start, end, new_inheritance);
1840 1.10 mrg
1841 1.10 mrg switch (new_inheritance) {
1842 1.10 mrg case VM_INHERIT_NONE:
1843 1.10 mrg case VM_INHERIT_COPY:
1844 1.10 mrg case VM_INHERIT_SHARE:
1845 1.10 mrg break;
1846 1.10 mrg default:
1847 1.10 mrg UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1848 1.38 mrg return (KERN_INVALID_ARGUMENT);
1849 1.10 mrg }
1850 1.1 mrg
1851 1.10 mrg vm_map_lock(map);
1852 1.10 mrg
1853 1.10 mrg VM_MAP_RANGE_CHECK(map, start, end);
1854 1.10 mrg
1855 1.10 mrg if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1856 1.10 mrg entry = temp_entry;
1857 1.10 mrg UVM_MAP_CLIP_START(map, entry, start);
1858 1.10 mrg } else {
1859 1.10 mrg entry = temp_entry->next;
1860 1.10 mrg }
1861 1.10 mrg
1862 1.10 mrg while ((entry != &map->header) && (entry->start < end)) {
1863 1.10 mrg UVM_MAP_CLIP_END(map, entry, end);
1864 1.10 mrg
1865 1.10 mrg entry->inheritance = new_inheritance;
1866 1.10 mrg
1867 1.10 mrg entry = entry->next;
1868 1.10 mrg }
1869 1.10 mrg
1870 1.10 mrg vm_map_unlock(map);
1871 1.10 mrg UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1872 1.10 mrg return(KERN_SUCCESS);
1873 1.41 mrg }
1874 1.41 mrg
1875 1.41 mrg /*
1876 1.41 mrg * uvm_map_advice: set advice code for range of addrs in map.
1877 1.41 mrg *
1878 1.41 mrg * => map must be unlocked
1879 1.41 mrg */
1880 1.41 mrg
1881 1.41 mrg int
1882 1.41 mrg uvm_map_advice(map, start, end, new_advice)
1883 1.41 mrg vm_map_t map;
1884 1.41 mrg vaddr_t start;
1885 1.41 mrg vaddr_t end;
1886 1.41 mrg int new_advice;
1887 1.41 mrg {
1888 1.41 mrg vm_map_entry_t entry, temp_entry;
1889 1.41 mrg UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
1890 1.41 mrg UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
1891 1.41 mrg map, start, end, new_advice);
1892 1.41 mrg
1893 1.41 mrg vm_map_lock(map);
1894 1.41 mrg
1895 1.41 mrg VM_MAP_RANGE_CHECK(map, start, end);
1896 1.41 mrg
1897 1.41 mrg if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1898 1.41 mrg entry = temp_entry;
1899 1.41 mrg UVM_MAP_CLIP_START(map, entry, start);
1900 1.41 mrg } else {
1901 1.41 mrg entry = temp_entry->next;
1902 1.41 mrg }
1903 1.61 thorpej
1904 1.61 thorpej /*
1905 1.61 thorpej * XXXJRT: disallow holes?
1906 1.61 thorpej */
1907 1.61 thorpej
1908 1.41 mrg while ((entry != &map->header) && (entry->start < end)) {
1909 1.41 mrg UVM_MAP_CLIP_END(map, entry, end);
1910 1.41 mrg
1911 1.41 mrg switch (new_advice) {
1912 1.41 mrg case MADV_NORMAL:
1913 1.41 mrg case MADV_RANDOM:
1914 1.41 mrg case MADV_SEQUENTIAL:
1915 1.41 mrg /* nothing special here */
1916 1.41 mrg break;
1917 1.41 mrg
1918 1.41 mrg default:
1919 1.50 mrg vm_map_unlock(map);
1920 1.41 mrg UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1921 1.41 mrg return (KERN_INVALID_ARGUMENT);
1922 1.41 mrg }
1923 1.41 mrg
1924 1.41 mrg
1925 1.41 mrg entry->advice = new_advice;
1926 1.41 mrg
1927 1.41 mrg entry = entry->next;
1928 1.41 mrg }
1929 1.41 mrg
1930 1.41 mrg vm_map_unlock(map);
1931 1.41 mrg UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1932 1.41 mrg return (KERN_SUCCESS);
1933 1.1 mrg }
1934 1.1 mrg
1935 1.1 mrg /*
1936 1.1 mrg * uvm_map_pageable: sets the pageability of a range in a map.
1937 1.1 mrg *
1938 1.56 thorpej * => wires map entries. should not be used for transient page locking.
1939 1.56 thorpej * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
1940 1.1 mrg * => regions sepcified as not pageable require lock-down (wired) memory
1941 1.1 mrg * and page tables.
1942 1.59 thorpej * => map must never be read-locked
1943 1.59 thorpej * => if islocked is TRUE, map is already write-locked
1944 1.59 thorpej * => we always unlock the map, since we must downgrade to a read-lock
1945 1.59 thorpej * to call uvm_fault_wire()
1946 1.1 mrg * => XXXCDC: check this and try and clean it up.
1947 1.1 mrg */
1948 1.1 mrg
1949 1.19 kleink int
1950 1.64 thorpej uvm_map_pageable(map, start, end, new_pageable, lockflags)
1951 1.11 mrg vm_map_t map;
1952 1.24 eeh vaddr_t start, end;
1953 1.64 thorpej boolean_t new_pageable;
1954 1.64 thorpej int lockflags;
1955 1.1 mrg {
1956 1.54 thorpej vm_map_entry_t entry, start_entry, failed_entry;
1957 1.10 mrg int rv;
1958 1.60 thorpej #ifdef DIAGNOSTIC
1959 1.60 thorpej u_int timestamp_save;
1960 1.60 thorpej #endif
1961 1.10 mrg UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
1962 1.10 mrg UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1963 1.1 mrg map, start, end, new_pageable);
1964 1.1 mrg
1965 1.45 thorpej #ifdef DIAGNOSTIC
1966 1.45 thorpej if ((map->flags & VM_MAP_PAGEABLE) == 0)
1967 1.45 thorpej panic("uvm_map_pageable: map %p not pageable", map);
1968 1.45 thorpej #endif
1969 1.45 thorpej
1970 1.64 thorpej if ((lockflags & UVM_LK_ENTER) == 0)
1971 1.59 thorpej vm_map_lock(map);
1972 1.64 thorpej
1973 1.10 mrg VM_MAP_RANGE_CHECK(map, start, end);
1974 1.10 mrg
1975 1.10 mrg /*
1976 1.10 mrg * only one pageability change may take place at one time, since
1977 1.10 mrg * uvm_fault_wire assumes it will be called only once for each
1978 1.10 mrg * wiring/unwiring. therefore, we have to make sure we're actually
1979 1.10 mrg * changing the pageability for the entire region. we do so before
1980 1.10 mrg * making any changes.
1981 1.10 mrg */
1982 1.10 mrg
1983 1.10 mrg if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1984 1.64 thorpej if ((lockflags & UVM_LK_EXIT) == 0)
1985 1.64 thorpej vm_map_unlock(map);
1986 1.10 mrg
1987 1.10 mrg UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1988 1.10 mrg return (KERN_INVALID_ADDRESS);
1989 1.10 mrg }
1990 1.10 mrg entry = start_entry;
1991 1.10 mrg
1992 1.10 mrg /*
1993 1.10 mrg * handle wiring and unwiring seperately.
1994 1.10 mrg */
1995 1.1 mrg
1996 1.56 thorpej if (new_pageable) { /* unwire */
1997 1.10 mrg UVM_MAP_CLIP_START(map, entry, start);
1998 1.10 mrg /*
1999 1.10 mrg * unwiring. first ensure that the range to be unwired is
2000 1.10 mrg * really wired down and that there are no holes.
2001 1.10 mrg */
2002 1.10 mrg while ((entry != &map->header) && (entry->start < end)) {
2003 1.10 mrg if (entry->wired_count == 0 ||
2004 1.10 mrg (entry->end < end &&
2005 1.55 thorpej (entry->next == &map->header ||
2006 1.55 thorpej entry->next->start > entry->end))) {
2007 1.64 thorpej if ((lockflags & UVM_LK_EXIT) == 0)
2008 1.64 thorpej vm_map_unlock(map);
2009 1.10 mrg UVMHIST_LOG(maphist,
2010 1.10 mrg "<- done (INVALID UNWIRE ARG)",0,0,0,0);
2011 1.10 mrg return (KERN_INVALID_ARGUMENT);
2012 1.10 mrg }
2013 1.10 mrg entry = entry->next;
2014 1.10 mrg }
2015 1.10 mrg
2016 1.10 mrg /*
2017 1.56 thorpej * POSIX 1003.1b - a single munlock call unlocks a region,
2018 1.56 thorpej * regardless of the number of mlock calls made on that
2019 1.56 thorpej * region.
2020 1.10 mrg */
2021 1.10 mrg entry = start_entry;
2022 1.10 mrg while ((entry != &map->header) && (entry->start < end)) {
2023 1.10 mrg UVM_MAP_CLIP_END(map, entry, end);
2024 1.56 thorpej if (VM_MAPENT_ISWIRED(entry))
2025 1.10 mrg uvm_map_entry_unwire(map, entry);
2026 1.10 mrg entry = entry->next;
2027 1.10 mrg }
2028 1.64 thorpej if ((lockflags & UVM_LK_EXIT) == 0)
2029 1.64 thorpej vm_map_unlock(map);
2030 1.10 mrg UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2031 1.10 mrg return(KERN_SUCCESS);
2032 1.10 mrg
2033 1.10 mrg /*
2034 1.10 mrg * end of unwire case!
2035 1.10 mrg */
2036 1.10 mrg }
2037 1.10 mrg
2038 1.10 mrg /*
2039 1.10 mrg * wire case: in two passes [XXXCDC: ugly block of code here]
2040 1.10 mrg *
2041 1.10 mrg * 1: holding the write lock, we create any anonymous maps that need
2042 1.10 mrg * to be created. then we clip each map entry to the region to
2043 1.10 mrg * be wired and increment its wiring count.
2044 1.10 mrg *
2045 1.10 mrg * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2046 1.56 thorpej * in the pages for any newly wired area (wired_count == 1).
2047 1.10 mrg *
2048 1.10 mrg * downgrading to a read lock for uvm_fault_wire avoids a possible
2049 1.10 mrg * deadlock with another thread that may have faulted on one of
2050 1.10 mrg * the pages to be wired (it would mark the page busy, blocking
2051 1.10 mrg * us, then in turn block on the map lock that we hold). because
2052 1.10 mrg * of problems in the recursive lock package, we cannot upgrade
2053 1.10 mrg * to a write lock in vm_map_lookup. thus, any actions that
2054 1.10 mrg * require the write lock must be done beforehand. because we
2055 1.10 mrg * keep the read lock on the map, the copy-on-write status of the
2056 1.10 mrg * entries we modify here cannot change.
2057 1.10 mrg */
2058 1.10 mrg
2059 1.10 mrg while ((entry != &map->header) && (entry->start < end)) {
2060 1.55 thorpej if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2061 1.10 mrg /*
2062 1.10 mrg * perform actions of vm_map_lookup that need the
2063 1.10 mrg * write lock on the map: create an anonymous map
2064 1.10 mrg * for a copy-on-write region, or an anonymous map
2065 1.29 chuck * for a zero-fill region. (XXXCDC: submap case
2066 1.29 chuck * ok?)
2067 1.10 mrg */
2068 1.29 chuck if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2069 1.54 thorpej if (UVM_ET_ISNEEDSCOPY(entry) &&
2070 1.54 thorpej ((entry->protection & VM_PROT_WRITE) ||
2071 1.54 thorpej (entry->object.uvm_obj == NULL))) {
2072 1.10 mrg amap_copy(map, entry, M_WAITOK, TRUE,
2073 1.10 mrg start, end);
2074 1.10 mrg /* XXXCDC: wait OK? */
2075 1.10 mrg }
2076 1.10 mrg }
2077 1.55 thorpej }
2078 1.10 mrg UVM_MAP_CLIP_START(map, entry, start);
2079 1.10 mrg UVM_MAP_CLIP_END(map, entry, end);
2080 1.10 mrg entry->wired_count++;
2081 1.10 mrg
2082 1.10 mrg /*
2083 1.10 mrg * Check for holes
2084 1.10 mrg */
2085 1.54 thorpej if (entry->protection == VM_PROT_NONE ||
2086 1.54 thorpej (entry->end < end &&
2087 1.54 thorpej (entry->next == &map->header ||
2088 1.54 thorpej entry->next->start > entry->end))) {
2089 1.10 mrg /*
2090 1.10 mrg * found one. amap creation actions do not need to
2091 1.10 mrg * be undone, but the wired counts need to be restored.
2092 1.10 mrg */
2093 1.10 mrg while (entry != &map->header && entry->end > start) {
2094 1.10 mrg entry->wired_count--;
2095 1.10 mrg entry = entry->prev;
2096 1.10 mrg }
2097 1.64 thorpej if ((lockflags & UVM_LK_EXIT) == 0)
2098 1.64 thorpej vm_map_unlock(map);
2099 1.10 mrg UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2100 1.38 mrg return (KERN_INVALID_ARGUMENT);
2101 1.10 mrg }
2102 1.10 mrg entry = entry->next;
2103 1.10 mrg }
2104 1.10 mrg
2105 1.10 mrg /*
2106 1.10 mrg * Pass 2.
2107 1.10 mrg */
2108 1.51 thorpej
2109 1.60 thorpej #ifdef DIAGNOSTIC
2110 1.60 thorpej timestamp_save = map->timestamp;
2111 1.60 thorpej #endif
2112 1.60 thorpej vm_map_busy(map);
2113 1.51 thorpej vm_map_downgrade(map);
2114 1.10 mrg
2115 1.10 mrg rv = 0;
2116 1.10 mrg entry = start_entry;
2117 1.10 mrg while (entry != &map->header && entry->start < end) {
2118 1.51 thorpej if (entry->wired_count == 1) {
2119 1.44 thorpej rv = uvm_fault_wire(map, entry->start, entry->end,
2120 1.46 thorpej entry->protection);
2121 1.10 mrg if (rv) {
2122 1.51 thorpej /*
2123 1.51 thorpej * wiring failed. break out of the loop.
2124 1.51 thorpej * we'll clean up the map below, once we
2125 1.51 thorpej * have a write lock again.
2126 1.51 thorpej */
2127 1.51 thorpej break;
2128 1.10 mrg }
2129 1.10 mrg }
2130 1.10 mrg entry = entry->next;
2131 1.10 mrg }
2132 1.10 mrg
2133 1.52 thorpej if (rv) { /* failed? */
2134 1.52 thorpej /*
2135 1.52 thorpej * Get back to an exclusive (write) lock.
2136 1.52 thorpej */
2137 1.52 thorpej vm_map_upgrade(map);
2138 1.60 thorpej vm_map_unbusy(map);
2139 1.60 thorpej
2140 1.60 thorpej #ifdef DIAGNOSTIC
2141 1.60 thorpej if (timestamp_save != map->timestamp)
2142 1.60 thorpej panic("uvm_map_pageable: stale map");
2143 1.60 thorpej #endif
2144 1.10 mrg
2145 1.51 thorpej /*
2146 1.51 thorpej * first drop the wiring count on all the entries
2147 1.51 thorpej * which haven't actually been wired yet.
2148 1.51 thorpej */
2149 1.54 thorpej failed_entry = entry;
2150 1.54 thorpej while (entry != &map->header && entry->start < end) {
2151 1.51 thorpej entry->wired_count--;
2152 1.54 thorpej entry = entry->next;
2153 1.54 thorpej }
2154 1.51 thorpej
2155 1.51 thorpej /*
2156 1.54 thorpej * now, unwire all the entries that were successfully
2157 1.54 thorpej * wired above.
2158 1.51 thorpej */
2159 1.54 thorpej entry = start_entry;
2160 1.54 thorpej while (entry != failed_entry) {
2161 1.54 thorpej entry->wired_count--;
2162 1.55 thorpej if (VM_MAPENT_ISWIRED(entry) == 0)
2163 1.54 thorpej uvm_map_entry_unwire(map, entry);
2164 1.54 thorpej entry = entry->next;
2165 1.54 thorpej }
2166 1.64 thorpej if ((lockflags & UVM_LK_EXIT) == 0)
2167 1.64 thorpej vm_map_unlock(map);
2168 1.10 mrg UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2169 1.10 mrg return(rv);
2170 1.10 mrg }
2171 1.51 thorpej
2172 1.52 thorpej /* We are holding a read lock here. */
2173 1.64 thorpej if ((lockflags & UVM_LK_EXIT) == 0) {
2174 1.64 thorpej vm_map_unbusy(map);
2175 1.64 thorpej vm_map_unlock_read(map);
2176 1.64 thorpej } else {
2177 1.64 thorpej /*
2178 1.64 thorpej * Get back to an exclusive (write) lock.
2179 1.64 thorpej */
2180 1.64 thorpej vm_map_upgrade(map);
2181 1.64 thorpej vm_map_unbusy(map);
2182 1.64 thorpej }
2183 1.64 thorpej
2184 1.10 mrg UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2185 1.10 mrg return(KERN_SUCCESS);
2186 1.1 mrg }
2187 1.1 mrg
2188 1.1 mrg /*
2189 1.54 thorpej * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2190 1.54 thorpej * all mapped regions.
2191 1.54 thorpej *
2192 1.54 thorpej * => map must not be locked.
2193 1.54 thorpej * => if no flags are specified, all regions are unwired.
2194 1.54 thorpej * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2195 1.54 thorpej */
2196 1.54 thorpej
2197 1.54 thorpej int
2198 1.54 thorpej uvm_map_pageable_all(map, flags, limit)
2199 1.54 thorpej vm_map_t map;
2200 1.54 thorpej int flags;
2201 1.54 thorpej vsize_t limit;
2202 1.54 thorpej {
2203 1.54 thorpej vm_map_entry_t entry, failed_entry;
2204 1.54 thorpej vsize_t size;
2205 1.54 thorpej int rv;
2206 1.60 thorpej #ifdef DIAGNOSTIC
2207 1.60 thorpej u_int timestamp_save;
2208 1.60 thorpej #endif
2209 1.54 thorpej UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2210 1.54 thorpej UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2211 1.54 thorpej
2212 1.54 thorpej #ifdef DIAGNOSTIC
2213 1.54 thorpej if ((map->flags & VM_MAP_PAGEABLE) == 0)
2214 1.54 thorpej panic("uvm_map_pageable_all: map %p not pageable", map);
2215 1.54 thorpej #endif
2216 1.54 thorpej
2217 1.54 thorpej vm_map_lock(map);
2218 1.54 thorpej
2219 1.54 thorpej /*
2220 1.54 thorpej * handle wiring and unwiring separately.
2221 1.54 thorpej */
2222 1.54 thorpej
2223 1.54 thorpej if (flags == 0) { /* unwire */
2224 1.54 thorpej /*
2225 1.56 thorpej * POSIX 1003.1b -- munlockall unlocks all regions,
2226 1.56 thorpej * regardless of how many times mlockall has been called.
2227 1.54 thorpej */
2228 1.54 thorpej for (entry = map->header.next; entry != &map->header;
2229 1.54 thorpej entry = entry->next) {
2230 1.56 thorpej if (VM_MAPENT_ISWIRED(entry))
2231 1.56 thorpej uvm_map_entry_unwire(map, entry);
2232 1.54 thorpej }
2233 1.61 thorpej vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2234 1.54 thorpej vm_map_unlock(map);
2235 1.54 thorpej UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2236 1.54 thorpej return (KERN_SUCCESS);
2237 1.54 thorpej
2238 1.54 thorpej /*
2239 1.54 thorpej * end of unwire case!
2240 1.54 thorpej */
2241 1.54 thorpej }
2242 1.54 thorpej
2243 1.54 thorpej if (flags & MCL_FUTURE) {
2244 1.54 thorpej /*
2245 1.54 thorpej * must wire all future mappings; remember this.
2246 1.54 thorpej */
2247 1.61 thorpej vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2248 1.54 thorpej }
2249 1.54 thorpej
2250 1.54 thorpej if ((flags & MCL_CURRENT) == 0) {
2251 1.54 thorpej /*
2252 1.54 thorpej * no more work to do!
2253 1.54 thorpej */
2254 1.54 thorpej UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2255 1.54 thorpej vm_map_unlock(map);
2256 1.54 thorpej return (KERN_SUCCESS);
2257 1.54 thorpej }
2258 1.54 thorpej
2259 1.54 thorpej /*
2260 1.54 thorpej * wire case: in three passes [XXXCDC: ugly block of code here]
2261 1.54 thorpej *
2262 1.54 thorpej * 1: holding the write lock, count all pages mapped by non-wired
2263 1.54 thorpej * entries. if this would cause us to go over our limit, we fail.
2264 1.54 thorpej *
2265 1.54 thorpej * 2: still holding the write lock, we create any anonymous maps that
2266 1.54 thorpej * need to be created. then we increment its wiring count.
2267 1.54 thorpej *
2268 1.54 thorpej * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2269 1.56 thorpej * in the pages for any newly wired area (wired_count == 1).
2270 1.54 thorpej *
2271 1.54 thorpej * downgrading to a read lock for uvm_fault_wire avoids a possible
2272 1.54 thorpej * deadlock with another thread that may have faulted on one of
2273 1.54 thorpej * the pages to be wired (it would mark the page busy, blocking
2274 1.54 thorpej * us, then in turn block on the map lock that we hold). because
2275 1.54 thorpej * of problems in the recursive lock package, we cannot upgrade
2276 1.54 thorpej * to a write lock in vm_map_lookup. thus, any actions that
2277 1.54 thorpej * require the write lock must be done beforehand. because we
2278 1.54 thorpej * keep the read lock on the map, the copy-on-write status of the
2279 1.54 thorpej * entries we modify here cannot change.
2280 1.54 thorpej */
2281 1.54 thorpej
2282 1.54 thorpej for (size = 0, entry = map->header.next; entry != &map->header;
2283 1.54 thorpej entry = entry->next) {
2284 1.54 thorpej if (entry->protection != VM_PROT_NONE &&
2285 1.55 thorpej VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2286 1.54 thorpej size += entry->end - entry->start;
2287 1.54 thorpej }
2288 1.54 thorpej }
2289 1.54 thorpej
2290 1.54 thorpej if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2291 1.54 thorpej vm_map_unlock(map);
2292 1.54 thorpej return (KERN_NO_SPACE); /* XXX overloaded */
2293 1.54 thorpej }
2294 1.54 thorpej
2295 1.54 thorpej /* XXX non-pmap_wired_count case must be handled by caller */
2296 1.54 thorpej #ifdef pmap_wired_count
2297 1.54 thorpej if (limit != 0 &&
2298 1.54 thorpej (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2299 1.54 thorpej vm_map_unlock(map);
2300 1.54 thorpej return (KERN_NO_SPACE); /* XXX overloaded */
2301 1.54 thorpej }
2302 1.54 thorpej #endif
2303 1.54 thorpej
2304 1.54 thorpej /*
2305 1.54 thorpej * Pass 2.
2306 1.54 thorpej */
2307 1.54 thorpej
2308 1.54 thorpej for (entry = map->header.next; entry != &map->header;
2309 1.54 thorpej entry = entry->next) {
2310 1.54 thorpej if (entry->protection == VM_PROT_NONE)
2311 1.54 thorpej continue;
2312 1.55 thorpej if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2313 1.54 thorpej /*
2314 1.54 thorpej * perform actions of vm_map_lookup that need the
2315 1.54 thorpej * write lock on the map: create an anonymous map
2316 1.54 thorpej * for a copy-on-write region, or an anonymous map
2317 1.54 thorpej * for a zero-fill region. (XXXCDC: submap case
2318 1.54 thorpej * ok?)
2319 1.54 thorpej */
2320 1.54 thorpej if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */
2321 1.54 thorpej if (UVM_ET_ISNEEDSCOPY(entry) &&
2322 1.54 thorpej ((entry->protection & VM_PROT_WRITE) ||
2323 1.54 thorpej (entry->object.uvm_obj == NULL))) {
2324 1.54 thorpej amap_copy(map, entry, M_WAITOK, TRUE,
2325 1.54 thorpej entry->start, entry->end);
2326 1.54 thorpej /* XXXCDC: wait OK? */
2327 1.54 thorpej }
2328 1.54 thorpej }
2329 1.55 thorpej }
2330 1.54 thorpej entry->wired_count++;
2331 1.54 thorpej }
2332 1.54 thorpej
2333 1.54 thorpej /*
2334 1.54 thorpej * Pass 3.
2335 1.54 thorpej */
2336 1.54 thorpej
2337 1.60 thorpej #ifdef DIAGNOSTIC
2338 1.60 thorpej timestamp_save = map->timestamp;
2339 1.60 thorpej #endif
2340 1.60 thorpej vm_map_busy(map);
2341 1.54 thorpej vm_map_downgrade(map);
2342 1.54 thorpej
2343 1.54 thorpej rv = KERN_SUCCESS;
2344 1.54 thorpej for (entry = map->header.next; entry != &map->header;
2345 1.54 thorpej entry = entry->next) {
2346 1.54 thorpej if (entry->wired_count == 1) {
2347 1.54 thorpej rv = uvm_fault_wire(map, entry->start, entry->end,
2348 1.54 thorpej entry->protection);
2349 1.54 thorpej if (rv) {
2350 1.54 thorpej /*
2351 1.54 thorpej * wiring failed. break out of the loop.
2352 1.54 thorpej * we'll clean up the map below, once we
2353 1.54 thorpej * have a write lock again.
2354 1.54 thorpej */
2355 1.54 thorpej break;
2356 1.54 thorpej }
2357 1.54 thorpej }
2358 1.54 thorpej }
2359 1.54 thorpej
2360 1.54 thorpej if (rv) { /* failed? */
2361 1.54 thorpej /*
2362 1.54 thorpej * Get back an exclusive (write) lock.
2363 1.54 thorpej */
2364 1.54 thorpej vm_map_upgrade(map);
2365 1.60 thorpej vm_map_unbusy(map);
2366 1.60 thorpej
2367 1.60 thorpej #ifdef DIAGNOSTIC
2368 1.60 thorpej if (timestamp_save != map->timestamp)
2369 1.60 thorpej panic("uvm_map_pageable_all: stale map");
2370 1.60 thorpej #endif
2371 1.54 thorpej
2372 1.54 thorpej /*
2373 1.54 thorpej * first drop the wiring count on all the entries
2374 1.54 thorpej * which haven't actually been wired yet.
2375 1.67 thorpej *
2376 1.67 thorpej * Skip VM_PROT_NONE entries like we did above.
2377 1.54 thorpej */
2378 1.54 thorpej failed_entry = entry;
2379 1.54 thorpej for (/* nothing */; entry != &map->header;
2380 1.67 thorpej entry = entry->next) {
2381 1.67 thorpej if (entry->protection == VM_PROT_NONE)
2382 1.67 thorpej continue;
2383 1.54 thorpej entry->wired_count--;
2384 1.67 thorpej }
2385 1.54 thorpej
2386 1.54 thorpej /*
2387 1.54 thorpej * now, unwire all the entries that were successfully
2388 1.54 thorpej * wired above.
2389 1.67 thorpej *
2390 1.67 thorpej * Skip VM_PROT_NONE entries like we did above.
2391 1.54 thorpej */
2392 1.54 thorpej for (entry = map->header.next; entry != failed_entry;
2393 1.54 thorpej entry = entry->next) {
2394 1.67 thorpej if (entry->protection == VM_PROT_NONE)
2395 1.67 thorpej continue;
2396 1.54 thorpej entry->wired_count--;
2397 1.67 thorpej if (VM_MAPENT_ISWIRED(entry))
2398 1.54 thorpej uvm_map_entry_unwire(map, entry);
2399 1.54 thorpej }
2400 1.54 thorpej vm_map_unlock(map);
2401 1.54 thorpej UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2402 1.54 thorpej return (rv);
2403 1.54 thorpej }
2404 1.54 thorpej
2405 1.54 thorpej /* We are holding a read lock here. */
2406 1.60 thorpej vm_map_unbusy(map);
2407 1.54 thorpej vm_map_unlock_read(map);
2408 1.54 thorpej
2409 1.54 thorpej UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2410 1.54 thorpej return (KERN_SUCCESS);
2411 1.54 thorpej }
2412 1.54 thorpej
2413 1.54 thorpej /*
2414 1.61 thorpej * uvm_map_clean: clean out a map range
2415 1.1 mrg *
2416 1.1 mrg * => valid flags:
2417 1.61 thorpej * if (flags & PGO_CLEANIT): dirty pages are cleaned first
2418 1.1 mrg * if (flags & PGO_SYNCIO): dirty pages are written synchronously
2419 1.1 mrg * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2420 1.1 mrg * if (flags & PGO_FREE): any cached pages are freed after clean
2421 1.1 mrg * => returns an error if any part of the specified range isn't mapped
2422 1.1 mrg * => never a need to flush amap layer since the anonymous memory has
2423 1.61 thorpej * no permanent home, but may deactivate pages there
2424 1.61 thorpej * => called from sys_msync() and sys_madvise()
2425 1.1 mrg * => caller must not write-lock map (read OK).
2426 1.1 mrg * => we may sleep while cleaning if SYNCIO [with map read-locked]
2427 1.1 mrg */
2428 1.1 mrg
2429 1.61 thorpej int amap_clean_works = 1; /* XXX for now, just in case... */
2430 1.61 thorpej
2431 1.10 mrg int
2432 1.10 mrg uvm_map_clean(map, start, end, flags)
2433 1.10 mrg vm_map_t map;
2434 1.24 eeh vaddr_t start, end;
2435 1.10 mrg int flags;
2436 1.10 mrg {
2437 1.61 thorpej vm_map_entry_t current, entry;
2438 1.61 thorpej struct uvm_object *uobj;
2439 1.61 thorpej struct vm_amap *amap;
2440 1.61 thorpej struct vm_anon *anon;
2441 1.61 thorpej struct vm_page *pg;
2442 1.61 thorpej vaddr_t offset;
2443 1.24 eeh vsize_t size;
2444 1.61 thorpej int rv, error, refs;
2445 1.10 mrg UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2446 1.10 mrg UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2447 1.10 mrg map, start, end, flags);
2448 1.10 mrg
2449 1.61 thorpej #ifdef DIAGNOSTIC
2450 1.61 thorpej if ((flags & (PGO_FREE|PGO_DEACTIVATE)) == (PGO_FREE|PGO_DEACTIVATE))
2451 1.61 thorpej panic("uvm_map_clean: FREE and DEACTIVATE");
2452 1.61 thorpej #endif
2453 1.61 thorpej
2454 1.10 mrg vm_map_lock_read(map);
2455 1.10 mrg VM_MAP_RANGE_CHECK(map, start, end);
2456 1.61 thorpej if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2457 1.10 mrg vm_map_unlock_read(map);
2458 1.10 mrg return(KERN_INVALID_ADDRESS);
2459 1.10 mrg }
2460 1.10 mrg
2461 1.10 mrg /*
2462 1.10 mrg * Make a first pass to check for holes.
2463 1.10 mrg */
2464 1.10 mrg for (current = entry; current->start < end; current = current->next) {
2465 1.10 mrg if (UVM_ET_ISSUBMAP(current)) {
2466 1.10 mrg vm_map_unlock_read(map);
2467 1.38 mrg return (KERN_INVALID_ARGUMENT);
2468 1.10 mrg }
2469 1.10 mrg if (end > current->end && (current->next == &map->header ||
2470 1.10 mrg current->end != current->next->start)) {
2471 1.10 mrg vm_map_unlock_read(map);
2472 1.38 mrg return (KERN_INVALID_ADDRESS);
2473 1.10 mrg }
2474 1.10 mrg }
2475 1.10 mrg
2476 1.61 thorpej error = KERN_SUCCESS;
2477 1.61 thorpej
2478 1.10 mrg for (current = entry; current->start < end; current = current->next) {
2479 1.61 thorpej amap = current->aref.ar_amap; /* top layer */
2480 1.61 thorpej uobj = current->object.uvm_obj; /* bottom layer */
2481 1.61 thorpej
2482 1.61 thorpej #ifdef DIAGNOSTIC
2483 1.61 thorpej if (start < current->start)
2484 1.61 thorpej panic("uvm_map_clean: hole");
2485 1.61 thorpej #endif
2486 1.1 mrg
2487 1.10 mrg /*
2488 1.61 thorpej * No amap cleaning necessary if:
2489 1.61 thorpej *
2490 1.61 thorpej * (1) There's no amap.
2491 1.61 thorpej *
2492 1.61 thorpej * (2) We're not deactivating or freeing pages.
2493 1.10 mrg */
2494 1.61 thorpej if (amap == NULL ||
2495 1.61 thorpej (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2496 1.61 thorpej goto flush_object;
2497 1.61 thorpej
2498 1.61 thorpej /* XXX for now, just in case... */
2499 1.61 thorpej if (amap_clean_works == 0)
2500 1.61 thorpej goto flush_object;
2501 1.61 thorpej
2502 1.61 thorpej amap_lock(amap);
2503 1.61 thorpej
2504 1.61 thorpej offset = start - current->start;
2505 1.61 thorpej size = (end <= current->end ? end : current->end) -
2506 1.61 thorpej start;
2507 1.61 thorpej
2508 1.61 thorpej for (/* nothing */; size != 0; size -= PAGE_SIZE,
2509 1.61 thorpej offset += PAGE_SIZE) {
2510 1.61 thorpej anon = amap_lookup(¤t->aref, offset);
2511 1.61 thorpej if (anon == NULL)
2512 1.61 thorpej continue;
2513 1.61 thorpej
2514 1.61 thorpej simple_lock(&anon->an_lock);
2515 1.61 thorpej
2516 1.63 thorpej pg = anon->u.an_page;
2517 1.63 thorpej if (pg == NULL) {
2518 1.63 thorpej simple_unlock(&anon->an_lock);
2519 1.63 thorpej continue;
2520 1.63 thorpej }
2521 1.63 thorpej
2522 1.61 thorpej switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2523 1.61 thorpej /*
2524 1.61 thorpej * XXX In these first 3 cases, we always just
2525 1.61 thorpej * XXX deactivate the page. We may want to
2526 1.61 thorpej * XXX handle the different cases more
2527 1.61 thorpej * XXX specifically, in the future.
2528 1.61 thorpej */
2529 1.61 thorpej case PGO_CLEANIT|PGO_FREE:
2530 1.61 thorpej case PGO_CLEANIT|PGO_DEACTIVATE:
2531 1.61 thorpej case PGO_DEACTIVATE:
2532 1.68 thorpej deactivate_it:
2533 1.61 thorpej /* skip the page if it's loaned or wired */
2534 1.61 thorpej if (pg->loan_count != 0 ||
2535 1.61 thorpej pg->wire_count != 0) {
2536 1.61 thorpej simple_unlock(&anon->an_lock);
2537 1.61 thorpej continue;
2538 1.61 thorpej }
2539 1.61 thorpej
2540 1.61 thorpej uvm_lock_pageq();
2541 1.61 thorpej
2542 1.61 thorpej /*
2543 1.61 thorpej * skip the page if it's not actually owned
2544 1.61 thorpej * by the anon (may simply be loaned to the
2545 1.61 thorpej * anon).
2546 1.61 thorpej */
2547 1.61 thorpej if ((pg->pqflags & PQ_ANON) == 0) {
2548 1.61 thorpej #ifdef DIAGNOSTIC
2549 1.61 thorpej if (pg->uobject != NULL)
2550 1.61 thorpej panic("uvm_map_clean: "
2551 1.61 thorpej "page anon vs. object "
2552 1.61 thorpej "inconsistency");
2553 1.61 thorpej #endif
2554 1.61 thorpej uvm_unlock_pageq();
2555 1.61 thorpej simple_unlock(&anon->an_lock);
2556 1.61 thorpej continue;
2557 1.61 thorpej }
2558 1.61 thorpej
2559 1.61 thorpej #ifdef DIAGNOSTIC
2560 1.61 thorpej if (pg->uanon != anon)
2561 1.61 thorpej panic("uvm_map_clean: anon "
2562 1.61 thorpej "inconsistency");
2563 1.61 thorpej #endif
2564 1.61 thorpej
2565 1.61 thorpej /* zap all mappings for the page. */
2566 1.69 chs pmap_page_protect(pg, VM_PROT_NONE);
2567 1.61 thorpej
2568 1.61 thorpej /* ...and deactivate the page. */
2569 1.61 thorpej uvm_pagedeactivate(pg);
2570 1.61 thorpej
2571 1.61 thorpej uvm_unlock_pageq();
2572 1.61 thorpej simple_unlock(&anon->an_lock);
2573 1.61 thorpej continue;
2574 1.61 thorpej
2575 1.61 thorpej case PGO_FREE:
2576 1.68 thorpej /*
2577 1.68 thorpej * If there are multiple references to
2578 1.68 thorpej * the amap, just deactivate the page.
2579 1.68 thorpej */
2580 1.68 thorpej if (amap_refs(amap) > 1)
2581 1.68 thorpej goto deactivate_it;
2582 1.68 thorpej
2583 1.62 thorpej /* XXX skip the page if it's wired */
2584 1.62 thorpej if (pg->wire_count != 0) {
2585 1.62 thorpej simple_unlock(&anon->an_lock);
2586 1.62 thorpej continue;
2587 1.62 thorpej }
2588 1.66 thorpej amap_unadd(¤t->aref, offset);
2589 1.61 thorpej refs = --anon->an_ref;
2590 1.61 thorpej simple_unlock(&anon->an_lock);
2591 1.61 thorpej if (refs == 0)
2592 1.61 thorpej uvm_anfree(anon);
2593 1.61 thorpej continue;
2594 1.61 thorpej
2595 1.61 thorpej default:
2596 1.61 thorpej panic("uvm_map_clean: wierd flags");
2597 1.61 thorpej }
2598 1.61 thorpej #ifdef DIAGNOSTIC
2599 1.61 thorpej panic("uvm_map_clean: unreachable code");
2600 1.61 thorpej #endif
2601 1.61 thorpej }
2602 1.61 thorpej
2603 1.61 thorpej amap_unlock(amap);
2604 1.1 mrg
2605 1.61 thorpej flush_object:
2606 1.10 mrg /*
2607 1.33 chuck * flush pages if we've got a valid backing object.
2608 1.10 mrg */
2609 1.1 mrg
2610 1.61 thorpej offset = current->offset + (start - current->start);
2611 1.61 thorpej size = (end <= current->end ? end : current->end) - start;
2612 1.61 thorpej
2613 1.61 thorpej if (uobj != NULL) {
2614 1.61 thorpej simple_lock(&uobj->vmobjlock);
2615 1.61 thorpej rv = uobj->pgops->pgo_flush(uobj, offset,
2616 1.61 thorpej offset + size, flags);
2617 1.61 thorpej simple_unlock(&uobj->vmobjlock);
2618 1.61 thorpej
2619 1.61 thorpej if (rv == FALSE)
2620 1.61 thorpej error = KERN_FAILURE;
2621 1.10 mrg }
2622 1.10 mrg start += size;
2623 1.10 mrg }
2624 1.61 thorpej
2625 1.1 mrg vm_map_unlock_read(map);
2626 1.61 thorpej return (error);
2627 1.1 mrg }
2628 1.1 mrg
2629 1.1 mrg
2630 1.1 mrg /*
2631 1.1 mrg * uvm_map_checkprot: check protection in map
2632 1.1 mrg *
2633 1.1 mrg * => must allow specified protection in a fully allocated region.
2634 1.1 mrg * => map must be read or write locked by caller.
2635 1.1 mrg */
2636 1.1 mrg
2637 1.10 mrg boolean_t
2638 1.10 mrg uvm_map_checkprot(map, start, end, protection)
2639 1.10 mrg vm_map_t map;
2640 1.24 eeh vaddr_t start, end;
2641 1.10 mrg vm_prot_t protection;
2642 1.10 mrg {
2643 1.10 mrg vm_map_entry_t entry;
2644 1.10 mrg vm_map_entry_t tmp_entry;
2645 1.10 mrg
2646 1.10 mrg if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2647 1.10 mrg return(FALSE);
2648 1.10 mrg }
2649 1.10 mrg
2650 1.10 mrg entry = tmp_entry;
2651 1.10 mrg
2652 1.10 mrg while (start < end) {
2653 1.10 mrg if (entry == &map->header) {
2654 1.10 mrg return(FALSE);
2655 1.10 mrg }
2656 1.10 mrg
2657 1.10 mrg /*
2658 1.10 mrg * no holes allowed
2659 1.10 mrg */
2660 1.10 mrg
2661 1.10 mrg if (start < entry->start) {
2662 1.10 mrg return(FALSE);
2663 1.10 mrg }
2664 1.10 mrg
2665 1.10 mrg /*
2666 1.10 mrg * check protection associated with entry
2667 1.10 mrg */
2668 1.1 mrg
2669 1.10 mrg if ((entry->protection & protection) != protection) {
2670 1.10 mrg return(FALSE);
2671 1.10 mrg }
2672 1.10 mrg
2673 1.10 mrg /* go to next entry */
2674 1.10 mrg
2675 1.10 mrg start = entry->end;
2676 1.10 mrg entry = entry->next;
2677 1.10 mrg }
2678 1.10 mrg return(TRUE);
2679 1.1 mrg }
2680 1.1 mrg
2681 1.1 mrg /*
2682 1.1 mrg * uvmspace_alloc: allocate a vmspace structure.
2683 1.1 mrg *
2684 1.1 mrg * - structure includes vm_map and pmap
2685 1.1 mrg * - XXX: no locking on this structure
2686 1.1 mrg * - refcnt set to 1, rest must be init'd by caller
2687 1.1 mrg */
2688 1.10 mrg struct vmspace *
2689 1.10 mrg uvmspace_alloc(min, max, pageable)
2690 1.24 eeh vaddr_t min, max;
2691 1.10 mrg int pageable;
2692 1.10 mrg {
2693 1.10 mrg struct vmspace *vm;
2694 1.10 mrg UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2695 1.10 mrg
2696 1.25 thorpej vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2697 1.15 thorpej uvmspace_init(vm, NULL, min, max, pageable);
2698 1.15 thorpej UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2699 1.15 thorpej return (vm);
2700 1.15 thorpej }
2701 1.15 thorpej
2702 1.15 thorpej /*
2703 1.15 thorpej * uvmspace_init: initialize a vmspace structure.
2704 1.15 thorpej *
2705 1.15 thorpej * - XXX: no locking on this structure
2706 1.15 thorpej * - refcnt set to 1, rest must me init'd by caller
2707 1.15 thorpej */
2708 1.15 thorpej void
2709 1.15 thorpej uvmspace_init(vm, pmap, min, max, pageable)
2710 1.15 thorpej struct vmspace *vm;
2711 1.15 thorpej struct pmap *pmap;
2712 1.24 eeh vaddr_t min, max;
2713 1.15 thorpej boolean_t pageable;
2714 1.15 thorpej {
2715 1.15 thorpej UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2716 1.15 thorpej
2717 1.23 perry memset(vm, 0, sizeof(*vm));
2718 1.15 thorpej
2719 1.45 thorpej uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0);
2720 1.15 thorpej
2721 1.15 thorpej if (pmap)
2722 1.15 thorpej pmap_reference(pmap);
2723 1.15 thorpej else
2724 1.15 thorpej pmap = pmap_create();
2725 1.15 thorpej vm->vm_map.pmap = pmap;
2726 1.15 thorpej
2727 1.10 mrg vm->vm_refcnt = 1;
2728 1.15 thorpej UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2729 1.1 mrg }
2730 1.1 mrg
2731 1.1 mrg /*
2732 1.1 mrg * uvmspace_share: share a vmspace between two proceses
2733 1.1 mrg *
2734 1.1 mrg * - XXX: no locking on vmspace
2735 1.1 mrg * - used for vfork, threads(?)
2736 1.1 mrg */
2737 1.1 mrg
2738 1.10 mrg void
2739 1.10 mrg uvmspace_share(p1, p2)
2740 1.10 mrg struct proc *p1, *p2;
2741 1.1 mrg {
2742 1.10 mrg p2->p_vmspace = p1->p_vmspace;
2743 1.10 mrg p1->p_vmspace->vm_refcnt++;
2744 1.1 mrg }
2745 1.1 mrg
2746 1.1 mrg /*
2747 1.1 mrg * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2748 1.1 mrg *
2749 1.1 mrg * - XXX: no locking on vmspace
2750 1.1 mrg */
2751 1.1 mrg
2752 1.10 mrg void
2753 1.10 mrg uvmspace_unshare(p)
2754 1.10 mrg struct proc *p;
2755 1.10 mrg {
2756 1.10 mrg struct vmspace *nvm, *ovm = p->p_vmspace;
2757 1.1 mrg
2758 1.10 mrg if (ovm->vm_refcnt == 1)
2759 1.10 mrg /* nothing to do: vmspace isn't shared in the first place */
2760 1.10 mrg return;
2761 1.1 mrg
2762 1.10 mrg /* make a new vmspace, still holding old one */
2763 1.10 mrg nvm = uvmspace_fork(ovm);
2764 1.10 mrg
2765 1.12 thorpej pmap_deactivate(p); /* unbind old vmspace */
2766 1.10 mrg p->p_vmspace = nvm;
2767 1.10 mrg pmap_activate(p); /* switch to new vmspace */
2768 1.13 thorpej
2769 1.10 mrg uvmspace_free(ovm); /* drop reference to old vmspace */
2770 1.1 mrg }
2771 1.1 mrg
2772 1.1 mrg /*
2773 1.1 mrg * uvmspace_exec: the process wants to exec a new program
2774 1.1 mrg *
2775 1.1 mrg * - XXX: no locking on vmspace
2776 1.1 mrg */
2777 1.1 mrg
2778 1.10 mrg void
2779 1.10 mrg uvmspace_exec(p)
2780 1.10 mrg struct proc *p;
2781 1.1 mrg {
2782 1.10 mrg struct vmspace *nvm, *ovm = p->p_vmspace;
2783 1.10 mrg vm_map_t map = &ovm->vm_map;
2784 1.1 mrg
2785 1.71 chs #ifdef __sparc__
2786 1.10 mrg /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2787 1.10 mrg kill_user_windows(p); /* before stack addresses go away */
2788 1.1 mrg #endif
2789 1.1 mrg
2790 1.10 mrg /*
2791 1.10 mrg * see if more than one process is using this vmspace...
2792 1.10 mrg */
2793 1.1 mrg
2794 1.10 mrg if (ovm->vm_refcnt == 1) {
2795 1.1 mrg
2796 1.10 mrg /*
2797 1.10 mrg * if p is the only process using its vmspace then we can safely
2798 1.10 mrg * recycle that vmspace for the program that is being exec'd.
2799 1.10 mrg */
2800 1.1 mrg
2801 1.1 mrg #ifdef SYSVSHM
2802 1.10 mrg /*
2803 1.10 mrg * SYSV SHM semantics require us to kill all segments on an exec
2804 1.10 mrg */
2805 1.10 mrg if (ovm->vm_shm)
2806 1.10 mrg shmexit(ovm);
2807 1.10 mrg #endif
2808 1.54 thorpej
2809 1.54 thorpej /*
2810 1.54 thorpej * POSIX 1003.1b -- "lock future mappings" is revoked
2811 1.54 thorpej * when a process execs another program image.
2812 1.54 thorpej */
2813 1.54 thorpej vm_map_lock(map);
2814 1.61 thorpej vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2815 1.54 thorpej vm_map_unlock(map);
2816 1.10 mrg
2817 1.10 mrg /*
2818 1.10 mrg * now unmap the old program
2819 1.10 mrg */
2820 1.29 chuck uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
2821 1.10 mrg
2822 1.10 mrg } else {
2823 1.10 mrg
2824 1.10 mrg /*
2825 1.10 mrg * p's vmspace is being shared, so we can't reuse it for p since
2826 1.10 mrg * it is still being used for others. allocate a new vmspace
2827 1.10 mrg * for p
2828 1.10 mrg */
2829 1.10 mrg nvm = uvmspace_alloc(map->min_offset, map->max_offset,
2830 1.45 thorpej (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2831 1.1 mrg
2832 1.10 mrg /*
2833 1.10 mrg * install new vmspace and drop our ref to the old one.
2834 1.10 mrg */
2835 1.10 mrg
2836 1.12 thorpej pmap_deactivate(p);
2837 1.10 mrg p->p_vmspace = nvm;
2838 1.10 mrg pmap_activate(p);
2839 1.13 thorpej
2840 1.10 mrg uvmspace_free(ovm);
2841 1.10 mrg }
2842 1.1 mrg }
2843 1.1 mrg
2844 1.1 mrg /*
2845 1.1 mrg * uvmspace_free: free a vmspace data structure
2846 1.1 mrg *
2847 1.1 mrg * - XXX: no locking on vmspace
2848 1.1 mrg */
2849 1.1 mrg
2850 1.10 mrg void
2851 1.10 mrg uvmspace_free(vm)
2852 1.10 mrg struct vmspace *vm;
2853 1.1 mrg {
2854 1.10 mrg vm_map_entry_t dead_entries;
2855 1.10 mrg UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2856 1.1 mrg
2857 1.10 mrg UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2858 1.10 mrg if (--vm->vm_refcnt == 0) {
2859 1.10 mrg /*
2860 1.10 mrg * lock the map, to wait out all other references to it. delete
2861 1.10 mrg * all of the mappings and pages they hold, then call the pmap
2862 1.10 mrg * module to reclaim anything left.
2863 1.10 mrg */
2864 1.10 mrg vm_map_lock(&vm->vm_map);
2865 1.10 mrg if (vm->vm_map.nentries) {
2866 1.10 mrg (void)uvm_unmap_remove(&vm->vm_map,
2867 1.10 mrg vm->vm_map.min_offset, vm->vm_map.max_offset,
2868 1.29 chuck &dead_entries);
2869 1.10 mrg if (dead_entries != NULL)
2870 1.10 mrg uvm_unmap_detach(dead_entries, 0);
2871 1.10 mrg }
2872 1.10 mrg pmap_destroy(vm->vm_map.pmap);
2873 1.10 mrg vm->vm_map.pmap = NULL;
2874 1.25 thorpej pool_put(&uvm_vmspace_pool, vm);
2875 1.10 mrg }
2876 1.10 mrg UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2877 1.1 mrg }
2878 1.1 mrg
2879 1.1 mrg /*
2880 1.1 mrg * F O R K - m a i n e n t r y p o i n t
2881 1.1 mrg */
2882 1.1 mrg /*
2883 1.1 mrg * uvmspace_fork: fork a process' main map
2884 1.1 mrg *
2885 1.1 mrg * => create a new vmspace for child process from parent.
2886 1.1 mrg * => parent's map must not be locked.
2887 1.1 mrg */
2888 1.1 mrg
2889 1.10 mrg struct vmspace *
2890 1.10 mrg uvmspace_fork(vm1)
2891 1.10 mrg struct vmspace *vm1;
2892 1.10 mrg {
2893 1.10 mrg struct vmspace *vm2;
2894 1.10 mrg vm_map_t old_map = &vm1->vm_map;
2895 1.10 mrg vm_map_t new_map;
2896 1.10 mrg vm_map_entry_t old_entry;
2897 1.10 mrg vm_map_entry_t new_entry;
2898 1.10 mrg pmap_t new_pmap;
2899 1.14 chuck boolean_t protect_child;
2900 1.10 mrg UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2901 1.1 mrg
2902 1.10 mrg vm_map_lock(old_map);
2903 1.1 mrg
2904 1.10 mrg vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset,
2905 1.45 thorpej (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2906 1.23 perry memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2907 1.1 mrg (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2908 1.10 mrg new_map = &vm2->vm_map; /* XXX */
2909 1.10 mrg new_pmap = new_map->pmap;
2910 1.10 mrg
2911 1.10 mrg old_entry = old_map->header.next;
2912 1.10 mrg
2913 1.10 mrg /*
2914 1.10 mrg * go entry-by-entry
2915 1.10 mrg */
2916 1.1 mrg
2917 1.10 mrg while (old_entry != &old_map->header) {
2918 1.1 mrg
2919 1.10 mrg /*
2920 1.10 mrg * first, some sanity checks on the old entry
2921 1.10 mrg */
2922 1.10 mrg if (UVM_ET_ISSUBMAP(old_entry))
2923 1.29 chuck panic("fork: encountered a submap during fork (illegal)");
2924 1.29 chuck
2925 1.29 chuck if (!UVM_ET_ISCOPYONWRITE(old_entry) &&
2926 1.10 mrg UVM_ET_ISNEEDSCOPY(old_entry))
2927 1.1 mrg panic("fork: non-copy_on_write map entry marked needs_copy (illegal)");
2928 1.1 mrg
2929 1.1 mrg
2930 1.10 mrg switch (old_entry->inheritance) {
2931 1.10 mrg case VM_INHERIT_NONE:
2932 1.10 mrg /*
2933 1.10 mrg * drop the mapping
2934 1.10 mrg */
2935 1.10 mrg break;
2936 1.10 mrg
2937 1.10 mrg case VM_INHERIT_SHARE:
2938 1.10 mrg /*
2939 1.10 mrg * share the mapping: this means we want the old and
2940 1.10 mrg * new entries to share amaps and backing objects.
2941 1.10 mrg */
2942 1.10 mrg
2943 1.10 mrg /*
2944 1.10 mrg * if the old_entry needs a new amap (due to prev fork)
2945 1.10 mrg * then we need to allocate it now so that we have
2946 1.10 mrg * something we own to share with the new_entry. [in
2947 1.10 mrg * other words, we need to clear needs_copy]
2948 1.10 mrg */
2949 1.10 mrg
2950 1.10 mrg if (UVM_ET_ISNEEDSCOPY(old_entry)) {
2951 1.10 mrg /* get our own amap, clears needs_copy */
2952 1.10 mrg amap_copy(old_map, old_entry, M_WAITOK, FALSE,
2953 1.10 mrg 0, 0);
2954 1.10 mrg /* XXXCDC: WAITOK??? */
2955 1.10 mrg }
2956 1.10 mrg
2957 1.10 mrg new_entry = uvm_mapent_alloc(new_map);
2958 1.10 mrg /* old_entry -> new_entry */
2959 1.10 mrg uvm_mapent_copy(old_entry, new_entry);
2960 1.10 mrg
2961 1.10 mrg /* new pmap has nothing wired in it */
2962 1.10 mrg new_entry->wired_count = 0;
2963 1.10 mrg
2964 1.10 mrg /*
2965 1.29 chuck * gain reference to object backing the map (can't
2966 1.29 chuck * be a submap, already checked this case).
2967 1.10 mrg */
2968 1.10 mrg if (new_entry->aref.ar_amap)
2969 1.10 mrg /* share reference */
2970 1.10 mrg amap_ref(new_entry, AMAP_SHARED);
2971 1.10 mrg
2972 1.10 mrg if (new_entry->object.uvm_obj &&
2973 1.10 mrg new_entry->object.uvm_obj->pgops->pgo_reference)
2974 1.10 mrg new_entry->object.uvm_obj->
2975 1.10 mrg pgops->pgo_reference(
2976 1.10 mrg new_entry->object.uvm_obj);
2977 1.10 mrg
2978 1.10 mrg /* insert entry at end of new_map's entry list */
2979 1.10 mrg uvm_map_entry_link(new_map, new_map->header.prev,
2980 1.10 mrg new_entry);
2981 1.10 mrg
2982 1.10 mrg /*
2983 1.10 mrg * pmap_copy the mappings: this routine is optional
2984 1.10 mrg * but if it is there it will reduce the number of
2985 1.10 mrg * page faults in the new proc.
2986 1.10 mrg */
2987 1.10 mrg
2988 1.10 mrg pmap_copy(new_pmap, old_map->pmap, new_entry->start,
2989 1.10 mrg (old_entry->end - old_entry->start),
2990 1.10 mrg old_entry->start);
2991 1.10 mrg
2992 1.10 mrg break;
2993 1.10 mrg
2994 1.10 mrg case VM_INHERIT_COPY:
2995 1.10 mrg
2996 1.10 mrg /*
2997 1.10 mrg * copy-on-write the mapping (using mmap's
2998 1.10 mrg * MAP_PRIVATE semantics)
2999 1.29 chuck *
3000 1.29 chuck * allocate new_entry, adjust reference counts.
3001 1.29 chuck * (note that new references are read-only).
3002 1.10 mrg */
3003 1.10 mrg
3004 1.10 mrg new_entry = uvm_mapent_alloc(new_map);
3005 1.10 mrg /* old_entry -> new_entry */
3006 1.10 mrg uvm_mapent_copy(old_entry, new_entry);
3007 1.10 mrg
3008 1.10 mrg if (new_entry->aref.ar_amap)
3009 1.10 mrg amap_ref(new_entry, 0);
3010 1.10 mrg
3011 1.10 mrg if (new_entry->object.uvm_obj &&
3012 1.10 mrg new_entry->object.uvm_obj->pgops->pgo_reference)
3013 1.10 mrg new_entry->object.uvm_obj->pgops->pgo_reference
3014 1.10 mrg (new_entry->object.uvm_obj);
3015 1.10 mrg
3016 1.10 mrg /* new pmap has nothing wired in it */
3017 1.10 mrg new_entry->wired_count = 0;
3018 1.10 mrg
3019 1.10 mrg new_entry->etype |=
3020 1.10 mrg (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3021 1.10 mrg uvm_map_entry_link(new_map, new_map->header.prev,
3022 1.10 mrg new_entry);
3023 1.10 mrg
3024 1.14 chuck /*
3025 1.10 mrg * the new entry will need an amap. it will either
3026 1.10 mrg * need to be copied from the old entry or created
3027 1.14 chuck * from scratch (if the old entry does not have an
3028 1.14 chuck * amap). can we defer this process until later
3029 1.14 chuck * (by setting "needs_copy") or do we need to copy
3030 1.14 chuck * the amap now?
3031 1.10 mrg *
3032 1.14 chuck * we must copy the amap now if any of the following
3033 1.10 mrg * conditions hold:
3034 1.14 chuck * 1. the old entry has an amap and that amap is
3035 1.14 chuck * being shared. this means that the old (parent)
3036 1.14 chuck * process is sharing the amap with another
3037 1.14 chuck * process. if we do not clear needs_copy here
3038 1.14 chuck * we will end up in a situation where both the
3039 1.14 chuck * parent and child process are refering to the
3040 1.14 chuck * same amap with "needs_copy" set. if the
3041 1.14 chuck * parent write-faults, the fault routine will
3042 1.14 chuck * clear "needs_copy" in the parent by allocating
3043 1.14 chuck * a new amap. this is wrong because the
3044 1.14 chuck * parent is supposed to be sharing the old amap
3045 1.14 chuck * and the new amap will break that.
3046 1.10 mrg *
3047 1.14 chuck * 2. if the old entry has an amap and a non-zero
3048 1.14 chuck * wire count then we are going to have to call
3049 1.14 chuck * amap_cow_now to avoid page faults in the
3050 1.14 chuck * parent process. since amap_cow_now requires
3051 1.14 chuck * "needs_copy" to be clear we might as well
3052 1.14 chuck * clear it here as well.
3053 1.10 mrg *
3054 1.10 mrg */
3055 1.10 mrg
3056 1.14 chuck if (old_entry->aref.ar_amap != NULL) {
3057 1.14 chuck
3058 1.34 chuck if ((amap_flags(old_entry->aref.ar_amap) &
3059 1.14 chuck AMAP_SHARED) != 0 ||
3060 1.55 thorpej VM_MAPENT_ISWIRED(old_entry)) {
3061 1.14 chuck
3062 1.14 chuck amap_copy(new_map, new_entry, M_WAITOK, FALSE,
3063 1.14 chuck 0, 0);
3064 1.14 chuck /* XXXCDC: M_WAITOK ... ok? */
3065 1.14 chuck }
3066 1.10 mrg }
3067 1.14 chuck
3068 1.10 mrg /*
3069 1.14 chuck * if the parent's entry is wired down, then the
3070 1.14 chuck * parent process does not want page faults on
3071 1.14 chuck * access to that memory. this means that we
3072 1.14 chuck * cannot do copy-on-write because we can't write
3073 1.14 chuck * protect the old entry. in this case we
3074 1.14 chuck * resolve all copy-on-write faults now, using
3075 1.14 chuck * amap_cow_now. note that we have already
3076 1.14 chuck * allocated any needed amap (above).
3077 1.10 mrg */
3078 1.10 mrg
3079 1.55 thorpej if (VM_MAPENT_ISWIRED(old_entry)) {
3080 1.1 mrg
3081 1.14 chuck /*
3082 1.14 chuck * resolve all copy-on-write faults now
3083 1.14 chuck * (note that there is nothing to do if
3084 1.14 chuck * the old mapping does not have an amap).
3085 1.14 chuck * XXX: is it worthwhile to bother with pmap_copy
3086 1.14 chuck * in this case?
3087 1.14 chuck */
3088 1.14 chuck if (old_entry->aref.ar_amap)
3089 1.14 chuck amap_cow_now(new_map, new_entry);
3090 1.14 chuck
3091 1.14 chuck } else {
3092 1.14 chuck
3093 1.14 chuck /*
3094 1.14 chuck * setup mappings to trigger copy-on-write faults
3095 1.14 chuck * we must write-protect the parent if it has
3096 1.14 chuck * an amap and it is not already "needs_copy"...
3097 1.14 chuck * if it is already "needs_copy" then the parent
3098 1.14 chuck * has already been write-protected by a previous
3099 1.14 chuck * fork operation.
3100 1.14 chuck *
3101 1.14 chuck * if we do not write-protect the parent, then
3102 1.14 chuck * we must be sure to write-protect the child
3103 1.14 chuck * after the pmap_copy() operation.
3104 1.14 chuck *
3105 1.14 chuck * XXX: pmap_copy should have some way of telling
3106 1.14 chuck * us that it didn't do anything so we can avoid
3107 1.14 chuck * calling pmap_protect needlessly.
3108 1.14 chuck */
3109 1.14 chuck
3110 1.14 chuck if (old_entry->aref.ar_amap) {
3111 1.14 chuck
3112 1.14 chuck if (!UVM_ET_ISNEEDSCOPY(old_entry)) {
3113 1.14 chuck if (old_entry->max_protection & VM_PROT_WRITE) {
3114 1.14 chuck pmap_protect(old_map->pmap,
3115 1.14 chuck old_entry->start,
3116 1.14 chuck old_entry->end,
3117 1.14 chuck old_entry->protection &
3118 1.14 chuck ~VM_PROT_WRITE);
3119 1.14 chuck }
3120 1.14 chuck old_entry->etype |= UVM_ET_NEEDSCOPY;
3121 1.14 chuck }
3122 1.14 chuck
3123 1.14 chuck /*
3124 1.14 chuck * parent must now be write-protected
3125 1.14 chuck */
3126 1.14 chuck protect_child = FALSE;
3127 1.14 chuck } else {
3128 1.14 chuck
3129 1.14 chuck /*
3130 1.14 chuck * we only need to protect the child if the
3131 1.14 chuck * parent has write access.
3132 1.14 chuck */
3133 1.14 chuck if (old_entry->max_protection & VM_PROT_WRITE)
3134 1.14 chuck protect_child = TRUE;
3135 1.14 chuck else
3136 1.14 chuck protect_child = FALSE;
3137 1.14 chuck
3138 1.14 chuck }
3139 1.14 chuck
3140 1.14 chuck /*
3141 1.14 chuck * copy the mappings
3142 1.14 chuck * XXX: need a way to tell if this does anything
3143 1.14 chuck */
3144 1.1 mrg
3145 1.14 chuck pmap_copy(new_pmap, old_map->pmap,
3146 1.10 mrg new_entry->start,
3147 1.14 chuck (old_entry->end - old_entry->start),
3148 1.10 mrg old_entry->start);
3149 1.14 chuck
3150 1.14 chuck /*
3151 1.14 chuck * protect the child's mappings if necessary
3152 1.14 chuck */
3153 1.14 chuck if (protect_child) {
3154 1.14 chuck pmap_protect(new_pmap, new_entry->start,
3155 1.14 chuck new_entry->end,
3156 1.14 chuck new_entry->protection &
3157 1.14 chuck ~VM_PROT_WRITE);
3158 1.14 chuck }
3159 1.10 mrg
3160 1.10 mrg }
3161 1.10 mrg break;
3162 1.14 chuck } /* end of switch statement */
3163 1.10 mrg old_entry = old_entry->next;
3164 1.1 mrg }
3165 1.1 mrg
3166 1.10 mrg new_map->size = old_map->size;
3167 1.10 mrg vm_map_unlock(old_map);
3168 1.1 mrg
3169 1.1 mrg #ifdef SYSVSHM
3170 1.10 mrg if (vm1->vm_shm)
3171 1.10 mrg shmfork(vm1, vm2);
3172 1.39 thorpej #endif
3173 1.39 thorpej
3174 1.39 thorpej #ifdef PMAP_FORK
3175 1.39 thorpej pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3176 1.1 mrg #endif
3177 1.1 mrg
3178 1.10 mrg UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3179 1.10 mrg return(vm2);
3180 1.1 mrg }
3181 1.1 mrg
3182 1.1 mrg
3183 1.1 mrg #if defined(DDB)
3184 1.1 mrg
3185 1.1 mrg /*
3186 1.1 mrg * DDB hooks
3187 1.1 mrg */
3188 1.1 mrg
3189 1.1 mrg /*
3190 1.1 mrg * uvm_map_print: print out a map
3191 1.1 mrg */
3192 1.1 mrg
3193 1.10 mrg void
3194 1.10 mrg uvm_map_print(map, full)
3195 1.10 mrg vm_map_t map;
3196 1.10 mrg boolean_t full;
3197 1.10 mrg {
3198 1.1 mrg
3199 1.10 mrg uvm_map_printit(map, full, printf);
3200 1.1 mrg }
3201 1.1 mrg
3202 1.1 mrg /*
3203 1.1 mrg * uvm_map_printit: actually prints the map
3204 1.1 mrg */
3205 1.1 mrg
3206 1.10 mrg void
3207 1.10 mrg uvm_map_printit(map, full, pr)
3208 1.10 mrg vm_map_t map;
3209 1.10 mrg boolean_t full;
3210 1.10 mrg void (*pr) __P((const char *, ...));
3211 1.10 mrg {
3212 1.10 mrg vm_map_entry_t entry;
3213 1.10 mrg
3214 1.10 mrg (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3215 1.53 thorpej (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3216 1.53 thorpej map->nentries, map->size, map->ref_count, map->timestamp,
3217 1.53 thorpej map->flags);
3218 1.16 chuck #ifdef pmap_resident_count
3219 1.16 chuck (*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3220 1.16 chuck pmap_resident_count(map->pmap));
3221 1.16 chuck #else
3222 1.16 chuck /* XXXCDC: this should be required ... */
3223 1.16 chuck (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap);
3224 1.16 chuck #endif
3225 1.10 mrg if (!full)
3226 1.10 mrg return;
3227 1.10 mrg for (entry = map->header.next; entry != &map->header;
3228 1.10 mrg entry = entry->next) {
3229 1.70 kleink (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
3230 1.10 mrg entry, entry->start, entry->end, entry->object.uvm_obj,
3231 1.70 kleink (long long)entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff);
3232 1.10 mrg (*pr)(
3233 1.29 chuck "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n",
3234 1.10 mrg (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3235 1.10 mrg (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3236 1.10 mrg (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3237 1.10 mrg entry->protection, entry->max_protection,
3238 1.10 mrg entry->inheritance, entry->wired_count, entry->advice);
3239 1.10 mrg }
3240 1.1 mrg }
3241 1.1 mrg
3242 1.1 mrg /*
3243 1.1 mrg * uvm_object_print: print out an object
3244 1.1 mrg */
3245 1.1 mrg
3246 1.10 mrg void
3247 1.10 mrg uvm_object_print(uobj, full)
3248 1.10 mrg struct uvm_object *uobj;
3249 1.10 mrg boolean_t full;
3250 1.10 mrg {
3251 1.1 mrg
3252 1.10 mrg uvm_object_printit(uobj, full, printf);
3253 1.1 mrg }
3254 1.1 mrg
3255 1.1 mrg /*
3256 1.1 mrg * uvm_object_printit: actually prints the object
3257 1.1 mrg */
3258 1.1 mrg
3259 1.10 mrg void
3260 1.10 mrg uvm_object_printit(uobj, full, pr)
3261 1.10 mrg struct uvm_object *uobj;
3262 1.10 mrg boolean_t full;
3263 1.10 mrg void (*pr) __P((const char *, ...));
3264 1.10 mrg {
3265 1.10 mrg struct vm_page *pg;
3266 1.10 mrg int cnt = 0;
3267 1.10 mrg
3268 1.71 chs (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
3269 1.71 chs uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
3270 1.42 thorpej if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3271 1.10 mrg (*pr)("refs=<SYSTEM>\n");
3272 1.10 mrg else
3273 1.10 mrg (*pr)("refs=%d\n", uobj->uo_refs);
3274 1.10 mrg
3275 1.71 chs if (!full) {
3276 1.71 chs return;
3277 1.71 chs }
3278 1.10 mrg (*pr)(" PAGES <pg,offset>:\n ");
3279 1.71 chs for (pg = TAILQ_FIRST(&uobj->memq);
3280 1.71 chs pg != NULL;
3281 1.71 chs pg = TAILQ_NEXT(pg, listq), cnt++) {
3282 1.10 mrg (*pr)("<%p,0x%lx> ", pg, pg->offset);
3283 1.71 chs if ((cnt % 3) == 2) {
3284 1.71 chs (*pr)("\n ");
3285 1.71 chs }
3286 1.71 chs }
3287 1.71 chs if ((cnt % 3) != 2) {
3288 1.71 chs (*pr)("\n");
3289 1.10 mrg }
3290 1.1 mrg }
3291 1.1 mrg
3292 1.71 chs const char page_flagbits[] =
3293 1.71 chs "\20\4CLEAN\5BUSY\6WANTED\7TABLED\12FAKE\13FILLED\14DIRTY\15RELEASED"
3294 1.71 chs "\16FAULTING\17CLEANCHK";
3295 1.71 chs const char page_pqflagbits[] =
3296 1.71 chs "\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ";
3297 1.71 chs
3298 1.1 mrg /*
3299 1.1 mrg * uvm_page_print: print out a page
3300 1.1 mrg */
3301 1.1 mrg
3302 1.10 mrg void
3303 1.10 mrg uvm_page_print(pg, full)
3304 1.10 mrg struct vm_page *pg;
3305 1.10 mrg boolean_t full;
3306 1.10 mrg {
3307 1.1 mrg
3308 1.10 mrg uvm_page_printit(pg, full, printf);
3309 1.1 mrg }
3310 1.1 mrg
3311 1.1 mrg /*
3312 1.1 mrg * uvm_page_printit: actually print the page
3313 1.1 mrg */
3314 1.1 mrg
3315 1.10 mrg void
3316 1.10 mrg uvm_page_printit(pg, full, pr)
3317 1.10 mrg struct vm_page *pg;
3318 1.10 mrg boolean_t full;
3319 1.10 mrg void (*pr) __P((const char *, ...));
3320 1.10 mrg {
3321 1.10 mrg struct vm_page *lcv;
3322 1.10 mrg struct uvm_object *uobj;
3323 1.10 mrg struct pglist *pgl;
3324 1.71 chs char pgbuf[128];
3325 1.71 chs char pqbuf[128];
3326 1.1 mrg
3327 1.10 mrg (*pr)("PAGE %p:\n", pg);
3328 1.71 chs bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
3329 1.71 chs bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
3330 1.71 chs (*pr)(" flags=%s, pqflags=%s, vers=%d, wire_count=%d, pa=0x%lx\n",
3331 1.71 chs pgbuf, pqbuf, pg->version, pg->wire_count, (long)pg->phys_addr);
3332 1.10 mrg (*pr)(" uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
3333 1.71 chs pg->uobject, pg->uanon, pg->offset, pg->loan_count);
3334 1.1 mrg #if defined(UVM_PAGE_TRKOWN)
3335 1.10 mrg if (pg->flags & PG_BUSY)
3336 1.10 mrg (*pr)(" owning process = %d, tag=%s\n",
3337 1.10 mrg pg->owner, pg->owner_tag);
3338 1.10 mrg else
3339 1.10 mrg (*pr)(" page not busy, no owner\n");
3340 1.1 mrg #else
3341 1.10 mrg (*pr)(" [page ownership tracking disabled]\n");
3342 1.1 mrg #endif
3343 1.1 mrg
3344 1.10 mrg if (!full)
3345 1.10 mrg return;
3346 1.10 mrg
3347 1.10 mrg /* cross-verify object/anon */
3348 1.10 mrg if ((pg->pqflags & PQ_FREE) == 0) {
3349 1.10 mrg if (pg->pqflags & PQ_ANON) {
3350 1.10 mrg if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3351 1.10 mrg (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3352 1.10 mrg (pg->uanon) ? pg->uanon->u.an_page : NULL);
3353 1.10 mrg else
3354 1.10 mrg (*pr)(" anon backpointer is OK\n");
3355 1.10 mrg } else {
3356 1.10 mrg uobj = pg->uobject;
3357 1.10 mrg if (uobj) {
3358 1.10 mrg (*pr)(" checking object list\n");
3359 1.10 mrg for (lcv = uobj->memq.tqh_first ; lcv ;
3360 1.10 mrg lcv = lcv->listq.tqe_next) {
3361 1.10 mrg if (lcv == pg) break;
3362 1.10 mrg }
3363 1.10 mrg if (lcv)
3364 1.10 mrg (*pr)(" page found on object list\n");
3365 1.10 mrg else
3366 1.10 mrg (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3367 1.10 mrg }
3368 1.10 mrg }
3369 1.10 mrg }
3370 1.1 mrg
3371 1.10 mrg /* cross-verify page queue */
3372 1.73 thorpej if (pg->pqflags & PQ_FREE) {
3373 1.73 thorpej int fl = uvm_page_lookup_freelist(pg);
3374 1.73 thorpej pgl = &uvm.page_free[fl].pgfl_queues[((pg)->flags & PG_ZERO) ?
3375 1.73 thorpej PGFL_ZEROS : PGFL_UNKNOWN];
3376 1.73 thorpej }
3377 1.10 mrg else if (pg->pqflags & PQ_INACTIVE)
3378 1.10 mrg pgl = (pg->pqflags & PQ_SWAPBACKED) ?
3379 1.10 mrg &uvm.page_inactive_swp : &uvm.page_inactive_obj;
3380 1.10 mrg else if (pg->pqflags & PQ_ACTIVE)
3381 1.10 mrg pgl = &uvm.page_active;
3382 1.10 mrg else
3383 1.10 mrg pgl = NULL;
3384 1.10 mrg
3385 1.10 mrg if (pgl) {
3386 1.10 mrg (*pr)(" checking pageq list\n");
3387 1.10 mrg for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) {
3388 1.10 mrg if (lcv == pg) break;
3389 1.10 mrg }
3390 1.10 mrg if (lcv)
3391 1.10 mrg (*pr)(" page found on pageq list\n");
3392 1.10 mrg else
3393 1.10 mrg (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3394 1.10 mrg }
3395 1.1 mrg }
3396 1.1 mrg #endif
3397