FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_kern.c
1 /*
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94
37 *
38 *
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
41 *
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43 *
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
49 *
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53 *
54 * Carnegie Mellon requests users of this software to return to
55 *
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
60 *
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
63 *
64 * $FreeBSD$
65 */
66
67 /*
68 * Kernel memory management.
69 */
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/proc.h>
74 #include <sys/malloc.h>
75
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <sys/lock.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_pageout.h>
84 #include <vm/vm_extern.h>
85
86 vm_map_t kernel_map=0;
87 vm_map_t kmem_map=0;
88 vm_map_t exec_map=0;
89 vm_map_t clean_map=0;
90 vm_map_t buffer_map=0;
91 vm_map_t mb_map=0;
92 int mb_map_full=0;
93
94 /*
95 * kmem_alloc_pageable:
96 *
97 * Allocate pageable memory to the kernel's address map.
98 * "map" must be kernel_map or a submap of kernel_map.
99 */
100
101 vm_offset_t
102 kmem_alloc_pageable(map, size)
103 vm_map_t map;
104 register vm_size_t size;
105 {
106 vm_offset_t addr;
107 register int result;
108
109 size = round_page(size);
110 addr = vm_map_min(map);
111 result = vm_map_find(map, NULL, (vm_offset_t) 0,
112 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
113 if (result != KERN_SUCCESS) {
114 return (0);
115 }
116 return (addr);
117 }
118
119 /*
120 * kmem_alloc_nofault:
121 *
122 * Same as kmem_alloc_pageable, except that it create a nofault entry.
123 */
124
125 vm_offset_t
126 kmem_alloc_nofault(map, size)
127 vm_map_t map;
128 register vm_size_t size;
129 {
130 vm_offset_t addr;
131 register int result;
132
133 size = round_page(size);
134 addr = vm_map_min(map);
135 result = vm_map_find(map, NULL, (vm_offset_t) 0,
136 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
137 if (result != KERN_SUCCESS) {
138 return (0);
139 }
140 return (addr);
141 }
142
143 /*
144 * Allocate wired-down memory in the kernel's address map
145 * or a submap.
146 */
147 vm_offset_t
148 kmem_alloc(map, size)
149 register vm_map_t map;
150 register vm_size_t size;
151 {
152 vm_offset_t addr;
153 register vm_offset_t offset;
154 vm_offset_t i;
155
156 size = round_page(size);
157
158 /*
159 * Use the kernel object for wired-down kernel pages. Assume that no
160 * region of the kernel object is referenced more than once.
161 */
162
163 /*
164 * Locate sufficient space in the map. This will give us the final
165 * virtual address for the new memory, and thus will tell us the
166 * offset within the kernel map.
167 */
168 vm_map_lock(map);
169 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
170 vm_map_unlock(map);
171 return (0);
172 }
173 offset = addr - VM_MIN_KERNEL_ADDRESS;
174 vm_object_reference(kernel_object);
175 vm_map_insert(map, kernel_object, offset, addr, addr + size,
176 VM_PROT_ALL, VM_PROT_ALL, 0);
177 vm_map_unlock(map);
178
179 /*
180 * Guarantee that there are pages already in this object before
181 * calling vm_map_pageable. This is to prevent the following
182 * scenario:
183 *
184 * 1) Threads have swapped out, so that there is a pager for the
185 * kernel_object. 2) The kmsg zone is empty, and so we are
186 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault;
187 * there is no page, but there is a pager, so we call
188 * pager_data_request. But the kmsg zone is empty, so we must
189 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
190 * we get the data back from the pager, it will be (very stale)
191 * non-zero data. kmem_alloc is defined to return zero-filled memory.
192 *
193 * We're intentionally not activating the pages we allocate to prevent a
194 * race with page-out. vm_map_pageable will wire the pages.
195 */
196
197 for (i = 0; i < size; i += PAGE_SIZE) {
198 vm_page_t mem;
199
200 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
201 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
202 if ((mem->flags & PG_ZERO) == 0)
203 vm_page_zero_fill(mem);
204 mem->valid = VM_PAGE_BITS_ALL;
205 vm_page_flag_clear(mem, PG_ZERO);
206 vm_page_wakeup(mem);
207 }
208
209 /*
210 * And finally, mark the data as non-pageable.
211 */
212
213 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE);
214
215 return (addr);
216 }
217
218 /*
219 * kmem_free:
220 *
221 * Release a region of kernel virtual memory allocated
222 * with kmem_alloc, and return the physical pages
223 * associated with that region.
224 *
225 * This routine may not block on kernel maps.
226 */
227 void
228 kmem_free(map, addr, size)
229 vm_map_t map;
230 register vm_offset_t addr;
231 vm_size_t size;
232 {
233 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
234 }
235
236 /*
237 * kmem_suballoc:
238 *
239 * Allocates a map to manage a subrange
240 * of the kernel virtual address space.
241 *
242 * Arguments are as follows:
243 *
244 * parent Map to take range from
245 * size Size of range to find
246 * min, max Returned endpoints of map
247 * pageable Can the region be paged
248 */
249 vm_map_t
250 kmem_suballoc(parent, min, max, size)
251 register vm_map_t parent;
252 vm_offset_t *min, *max;
253 register vm_size_t size;
254 {
255 register int ret;
256 vm_map_t result;
257
258 size = round_page(size);
259
260 *min = (vm_offset_t) vm_map_min(parent);
261 ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
262 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
263 if (ret != KERN_SUCCESS) {
264 printf("kmem_suballoc: bad status return of %d.\n", ret);
265 panic("kmem_suballoc");
266 }
267 *max = *min + size;
268 result = vm_map_create(vm_map_pmap(parent), *min, *max);
269 if (result == NULL)
270 panic("kmem_suballoc: cannot create submap");
271 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
272 panic("kmem_suballoc: unable to change range to submap");
273 return (result);
274 }
275
276 /*
277 * kmem_malloc:
278 *
279 * Allocate wired-down memory in the kernel's address map for the higher
280 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
281 * kmem_alloc() because we may need to allocate memory at interrupt
282 * level where we cannot block (canwait == FALSE).
283 *
284 * This routine has its own private kernel submap (kmem_map) and object
285 * (kmem_object). This, combined with the fact that only malloc uses
286 * this routine, ensures that we will never block in map or object waits.
287 *
288 * Note that this still only works in a uni-processor environment and
289 * when called at splhigh().
290 *
291 * We don't worry about expanding the map (adding entries) since entries
292 * for wired maps are statically allocated.
293 *
294 * NOTE: This routine is not supposed to block if M_NOWAIT is set, but
295 * I have not verified that it actually does not block.
296 */
297 vm_offset_t
298 kmem_malloc(map, size, flags)
299 register vm_map_t map;
300 register vm_size_t size;
301 int flags;
302 {
303 register vm_offset_t offset, i;
304 vm_map_entry_t entry;
305 vm_offset_t addr;
306 vm_page_t m;
307
308 if (map != kmem_map && map != mb_map)
309 panic("kmem_malloc: map != {kmem,mb}_map");
310
311 size = round_page(size);
312 addr = vm_map_min(map);
313
314 /*
315 * Locate sufficient space in the map. This will give us the final
316 * virtual address for the new memory, and thus will tell us the
317 * offset within the kernel map.
318 */
319 vm_map_lock(map);
320 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
321 vm_map_unlock(map);
322 if (map == mb_map) {
323 mb_map_full = TRUE;
324 printf("Out of mbuf clusters - adjust NMBCLUSTERS or increase maxusers!\n");
325 return (0);
326 }
327 if ((flags & M_NOWAIT) == 0)
328 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
329 (long)size, (long)map->size);
330 return (0);
331 }
332 offset = addr - VM_MIN_KERNEL_ADDRESS;
333 vm_object_reference(kmem_object);
334 vm_map_insert(map, kmem_object, offset, addr, addr + size,
335 VM_PROT_ALL, VM_PROT_ALL, 0);
336
337 for (i = 0; i < size; i += PAGE_SIZE) {
338 /*
339 * Note: if M_NOWAIT specified alone, allocate from
340 * interrupt-safe queues only (just the free list). If
341 * M_ASLEEP or M_USE_RESERVE is also specified, we can also
342 * allocate from the cache. Neither of the latter two
343 * flags may be specified from an interrupt since interrupts
344 * are not allowed to mess with the cache queue.
345 */
346 retry:
347 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i),
348 ((flags & (M_NOWAIT|M_ASLEEP|M_USE_RESERVE)) == M_NOWAIT) ?
349 VM_ALLOC_INTERRUPT :
350 VM_ALLOC_SYSTEM);
351
352 /*
353 * Ran out of space, free everything up and return. Don't need
354 * to lock page queues here as we know that the pages we got
355 * aren't on any queues.
356 */
357 if (m == NULL) {
358 if ((flags & M_NOWAIT) == 0) {
359 vm_map_unlock(map);
360 VM_WAIT;
361 vm_map_lock(map);
362 goto retry;
363 }
364 /*
365 * Free the pages before removing the map entry.
366 * They are already marked busy. Calling
367 * vm_map_delete before the pages has been freed or
368 * unbusied will cause a deadlock.
369 */
370 while (i != 0) {
371 i -= PAGE_SIZE;
372 m = vm_page_lookup(kmem_object,
373 OFF_TO_IDX(offset + i));
374 vm_page_free(m);
375 }
376 vm_map_delete(map, addr, addr + size);
377 vm_map_unlock(map);
378 if (flags & M_ASLEEP) {
379 VM_AWAIT;
380 }
381 return (0);
382 }
383 vm_page_flag_clear(m, PG_ZERO);
384 m->valid = VM_PAGE_BITS_ALL;
385 }
386
387 /*
388 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
389 * be able to extend the previous entry so there will be a new entry
390 * exactly corresponding to this address range and it will have
391 * wired_count == 0.
392 */
393 if (!vm_map_lookup_entry(map, addr, &entry) ||
394 entry->start != addr || entry->end != addr + size ||
395 entry->wired_count != 0)
396 panic("kmem_malloc: entry not found or misaligned");
397 entry->wired_count = 1;
398
399 vm_map_simplify_entry(map, entry);
400
401 /*
402 * Loop thru pages, entering them in the pmap. (We cannot add them to
403 * the wired count without wrapping the vm_page_queue_lock in
404 * splimp...)
405 */
406 for (i = 0; i < size; i += PAGE_SIZE) {
407 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
408 vm_page_wire(m);
409 vm_page_wakeup(m);
410 /*
411 * Because this is kernel_pmap, this call will not block.
412 */
413 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
414 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED);
415 }
416 vm_map_unlock(map);
417
418 return (addr);
419 }
420
421 /*
422 * kmem_alloc_wait:
423 *
424 * Allocates pageable memory from a sub-map of the kernel. If the submap
425 * has no room, the caller sleeps waiting for more memory in the submap.
426 *
427 * This routine may block.
428 */
429
430 vm_offset_t
431 kmem_alloc_wait(map, size)
432 vm_map_t map;
433 vm_size_t size;
434 {
435 vm_offset_t addr;
436
437 size = round_page(size);
438
439 for (;;) {
440 /*
441 * To make this work for more than one map, use the map's lock
442 * to lock out sleepers/wakers.
443 */
444 vm_map_lock(map);
445 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
446 break;
447 /* no space now; see if we can ever get space */
448 if (vm_map_max(map) - vm_map_min(map) < size) {
449 vm_map_unlock(map);
450 return (0);
451 }
452 vm_map_unlock(map);
453 tsleep(map, PVM, "kmaw", 0);
454 }
455 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
456 vm_map_unlock(map);
457 return (addr);
458 }
459
460 /*
461 * kmem_free_wakeup:
462 *
463 * Returns memory to a submap of the kernel, and wakes up any processes
464 * waiting for memory in that map.
465 */
466 void
467 kmem_free_wakeup(map, addr, size)
468 vm_map_t map;
469 vm_offset_t addr;
470 vm_size_t size;
471 {
472 vm_map_lock(map);
473 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
474 wakeup(map);
475 vm_map_unlock(map);
476 }
477
478 /*
479 * kmem_init:
480 *
481 * Create the kernel map; insert a mapping covering kernel text,
482 * data, bss, and all space allocated thus far (`boostrap' data). The
483 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
484 * `start' as allocated, and the range between `start' and `end' as free.
485 */
486
487 void
488 kmem_init(start, end)
489 vm_offset_t start, end;
490 {
491 register vm_map_t m;
492
493 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
494 vm_map_lock(m);
495 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
496 kernel_map = m;
497 kernel_map->system_map = 1;
498 (void) vm_map_insert(m, NULL, (vm_offset_t) 0,
499 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
500 /* ... and ending with the completion of the above `insert' */
501 vm_map_unlock(m);
502 }
503
Cache object: 6eea28237ccfa8a318a2cead34b56ee8
|