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 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94
33 *
34 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
45 *
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61 /*
62 * Kernel memory management.
63 */
64
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h> /* for ticks and hz */
71 #include <sys/lock.h>
72 #include <sys/mutex.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 <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_pageout.h>
83 #include <vm/vm_extern.h>
84
85 vm_map_t kernel_map=0;
86 vm_map_t kmem_map=0;
87 vm_map_t exec_map=0;
88 vm_map_t pipe_map;
89 vm_map_t buffer_map=0;
90
91 /*
92 * kmem_alloc_nofault:
93 *
94 * Allocate a virtual address range with no underlying object and
95 * no initial mapping to physical memory. Any mapping from this
96 * range to physical memory must be explicitly created prior to
97 * its use, typically with pmap_qenter(). Any attempt to create
98 * a mapping on demand through vm_fault() will result in a panic.
99 */
100 vm_offset_t
101 kmem_alloc_nofault(map, size)
102 vm_map_t map;
103 vm_size_t size;
104 {
105 vm_offset_t addr;
106 int result;
107
108 size = round_page(size);
109 addr = vm_map_min(map);
110 result = vm_map_find(map, NULL, 0,
111 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
112 if (result != KERN_SUCCESS) {
113 return (0);
114 }
115 return (addr);
116 }
117
118 /*
119 * Allocate wired-down memory in the kernel's address map
120 * or a submap.
121 */
122 vm_offset_t
123 kmem_alloc(map, size)
124 vm_map_t map;
125 vm_size_t size;
126 {
127 vm_offset_t addr;
128 vm_offset_t offset;
129 vm_offset_t i;
130
131 size = round_page(size);
132
133 /*
134 * Use the kernel object for wired-down kernel pages. Assume that no
135 * region of the kernel object is referenced more than once.
136 */
137
138 /*
139 * Locate sufficient space in the map. This will give us the final
140 * virtual address for the new memory, and thus will tell us the
141 * offset within the kernel map.
142 */
143 vm_map_lock(map);
144 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
145 vm_map_unlock(map);
146 return (0);
147 }
148 offset = addr - VM_MIN_KERNEL_ADDRESS;
149 vm_object_reference(kernel_object);
150 vm_map_insert(map, kernel_object, offset, addr, addr + size,
151 VM_PROT_ALL, VM_PROT_ALL, 0);
152 vm_map_unlock(map);
153
154 /*
155 * Guarantee that there are pages already in this object before
156 * calling vm_map_wire. This is to prevent the following
157 * scenario:
158 *
159 * 1) Threads have swapped out, so that there is a pager for the
160 * kernel_object. 2) The kmsg zone is empty, and so we are
161 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
162 * there is no page, but there is a pager, so we call
163 * pager_data_request. But the kmsg zone is empty, so we must
164 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
165 * we get the data back from the pager, it will be (very stale)
166 * non-zero data. kmem_alloc is defined to return zero-filled memory.
167 *
168 * We're intentionally not activating the pages we allocate to prevent a
169 * race with page-out. vm_map_wire will wire the pages.
170 */
171 VM_OBJECT_LOCK(kernel_object);
172 for (i = 0; i < size; i += PAGE_SIZE) {
173 vm_page_t mem;
174
175 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
176 VM_ALLOC_NOBUSY | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
177 mem->valid = VM_PAGE_BITS_ALL;
178 vm_page_lock_queues();
179 vm_page_unmanage(mem);
180 vm_page_unlock_queues();
181 }
182 VM_OBJECT_UNLOCK(kernel_object);
183
184 /*
185 * And finally, mark the data as non-pageable.
186 */
187 (void) vm_map_wire(map, addr, addr + size,
188 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
189
190 return (addr);
191 }
192
193 /*
194 * kmem_free:
195 *
196 * Release a region of kernel virtual memory allocated
197 * with kmem_alloc, and return the physical pages
198 * associated with that region.
199 *
200 * This routine may not block on kernel maps.
201 */
202 void
203 kmem_free(map, addr, size)
204 vm_map_t map;
205 vm_offset_t addr;
206 vm_size_t size;
207 {
208
209 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
210 }
211
212 /*
213 * kmem_suballoc:
214 *
215 * Allocates a map to manage a subrange
216 * of the kernel virtual address space.
217 *
218 * Arguments are as follows:
219 *
220 * parent Map to take range from
221 * min, max Returned endpoints of map
222 * size Size of range to find
223 */
224 vm_map_t
225 kmem_suballoc(parent, min, max, size)
226 vm_map_t parent;
227 vm_offset_t *min, *max;
228 vm_size_t size;
229 {
230 int ret;
231 vm_map_t result;
232
233 size = round_page(size);
234
235 *min = (vm_offset_t) vm_map_min(parent);
236 ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
237 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
238 if (ret != KERN_SUCCESS) {
239 printf("kmem_suballoc: bad status return of %d.\n", ret);
240 panic("kmem_suballoc");
241 }
242 *max = *min + size;
243 result = vm_map_create(vm_map_pmap(parent), *min, *max);
244 if (result == NULL)
245 panic("kmem_suballoc: cannot create submap");
246 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
247 panic("kmem_suballoc: unable to change range to submap");
248 return (result);
249 }
250
251 /*
252 * kmem_malloc:
253 *
254 * Allocate wired-down memory in the kernel's address map for the higher
255 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
256 * kmem_alloc() because we may need to allocate memory at interrupt
257 * level where we cannot block (canwait == FALSE).
258 *
259 * This routine has its own private kernel submap (kmem_map) and object
260 * (kmem_object). This, combined with the fact that only malloc uses
261 * this routine, ensures that we will never block in map or object waits.
262 *
263 * Note that this still only works in a uni-processor environment and
264 * when called at splhigh().
265 *
266 * We don't worry about expanding the map (adding entries) since entries
267 * for wired maps are statically allocated.
268 *
269 * NOTE: This routine is not supposed to block if M_NOWAIT is set, but
270 * I have not verified that it actually does not block.
271 *
272 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
273 * which we never free.
274 */
275 vm_offset_t
276 kmem_malloc(map, size, flags)
277 vm_map_t map;
278 vm_size_t size;
279 int flags;
280 {
281 vm_offset_t offset, i;
282 vm_map_entry_t entry;
283 vm_offset_t addr;
284 vm_page_t m;
285 int pflags;
286
287 size = round_page(size);
288 addr = vm_map_min(map);
289
290 /*
291 * Locate sufficient space in the map. This will give us the final
292 * virtual address for the new memory, and thus will tell us the
293 * offset within the kernel map.
294 */
295 vm_map_lock(map);
296 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
297 vm_map_unlock(map);
298 if ((flags & M_NOWAIT) == 0)
299 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
300 (long)size, (long)map->size);
301 return (0);
302 }
303 offset = addr - VM_MIN_KERNEL_ADDRESS;
304 vm_object_reference(kmem_object);
305 vm_map_insert(map, kmem_object, offset, addr, addr + size,
306 VM_PROT_ALL, VM_PROT_ALL, 0);
307
308 /*
309 * Note: if M_NOWAIT specified alone, allocate from
310 * interrupt-safe queues only (just the free list). If
311 * M_USE_RESERVE is also specified, we can also
312 * allocate from the cache. Neither of the latter two
313 * flags may be specified from an interrupt since interrupts
314 * are not allowed to mess with the cache queue.
315 */
316
317 if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
318 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
319 else
320 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
321
322 if (flags & M_ZERO)
323 pflags |= VM_ALLOC_ZERO;
324
325 VM_OBJECT_LOCK(kmem_object);
326 for (i = 0; i < size; i += PAGE_SIZE) {
327 retry:
328 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
329
330 /*
331 * Ran out of space, free everything up and return. Don't need
332 * to lock page queues here as we know that the pages we got
333 * aren't on any queues.
334 */
335 if (m == NULL) {
336 if ((flags & M_NOWAIT) == 0) {
337 VM_OBJECT_UNLOCK(kmem_object);
338 vm_map_unlock(map);
339 VM_WAIT;
340 vm_map_lock(map);
341 VM_OBJECT_LOCK(kmem_object);
342 goto retry;
343 }
344 /*
345 * Free the pages before removing the map entry.
346 * They are already marked busy. Calling
347 * vm_map_delete before the pages has been freed or
348 * unbusied will cause a deadlock.
349 */
350 while (i != 0) {
351 i -= PAGE_SIZE;
352 m = vm_page_lookup(kmem_object,
353 OFF_TO_IDX(offset + i));
354 vm_page_lock_queues();
355 vm_page_unwire(m, 0);
356 vm_page_free(m);
357 vm_page_unlock_queues();
358 }
359 VM_OBJECT_UNLOCK(kmem_object);
360 vm_map_delete(map, addr, addr + size);
361 vm_map_unlock(map);
362 return (0);
363 }
364 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
365 pmap_zero_page(m);
366 m->valid = VM_PAGE_BITS_ALL;
367 vm_page_lock_queues();
368 vm_page_unmanage(m);
369 vm_page_unlock_queues();
370 }
371 VM_OBJECT_UNLOCK(kmem_object);
372
373 /*
374 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
375 * be able to extend the previous entry so there will be a new entry
376 * exactly corresponding to this address range and it will have
377 * wired_count == 0.
378 */
379 if (!vm_map_lookup_entry(map, addr, &entry) ||
380 entry->start != addr || entry->end != addr + size ||
381 entry->wired_count != 0)
382 panic("kmem_malloc: entry not found or misaligned");
383 entry->wired_count = 1;
384
385 /*
386 * At this point, the kmem_object must be unlocked because
387 * vm_map_simplify_entry() calls vm_object_deallocate(), which
388 * locks the kmem_object.
389 */
390 vm_map_simplify_entry(map, entry);
391
392 /*
393 * Loop thru pages, entering them in the pmap. (We cannot add them to
394 * the wired count without wrapping the vm_page_queue_lock in
395 * splimp...)
396 */
397 VM_OBJECT_LOCK(kmem_object);
398 for (i = 0; i < size; i += PAGE_SIZE) {
399 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
400 /*
401 * Because this is kernel_pmap, this call will not block.
402 */
403 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
404 vm_page_lock_queues();
405 vm_page_flag_set(m, PG_WRITEABLE | PG_REFERENCED);
406 vm_page_wakeup(m);
407 vm_page_unlock_queues();
408 }
409 VM_OBJECT_UNLOCK(kmem_object);
410 vm_map_unlock(map);
411
412 return (addr);
413 }
414
415 /*
416 * kmem_alloc_wait:
417 *
418 * Allocates pageable memory from a sub-map of the kernel. If the submap
419 * has no room, the caller sleeps waiting for more memory in the submap.
420 *
421 * This routine may block.
422 */
423 vm_offset_t
424 kmem_alloc_wait(map, size)
425 vm_map_t map;
426 vm_size_t size;
427 {
428 vm_offset_t addr;
429
430 size = round_page(size);
431
432 for (;;) {
433 /*
434 * To make this work for more than one map, use the map's lock
435 * to lock out sleepers/wakers.
436 */
437 vm_map_lock(map);
438 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
439 break;
440 /* no space now; see if we can ever get space */
441 if (vm_map_max(map) - vm_map_min(map) < size) {
442 vm_map_unlock(map);
443 return (0);
444 }
445 map->needs_wakeup = TRUE;
446 vm_map_unlock_and_wait(map, FALSE);
447 }
448 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
449 vm_map_unlock(map);
450 return (addr);
451 }
452
453 /*
454 * kmem_free_wakeup:
455 *
456 * Returns memory to a submap of the kernel, and wakes up any processes
457 * waiting for memory in that map.
458 */
459 void
460 kmem_free_wakeup(map, addr, size)
461 vm_map_t map;
462 vm_offset_t addr;
463 vm_size_t size;
464 {
465
466 vm_map_lock(map);
467 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
468 if (map->needs_wakeup) {
469 map->needs_wakeup = FALSE;
470 vm_map_wakeup(map);
471 }
472 vm_map_unlock(map);
473 }
474
475 /*
476 * kmem_init:
477 *
478 * Create the kernel map; insert a mapping covering kernel text,
479 * data, bss, and all space allocated thus far (`boostrap' data). The
480 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
481 * `start' as allocated, and the range between `start' and `end' as free.
482 */
483 void
484 kmem_init(start, end)
485 vm_offset_t start, end;
486 {
487 vm_map_t m;
488
489 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
490 m->system_map = 1;
491 vm_map_lock(m);
492 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
493 kernel_map = m;
494 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
495 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL,
496 MAP_NOFAULT);
497 /* ... and ending with the completion of the above `insert' */
498 vm_map_unlock(m);
499 }
Cache object: 02dad2a06c2e6cfa7c14f69221094f6d
|