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: releng/8.1/sys/vm/vm_kern.c 199583 2009-11-20 15:27:52Z jhb $");
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h> /* for ticks and hz */
71 #include <sys/eventhandler.h>
72 #include <sys/lock.h>
73 #include <sys/mutex.h>
74 #include <sys/proc.h>
75 #include <sys/malloc.h>
76 #include <sys/sysctl.h>
77
78 #include <vm/vm.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pageout.h>
85 #include <vm/vm_extern.h>
86 #include <vm/uma.h>
87
88 vm_map_t kernel_map=0;
89 vm_map_t kmem_map=0;
90 vm_map_t exec_map=0;
91 vm_map_t pipe_map;
92 vm_map_t buffer_map=0;
93
94 /*
95 * kmem_alloc_nofault:
96 *
97 * Allocate a virtual address range with no underlying object and
98 * no initial mapping to physical memory. Any mapping from this
99 * range to physical memory must be explicitly created prior to
100 * its use, typically with pmap_qenter(). Any attempt to create
101 * a mapping on demand through vm_fault() will result in a panic.
102 */
103 vm_offset_t
104 kmem_alloc_nofault(map, size)
105 vm_map_t map;
106 vm_size_t size;
107 {
108 vm_offset_t addr;
109 int result;
110
111 size = round_page(size);
112 addr = vm_map_min(map);
113 result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
114 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
115 if (result != KERN_SUCCESS) {
116 return (0);
117 }
118 return (addr);
119 }
120
121 /*
122 * Allocate wired-down memory in the kernel's address map
123 * or a submap.
124 */
125 vm_offset_t
126 kmem_alloc(map, size)
127 vm_map_t map;
128 vm_size_t size;
129 {
130 vm_offset_t addr;
131 vm_offset_t offset;
132 vm_offset_t i;
133
134 size = round_page(size);
135
136 /*
137 * Use the kernel object for wired-down kernel pages. Assume that no
138 * region of the kernel object is referenced more than once.
139 */
140
141 /*
142 * Locate sufficient space in the map. This will give us the final
143 * virtual address for the new memory, and thus will tell us the
144 * offset within the kernel map.
145 */
146 vm_map_lock(map);
147 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
148 vm_map_unlock(map);
149 return (0);
150 }
151 offset = addr - VM_MIN_KERNEL_ADDRESS;
152 vm_object_reference(kernel_object);
153 vm_map_insert(map, kernel_object, offset, addr, addr + size,
154 VM_PROT_ALL, VM_PROT_ALL, 0);
155 vm_map_unlock(map);
156
157 /*
158 * Guarantee that there are pages already in this object before
159 * calling vm_map_wire. This is to prevent the following
160 * scenario:
161 *
162 * 1) Threads have swapped out, so that there is a pager for the
163 * kernel_object. 2) The kmsg zone is empty, and so we are
164 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
165 * there is no page, but there is a pager, so we call
166 * pager_data_request. But the kmsg zone is empty, so we must
167 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
168 * we get the data back from the pager, it will be (very stale)
169 * non-zero data. kmem_alloc is defined to return zero-filled memory.
170 *
171 * We're intentionally not activating the pages we allocate to prevent a
172 * race with page-out. vm_map_wire will wire the pages.
173 */
174 VM_OBJECT_LOCK(kernel_object);
175 for (i = 0; i < size; i += PAGE_SIZE) {
176 vm_page_t mem;
177
178 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
179 VM_ALLOC_NOBUSY | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
180 mem->valid = VM_PAGE_BITS_ALL;
181 KASSERT((mem->flags & PG_UNMANAGED) != 0,
182 ("kmem_alloc: page %p is managed", mem));
183 }
184 VM_OBJECT_UNLOCK(kernel_object);
185
186 /*
187 * And finally, mark the data as non-pageable.
188 */
189 (void) vm_map_wire(map, addr, addr + size,
190 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
191
192 return (addr);
193 }
194
195 /*
196 * kmem_free:
197 *
198 * Release a region of kernel virtual memory allocated
199 * with kmem_alloc, and return the physical pages
200 * associated with that region.
201 *
202 * This routine may not block on kernel maps.
203 */
204 void
205 kmem_free(map, addr, size)
206 vm_map_t map;
207 vm_offset_t addr;
208 vm_size_t size;
209 {
210
211 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
212 }
213
214 /*
215 * kmem_suballoc:
216 *
217 * Allocates a map to manage a subrange
218 * of the kernel virtual address space.
219 *
220 * Arguments are as follows:
221 *
222 * parent Map to take range from
223 * min, max Returned endpoints of map
224 * size Size of range to find
225 * superpage_align Request that min is superpage aligned
226 */
227 vm_map_t
228 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
229 vm_size_t size, boolean_t superpage_align)
230 {
231 int ret;
232 vm_map_t result;
233
234 size = round_page(size);
235
236 *min = vm_map_min(parent);
237 ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
238 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
239 MAP_ACC_NO_CHARGE);
240 if (ret != KERN_SUCCESS)
241 panic("kmem_suballoc: bad status return of %d", ret);
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 * We don't worry about expanding the map (adding entries) since entries
264 * for wired maps are statically allocated.
265 *
266 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
267 * which we never free.
268 */
269 vm_offset_t
270 kmem_malloc(map, size, flags)
271 vm_map_t map;
272 vm_size_t size;
273 int flags;
274 {
275 vm_offset_t offset, i;
276 vm_map_entry_t entry;
277 vm_offset_t addr;
278 vm_page_t m;
279 int pflags;
280
281 size = round_page(size);
282 addr = vm_map_min(map);
283
284 /*
285 * Locate sufficient space in the map. This will give us the final
286 * virtual address for the new memory, and thus will tell us the
287 * offset within the kernel map.
288 */
289 vm_map_lock(map);
290 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
291 vm_map_unlock(map);
292 if ((flags & M_NOWAIT) == 0) {
293 for (i = 0; i < 8; i++) {
294 EVENTHANDLER_INVOKE(vm_lowmem, 0);
295 uma_reclaim();
296 vm_map_lock(map);
297 if (vm_map_findspace(map, vm_map_min(map),
298 size, &addr) == 0) {
299 break;
300 }
301 vm_map_unlock(map);
302 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
303 }
304 if (i == 8) {
305 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
306 (long)size, (long)map->size);
307 }
308 } else {
309 return (0);
310 }
311 }
312 offset = addr - VM_MIN_KERNEL_ADDRESS;
313 vm_object_reference(kmem_object);
314 vm_map_insert(map, kmem_object, offset, addr, addr + size,
315 VM_PROT_ALL, VM_PROT_ALL, 0);
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 KASSERT((m->flags & PG_UNMANAGED) != 0,
368 ("kmem_malloc: page %p is managed", m));
369 }
370 VM_OBJECT_UNLOCK(kmem_object);
371
372 /*
373 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
374 * be able to extend the previous entry so there will be a new entry
375 * exactly corresponding to this address range and it will have
376 * wired_count == 0.
377 */
378 if (!vm_map_lookup_entry(map, addr, &entry) ||
379 entry->start != addr || entry->end != addr + size ||
380 entry->wired_count != 0)
381 panic("kmem_malloc: entry not found or misaligned");
382 entry->wired_count = 1;
383
384 /*
385 * At this point, the kmem_object must be unlocked because
386 * vm_map_simplify_entry() calls vm_object_deallocate(), which
387 * locks the kmem_object.
388 */
389 vm_map_simplify_entry(map, entry);
390
391 /*
392 * Loop thru pages, entering them in the pmap.
393 */
394 VM_OBJECT_LOCK(kmem_object);
395 for (i = 0; i < size; i += PAGE_SIZE) {
396 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
397 /*
398 * Because this is kernel_pmap, this call will not block.
399 */
400 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
401 TRUE);
402 vm_page_wakeup(m);
403 }
404 VM_OBJECT_UNLOCK(kmem_object);
405 vm_map_unlock(map);
406
407 return (addr);
408 }
409
410 /*
411 * kmem_alloc_wait:
412 *
413 * Allocates pageable memory from a sub-map of the kernel. If the submap
414 * has no room, the caller sleeps waiting for more memory in the submap.
415 *
416 * This routine may block.
417 */
418 vm_offset_t
419 kmem_alloc_wait(map, size)
420 vm_map_t map;
421 vm_size_t size;
422 {
423 vm_offset_t addr;
424
425 size = round_page(size);
426 if (!swap_reserve(size))
427 return (0);
428
429 for (;;) {
430 /*
431 * To make this work for more than one map, use the map's lock
432 * to lock out sleepers/wakers.
433 */
434 vm_map_lock(map);
435 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
436 break;
437 /* no space now; see if we can ever get space */
438 if (vm_map_max(map) - vm_map_min(map) < size) {
439 vm_map_unlock(map);
440 swap_release(size);
441 return (0);
442 }
443 map->needs_wakeup = TRUE;
444 vm_map_unlock_and_wait(map, 0);
445 }
446 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
447 VM_PROT_ALL, MAP_ACC_CHARGED);
448 vm_map_unlock(map);
449 return (addr);
450 }
451
452 /*
453 * kmem_free_wakeup:
454 *
455 * Returns memory to a submap of the kernel, and wakes up any processes
456 * waiting for memory in that map.
457 */
458 void
459 kmem_free_wakeup(map, addr, size)
460 vm_map_t map;
461 vm_offset_t addr;
462 vm_size_t size;
463 {
464
465 vm_map_lock(map);
466 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
467 if (map->needs_wakeup) {
468 map->needs_wakeup = FALSE;
469 vm_map_wakeup(map);
470 }
471 vm_map_unlock(map);
472 }
473
474 /*
475 * kmem_init:
476 *
477 * Create the kernel map; insert a mapping covering kernel text,
478 * data, bss, and all space allocated thus far (`boostrap' data). The
479 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
480 * `start' as allocated, and the range between `start' and `end' as free.
481 */
482 void
483 kmem_init(start, end)
484 vm_offset_t start, end;
485 {
486 vm_map_t m;
487
488 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
489 m->system_map = 1;
490 vm_map_lock(m);
491 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
492 kernel_map = m;
493 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
494 #ifdef __amd64__
495 KERNBASE,
496 #else
497 VM_MIN_KERNEL_ADDRESS,
498 #endif
499 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
500 /* ... and ending with the completion of the above `insert' */
501 vm_map_unlock(m);
502 }
503
504 #ifdef DIAGNOSTIC
505 /*
506 * Allow userspace to directly trigger the VM drain routine for testing
507 * purposes.
508 */
509 static int
510 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
511 {
512 int error, i;
513
514 i = 0;
515 error = sysctl_handle_int(oidp, &i, 0, req);
516 if (error)
517 return (error);
518 if (i)
519 EVENTHANDLER_INVOKE(vm_lowmem, 0);
520 return (0);
521 }
522
523 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
524 debug_vm_lowmem, "I", "set to trigger vm_lowmem event");
525 #endif
Cache object: 19a6afab68b960779575eb35aea4db95
|