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/9.1/sys/vm/vm_kern.c 237949 2012-07-02 03:49:52Z alc $");
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 const void *zero_region;
95 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
96
97 /*
98 * kmem_alloc_nofault:
99 *
100 * Allocate a virtual address range with no underlying object and
101 * no initial mapping to physical memory. Any mapping from this
102 * range to physical memory must be explicitly created prior to
103 * its use, typically with pmap_qenter(). Any attempt to create
104 * a mapping on demand through vm_fault() will result in a panic.
105 */
106 vm_offset_t
107 kmem_alloc_nofault(map, size)
108 vm_map_t map;
109 vm_size_t size;
110 {
111 vm_offset_t addr;
112 int result;
113
114 size = round_page(size);
115 addr = vm_map_min(map);
116 result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
117 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
118 if (result != KERN_SUCCESS) {
119 return (0);
120 }
121 return (addr);
122 }
123
124 /*
125 * kmem_alloc_nofault_space:
126 *
127 * Allocate a virtual address range with no underlying object and
128 * no initial mapping to physical memory within the specified
129 * address space. Any mapping from this range to physical memory
130 * must be explicitly created prior to its use, typically with
131 * pmap_qenter(). Any attempt to create a mapping on demand
132 * through vm_fault() will result in a panic.
133 */
134 vm_offset_t
135 kmem_alloc_nofault_space(map, size, find_space)
136 vm_map_t map;
137 vm_size_t size;
138 int find_space;
139 {
140 vm_offset_t addr;
141 int result;
142
143 size = round_page(size);
144 addr = vm_map_min(map);
145 result = vm_map_find(map, NULL, 0, &addr, size, find_space,
146 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
147 if (result != KERN_SUCCESS) {
148 return (0);
149 }
150 return (addr);
151 }
152
153 /*
154 * Allocate wired-down memory in the kernel's address map
155 * or a submap.
156 */
157 vm_offset_t
158 kmem_alloc(map, size)
159 vm_map_t map;
160 vm_size_t size;
161 {
162 vm_offset_t addr;
163 vm_offset_t offset;
164
165 size = round_page(size);
166
167 /*
168 * Use the kernel object for wired-down kernel pages. Assume that no
169 * region of the kernel object is referenced more than once.
170 */
171
172 /*
173 * Locate sufficient space in the map. This will give us the final
174 * virtual address for the new memory, and thus will tell us the
175 * offset within the kernel map.
176 */
177 vm_map_lock(map);
178 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
179 vm_map_unlock(map);
180 return (0);
181 }
182 offset = addr - VM_MIN_KERNEL_ADDRESS;
183 vm_object_reference(kernel_object);
184 vm_map_insert(map, kernel_object, offset, addr, addr + size,
185 VM_PROT_ALL, VM_PROT_ALL, 0);
186 vm_map_unlock(map);
187
188 /*
189 * And finally, mark the data as non-pageable.
190 */
191 (void) vm_map_wire(map, addr, addr + size,
192 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
193
194 return (addr);
195 }
196
197 /*
198 * kmem_free:
199 *
200 * Release a region of kernel virtual memory allocated
201 * with kmem_alloc, and return the physical pages
202 * associated with that region.
203 *
204 * This routine may not block on kernel maps.
205 */
206 void
207 kmem_free(map, addr, size)
208 vm_map_t map;
209 vm_offset_t addr;
210 vm_size_t size;
211 {
212
213 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
214 }
215
216 /*
217 * kmem_suballoc:
218 *
219 * Allocates a map to manage a subrange
220 * of the kernel virtual address space.
221 *
222 * Arguments are as follows:
223 *
224 * parent Map to take range from
225 * min, max Returned endpoints of map
226 * size Size of range to find
227 * superpage_align Request that min is superpage aligned
228 */
229 vm_map_t
230 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
231 vm_size_t size, boolean_t superpage_align)
232 {
233 int ret;
234 vm_map_t result;
235
236 size = round_page(size);
237
238 *min = vm_map_min(parent);
239 ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
240 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
241 MAP_ACC_NO_CHARGE);
242 if (ret != KERN_SUCCESS)
243 panic("kmem_suballoc: bad status return of %d", ret);
244 *max = *min + size;
245 result = vm_map_create(vm_map_pmap(parent), *min, *max);
246 if (result == NULL)
247 panic("kmem_suballoc: cannot create submap");
248 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
249 panic("kmem_suballoc: unable to change range to submap");
250 return (result);
251 }
252
253 /*
254 * kmem_malloc:
255 *
256 * Allocate wired-down memory in the kernel's address map for the higher
257 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
258 * kmem_alloc() because we may need to allocate memory at interrupt
259 * level where we cannot block (canwait == FALSE).
260 *
261 * This routine has its own private kernel submap (kmem_map) and object
262 * (kmem_object). This, combined with the fact that only malloc uses
263 * this routine, ensures that we will never block in map or object waits.
264 *
265 * We don't worry about expanding the map (adding entries) since entries
266 * for wired maps are statically allocated.
267 *
268 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
269 * which we never free.
270 */
271 vm_offset_t
272 kmem_malloc(map, size, flags)
273 vm_map_t map;
274 vm_size_t size;
275 int flags;
276 {
277 vm_offset_t addr;
278 int i, rv;
279
280 size = round_page(size);
281 addr = vm_map_min(map);
282
283 /*
284 * Locate sufficient space in the map. This will give us the final
285 * virtual address for the new memory, and thus will tell us the
286 * offset within the kernel map.
287 */
288 vm_map_lock(map);
289 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
290 vm_map_unlock(map);
291 if ((flags & M_NOWAIT) == 0) {
292 for (i = 0; i < 8; i++) {
293 EVENTHANDLER_INVOKE(vm_lowmem, 0);
294 uma_reclaim();
295 vm_map_lock(map);
296 if (vm_map_findspace(map, vm_map_min(map),
297 size, &addr) == 0) {
298 break;
299 }
300 vm_map_unlock(map);
301 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
302 }
303 if (i == 8) {
304 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
305 (long)size, (long)map->size);
306 }
307 } else {
308 return (0);
309 }
310 }
311
312 rv = kmem_back(map, addr, size, flags);
313 vm_map_unlock(map);
314 return (rv == KERN_SUCCESS ? addr : 0);
315 }
316
317 /*
318 * kmem_back:
319 *
320 * Allocate physical pages for the specified virtual address range.
321 */
322 int
323 kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
324 {
325 vm_offset_t offset, i;
326 vm_map_entry_t entry;
327 vm_page_t m;
328 int pflags;
329 boolean_t found;
330
331 KASSERT(vm_map_locked(map), ("kmem_back: map %p is not locked", map));
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 /*
338 * Assert: vm_map_insert() will never be able to extend the
339 * previous entry so vm_map_lookup_entry() will find a new
340 * entry exactly corresponding to this address range and it
341 * will have wired_count == 0.
342 */
343 found = vm_map_lookup_entry(map, addr, &entry);
344 KASSERT(found && entry->start == addr && entry->end == addr + size &&
345 entry->wired_count == 0 && (entry->eflags & MAP_ENTRY_IN_TRANSITION)
346 == 0, ("kmem_back: entry not found or misaligned"));
347
348 if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
349 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
350 else
351 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
352
353 if (flags & M_ZERO)
354 pflags |= VM_ALLOC_ZERO;
355 if (flags & M_NODUMP)
356 pflags |= VM_ALLOC_NODUMP;
357
358 VM_OBJECT_LOCK(kmem_object);
359 for (i = 0; i < size; i += PAGE_SIZE) {
360 retry:
361 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
362
363 /*
364 * Ran out of space, free everything up and return. Don't need
365 * to lock page queues here as we know that the pages we got
366 * aren't on any queues.
367 */
368 if (m == NULL) {
369 if ((flags & M_NOWAIT) == 0) {
370 VM_OBJECT_UNLOCK(kmem_object);
371 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
372 vm_map_unlock(map);
373 VM_WAIT;
374 vm_map_lock(map);
375 KASSERT(
376 (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_NEEDS_WAKEUP)) ==
377 MAP_ENTRY_IN_TRANSITION,
378 ("kmem_back: volatile entry"));
379 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
380 VM_OBJECT_LOCK(kmem_object);
381 goto retry;
382 }
383 /*
384 * Free the pages before removing the map entry.
385 * They are already marked busy. Calling
386 * vm_map_delete before the pages has been freed or
387 * unbusied will cause a deadlock.
388 */
389 while (i != 0) {
390 i -= PAGE_SIZE;
391 m = vm_page_lookup(kmem_object,
392 OFF_TO_IDX(offset + i));
393 vm_page_unwire(m, 0);
394 vm_page_free(m);
395 }
396 VM_OBJECT_UNLOCK(kmem_object);
397 vm_map_delete(map, addr, addr + size);
398 return (KERN_NO_SPACE);
399 }
400 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
401 pmap_zero_page(m);
402 m->valid = VM_PAGE_BITS_ALL;
403 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
404 ("kmem_malloc: page %p is managed", m));
405 }
406 VM_OBJECT_UNLOCK(kmem_object);
407
408 /*
409 * Mark map entry as non-pageable. Repeat the assert.
410 */
411 KASSERT(entry->start == addr && entry->end == addr + size &&
412 entry->wired_count == 0,
413 ("kmem_back: entry not found or misaligned after allocation"));
414 entry->wired_count = 1;
415
416 /*
417 * At this point, the kmem_object must be unlocked because
418 * vm_map_simplify_entry() calls vm_object_deallocate(), which
419 * locks the kmem_object.
420 */
421 vm_map_simplify_entry(map, entry);
422
423 /*
424 * Loop thru pages, entering them in the pmap.
425 */
426 VM_OBJECT_LOCK(kmem_object);
427 for (i = 0; i < size; i += PAGE_SIZE) {
428 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
429 /*
430 * Because this is kernel_pmap, this call will not block.
431 */
432 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
433 TRUE);
434 vm_page_wakeup(m);
435 }
436 VM_OBJECT_UNLOCK(kmem_object);
437
438 return (KERN_SUCCESS);
439 }
440
441 /*
442 * kmem_alloc_wait:
443 *
444 * Allocates pageable memory from a sub-map of the kernel. If the submap
445 * has no room, the caller sleeps waiting for more memory in the submap.
446 *
447 * This routine may block.
448 */
449 vm_offset_t
450 kmem_alloc_wait(map, size)
451 vm_map_t map;
452 vm_size_t size;
453 {
454 vm_offset_t addr;
455
456 size = round_page(size);
457 if (!swap_reserve(size))
458 return (0);
459
460 for (;;) {
461 /*
462 * To make this work for more than one map, use the map's lock
463 * to lock out sleepers/wakers.
464 */
465 vm_map_lock(map);
466 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
467 break;
468 /* no space now; see if we can ever get space */
469 if (vm_map_max(map) - vm_map_min(map) < size) {
470 vm_map_unlock(map);
471 swap_release(size);
472 return (0);
473 }
474 map->needs_wakeup = TRUE;
475 vm_map_unlock_and_wait(map, 0);
476 }
477 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
478 VM_PROT_ALL, MAP_ACC_CHARGED);
479 vm_map_unlock(map);
480 return (addr);
481 }
482
483 /*
484 * kmem_free_wakeup:
485 *
486 * Returns memory to a submap of the kernel, and wakes up any processes
487 * waiting for memory in that map.
488 */
489 void
490 kmem_free_wakeup(map, addr, size)
491 vm_map_t map;
492 vm_offset_t addr;
493 vm_size_t size;
494 {
495
496 vm_map_lock(map);
497 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
498 if (map->needs_wakeup) {
499 map->needs_wakeup = FALSE;
500 vm_map_wakeup(map);
501 }
502 vm_map_unlock(map);
503 }
504
505 static void
506 kmem_init_zero_region(void)
507 {
508 vm_offset_t addr, i;
509 vm_page_t m;
510 int error;
511
512 /*
513 * Map a single physical page of zeros to a larger virtual range.
514 * This requires less looping in places that want large amounts of
515 * zeros, while not using much more physical resources.
516 */
517 addr = kmem_alloc_nofault(kernel_map, ZERO_REGION_SIZE);
518 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
519 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
520 if ((m->flags & PG_ZERO) == 0)
521 pmap_zero_page(m);
522 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
523 pmap_qenter(addr + i, &m, 1);
524 error = vm_map_protect(kernel_map, addr, addr + ZERO_REGION_SIZE,
525 VM_PROT_READ, TRUE);
526 KASSERT(error == 0, ("error=%d", error));
527
528 zero_region = (const void *)addr;
529 }
530
531 /*
532 * kmem_init:
533 *
534 * Create the kernel map; insert a mapping covering kernel text,
535 * data, bss, and all space allocated thus far (`boostrap' data). The
536 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
537 * `start' as allocated, and the range between `start' and `end' as free.
538 */
539 void
540 kmem_init(start, end)
541 vm_offset_t start, end;
542 {
543 vm_map_t m;
544
545 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
546 m->system_map = 1;
547 vm_map_lock(m);
548 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
549 kernel_map = m;
550 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
551 #ifdef __amd64__
552 KERNBASE,
553 #else
554 VM_MIN_KERNEL_ADDRESS,
555 #endif
556 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
557 /* ... and ending with the completion of the above `insert' */
558 vm_map_unlock(m);
559
560 kmem_init_zero_region();
561 }
562
563 #ifdef DIAGNOSTIC
564 /*
565 * Allow userspace to directly trigger the VM drain routine for testing
566 * purposes.
567 */
568 static int
569 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
570 {
571 int error, i;
572
573 i = 0;
574 error = sysctl_handle_int(oidp, &i, 0, req);
575 if (error)
576 return (error);
577 if (i)
578 EVENTHANDLER_INVOKE(vm_lowmem, 0);
579 return (0);
580 }
581
582 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
583 debug_vm_lowmem, "I", "set to trigger vm_lowmem event");
584 #endif
Cache object: cf22c6f4a5ba5fc55d05c74e9b347a48
|