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/11.1/sys/vm/vm_kern.c 316073 2017-03-28 06:07:59Z kib $");
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/proc.h>
74 #include <sys/malloc.h>
75 #include <sys/rwlock.h>
76 #include <sys/sysctl.h>
77 #include <sys/vmem.h>
78
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/vm_kern.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pageout.h>
87 #include <vm/vm_extern.h>
88 #include <vm/uma.h>
89
90 vm_map_t kernel_map;
91 vm_map_t exec_map;
92 vm_map_t pipe_map;
93
94 const void *zero_region;
95 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
96
97 /* NB: Used by kernel debuggers. */
98 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
99
100 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
101 SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
102
103 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
104 #if defined(__arm__) || defined(__sparc64__)
105 &vm_max_kernel_address, 0,
106 #else
107 SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
108 #endif
109 "Max kernel address");
110
111 /*
112 * kva_alloc:
113 *
114 * Allocate a virtual address range with no underlying object and
115 * no initial mapping to physical memory. Any mapping from this
116 * range to physical memory must be explicitly created prior to
117 * its use, typically with pmap_qenter(). Any attempt to create
118 * a mapping on demand through vm_fault() will result in a panic.
119 */
120 vm_offset_t
121 kva_alloc(size)
122 vm_size_t size;
123 {
124 vm_offset_t addr;
125
126 size = round_page(size);
127 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
128 return (0);
129
130 return (addr);
131 }
132
133 /*
134 * kva_free:
135 *
136 * Release a region of kernel virtual memory allocated
137 * with kva_alloc, and return the physical pages
138 * associated with that region.
139 *
140 * This routine may not block on kernel maps.
141 */
142 void
143 kva_free(addr, size)
144 vm_offset_t addr;
145 vm_size_t size;
146 {
147
148 size = round_page(size);
149 vmem_free(kernel_arena, addr, size);
150 }
151
152 /*
153 * Allocates a region from the kernel address map and physical pages
154 * within the specified address range to the kernel object. Creates a
155 * wired mapping from this region to these pages, and returns the
156 * region's starting virtual address. The allocated pages are not
157 * necessarily physically contiguous. If M_ZERO is specified through the
158 * given flags, then the pages are zeroed before they are mapped.
159 */
160 vm_offset_t
161 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
162 vm_paddr_t high, vm_memattr_t memattr)
163 {
164 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
165 vm_offset_t addr, i, offset;
166 vm_page_t m;
167 int pflags, tries;
168
169 size = round_page(size);
170 if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
171 return (0);
172 offset = addr - VM_MIN_KERNEL_ADDRESS;
173 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
174 VM_OBJECT_WLOCK(object);
175 for (i = 0; i < size; i += PAGE_SIZE) {
176 tries = 0;
177 retry:
178 m = vm_page_alloc_contig(object, atop(offset + i),
179 pflags, 1, low, high, PAGE_SIZE, 0, memattr);
180 if (m == NULL) {
181 VM_OBJECT_WUNLOCK(object);
182 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
183 if (!vm_page_reclaim_contig(pflags, 1,
184 low, high, PAGE_SIZE, 0) &&
185 (flags & M_WAITOK) != 0)
186 VM_WAIT;
187 VM_OBJECT_WLOCK(object);
188 tries++;
189 goto retry;
190 }
191 kmem_unback(object, addr, i);
192 vmem_free(vmem, addr, size);
193 return (0);
194 }
195 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
196 pmap_zero_page(m);
197 m->valid = VM_PAGE_BITS_ALL;
198 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
199 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
200 }
201 VM_OBJECT_WUNLOCK(object);
202 return (addr);
203 }
204
205 /*
206 * Allocates a region from the kernel address map and physically
207 * contiguous pages within the specified address range to the kernel
208 * object. Creates a wired mapping from this region to these pages, and
209 * returns the region's starting virtual address. If M_ZERO is specified
210 * through the given flags, then the pages are zeroed before they are
211 * mapped.
212 */
213 vm_offset_t
214 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
215 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
216 vm_memattr_t memattr)
217 {
218 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
219 vm_offset_t addr, offset, tmp;
220 vm_page_t end_m, m;
221 u_long npages;
222 int pflags, tries;
223
224 size = round_page(size);
225 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
226 return (0);
227 offset = addr - VM_MIN_KERNEL_ADDRESS;
228 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
229 npages = atop(size);
230 VM_OBJECT_WLOCK(object);
231 tries = 0;
232 retry:
233 m = vm_page_alloc_contig(object, atop(offset), pflags,
234 npages, low, high, alignment, boundary, memattr);
235 if (m == NULL) {
236 VM_OBJECT_WUNLOCK(object);
237 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
238 if (!vm_page_reclaim_contig(pflags, npages, low, high,
239 alignment, boundary) && (flags & M_WAITOK) != 0)
240 VM_WAIT;
241 VM_OBJECT_WLOCK(object);
242 tries++;
243 goto retry;
244 }
245 vmem_free(vmem, addr, size);
246 return (0);
247 }
248 end_m = m + npages;
249 tmp = addr;
250 for (; m < end_m; m++) {
251 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
252 pmap_zero_page(m);
253 m->valid = VM_PAGE_BITS_ALL;
254 pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
255 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
256 tmp += PAGE_SIZE;
257 }
258 VM_OBJECT_WUNLOCK(object);
259 return (addr);
260 }
261
262 /*
263 * kmem_suballoc:
264 *
265 * Allocates a map to manage a subrange
266 * of the kernel virtual address space.
267 *
268 * Arguments are as follows:
269 *
270 * parent Map to take range from
271 * min, max Returned endpoints of map
272 * size Size of range to find
273 * superpage_align Request that min is superpage aligned
274 */
275 vm_map_t
276 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
277 vm_size_t size, boolean_t superpage_align)
278 {
279 int ret;
280 vm_map_t result;
281
282 size = round_page(size);
283
284 *min = vm_map_min(parent);
285 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
286 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
287 MAP_ACC_NO_CHARGE);
288 if (ret != KERN_SUCCESS)
289 panic("kmem_suballoc: bad status return of %d", ret);
290 *max = *min + size;
291 result = vm_map_create(vm_map_pmap(parent), *min, *max);
292 if (result == NULL)
293 panic("kmem_suballoc: cannot create submap");
294 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
295 panic("kmem_suballoc: unable to change range to submap");
296 return (result);
297 }
298
299 /*
300 * kmem_malloc:
301 *
302 * Allocate wired-down pages in the kernel's address space.
303 */
304 vm_offset_t
305 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
306 {
307 vm_offset_t addr;
308 int rv;
309
310 size = round_page(size);
311 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
312 return (0);
313
314 rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
315 addr, size, flags);
316 if (rv != KERN_SUCCESS) {
317 vmem_free(vmem, addr, size);
318 return (0);
319 }
320 return (addr);
321 }
322
323 /*
324 * kmem_back:
325 *
326 * Allocate physical pages for the specified virtual address range.
327 */
328 int
329 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
330 {
331 vm_offset_t offset, i;
332 vm_page_t m;
333 int pflags;
334
335 KASSERT(object == kmem_object || object == kernel_object,
336 ("kmem_back: only supports kernel objects."));
337
338 offset = addr - VM_MIN_KERNEL_ADDRESS;
339 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
340
341 VM_OBJECT_WLOCK(object);
342 for (i = 0; i < size; i += PAGE_SIZE) {
343 retry:
344 m = vm_page_alloc(object, atop(offset + i), pflags);
345
346 /*
347 * Ran out of space, free everything up and return. Don't need
348 * to lock page queues here as we know that the pages we got
349 * aren't on any queues.
350 */
351 if (m == NULL) {
352 VM_OBJECT_WUNLOCK(object);
353 if ((flags & M_NOWAIT) == 0) {
354 VM_WAIT;
355 VM_OBJECT_WLOCK(object);
356 goto retry;
357 }
358 kmem_unback(object, addr, i);
359 return (KERN_NO_SPACE);
360 }
361 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
362 pmap_zero_page(m);
363 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
364 ("kmem_malloc: page %p is managed", m));
365 m->valid = VM_PAGE_BITS_ALL;
366 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
367 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
368 }
369 VM_OBJECT_WUNLOCK(object);
370
371 return (KERN_SUCCESS);
372 }
373
374 /*
375 * kmem_unback:
376 *
377 * Unmap and free the physical pages underlying the specified virtual
378 * address range.
379 *
380 * A physical page must exist within the specified object at each index
381 * that is being unmapped.
382 */
383 void
384 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
385 {
386 vm_page_t m;
387 vm_offset_t i, offset;
388
389 KASSERT(object == kmem_object || object == kernel_object,
390 ("kmem_unback: only supports kernel objects."));
391
392 pmap_remove(kernel_pmap, addr, addr + size);
393 offset = addr - VM_MIN_KERNEL_ADDRESS;
394 VM_OBJECT_WLOCK(object);
395 for (i = 0; i < size; i += PAGE_SIZE) {
396 m = vm_page_lookup(object, atop(offset + i));
397 vm_page_unwire(m, PQ_NONE);
398 vm_page_free(m);
399 }
400 VM_OBJECT_WUNLOCK(object);
401 }
402
403 /*
404 * kmem_free:
405 *
406 * Free memory allocated with kmem_malloc. The size must match the
407 * original allocation.
408 */
409 void
410 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
411 {
412
413 size = round_page(size);
414 kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
415 addr, size);
416 vmem_free(vmem, addr, size);
417 }
418
419 /*
420 * kmap_alloc_wait:
421 *
422 * Allocates pageable memory from a sub-map of the kernel. If the submap
423 * has no room, the caller sleeps waiting for more memory in the submap.
424 *
425 * This routine may block.
426 */
427 vm_offset_t
428 kmap_alloc_wait(map, size)
429 vm_map_t map;
430 vm_size_t size;
431 {
432 vm_offset_t addr;
433
434 size = round_page(size);
435 if (!swap_reserve(size))
436 return (0);
437
438 for (;;) {
439 /*
440 * To make this work for more than one map, use the map's lock
441 * to lock out sleepers/wakers.
442 */
443 vm_map_lock(map);
444 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
445 break;
446 /* no space now; see if we can ever get space */
447 if (vm_map_max(map) - vm_map_min(map) < size) {
448 vm_map_unlock(map);
449 swap_release(size);
450 return (0);
451 }
452 map->needs_wakeup = TRUE;
453 vm_map_unlock_and_wait(map, 0);
454 }
455 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
456 VM_PROT_ALL, MAP_ACC_CHARGED);
457 vm_map_unlock(map);
458 return (addr);
459 }
460
461 /*
462 * kmap_free_wakeup:
463 *
464 * Returns memory to a submap of the kernel, and wakes up any processes
465 * waiting for memory in that map.
466 */
467 void
468 kmap_free_wakeup(map, addr, size)
469 vm_map_t map;
470 vm_offset_t addr;
471 vm_size_t size;
472 {
473
474 vm_map_lock(map);
475 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
476 if (map->needs_wakeup) {
477 map->needs_wakeup = FALSE;
478 vm_map_wakeup(map);
479 }
480 vm_map_unlock(map);
481 }
482
483 void
484 kmem_init_zero_region(void)
485 {
486 vm_offset_t addr, i;
487 vm_page_t m;
488
489 /*
490 * Map a single physical page of zeros to a larger virtual range.
491 * This requires less looping in places that want large amounts of
492 * zeros, while not using much more physical resources.
493 */
494 addr = kva_alloc(ZERO_REGION_SIZE);
495 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
496 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
497 if ((m->flags & PG_ZERO) == 0)
498 pmap_zero_page(m);
499 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
500 pmap_qenter(addr + i, &m, 1);
501 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
502
503 zero_region = (const void *)addr;
504 }
505
506 /*
507 * kmem_init:
508 *
509 * Create the kernel map; insert a mapping covering kernel text,
510 * data, bss, and all space allocated thus far (`boostrap' data). The
511 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
512 * `start' as allocated, and the range between `start' and `end' as free.
513 */
514 void
515 kmem_init(start, end)
516 vm_offset_t start, end;
517 {
518 vm_map_t m;
519
520 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
521 m->system_map = 1;
522 vm_map_lock(m);
523 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
524 kernel_map = m;
525 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
526 #ifdef __amd64__
527 KERNBASE,
528 #else
529 VM_MIN_KERNEL_ADDRESS,
530 #endif
531 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
532 /* ... and ending with the completion of the above `insert' */
533 vm_map_unlock(m);
534 }
535
536 #ifdef DIAGNOSTIC
537 /*
538 * Allow userspace to directly trigger the VM drain routine for testing
539 * purposes.
540 */
541 static int
542 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
543 {
544 int error, i;
545
546 i = 0;
547 error = sysctl_handle_int(oidp, &i, 0, req);
548 if (error)
549 return (error);
550 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
551 return (EINVAL);
552 if (i != 0)
553 EVENTHANDLER_INVOKE(vm_lowmem, i);
554 return (0);
555 }
556
557 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
558 debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");
559 #endif
Cache object: ca1b7491ef2513b8c939a51c03f860c9
|