2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
36 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
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.
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.
54 * Carnegie Mellon requests users of this software to return to
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
64 * $FreeBSD: src/sys/vm/vm_kern.c,v 1.61.2.2 2002/03/12 18:25:26 tegge Exp $
65 * $DragonFly: src/sys/vm/vm_kern.c,v 1.17 2004/05/20 22:42:25 dillon Exp $
69 * Kernel memory management.
72 #include <sys/param.h>
73 #include <sys/systm.h>
75 #include <sys/malloc.h>
78 #include <vm/vm_param.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_kern.h>
86 #include <vm/vm_extern.h>
88 vm_map_t kernel_map=0;
91 vm_map_t buffer_map=0;
96 * kmem_alloc_pageable:
98 * Allocate pageable memory to the kernel's address map.
99 * "map" must be kernel_map or a submap of kernel_map.
102 kmem_alloc_pageable(vm_map_t map, vm_size_t size)
107 size = round_page(size);
108 addr = vm_map_min(map);
109 result = vm_map_find(map, NULL, (vm_offset_t) 0,
110 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
111 if (result != KERN_SUCCESS) {
118 * kmem_alloc_nofault:
120 * Same as kmem_alloc_pageable, except that it create a nofault entry.
123 kmem_alloc_nofault(vm_map_t map, vm_size_t size)
128 size = round_page(size);
129 addr = vm_map_min(map);
130 result = vm_map_find(map, NULL, (vm_offset_t) 0,
131 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
132 if (result != KERN_SUCCESS) {
139 * Allocate wired-down memory in the kernel's address map
143 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags)
150 size = round_page(size);
152 if (kmflags & KM_KRESERVE)
153 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
155 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
158 * Use the kernel object for wired-down kernel pages. Assume that no
159 * region of the kernel object is referenced more than once.
161 * Locate sufficient space in the map. This will give us the final
162 * virtual address for the new memory, and thus will tell us the
163 * offset within the kernel map.
166 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
168 if (kmflags & KM_KRESERVE)
169 vm_map_entry_krelease(count);
171 vm_map_entry_release(count);
174 offset = addr - VM_MIN_KERNEL_ADDRESS;
175 vm_object_reference(kernel_object);
176 vm_map_insert(map, &count,
177 kernel_object, offset, addr, addr + size,
178 VM_PROT_ALL, VM_PROT_ALL, 0);
180 if (kmflags & KM_KRESERVE)
181 vm_map_entry_krelease(count);
183 vm_map_entry_release(count);
186 * Guarantee that there are pages already in this object before
187 * calling vm_map_wire. This is to prevent the following
190 * 1) Threads have swapped out, so that there is a pager for the
191 * kernel_object. 2) The kmsg zone is empty, and so we are
192 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
193 * there is no page, but there is a pager, so we call
194 * pager_data_request. But the kmsg zone is empty, so we must
195 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
196 * we get the data back from the pager, it will be (very stale)
197 * non-zero data. kmem_alloc is defined to return zero-filled memory.
199 * We're intentionally not activating the pages we allocate to prevent a
200 * race with page-out. vm_map_wire will wire the pages.
203 for (i = 0; i < size; i += PAGE_SIZE) {
206 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
207 VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
208 if ((mem->flags & PG_ZERO) == 0)
209 vm_page_zero_fill(mem);
210 mem->valid = VM_PAGE_BITS_ALL;
211 vm_page_flag_clear(mem, PG_ZERO);
216 * And finally, mark the data as non-pageable.
219 (void) vm_map_wire(map, (vm_offset_t) addr, addr + size, kmflags);
227 * Release a region of kernel virtual memory allocated
228 * with kmem_alloc, and return the physical pages
229 * associated with that region.
231 * This routine may not block on kernel maps.
234 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size)
236 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
242 * Allocates a map to manage a subrange
243 * of the kernel virtual address space.
245 * Arguments are as follows:
247 * parent Map to take range from
248 * size Size of range to find
249 * min, max Returned endpoints of map
250 * pageable Can the region be paged
253 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
259 size = round_page(size);
261 *min = (vm_offset_t) vm_map_min(parent);
262 ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
263 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
264 if (ret != KERN_SUCCESS) {
265 printf("kmem_suballoc: bad status return of %d.\n", ret);
266 panic("kmem_suballoc");
269 pmap_reference(vm_map_pmap(parent));
270 result = vm_map_create(vm_map_pmap(parent), *min, *max);
272 panic("kmem_suballoc: cannot create submap");
273 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
274 panic("kmem_suballoc: unable to change range to submap");
281 * Allocate wired-down memory in the kernel's address map for the higher
282 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
283 * kmem_alloc() because we may need to allocate memory at interrupt
284 * level where we cannot block (canwait == FALSE).
286 * We don't worry about expanding the map (adding entries) since entries
287 * for wired maps are statically allocated.
289 * NOTE: Please see kmem_slab_alloc() for a better explanation of the
293 kmem_malloc(vm_map_t map, vm_size_t size, int flags)
295 vm_offset_t offset, i;
296 vm_map_entry_t entry;
303 if (map != kernel_map && map != mb_map)
304 panic("kmem_malloc: map != {kmem,mb}_map");
306 size = round_page(size);
307 addr = vm_map_min(map);
310 * Locate sufficient space in the map. This will give us the final
311 * virtual address for the new memory, and thus will tell us the
312 * offset within the kernel map.
314 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
316 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
318 vm_map_entry_release(count);
321 printf("Out of mbuf clusters - adjust NMBCLUSTERS or increase maxusers!\n");
324 if ((flags & (M_RNOWAIT|M_NULLOK)) == 0 ||
325 (flags & (M_FAILSAFE|M_NULLOK)) == M_FAILSAFE
327 panic("kmem_malloc(%ld): kernel_map too small: "
328 "%ld total allocated",
329 (long)size, (long)map->size);
333 offset = addr - VM_MIN_KERNEL_ADDRESS;
334 vm_object_reference(kmem_object);
335 vm_map_insert(map, &count,
336 kmem_object, offset, addr, addr + size,
337 VM_PROT_ALL, VM_PROT_ALL, 0);
342 for (i = 0; i < size; i += PAGE_SIZE) {
345 vmflags = VM_ALLOC_SYSTEM; /* XXX M_USE_RESERVE? */
346 if ((flags & (M_WAITOK|M_RNOWAIT)) == 0)
347 printf("kmem_malloc: bad flags %08x (%p)\n", flags, ((int **)&map)[-1]);
348 if (flags & M_USE_INTERRUPT_RESERVE)
349 vmflags |= VM_ALLOC_INTERRUPT;
350 if (flags & (M_FAILSAFE|M_WAITOK)) {
351 if (td->td_preempted) {
354 vmflags |= VM_ALLOC_NORMAL;
359 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), vmflags);
362 * Ran out of space, free everything up and return. Don't need
363 * to lock page queues here as we know that the pages we got
364 * aren't on any queues.
366 * If M_WAITOK or M_FAILSAFE is set we can yield or block.
369 if (flags & (M_FAILSAFE|M_WAITOK)) {
370 if (wanted_reserve) {
371 if (flags & M_FAILSAFE)
372 printf("kmem_malloc: no memory, try failsafe\n");
377 if (flags & M_FAILSAFE)
378 printf("kmem_malloc: no memory, block even though we shouldn't\n");
383 i -= PAGE_SIZE; /* retry */
387 * Free the pages before removing the map entry.
388 * They are already marked busy. Calling
389 * vm_map_delete before the pages has been freed or
390 * unbusied will cause a deadlock.
394 m = vm_page_lookup(kmem_object,
395 OFF_TO_IDX(offset + i));
398 vm_map_delete(map, addr, addr + size, &count);
400 vm_map_entry_release(count);
403 vm_page_flag_clear(m, PG_ZERO);
404 m->valid = VM_PAGE_BITS_ALL;
408 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
409 * be able to extend the previous entry so there will be a new entry
410 * exactly corresponding to this address range and it will have
413 if (!vm_map_lookup_entry(map, addr, &entry) ||
414 entry->start != addr || entry->end != addr + size ||
415 entry->wired_count != 0)
416 panic("kmem_malloc: entry not found or misaligned");
417 entry->wired_count = 1;
419 vm_map_simplify_entry(map, entry, &count);
422 * Loop thru pages, entering them in the pmap. (We cannot add them to
423 * the wired count without wrapping the vm_page_queue_lock in
426 for (i = 0; i < size; i += PAGE_SIZE) {
427 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
431 * Because this is kernel_pmap, this call will not block.
433 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
434 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED);
437 vm_map_entry_release(count);
445 * Allocates pageable memory from a sub-map of the kernel. If the submap
446 * has no room, the caller sleeps waiting for more memory in the submap.
448 * This routine may block.
452 kmem_alloc_wait(vm_map_t map, vm_size_t size)
457 size = round_page(size);
459 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
463 * To make this work for more than one map, use the map's lock
464 * to lock out sleepers/wakers.
467 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) == 0)
469 /* no space now; see if we can ever get space */
470 if (vm_map_max(map) - vm_map_min(map) < size) {
471 vm_map_entry_release(count);
476 tsleep(map, 0, "kmaw", 0);
478 vm_map_insert(map, &count,
479 NULL, (vm_offset_t) 0,
480 addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
482 vm_map_entry_release(count);
489 * Returns memory to a submap of the kernel, and wakes up any processes
490 * waiting for memory in that map.
493 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
497 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
499 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count);
502 vm_map_entry_release(count);
508 * Create the kernel map; insert a mapping covering kernel text,
509 * data, bss, and all space allocated thus far (`boostrap' data). The
510 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
511 * `start' as allocated, and the range between `start' and `end' as free.
513 * Depend on the zalloc bootstrap cache to get our vm_map_entry_t.
516 kmem_init(vm_offset_t start, vm_offset_t end)
521 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
523 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
525 kernel_map->system_map = 1;
526 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
527 (void) vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
528 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
529 /* ... and ending with the completion of the above `insert' */
531 vm_map_entry_release(count);