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 * 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.
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
32 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
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.
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.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Kernel memory management.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h> /* for ticks and hz */
71 #include <sys/eventhandler.h>
73 #include <sys/mutex.h>
75 #include <sys/malloc.h>
76 #include <sys/sysctl.h>
79 #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_extern.h>
88 vm_map_t kernel_map=0;
92 vm_map_t buffer_map=0;
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.
104 kmem_alloc_nofault(map, size)
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) {
122 * kmem_alloc_nofault_space:
124 * Allocate a virtual address range with no underlying object and
125 * no initial mapping to physical memory within the specified
126 * address space. Any mapping from this range to physical memory
127 * must be explicitly created prior to its use, typically with
128 * pmap_qenter(). Any attempt to create a mapping on demand
129 * through vm_fault() will result in a panic.
132 kmem_alloc_nofault_space(map, size, find_space)
140 size = round_page(size);
141 addr = vm_map_min(map);
142 result = vm_map_find(map, NULL, 0, &addr, size, find_space,
143 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
144 if (result != KERN_SUCCESS) {
151 * Allocate wired-down memory in the kernel's address map
155 kmem_alloc(map, size)
163 size = round_page(size);
166 * Use the kernel object for wired-down kernel pages. Assume that no
167 * region of the kernel object is referenced more than once.
171 * Locate sufficient space in the map. This will give us the final
172 * virtual address for the new memory, and thus will tell us the
173 * offset within the kernel map.
176 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
180 offset = addr - VM_MIN_KERNEL_ADDRESS;
181 vm_object_reference(kernel_object);
182 vm_map_insert(map, kernel_object, offset, addr, addr + size,
183 VM_PROT_ALL, VM_PROT_ALL, 0);
187 * Guarantee that there are pages already in this object before
188 * calling vm_map_wire. This is to prevent the following
191 * 1) Threads have swapped out, so that there is a pager for the
192 * kernel_object. 2) The kmsg zone is empty, and so we are
193 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
194 * there is no page, but there is a pager, so we call
195 * pager_data_request. But the kmsg zone is empty, so we must
196 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
197 * we get the data back from the pager, it will be (very stale)
198 * non-zero data. kmem_alloc is defined to return zero-filled memory.
200 * We're intentionally not activating the pages we allocate to prevent a
201 * race with page-out. vm_map_wire will wire the pages.
203 VM_OBJECT_LOCK(kernel_object);
204 for (i = 0; i < size; i += PAGE_SIZE) {
207 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
208 VM_ALLOC_NOBUSY | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
209 mem->valid = VM_PAGE_BITS_ALL;
210 KASSERT((mem->flags & PG_UNMANAGED) != 0,
211 ("kmem_alloc: page %p is managed", mem));
213 VM_OBJECT_UNLOCK(kernel_object);
216 * And finally, mark the data as non-pageable.
218 (void) vm_map_wire(map, addr, addr + size,
219 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
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(map, addr, size)
240 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
246 * Allocates a map to manage a subrange
247 * of the kernel virtual address space.
249 * Arguments are as follows:
251 * parent Map to take range from
252 * min, max Returned endpoints of map
253 * size Size of range to find
254 * superpage_align Request that min is superpage aligned
257 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
258 vm_size_t size, boolean_t superpage_align)
263 size = round_page(size);
265 *min = vm_map_min(parent);
266 ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
267 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
269 if (ret != KERN_SUCCESS)
270 panic("kmem_suballoc: bad status return of %d", ret);
272 result = vm_map_create(vm_map_pmap(parent), *min, *max);
274 panic("kmem_suballoc: cannot create submap");
275 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
276 panic("kmem_suballoc: unable to change range to submap");
283 * Allocate wired-down memory in the kernel's address map for the higher
284 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
285 * kmem_alloc() because we may need to allocate memory at interrupt
286 * level where we cannot block (canwait == FALSE).
288 * This routine has its own private kernel submap (kmem_map) and object
289 * (kmem_object). This, combined with the fact that only malloc uses
290 * this routine, ensures that we will never block in map or object waits.
292 * We don't worry about expanding the map (adding entries) since entries
293 * for wired maps are statically allocated.
295 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
296 * which we never free.
299 kmem_malloc(map, size, flags)
307 size = round_page(size);
308 addr = vm_map_min(map);
311 * Locate sufficient space in the map. This will give us the final
312 * virtual address for the new memory, and thus will tell us the
313 * offset within the kernel map.
316 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
318 if ((flags & M_NOWAIT) == 0) {
319 for (i = 0; i < 8; i++) {
320 EVENTHANDLER_INVOKE(vm_lowmem, 0);
323 if (vm_map_findspace(map, vm_map_min(map),
328 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
331 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
332 (long)size, (long)map->size);
339 rv = kmem_back(map, addr, size, flags);
341 return (rv == KERN_SUCCESS ? addr : 0);
347 * Allocate physical pages for the specified virtual address range.
350 kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
352 vm_offset_t offset, i;
353 vm_map_entry_t entry;
358 * XXX the map must be locked for write on entry, but there's
359 * no easy way to assert that.
362 offset = addr - VM_MIN_KERNEL_ADDRESS;
363 vm_object_reference(kmem_object);
364 vm_map_insert(map, kmem_object, offset, addr, addr + size,
365 VM_PROT_ALL, VM_PROT_ALL, 0);
367 if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
368 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
370 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
373 pflags |= VM_ALLOC_ZERO;
375 VM_OBJECT_LOCK(kmem_object);
376 for (i = 0; i < size; i += PAGE_SIZE) {
378 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
381 * Ran out of space, free everything up and return. Don't need
382 * to lock page queues here as we know that the pages we got
383 * aren't on any queues.
386 if ((flags & M_NOWAIT) == 0) {
387 VM_OBJECT_UNLOCK(kmem_object);
391 VM_OBJECT_LOCK(kmem_object);
395 * Free the pages before removing the map entry.
396 * They are already marked busy. Calling
397 * vm_map_delete before the pages has been freed or
398 * unbusied will cause a deadlock.
402 m = vm_page_lookup(kmem_object,
403 OFF_TO_IDX(offset + i));
404 vm_page_lock_queues();
405 vm_page_unwire(m, 0);
407 vm_page_unlock_queues();
409 VM_OBJECT_UNLOCK(kmem_object);
410 vm_map_delete(map, addr, addr + size);
411 return (KERN_NO_SPACE);
413 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
415 m->valid = VM_PAGE_BITS_ALL;
416 KASSERT((m->flags & PG_UNMANAGED) != 0,
417 ("kmem_malloc: page %p is managed", m));
419 VM_OBJECT_UNLOCK(kmem_object);
422 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
423 * be able to extend the previous entry so there will be a new entry
424 * exactly corresponding to this address range and it will have
427 if (!vm_map_lookup_entry(map, addr, &entry) ||
428 entry->start != addr || entry->end != addr + size ||
429 entry->wired_count != 0)
430 panic("kmem_malloc: entry not found or misaligned");
431 entry->wired_count = 1;
434 * At this point, the kmem_object must be unlocked because
435 * vm_map_simplify_entry() calls vm_object_deallocate(), which
436 * locks the kmem_object.
438 vm_map_simplify_entry(map, entry);
441 * Loop thru pages, entering them in the pmap.
443 VM_OBJECT_LOCK(kmem_object);
444 for (i = 0; i < size; i += PAGE_SIZE) {
445 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
447 * Because this is kernel_pmap, this call will not block.
449 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
453 VM_OBJECT_UNLOCK(kmem_object);
455 return (KERN_SUCCESS);
461 * Allocates pageable memory from a sub-map of the kernel. If the submap
462 * has no room, the caller sleeps waiting for more memory in the submap.
464 * This routine may block.
467 kmem_alloc_wait(map, size)
473 size = round_page(size);
474 if (!swap_reserve(size))
479 * To make this work for more than one map, use the map's lock
480 * to lock out sleepers/wakers.
483 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
485 /* no space now; see if we can ever get space */
486 if (vm_map_max(map) - vm_map_min(map) < size) {
491 map->needs_wakeup = TRUE;
492 vm_map_unlock_and_wait(map, 0);
494 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
495 VM_PROT_ALL, MAP_ACC_CHARGED);
503 * Returns memory to a submap of the kernel, and wakes up any processes
504 * waiting for memory in that map.
507 kmem_free_wakeup(map, addr, size)
514 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
515 if (map->needs_wakeup) {
516 map->needs_wakeup = FALSE;
525 * Create the kernel map; insert a mapping covering kernel text,
526 * data, bss, and all space allocated thus far (`boostrap' data). The
527 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
528 * `start' as allocated, and the range between `start' and `end' as free.
531 kmem_init(start, end)
532 vm_offset_t start, end;
536 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
539 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
541 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
545 VM_MIN_KERNEL_ADDRESS,
547 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
548 /* ... and ending with the completion of the above `insert' */
554 * Allow userspace to directly trigger the VM drain routine for testing
558 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
563 error = sysctl_handle_int(oidp, &i, 0, req);
567 EVENTHANDLER_INVOKE(vm_lowmem, 0);
571 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
572 debug_vm_lowmem, "I", "set to trigger vm_lowmem event");