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. 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>
74 #include <sys/malloc.h>
75 #include <sys/rwlock.h>
76 #include <sys/sysctl.h>
80 #include <vm/vm_param.h>
81 #include <vm/vm_kern.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_radix.h>
88 #include <vm/vm_extern.h>
95 const void *zero_region;
96 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
98 /* NB: Used by kernel debuggers. */
99 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
101 u_int exec_map_entry_size;
102 u_int exec_map_entries;
104 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
105 SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
107 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
108 #if defined(__arm__) || defined(__sparc64__)
109 &vm_max_kernel_address, 0,
111 SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
113 "Max kernel address");
118 * Allocate a virtual address range with no underlying object and
119 * no initial mapping to physical memory. Any mapping from this
120 * range to physical memory must be explicitly created prior to
121 * its use, typically with pmap_qenter(). Any attempt to create
122 * a mapping on demand through vm_fault() will result in a panic.
130 size = round_page(size);
131 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
140 * Release a region of kernel virtual memory allocated
141 * with kva_alloc, and return the physical pages
142 * associated with that region.
144 * This routine may not block on kernel maps.
152 size = round_page(size);
153 vmem_free(kernel_arena, addr, size);
157 * Allocates a region from the kernel address map and physical pages
158 * within the specified address range to the kernel object. Creates a
159 * wired mapping from this region to these pages, and returns the
160 * region's starting virtual address. The allocated pages are not
161 * necessarily physically contiguous. If M_ZERO is specified through the
162 * given flags, then the pages are zeroed before they are mapped.
165 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
166 vm_paddr_t high, vm_memattr_t memattr)
168 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
169 vm_offset_t addr, i, offset;
173 size = round_page(size);
174 if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
176 offset = addr - VM_MIN_KERNEL_ADDRESS;
177 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
178 VM_OBJECT_WLOCK(object);
179 for (i = 0; i < size; i += PAGE_SIZE) {
182 m = vm_page_alloc_contig(object, atop(offset + i),
183 pflags, 1, low, high, PAGE_SIZE, 0, memattr);
185 VM_OBJECT_WUNLOCK(object);
186 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
187 if (!vm_page_reclaim_contig(pflags, 1,
188 low, high, PAGE_SIZE, 0) &&
189 (flags & M_WAITOK) != 0)
191 VM_OBJECT_WLOCK(object);
195 kmem_unback(object, addr, i);
196 vmem_free(vmem, addr, size);
199 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
201 m->valid = VM_PAGE_BITS_ALL;
202 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
203 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
205 VM_OBJECT_WUNLOCK(object);
210 * Allocates a region from the kernel address map and physically
211 * contiguous pages within the specified address range to the kernel
212 * object. Creates a wired mapping from this region to these pages, and
213 * returns the region's starting virtual address. If M_ZERO is specified
214 * through the given flags, then the pages are zeroed before they are
218 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
219 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
220 vm_memattr_t memattr)
222 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
223 vm_offset_t addr, offset, tmp;
228 size = round_page(size);
229 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
231 offset = addr - VM_MIN_KERNEL_ADDRESS;
232 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
234 VM_OBJECT_WLOCK(object);
237 m = vm_page_alloc_contig(object, atop(offset), pflags,
238 npages, low, high, alignment, boundary, memattr);
240 VM_OBJECT_WUNLOCK(object);
241 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
242 if (!vm_page_reclaim_contig(pflags, npages, low, high,
243 alignment, boundary) && (flags & M_WAITOK) != 0)
245 VM_OBJECT_WLOCK(object);
249 vmem_free(vmem, addr, size);
254 for (; m < end_m; m++) {
255 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
257 m->valid = VM_PAGE_BITS_ALL;
258 pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
259 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
262 VM_OBJECT_WUNLOCK(object);
269 * Allocates a map to manage a subrange
270 * of the kernel virtual address space.
272 * Arguments are as follows:
274 * parent Map to take range from
275 * min, max Returned endpoints of map
276 * size Size of range to find
277 * superpage_align Request that min is superpage aligned
280 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
281 vm_size_t size, boolean_t superpage_align)
286 size = round_page(size);
288 *min = vm_map_min(parent);
289 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
290 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
292 if (ret != KERN_SUCCESS)
293 panic("kmem_suballoc: bad status return of %d", ret);
295 result = vm_map_create(vm_map_pmap(parent), *min, *max);
297 panic("kmem_suballoc: cannot create submap");
298 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
299 panic("kmem_suballoc: unable to change range to submap");
306 * Allocate wired-down pages in the kernel's address space.
309 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
314 size = round_page(size);
315 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
318 rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
320 if (rv != KERN_SUCCESS) {
321 vmem_free(vmem, addr, size);
330 * Allocate physical pages for the specified virtual address range.
333 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
335 vm_offset_t offset, i;
339 KASSERT(object == kmem_object || object == kernel_object,
340 ("kmem_back: only supports kernel objects."));
342 offset = addr - VM_MIN_KERNEL_ADDRESS;
343 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
347 VM_OBJECT_WLOCK(object);
348 mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
349 for (; i < size; i += PAGE_SIZE, mpred = m) {
350 m = vm_page_alloc_after(object, atop(offset + i), pflags,
354 * Ran out of space, free everything up and return. Don't need
355 * to lock page queues here as we know that the pages we got
356 * aren't on any queues.
359 VM_OBJECT_WUNLOCK(object);
360 if ((flags & M_NOWAIT) == 0) {
364 kmem_unback(object, addr, i);
365 return (KERN_NO_SPACE);
367 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
369 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
370 ("kmem_malloc: page %p is managed", m));
371 m->valid = VM_PAGE_BITS_ALL;
372 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
373 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
375 VM_OBJECT_WUNLOCK(object);
377 return (KERN_SUCCESS);
383 * Unmap and free the physical pages underlying the specified virtual
386 * A physical page must exist within the specified object at each index
387 * that is being unmapped.
390 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
393 vm_offset_t end, offset;
395 KASSERT(object == kmem_object || object == kernel_object,
396 ("kmem_unback: only supports kernel objects."));
398 pmap_remove(kernel_pmap, addr, addr + size);
399 offset = addr - VM_MIN_KERNEL_ADDRESS;
401 VM_OBJECT_WLOCK(object);
402 for (m = vm_page_lookup(object, atop(offset)); offset < end;
403 offset += PAGE_SIZE, m = next) {
404 next = vm_page_next(m);
405 vm_page_unwire(m, PQ_NONE);
408 VM_OBJECT_WUNLOCK(object);
414 * Free memory allocated with kmem_malloc. The size must match the
415 * original allocation.
418 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
421 size = round_page(size);
422 kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
424 vmem_free(vmem, addr, size);
430 * Allocates pageable memory from a sub-map of the kernel. If the submap
431 * has no room, the caller sleeps waiting for more memory in the submap.
433 * This routine may block.
436 kmap_alloc_wait(map, size)
442 size = round_page(size);
443 if (!swap_reserve(size))
448 * To make this work for more than one map, use the map's lock
449 * to lock out sleepers/wakers.
452 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
454 /* no space now; see if we can ever get space */
455 if (vm_map_max(map) - vm_map_min(map) < size) {
460 map->needs_wakeup = TRUE;
461 vm_map_unlock_and_wait(map, 0);
463 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
464 VM_PROT_ALL, MAP_ACC_CHARGED);
472 * Returns memory to a submap of the kernel, and wakes up any processes
473 * waiting for memory in that map.
476 kmap_free_wakeup(map, addr, size)
483 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
484 if (map->needs_wakeup) {
485 map->needs_wakeup = FALSE;
492 kmem_init_zero_region(void)
498 * Map a single physical page of zeros to a larger virtual range.
499 * This requires less looping in places that want large amounts of
500 * zeros, while not using much more physical resources.
502 addr = kva_alloc(ZERO_REGION_SIZE);
503 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
504 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
505 if ((m->flags & PG_ZERO) == 0)
507 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
508 pmap_qenter(addr + i, &m, 1);
509 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
511 zero_region = (const void *)addr;
517 * Create the kernel map; insert a mapping covering kernel text,
518 * data, bss, and all space allocated thus far (`boostrap' data). The
519 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
520 * `start' as allocated, and the range between `start' and `end' as free.
523 kmem_init(start, end)
524 vm_offset_t start, end;
528 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
531 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
533 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
537 VM_MIN_KERNEL_ADDRESS,
539 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
540 /* ... and ending with the completion of the above `insert' */
546 * Allow userspace to directly trigger the VM drain routine for testing
550 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
555 error = sysctl_handle_int(oidp, &i, 0, req);
558 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
561 EVENTHANDLER_INVOKE(vm_lowmem, i);
565 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
566 debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");