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>
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_phys.h>
88 #include <vm/vm_radix.h>
89 #include <vm/vm_extern.h>
96 const void *zero_region;
97 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
99 /* NB: Used by kernel debuggers. */
100 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
102 u_int exec_map_entry_size;
103 u_int exec_map_entries;
105 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
106 SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
108 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
109 #if defined(__arm__) || defined(__sparc64__)
110 &vm_max_kernel_address, 0,
112 SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
114 "Max kernel address");
119 * Allocate a virtual address range with no underlying object and
120 * no initial mapping to physical memory. Any mapping from this
121 * range to physical memory must be explicitly created prior to
122 * its use, typically with pmap_qenter(). Any attempt to create
123 * a mapping on demand through vm_fault() will result in a panic.
126 kva_alloc(vm_size_t size)
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.
147 kva_free(vm_offset_t addr, vm_size_t size)
150 size = round_page(size);
151 vmem_free(kernel_arena, addr, size);
155 * Allocates a region from the kernel address map and physical pages
156 * within the specified address range to the kernel object. Creates a
157 * wired mapping from this region to these pages, and returns the
158 * region's starting virtual address. The allocated pages are not
159 * necessarily physically contiguous. If M_ZERO is specified through the
160 * given flags, then the pages are zeroed before they are mapped.
163 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
164 vm_paddr_t high, vm_memattr_t memattr)
166 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
167 vm_offset_t addr, i, offset;
171 size = round_page(size);
172 if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
174 offset = addr - VM_MIN_KERNEL_ADDRESS;
175 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
176 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
177 pflags |= VM_ALLOC_NOWAIT;
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;
233 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
234 pflags |= VM_ALLOC_NOWAIT;
236 VM_OBJECT_WLOCK(object);
239 m = vm_page_alloc_contig(object, atop(offset), pflags,
240 npages, low, high, alignment, boundary, memattr);
242 VM_OBJECT_WUNLOCK(object);
243 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
244 if (!vm_page_reclaim_contig(pflags, npages, low, high,
245 alignment, boundary) && (flags & M_WAITOK) != 0)
247 VM_OBJECT_WLOCK(object);
251 vmem_free(vmem, addr, size);
256 for (; m < end_m; m++) {
257 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
259 m->valid = VM_PAGE_BITS_ALL;
260 pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
261 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
264 VM_OBJECT_WUNLOCK(object);
271 * Allocates a map to manage a subrange
272 * of the kernel virtual address space.
274 * Arguments are as follows:
276 * parent Map to take range from
277 * min, max Returned endpoints of map
278 * size Size of range to find
279 * superpage_align Request that min is superpage aligned
282 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
283 vm_size_t size, boolean_t superpage_align)
288 size = round_page(size);
290 *min = vm_map_min(parent);
291 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
292 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
294 if (ret != KERN_SUCCESS)
295 panic("kmem_suballoc: bad status return of %d", ret);
297 result = vm_map_create(vm_map_pmap(parent), *min, *max);
299 panic("kmem_suballoc: cannot create submap");
300 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
301 panic("kmem_suballoc: unable to change range to submap");
308 * Allocate wired-down pages in the kernel's address space.
311 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
316 size = round_page(size);
317 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
320 rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
322 if (rv != KERN_SUCCESS) {
323 vmem_free(vmem, addr, size);
332 * Allocate physical pages for the specified virtual address range.
335 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
337 vm_offset_t offset, i;
341 KASSERT(object == kmem_object || object == kernel_object,
342 ("kmem_back: only supports kernel objects."));
344 offset = addr - VM_MIN_KERNEL_ADDRESS;
345 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
346 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
347 if (flags & M_WAITOK)
348 pflags |= VM_ALLOC_WAITFAIL;
351 VM_OBJECT_WLOCK(object);
353 mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
354 for (; i < size; i += PAGE_SIZE, mpred = m) {
355 m = vm_page_alloc_after(object, atop(offset + i), pflags,
359 * Ran out of space, free everything up and return. Don't need
360 * to lock page queues here as we know that the pages we got
361 * aren't on any queues.
364 if ((flags & M_NOWAIT) == 0)
366 VM_OBJECT_WUNLOCK(object);
367 kmem_unback(object, addr, i);
368 return (KERN_NO_SPACE);
370 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
372 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
373 ("kmem_malloc: page %p is managed", m));
374 m->valid = VM_PAGE_BITS_ALL;
375 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
376 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
378 VM_OBJECT_WUNLOCK(object);
380 return (KERN_SUCCESS);
386 * Unmap and free the physical pages underlying the specified virtual
389 * A physical page must exist within the specified object at each index
390 * that is being unmapped.
393 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
396 vm_offset_t end, offset;
398 KASSERT(object == kmem_object || object == kernel_object,
399 ("kmem_unback: only supports kernel objects."));
401 pmap_remove(kernel_pmap, addr, addr + size);
402 offset = addr - VM_MIN_KERNEL_ADDRESS;
404 VM_OBJECT_WLOCK(object);
405 for (m = vm_page_lookup(object, atop(offset)); offset < end;
406 offset += PAGE_SIZE, m = next) {
407 next = vm_page_next(m);
408 vm_page_unwire(m, PQ_NONE);
411 VM_OBJECT_WUNLOCK(object);
417 * Free memory allocated with kmem_malloc. The size must match the
418 * original allocation.
421 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
424 size = round_page(size);
425 kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
427 vmem_free(vmem, addr, size);
433 * Allocates pageable memory from a sub-map of the kernel. If the submap
434 * has no room, the caller sleeps waiting for more memory in the submap.
436 * This routine may block.
439 kmap_alloc_wait(vm_map_t map, vm_size_t size)
443 size = round_page(size);
444 if (!swap_reserve(size))
449 * To make this work for more than one map, use the map's lock
450 * to lock out sleepers/wakers.
453 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
455 /* no space now; see if we can ever get space */
456 if (vm_map_max(map) - vm_map_min(map) < size) {
461 map->needs_wakeup = TRUE;
462 vm_map_unlock_and_wait(map, 0);
464 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_RW, VM_PROT_RW,
473 * Returns memory to a submap of the kernel, and wakes up any processes
474 * waiting for memory in that map.
477 kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
481 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
482 if (map->needs_wakeup) {
483 map->needs_wakeup = FALSE;
490 kmem_init_zero_region(void)
496 * Map a single physical page of zeros to a larger virtual range.
497 * This requires less looping in places that want large amounts of
498 * zeros, while not using much more physical resources.
500 addr = kva_alloc(ZERO_REGION_SIZE);
501 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
502 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
503 if ((m->flags & PG_ZERO) == 0)
505 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
506 pmap_qenter(addr + i, &m, 1);
507 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
509 zero_region = (const void *)addr;
515 * Create the kernel map; insert a mapping covering kernel text,
516 * data, bss, and all space allocated thus far (`boostrap' data). The
517 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
518 * `start' as allocated, and the range between `start' and `end' as free.
521 kmem_init(vm_offset_t start, vm_offset_t end)
525 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
528 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
530 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
534 VM_MIN_KERNEL_ADDRESS,
536 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
537 /* ... and ending with the completion of the above `insert' */
542 * kmem_bootstrap_free:
544 * Free pages backing preloaded data (e.g., kernel modules) to the
545 * system. Currently only supported on platforms that create a
546 * vm_phys segment for preloaded data.
549 kmem_bootstrap_free(vm_offset_t start, vm_size_t size)
551 #if defined(__i386__) || defined(__amd64__)
552 struct vm_domain *vmd;
557 end = trunc_page(start + size);
558 start = round_page(start);
560 for (va = start; va < end; va += PAGE_SIZE) {
561 pa = pmap_kextract(va);
562 m = PHYS_TO_VM_PAGE(pa);
564 vmd = vm_phys_domain(m);
565 mtx_lock(&vm_page_queue_free_mtx);
566 vm_phys_free_pages(m, 0);
567 vmd->vmd_page_count++;
568 vm_phys_freecnt_adj(m, 1);
569 mtx_unlock(&vm_page_queue_free_mtx);
571 vm_cnt.v_page_count++;
573 pmap_remove(kernel_pmap, start, end);
574 (void)vmem_add(kernel_arena, start, end - start, M_WAITOK);
580 * Allow userspace to directly trigger the VM drain routine for testing
584 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
589 error = sysctl_handle_int(oidp, &i, 0, req);
592 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
595 EVENTHANDLER_INVOKE(vm_lowmem, i);
599 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
600 debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");