2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Kernel memory management.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/kernel.h> /* for ticks and hz */
75 #include <sys/domainset.h>
76 #include <sys/eventhandler.h>
79 #include <sys/malloc.h>
80 #include <sys/rwlock.h>
81 #include <sys/sysctl.h>
83 #include <sys/vmmeter.h>
86 #include <vm/vm_param.h>
87 #include <vm/vm_domainset.h>
88 #include <vm/vm_kern.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_pageout.h>
94 #include <vm/vm_phys.h>
95 #include <vm/vm_pagequeue.h>
96 #include <vm/vm_radix.h>
97 #include <vm/vm_extern.h>
104 const void *zero_region;
105 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
107 /* NB: Used by kernel debuggers. */
108 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
110 u_int exec_map_entry_size;
111 u_int exec_map_entries;
113 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
114 SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
116 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
117 #if defined(__arm__) || defined(__sparc64__)
118 &vm_max_kernel_address, 0,
120 SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
122 "Max kernel address");
127 * Allocate a virtual address range with no underlying object and
128 * no initial mapping to physical memory. Any mapping from this
129 * range to physical memory must be explicitly created prior to
130 * its use, typically with pmap_qenter(). Any attempt to create
131 * a mapping on demand through vm_fault() will result in a panic.
134 kva_alloc(vm_size_t size)
138 size = round_page(size);
139 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
148 * Release a region of kernel virtual memory allocated
149 * with kva_alloc, and return the physical pages
150 * associated with that region.
152 * This routine may not block on kernel maps.
155 kva_free(vm_offset_t addr, vm_size_t size)
158 size = round_page(size);
159 vmem_free(kernel_arena, addr, size);
163 * Allocates a region from the kernel address map and physical pages
164 * within the specified address range to the kernel object. Creates a
165 * wired mapping from this region to these pages, and returns the
166 * region's starting virtual address. The allocated pages are not
167 * necessarily physically contiguous. If M_ZERO is specified through the
168 * given flags, then the pages are zeroed before they are mapped.
171 kmem_alloc_attr_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
172 vm_paddr_t high, vm_memattr_t memattr)
175 vm_object_t object = kernel_object;
176 vm_offset_t addr, i, offset;
180 size = round_page(size);
181 vmem = vm_dom[domain].vmd_kernel_arena;
182 if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
184 offset = addr - VM_MIN_KERNEL_ADDRESS;
185 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
186 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
187 pflags |= VM_ALLOC_NOWAIT;
188 VM_OBJECT_WLOCK(object);
189 for (i = 0; i < size; i += PAGE_SIZE) {
192 m = vm_page_alloc_contig_domain(object, atop(offset + i),
193 domain, pflags, 1, low, high, PAGE_SIZE, 0, memattr);
195 VM_OBJECT_WUNLOCK(object);
196 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
197 if (!vm_page_reclaim_contig_domain(domain,
198 pflags, 1, low, high, PAGE_SIZE, 0) &&
199 (flags & M_WAITOK) != 0)
200 vm_wait_domain(domain);
201 VM_OBJECT_WLOCK(object);
205 kmem_unback(object, addr, i);
206 vmem_free(vmem, addr, size);
209 KASSERT(vm_phys_domain(m) == domain,
210 ("kmem_alloc_attr_domain: Domain mismatch %d != %d",
211 vm_phys_domain(m), domain));
212 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
214 m->valid = VM_PAGE_BITS_ALL;
215 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
216 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
218 VM_OBJECT_WUNLOCK(object);
223 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
224 vm_paddr_t high, vm_memattr_t memattr)
226 struct vm_domainset_iter di;
230 KASSERT(vmem == kernel_arena,
231 ("kmem_alloc_attr: Only kernel_arena is supported."));
233 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
235 addr = kmem_alloc_attr_domain(domain, size, flags, low, high,
239 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
245 * Allocates a region from the kernel address map and physically
246 * contiguous pages within the specified address range to the kernel
247 * object. Creates a wired mapping from this region to these pages, and
248 * returns the region's starting virtual address. If M_ZERO is specified
249 * through the given flags, then the pages are zeroed before they are
253 kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
254 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
255 vm_memattr_t memattr)
258 vm_object_t object = kernel_object;
259 vm_offset_t addr, offset, tmp;
264 size = round_page(size);
265 vmem = vm_dom[domain].vmd_kernel_arena;
266 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
268 offset = addr - VM_MIN_KERNEL_ADDRESS;
269 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
270 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
271 pflags |= VM_ALLOC_NOWAIT;
273 VM_OBJECT_WLOCK(object);
276 m = vm_page_alloc_contig_domain(object, atop(offset), domain, pflags,
277 npages, low, high, alignment, boundary, memattr);
279 VM_OBJECT_WUNLOCK(object);
280 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
281 if (!vm_page_reclaim_contig_domain(domain, pflags,
282 npages, low, high, alignment, boundary) &&
283 (flags & M_WAITOK) != 0)
284 vm_wait_domain(domain);
285 VM_OBJECT_WLOCK(object);
289 vmem_free(vmem, addr, size);
292 KASSERT(vm_phys_domain(m) == domain,
293 ("kmem_alloc_contig_domain: Domain mismatch %d != %d",
294 vm_phys_domain(m), domain));
297 for (; m < end_m; m++) {
298 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
300 m->valid = VM_PAGE_BITS_ALL;
301 pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
302 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
305 VM_OBJECT_WUNLOCK(object);
310 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
311 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
312 vm_memattr_t memattr)
314 struct vm_domainset_iter di;
318 KASSERT(vmem == kernel_arena,
319 ("kmem_alloc_contig: Only kernel_arena is supported."));
321 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
323 addr = kmem_alloc_contig_domain(domain, size, flags, low, high,
324 alignment, boundary, memattr);
327 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
335 * Allocates a map to manage a subrange
336 * of the kernel virtual address space.
338 * Arguments are as follows:
340 * parent Map to take range from
341 * min, max Returned endpoints of map
342 * size Size of range to find
343 * superpage_align Request that min is superpage aligned
346 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
347 vm_size_t size, boolean_t superpage_align)
352 size = round_page(size);
354 *min = vm_map_min(parent);
355 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
356 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
358 if (ret != KERN_SUCCESS)
359 panic("kmem_suballoc: bad status return of %d", ret);
361 result = vm_map_create(vm_map_pmap(parent), *min, *max);
363 panic("kmem_suballoc: cannot create submap");
364 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
365 panic("kmem_suballoc: unable to change range to submap");
372 * Allocate wired-down pages in the kernel's address space.
375 kmem_malloc_domain(int domain, vm_size_t size, int flags)
381 vmem = vm_dom[domain].vmd_kernel_arena;
382 size = round_page(size);
383 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
386 rv = kmem_back_domain(domain, kernel_object, addr, size, flags);
387 if (rv != KERN_SUCCESS) {
388 vmem_free(vmem, addr, size);
395 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
397 struct vm_domainset_iter di;
401 KASSERT(vmem == kernel_arena,
402 ("kmem_malloc: Only kernel_arena is supported."));
404 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
406 addr = kmem_malloc_domain(domain, size, flags);
409 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
417 * Allocate physical pages for the specified virtual address range.
420 kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr,
421 vm_size_t size, int flags)
423 vm_offset_t offset, i;
427 KASSERT(object == kernel_object,
428 ("kmem_back_domain: only supports kernel object."));
430 offset = addr - VM_MIN_KERNEL_ADDRESS;
431 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
432 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
433 if (flags & M_WAITOK)
434 pflags |= VM_ALLOC_WAITFAIL;
437 VM_OBJECT_WLOCK(object);
439 mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
440 for (; i < size; i += PAGE_SIZE, mpred = m) {
441 m = vm_page_alloc_domain_after(object, atop(offset + i),
442 domain, pflags, mpred);
445 * Ran out of space, free everything up and return. Don't need
446 * to lock page queues here as we know that the pages we got
447 * aren't on any queues.
450 if ((flags & M_NOWAIT) == 0)
452 VM_OBJECT_WUNLOCK(object);
453 kmem_unback(object, addr, i);
454 return (KERN_NO_SPACE);
456 KASSERT(vm_phys_domain(m) == domain,
457 ("kmem_back_domain: Domain mismatch %d != %d",
458 vm_phys_domain(m), domain));
459 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
461 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
462 ("kmem_malloc: page %p is managed", m));
463 m->valid = VM_PAGE_BITS_ALL;
464 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
465 VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
467 VM_OBJECT_WUNLOCK(object);
469 return (KERN_SUCCESS);
473 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
475 struct vm_domainset_iter di;
479 KASSERT(object == kernel_object,
480 ("kmem_back: only supports kernel object."));
482 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
484 ret = kmem_back_domain(domain, object, addr, size, flags);
485 if (ret == KERN_SUCCESS)
487 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
495 * Unmap and free the physical pages underlying the specified virtual
498 * A physical page must exist within the specified object at each index
499 * that is being unmapped.
502 _kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
505 vm_offset_t end, offset;
508 KASSERT(object == kernel_object,
509 ("kmem_unback: only supports kernel object."));
513 pmap_remove(kernel_pmap, addr, addr + size);
514 offset = addr - VM_MIN_KERNEL_ADDRESS;
516 VM_OBJECT_WLOCK(object);
517 m = vm_page_lookup(object, atop(offset));
518 domain = vm_phys_domain(m);
519 for (; offset < end; offset += PAGE_SIZE, m = next) {
520 next = vm_page_next(m);
521 vm_page_unwire(m, PQ_NONE);
524 VM_OBJECT_WUNLOCK(object);
530 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
533 _kmem_unback(object, addr, size);
539 * Free memory allocated with kmem_malloc. The size must match the
540 * original allocation.
543 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
547 KASSERT(vmem == kernel_arena,
548 ("kmem_free: Only kernel_arena is supported."));
549 size = round_page(size);
550 domain = _kmem_unback(kernel_object, addr, size);
551 vmem_free(vm_dom[domain].vmd_kernel_arena, addr, size);
557 * Allocates pageable memory from a sub-map of the kernel. If the submap
558 * has no room, the caller sleeps waiting for more memory in the submap.
560 * This routine may block.
563 kmap_alloc_wait(vm_map_t map, vm_size_t size)
567 size = round_page(size);
568 if (!swap_reserve(size))
573 * To make this work for more than one map, use the map's lock
574 * to lock out sleepers/wakers.
577 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
579 /* no space now; see if we can ever get space */
580 if (vm_map_max(map) - vm_map_min(map) < size) {
585 map->needs_wakeup = TRUE;
586 vm_map_unlock_and_wait(map, 0);
588 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
589 VM_PROT_ALL, MAP_ACC_CHARGED);
597 * Returns memory to a submap of the kernel, and wakes up any processes
598 * waiting for memory in that map.
601 kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
605 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
606 if (map->needs_wakeup) {
607 map->needs_wakeup = FALSE;
614 kmem_init_zero_region(void)
620 * Map a single physical page of zeros to a larger virtual range.
621 * This requires less looping in places that want large amounts of
622 * zeros, while not using much more physical resources.
624 addr = kva_alloc(ZERO_REGION_SIZE);
625 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
626 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
627 if ((m->flags & PG_ZERO) == 0)
629 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
630 pmap_qenter(addr + i, &m, 1);
631 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
633 zero_region = (const void *)addr;
639 * Create the kernel map; insert a mapping covering kernel text,
640 * data, bss, and all space allocated thus far (`boostrap' data). The
641 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
642 * `start' as allocated, and the range between `start' and `end' as free.
645 kmem_init(vm_offset_t start, vm_offset_t end)
649 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
652 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
654 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
658 VM_MIN_KERNEL_ADDRESS,
660 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
661 /* ... and ending with the completion of the above `insert' */
667 * Allow userspace to directly trigger the VM drain routine for testing
671 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
676 error = sysctl_handle_int(oidp, &i, 0, req);
679 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
682 EVENTHANDLER_INVOKE(vm_lowmem, i);
686 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
687 debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");