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_RW,
216 VM_PROT_RW | PMAP_ENTER_WIRED, 0);
218 VM_OBJECT_WUNLOCK(object);
223 kmem_alloc_attr(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
224 vm_memattr_t memattr)
226 struct vm_domainset_iter di;
230 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
232 addr = kmem_alloc_attr_domain(domain, size, flags, low, high,
236 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
242 * Allocates a region from the kernel address map and physically
243 * contiguous pages within the specified address range to the kernel
244 * object. Creates a wired mapping from this region to these pages, and
245 * returns the region's starting virtual address. If M_ZERO is specified
246 * through the given flags, then the pages are zeroed before they are
250 kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
251 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
252 vm_memattr_t memattr)
255 vm_object_t object = kernel_object;
256 vm_offset_t addr, offset, tmp;
261 size = round_page(size);
262 vmem = vm_dom[domain].vmd_kernel_arena;
263 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
265 offset = addr - VM_MIN_KERNEL_ADDRESS;
266 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
267 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
268 pflags |= VM_ALLOC_NOWAIT;
270 VM_OBJECT_WLOCK(object);
273 m = vm_page_alloc_contig_domain(object, atop(offset), domain, pflags,
274 npages, low, high, alignment, boundary, memattr);
276 VM_OBJECT_WUNLOCK(object);
277 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
278 if (!vm_page_reclaim_contig_domain(domain, pflags,
279 npages, low, high, alignment, boundary) &&
280 (flags & M_WAITOK) != 0)
281 vm_wait_domain(domain);
282 VM_OBJECT_WLOCK(object);
286 vmem_free(vmem, addr, size);
289 KASSERT(vm_phys_domain(m) == domain,
290 ("kmem_alloc_contig_domain: Domain mismatch %d != %d",
291 vm_phys_domain(m), domain));
294 for (; m < end_m; m++) {
295 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
297 m->valid = VM_PAGE_BITS_ALL;
298 pmap_enter(kernel_pmap, tmp, m, VM_PROT_RW,
299 VM_PROT_RW | PMAP_ENTER_WIRED, 0);
302 VM_OBJECT_WUNLOCK(object);
307 kmem_alloc_contig(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
308 u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
310 struct vm_domainset_iter di;
314 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
316 addr = kmem_alloc_contig_domain(domain, size, flags, low, high,
317 alignment, boundary, memattr);
320 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
328 * Allocates a map to manage a subrange
329 * of the kernel virtual address space.
331 * Arguments are as follows:
333 * parent Map to take range from
334 * min, max Returned endpoints of map
335 * size Size of range to find
336 * superpage_align Request that min is superpage aligned
339 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
340 vm_size_t size, boolean_t superpage_align)
345 size = round_page(size);
347 *min = vm_map_min(parent);
348 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
349 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
351 if (ret != KERN_SUCCESS)
352 panic("kmem_suballoc: bad status return of %d", ret);
354 result = vm_map_create(vm_map_pmap(parent), *min, *max);
356 panic("kmem_suballoc: cannot create submap");
357 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
358 panic("kmem_suballoc: unable to change range to submap");
365 * Allocate wired-down pages in the kernel's address space.
368 kmem_malloc_domain(int domain, vm_size_t size, int flags)
374 #if VM_NRESERVLEVEL > 0
375 if (__predict_true((flags & M_EXEC) == 0))
376 arena = vm_dom[domain].vmd_kernel_arena;
378 arena = vm_dom[domain].vmd_kernel_rwx_arena;
380 arena = vm_dom[domain].vmd_kernel_arena;
382 size = round_page(size);
383 if (vmem_alloc(arena, size, flags | M_BESTFIT, &addr))
386 rv = kmem_back_domain(domain, kernel_object, addr, size, flags);
387 if (rv != KERN_SUCCESS) {
388 vmem_free(arena, addr, size);
395 kmem_malloc(struct vmem *vmem __unused, vm_size_t size, int flags)
397 struct vm_domainset_iter di;
401 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
403 addr = kmem_malloc_domain(domain, size, flags);
406 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
414 * Allocate physical pages for the specified virtual address range.
417 kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr,
418 vm_size_t size, int flags)
420 vm_offset_t offset, i;
425 KASSERT(object == kernel_object,
426 ("kmem_back_domain: only supports kernel object."));
428 offset = addr - VM_MIN_KERNEL_ADDRESS;
429 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
430 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
431 if (flags & M_WAITOK)
432 pflags |= VM_ALLOC_WAITFAIL;
433 prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
436 VM_OBJECT_WLOCK(object);
438 mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
439 for (; i < size; i += PAGE_SIZE, mpred = m) {
440 m = vm_page_alloc_domain_after(object, atop(offset + i),
441 domain, pflags, mpred);
444 * Ran out of space, free everything up and return. Don't need
445 * to lock page queues here as we know that the pages we got
446 * aren't on any queues.
449 if ((flags & M_NOWAIT) == 0)
451 VM_OBJECT_WUNLOCK(object);
452 kmem_unback(object, addr, i);
453 return (KERN_NO_SPACE);
455 KASSERT(vm_phys_domain(m) == domain,
456 ("kmem_back_domain: Domain mismatch %d != %d",
457 vm_phys_domain(m), domain));
458 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
460 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
461 ("kmem_malloc: page %p is managed", m));
462 m->valid = VM_PAGE_BITS_ALL;
463 pmap_enter(kernel_pmap, addr + i, m, prot,
464 prot | PMAP_ENTER_WIRED, 0);
466 VM_OBJECT_WUNLOCK(object);
468 return (KERN_SUCCESS);
472 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
474 struct vm_domainset_iter di;
478 KASSERT(object == kernel_object,
479 ("kmem_back: only supports kernel object."));
481 vm_domainset_iter_malloc_init(&di, kernel_object, &domain, &flags);
483 ret = kmem_back_domain(domain, object, addr, size, flags);
484 if (ret == KERN_SUCCESS)
486 } while (vm_domainset_iter_malloc(&di, &domain, &flags) == 0);
494 * Unmap and free the physical pages underlying the specified virtual
497 * A physical page must exist within the specified object at each index
498 * that is being unmapped.
501 _kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
504 vm_offset_t end, offset;
507 KASSERT(object == kernel_object,
508 ("kmem_unback: only supports kernel object."));
512 pmap_remove(kernel_pmap, addr, addr + size);
513 offset = addr - VM_MIN_KERNEL_ADDRESS;
515 VM_OBJECT_WLOCK(object);
516 m = vm_page_lookup(object, atop(offset));
517 domain = vm_phys_domain(m);
518 for (; offset < end; offset += PAGE_SIZE, m = next) {
519 next = vm_page_next(m);
520 vm_page_unwire(m, PQ_NONE);
523 VM_OBJECT_WUNLOCK(object);
529 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
532 _kmem_unback(object, addr, size);
538 * Free memory allocated with kmem_malloc. The size must match the
539 * original allocation.
542 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
547 #if VM_NRESERVLEVEL > 0
548 KASSERT(vmem == kernel_arena || vmem == kernel_rwx_arena,
549 ("kmem_free: Only kernel_arena or kernel_rwx_arena are supported."));
551 KASSERT(vmem == kernel_arena,
552 ("kmem_free: Only kernel_arena is supported."));
555 size = round_page(size);
556 domain = _kmem_unback(kernel_object, addr, size);
557 #if VM_NRESERVLEVEL > 0
558 if (__predict_true(vmem == kernel_arena))
559 arena = vm_dom[domain].vmd_kernel_arena;
561 arena = vm_dom[domain].vmd_kernel_rwx_arena;
563 arena = vm_dom[domain].vmd_kernel_arena;
565 vmem_free(arena, addr, size);
571 * Allocates pageable memory from a sub-map of the kernel. If the submap
572 * has no room, the caller sleeps waiting for more memory in the submap.
574 * This routine may block.
577 kmap_alloc_wait(vm_map_t map, vm_size_t size)
581 size = round_page(size);
582 if (!swap_reserve(size))
587 * To make this work for more than one map, use the map's lock
588 * to lock out sleepers/wakers.
591 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
593 /* no space now; see if we can ever get space */
594 if (vm_map_max(map) - vm_map_min(map) < size) {
599 map->needs_wakeup = TRUE;
600 vm_map_unlock_and_wait(map, 0);
602 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
603 VM_PROT_ALL, MAP_ACC_CHARGED);
611 * Returns memory to a submap of the kernel, and wakes up any processes
612 * waiting for memory in that map.
615 kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
619 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
620 if (map->needs_wakeup) {
621 map->needs_wakeup = FALSE;
628 kmem_init_zero_region(void)
634 * Map a single physical page of zeros to a larger virtual range.
635 * This requires less looping in places that want large amounts of
636 * zeros, while not using much more physical resources.
638 addr = kva_alloc(ZERO_REGION_SIZE);
639 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
640 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
641 if ((m->flags & PG_ZERO) == 0)
643 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
644 pmap_qenter(addr + i, &m, 1);
645 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
647 zero_region = (const void *)addr;
653 * Create the kernel map; insert a mapping covering kernel text,
654 * data, bss, and all space allocated thus far (`boostrap' data). The
655 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
656 * `start' as allocated, and the range between `start' and `end' as free.
659 kmem_init(vm_offset_t start, vm_offset_t end)
663 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
666 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
668 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
672 VM_MIN_KERNEL_ADDRESS,
674 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
675 /* ... and ending with the completion of the above `insert' */
680 * kmem_bootstrap_free:
682 * Free pages backing preloaded data (e.g., kernel modules) to the
683 * system. Currently only supported on platforms that create a
684 * vm_phys segment for preloaded data.
687 kmem_bootstrap_free(vm_offset_t start, vm_size_t size)
689 #if defined(__i386__) || defined(__amd64__)
690 struct vm_domain *vmd;
695 end = trunc_page(start + size);
696 start = round_page(start);
698 for (va = start; va < end; va += PAGE_SIZE) {
699 pa = pmap_kextract(va);
700 m = PHYS_TO_VM_PAGE(pa);
702 vmd = vm_pagequeue_domain(m);
703 vm_domain_free_lock(vmd);
704 vm_phys_free_pages(m, 0);
705 vmd->vmd_page_count++;
706 vm_domain_free_unlock(vmd);
708 vm_domain_freecnt_inc(vmd, 1);
709 vm_cnt.v_page_count++;
711 pmap_remove(kernel_pmap, start, end);
712 (void)vmem_add(kernel_arena, start, end - start, M_WAITOK);
718 * Allow userspace to directly trigger the VM drain routine for testing
722 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
727 error = sysctl_handle_int(oidp, &i, 0, req);
730 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
733 EVENTHANDLER_INVOKE(vm_lowmem, i);
737 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
738 debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");