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>
70 #include <sys/param.h>
71 #include <sys/systm.h>
73 #include <sys/domainset.h>
74 #include <sys/eventhandler.h>
75 #include <sys/kernel.h>
77 #include <sys/malloc.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
84 #include <sys/vmmeter.h>
87 #include <vm/vm_param.h>
88 #include <vm/vm_domainset.h>
89 #include <vm/vm_kern.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_pagequeue.h>
96 #include <vm/vm_phys.h>
97 #include <vm/vm_radix.h>
98 #include <vm/vm_extern.h>
101 struct vm_map kernel_map_store;
102 struct vm_map exec_map_store;
103 struct vm_map pipe_map_store;
105 const void *zero_region;
106 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
108 /* NB: Used by kernel debuggers. */
109 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
111 u_int exec_map_entry_size;
112 u_int exec_map_entries;
114 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
115 SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
117 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
119 &vm_max_kernel_address, 0,
121 SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
123 "Max kernel address");
125 #if VM_NRESERVLEVEL > 0
126 #define KVA_QUANTUM_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT)
128 /* On non-superpage architectures we want large import sizes. */
129 #define KVA_QUANTUM_SHIFT (8 + PAGE_SHIFT)
131 #define KVA_QUANTUM (1ul << KVA_QUANTUM_SHIFT)
132 #define KVA_NUMA_IMPORT_QUANTUM (KVA_QUANTUM * 128)
134 extern void uma_startup2(void);
139 * Allocate a virtual address range with no underlying object and
140 * no initial mapping to physical memory. Any mapping from this
141 * range to physical memory must be explicitly created prior to
142 * its use, typically with pmap_qenter(). Any attempt to create
143 * a mapping on demand through vm_fault() will result in a panic.
146 kva_alloc(vm_size_t size)
151 size = round_page(size);
152 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
162 * Release a region of kernel virtual memory allocated
163 * with kva_alloc, and return the physical pages
164 * associated with that region.
166 * This routine may not block on kernel maps.
169 kva_free(vm_offset_t addr, vm_size_t size)
172 size = round_page(size);
173 vmem_free(kernel_arena, addr, size);
177 * Update sanitizer shadow state to reflect a new allocation. Force inlining to
178 * help make KMSAN origin tracking more precise.
180 static __always_inline void
181 kmem_alloc_san(vm_offset_t addr, vm_size_t size, vm_size_t asize, int flags)
183 if ((flags & M_ZERO) == 0) {
184 kmsan_mark((void *)addr, asize, KMSAN_STATE_UNINIT);
185 kmsan_orig((void *)addr, asize, KMSAN_TYPE_KMEM,
188 kmsan_mark((void *)addr, asize, KMSAN_STATE_INITED);
190 kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
194 kmem_alloc_contig_pages(vm_object_t object, vm_pindex_t pindex, int domain,
195 int pflags, u_long npages, vm_paddr_t low, vm_paddr_t high,
196 u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
202 VM_OBJECT_ASSERT_WLOCKED(object);
204 wait = (pflags & VM_ALLOC_WAITOK) != 0;
205 reclaim = (pflags & VM_ALLOC_NORECLAIM) == 0;
206 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
207 pflags |= VM_ALLOC_NOWAIT;
208 for (tries = wait ? 3 : 1;; tries--) {
209 m = vm_page_alloc_contig_domain(object, pindex, domain, pflags,
210 npages, low, high, alignment, boundary, memattr);
211 if (m != NULL || tries == 0 || !reclaim)
214 VM_OBJECT_WUNLOCK(object);
215 if (!vm_page_reclaim_contig_domain(domain, pflags, npages,
216 low, high, alignment, boundary) && wait)
217 vm_wait_domain(domain);
218 VM_OBJECT_WLOCK(object);
224 * Allocates a region from the kernel address map and physical pages
225 * within the specified address range to the kernel object. Creates a
226 * wired mapping from this region to these pages, and returns the
227 * region's starting virtual address. The allocated pages are not
228 * necessarily physically contiguous. If M_ZERO is specified through the
229 * given flags, then the pages are zeroed before they are mapped.
232 kmem_alloc_attr_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
233 vm_paddr_t high, vm_memattr_t memattr)
237 vm_offset_t addr, i, offset;
243 object = kernel_object;
244 asize = round_page(size);
245 vmem = vm_dom[domain].vmd_kernel_arena;
246 if (vmem_alloc(vmem, asize, M_BESTFIT | flags, &addr))
248 offset = addr - VM_MIN_KERNEL_ADDRESS;
249 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
250 prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
251 VM_OBJECT_WLOCK(object);
252 for (i = 0; i < asize; i += PAGE_SIZE) {
253 m = kmem_alloc_contig_pages(object, atop(offset + i),
254 domain, pflags, 1, low, high, PAGE_SIZE, 0, memattr);
256 VM_OBJECT_WUNLOCK(object);
257 kmem_unback(object, addr, i);
258 vmem_free(vmem, addr, asize);
261 KASSERT(vm_page_domain(m) == domain,
262 ("kmem_alloc_attr_domain: Domain mismatch %d != %d",
263 vm_page_domain(m), domain));
264 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
267 pmap_enter(kernel_pmap, addr + i, m, prot,
268 prot | PMAP_ENTER_WIRED, 0);
270 VM_OBJECT_WUNLOCK(object);
271 kmem_alloc_san(addr, size, asize, flags);
272 return ((void *)addr);
276 kmem_alloc_attr(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
277 vm_memattr_t memattr)
280 return (kmem_alloc_attr_domainset(DOMAINSET_RR(), size, flags, low,
285 kmem_alloc_attr_domainset(struct domainset *ds, vm_size_t size, int flags,
286 vm_paddr_t low, vm_paddr_t high, vm_memattr_t memattr)
288 struct vm_domainset_iter di;
292 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
294 addr = kmem_alloc_attr_domain(domain, size, flags, low, high,
298 } while (vm_domainset_iter_policy(&di, &domain) == 0);
304 * Allocates a region from the kernel address map and physically
305 * contiguous pages within the specified address range to the kernel
306 * object. Creates a wired mapping from this region to these pages, and
307 * returns the region's starting virtual address. If M_ZERO is specified
308 * through the given flags, then the pages are zeroed before they are
312 kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
313 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
314 vm_memattr_t memattr)
318 vm_offset_t addr, offset, tmp;
324 object = kernel_object;
325 asize = round_page(size);
326 vmem = vm_dom[domain].vmd_kernel_arena;
327 if (vmem_alloc(vmem, asize, flags | M_BESTFIT, &addr))
329 offset = addr - VM_MIN_KERNEL_ADDRESS;
330 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
331 npages = atop(asize);
332 VM_OBJECT_WLOCK(object);
333 m = kmem_alloc_contig_pages(object, atop(offset), domain,
334 pflags, npages, low, high, alignment, boundary, memattr);
336 VM_OBJECT_WUNLOCK(object);
337 vmem_free(vmem, addr, asize);
340 KASSERT(vm_page_domain(m) == domain,
341 ("kmem_alloc_contig_domain: Domain mismatch %d != %d",
342 vm_page_domain(m), domain));
345 for (; m < end_m; m++) {
346 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
349 pmap_enter(kernel_pmap, tmp, m, VM_PROT_RW,
350 VM_PROT_RW | PMAP_ENTER_WIRED, 0);
353 VM_OBJECT_WUNLOCK(object);
354 kmem_alloc_san(addr, size, asize, flags);
355 return ((void *)addr);
359 kmem_alloc_contig(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
360 u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
363 return (kmem_alloc_contig_domainset(DOMAINSET_RR(), size, flags, low,
364 high, alignment, boundary, memattr));
368 kmem_alloc_contig_domainset(struct domainset *ds, vm_size_t size, int flags,
369 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
370 vm_memattr_t memattr)
372 struct vm_domainset_iter di;
376 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
378 addr = kmem_alloc_contig_domain(domain, size, flags, low, high,
379 alignment, boundary, memattr);
382 } while (vm_domainset_iter_policy(&di, &domain) == 0);
390 * Initializes a map to manage a subrange
391 * of the kernel virtual address space.
393 * Arguments are as follows:
395 * parent Map to take range from
396 * min, max Returned endpoints of map
397 * size Size of range to find
398 * superpage_align Request that min is superpage aligned
401 kmem_subinit(vm_map_t map, vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
402 vm_size_t size, bool superpage_align)
406 size = round_page(size);
408 *min = vm_map_min(parent);
409 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
410 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
412 if (ret != KERN_SUCCESS)
413 panic("kmem_subinit: bad status return of %d", ret);
415 vm_map_init(map, vm_map_pmap(parent), *min, *max);
416 if (vm_map_submap(parent, *min, *max, map) != KERN_SUCCESS)
417 panic("kmem_subinit: unable to change range to submap");
421 * kmem_malloc_domain:
423 * Allocate wired-down pages in the kernel's address space.
426 kmem_malloc_domain(int domain, vm_size_t size, int flags)
433 if (__predict_true((flags & M_EXEC) == 0))
434 arena = vm_dom[domain].vmd_kernel_arena;
436 arena = vm_dom[domain].vmd_kernel_rwx_arena;
437 asize = round_page(size);
438 if (vmem_alloc(arena, asize, flags | M_BESTFIT, &addr))
441 rv = kmem_back_domain(domain, kernel_object, addr, asize, flags);
442 if (rv != KERN_SUCCESS) {
443 vmem_free(arena, addr, asize);
446 kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
447 return ((void *)addr);
451 kmem_malloc(vm_size_t size, int flags)
456 p = kmem_malloc_domainset(DOMAINSET_RR(), size, flags);
462 kmem_malloc_domainset(struct domainset *ds, vm_size_t size, int flags)
464 struct vm_domainset_iter di;
468 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
470 addr = kmem_malloc_domain(domain, size, flags);
473 } while (vm_domainset_iter_policy(&di, &domain) == 0);
481 * Allocate physical pages from the specified domain for the specified
482 * virtual address range.
485 kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr,
486 vm_size_t size, int flags)
488 vm_offset_t offset, i;
493 KASSERT(object == kernel_object,
494 ("kmem_back_domain: only supports kernel object."));
496 offset = addr - VM_MIN_KERNEL_ADDRESS;
497 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
498 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
499 if (flags & M_WAITOK)
500 pflags |= VM_ALLOC_WAITFAIL;
501 prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
504 VM_OBJECT_WLOCK(object);
506 mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
507 for (; i < size; i += PAGE_SIZE, mpred = m) {
508 m = vm_page_alloc_domain_after(object, atop(offset + i),
509 domain, pflags, mpred);
512 * Ran out of space, free everything up and return. Don't need
513 * to lock page queues here as we know that the pages we got
514 * aren't on any queues.
517 if ((flags & M_NOWAIT) == 0)
519 VM_OBJECT_WUNLOCK(object);
520 kmem_unback(object, addr, i);
521 return (KERN_NO_SPACE);
523 KASSERT(vm_page_domain(m) == domain,
524 ("kmem_back_domain: Domain mismatch %d != %d",
525 vm_page_domain(m), domain));
526 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
528 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
529 ("kmem_malloc: page %p is managed", m));
531 pmap_enter(kernel_pmap, addr + i, m, prot,
532 prot | PMAP_ENTER_WIRED, 0);
533 if (__predict_false((prot & VM_PROT_EXECUTE) != 0))
534 m->oflags |= VPO_KMEM_EXEC;
536 VM_OBJECT_WUNLOCK(object);
537 kmem_alloc_san(addr, size, size, flags);
538 return (KERN_SUCCESS);
544 * Allocate physical pages for the specified virtual address range.
547 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
549 vm_offset_t end, next, start;
552 KASSERT(object == kernel_object,
553 ("kmem_back: only supports kernel object."));
555 for (start = addr, end = addr + size; addr < end; addr = next) {
557 * We must ensure that pages backing a given large virtual page
558 * all come from the same physical domain.
560 if (vm_ndomains > 1) {
561 domain = (addr >> KVA_QUANTUM_SHIFT) % vm_ndomains;
562 while (VM_DOMAIN_EMPTY(domain))
564 next = roundup2(addr + 1, KVA_QUANTUM);
565 if (next > end || next < start)
571 rv = kmem_back_domain(domain, object, addr, next - addr, flags);
572 if (rv != KERN_SUCCESS) {
573 kmem_unback(object, start, addr - start);
583 * Unmap and free the physical pages underlying the specified virtual
586 * A physical page must exist within the specified object at each index
587 * that is being unmapped.
590 _kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
594 vm_offset_t end, offset;
597 KASSERT(object == kernel_object,
598 ("kmem_unback: only supports kernel object."));
602 pmap_remove(kernel_pmap, addr, addr + size);
603 offset = addr - VM_MIN_KERNEL_ADDRESS;
605 VM_OBJECT_WLOCK(object);
606 m = vm_page_lookup(object, atop(offset));
607 domain = vm_page_domain(m);
608 if (__predict_true((m->oflags & VPO_KMEM_EXEC) == 0))
609 arena = vm_dom[domain].vmd_kernel_arena;
611 arena = vm_dom[domain].vmd_kernel_rwx_arena;
612 for (; offset < end; offset += PAGE_SIZE, m = next) {
613 next = vm_page_next(m);
614 vm_page_xbusy_claim(m);
615 vm_page_unwire_noq(m);
618 VM_OBJECT_WUNLOCK(object);
624 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
627 (void)_kmem_unback(object, addr, size);
633 * Free memory allocated with kmem_malloc. The size must match the
634 * original allocation.
637 kmem_free(void *addr, vm_size_t size)
641 size = round_page(size);
642 kasan_mark(addr, size, size, 0);
643 arena = _kmem_unback(kernel_object, (uintptr_t)addr, size);
645 vmem_free(arena, (uintptr_t)addr, size);
651 * Allocates pageable memory from a sub-map of the kernel. If the submap
652 * has no room, the caller sleeps waiting for more memory in the submap.
654 * This routine may block.
657 kmap_alloc_wait(vm_map_t map, vm_size_t size)
661 size = round_page(size);
662 if (!swap_reserve(size))
667 * To make this work for more than one map, use the map's lock
668 * to lock out sleepers/wakers.
671 addr = vm_map_findspace(map, vm_map_min(map), size);
672 if (addr + size <= vm_map_max(map))
674 /* no space now; see if we can ever get space */
675 if (vm_map_max(map) - vm_map_min(map) < size) {
680 map->needs_wakeup = TRUE;
681 vm_map_unlock_and_wait(map, 0);
683 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_RW, VM_PROT_RW,
692 * Returns memory to a submap of the kernel, and wakes up any processes
693 * waiting for memory in that map.
696 kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
700 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
701 if (map->needs_wakeup) {
702 map->needs_wakeup = FALSE;
709 kmem_init_zero_region(void)
715 * Map a single physical page of zeros to a larger virtual range.
716 * This requires less looping in places that want large amounts of
717 * zeros, while not using much more physical resources.
719 addr = kva_alloc(ZERO_REGION_SIZE);
720 m = vm_page_alloc_noobj(VM_ALLOC_WIRED | VM_ALLOC_ZERO);
721 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
722 pmap_qenter(addr + i, &m, 1);
723 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
725 zero_region = (const void *)addr;
729 * Import KVA from the kernel map into the kernel arena.
732 kva_import(void *unused, vmem_size_t size, int flags, vmem_addr_t *addrp)
738 KASSERT((size % KVA_QUANTUM) == 0,
739 ("kva_import: Size %jd is not a multiple of %d",
740 (intmax_t)size, (int)KVA_QUANTUM));
741 addr = vm_map_min(kernel_map);
742 result = vm_map_find(kernel_map, NULL, 0, &addr, size, 0,
743 VMFS_SUPER_SPACE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
744 if (result != KERN_SUCCESS) {
756 * Import KVA from a parent arena into a per-domain arena. Imports must be
757 * KVA_QUANTUM-aligned and a multiple of KVA_QUANTUM in size.
760 kva_import_domain(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp)
763 KASSERT((size % KVA_QUANTUM) == 0,
764 ("kva_import_domain: Size %jd is not a multiple of %d",
765 (intmax_t)size, (int)KVA_QUANTUM));
766 return (vmem_xalloc(arena, size, KVA_QUANTUM, 0, 0, VMEM_ADDR_MIN,
767 VMEM_ADDR_MAX, flags, addrp));
773 * Create the kernel map; insert a mapping covering kernel text,
774 * data, bss, and all space allocated thus far (`boostrap' data). The
775 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
776 * `start' as allocated, and the range between `start' and `end' as free.
777 * Create the kernel vmem arena and its per-domain children.
780 kmem_init(vm_offset_t start, vm_offset_t end)
785 vm_map_init(kernel_map, kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
786 kernel_map->system_map = 1;
787 vm_map_lock(kernel_map);
788 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
789 (void)vm_map_insert(kernel_map, NULL, 0,
793 VM_MIN_KERNEL_ADDRESS,
795 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
796 /* ... and ending with the completion of the above `insert' */
800 * Mark KVA used for the page array as allocated. Other platforms
801 * that handle vm_page_array allocation can simply adjust virtual_avail
804 (void)vm_map_insert(kernel_map, NULL, 0, (vm_offset_t)vm_page_array,
805 (vm_offset_t)vm_page_array + round_2mpage(vm_page_array_size *
806 sizeof(struct vm_page)),
807 VM_PROT_RW, VM_PROT_RW, MAP_NOFAULT);
809 vm_map_unlock(kernel_map);
812 * Use a large import quantum on NUMA systems. This helps minimize
813 * interleaving of superpages, reducing internal fragmentation within
814 * the per-domain arenas.
816 if (vm_ndomains > 1 && PMAP_HAS_DMAP)
817 quantum = KVA_NUMA_IMPORT_QUANTUM;
819 quantum = KVA_QUANTUM;
822 * Initialize the kernel_arena. This can grow on demand.
824 vmem_init(kernel_arena, "kernel arena", 0, 0, PAGE_SIZE, 0, 0);
825 vmem_set_import(kernel_arena, kva_import, NULL, NULL, quantum);
827 for (domain = 0; domain < vm_ndomains; domain++) {
829 * Initialize the per-domain arenas. These are used to color
830 * the KVA space in a way that ensures that virtual large pages
831 * are backed by memory from the same physical domain,
832 * maximizing the potential for superpage promotion.
834 vm_dom[domain].vmd_kernel_arena = vmem_create(
835 "kernel arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
836 vmem_set_import(vm_dom[domain].vmd_kernel_arena,
837 kva_import_domain, NULL, kernel_arena, quantum);
840 * In architectures with superpages, maintain separate arenas
841 * for allocations with permissions that differ from the
842 * "standard" read/write permissions used for kernel memory,
843 * so as not to inhibit superpage promotion.
845 * Use the base import quantum since this arena is rarely used.
847 #if VM_NRESERVLEVEL > 0
848 vm_dom[domain].vmd_kernel_rwx_arena = vmem_create(
849 "kernel rwx arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
850 vmem_set_import(vm_dom[domain].vmd_kernel_rwx_arena,
851 kva_import_domain, (vmem_release_t *)vmem_xfree,
852 kernel_arena, KVA_QUANTUM);
854 vm_dom[domain].vmd_kernel_rwx_arena =
855 vm_dom[domain].vmd_kernel_arena;
860 * This must be the very first call so that the virtual address
861 * space used for early allocations is properly marked used in
868 * kmem_bootstrap_free:
870 * Free pages backing preloaded data (e.g., kernel modules) to the
871 * system. Currently only supported on platforms that create a
872 * vm_phys segment for preloaded data.
875 kmem_bootstrap_free(vm_offset_t start, vm_size_t size)
877 #if defined(__i386__) || defined(__amd64__)
878 struct vm_domain *vmd;
883 end = trunc_page(start + size);
884 start = round_page(start);
888 * Preloaded files do not have execute permissions by default on amd64.
889 * Restore the default permissions to ensure that the direct map alias
892 pmap_change_prot(start, end - start, VM_PROT_RW);
894 for (va = start; va < end; va += PAGE_SIZE) {
895 pa = pmap_kextract(va);
896 m = PHYS_TO_VM_PAGE(pa);
898 vmd = vm_pagequeue_domain(m);
899 vm_domain_free_lock(vmd);
900 vm_phys_free_pages(m, 0);
901 vm_domain_free_unlock(vmd);
903 vm_domain_freecnt_inc(vmd, 1);
904 vm_cnt.v_page_count++;
906 pmap_remove(kernel_pmap, start, end);
907 (void)vmem_add(kernel_arena, start, end - start, M_WAITOK);
911 #ifdef PMAP_WANT_ACTIVE_CPUS_NAIVE
913 pmap_active_cpus(pmap_t pmap, cpuset_t *res)
922 td = cpuid_to_pcpu[c]->pc_curthread;
926 vm = vmspace_acquire_ref(p);
929 if (pmap == vmspace_pmap(vm))
937 * Allow userspace to directly trigger the VM drain routine for testing
941 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
946 error = sysctl_handle_int(oidp, &i, 0, req);
949 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
952 EVENTHANDLER_INVOKE(vm_lowmem, i);
955 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem,
956 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_vm_lowmem, "I",
957 "set to trigger vm_lowmem event with given flags");
960 debug_uma_reclaim(SYSCTL_HANDLER_ARGS)
965 error = sysctl_handle_int(oidp, &i, 0, req);
966 if (error != 0 || req->newptr == NULL)
968 if (i != UMA_RECLAIM_TRIM && i != UMA_RECLAIM_DRAIN &&
969 i != UMA_RECLAIM_DRAIN_CPU)
974 SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim,
975 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_uma_reclaim, "I",
976 "set to generate request to reclaim uma caches");
979 debug_uma_reclaim_domain(SYSCTL_HANDLER_ARGS)
981 int domain, error, request;
984 error = sysctl_handle_int(oidp, &request, 0, req);
985 if (error != 0 || req->newptr == NULL)
988 domain = request >> 4;
990 if (request != UMA_RECLAIM_TRIM && request != UMA_RECLAIM_DRAIN &&
991 request != UMA_RECLAIM_DRAIN_CPU)
993 if (domain < 0 || domain >= vm_ndomains)
995 uma_reclaim_domain(request, domain);
998 SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim_domain,
999 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0,
1000 debug_uma_reclaim_domain, "I",