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>
75 #include <sys/domainset.h>
76 #include <sys/eventhandler.h>
77 #include <sys/kernel.h>
79 #include <sys/malloc.h>
81 #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)
150 size = round_page(size);
151 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
160 * Release a region of kernel virtual memory allocated
161 * with kva_alloc, and return the physical pages
162 * associated with that region.
164 * This routine may not block on kernel maps.
167 kva_free(vm_offset_t addr, vm_size_t size)
170 size = round_page(size);
171 vmem_free(kernel_arena, addr, size);
175 kmem_alloc_contig_pages(vm_object_t object, vm_pindex_t pindex, int domain,
176 int pflags, u_long npages, vm_paddr_t low, vm_paddr_t high,
177 u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
183 VM_OBJECT_ASSERT_WLOCKED(object);
185 /* Disallow an invalid combination of flags. */
186 MPASS((pflags & (VM_ALLOC_WAITOK | VM_ALLOC_NORECLAIM)) !=
187 (VM_ALLOC_WAITOK | VM_ALLOC_NORECLAIM));
189 wait = (pflags & VM_ALLOC_WAITOK) != 0;
190 reclaim = (pflags & VM_ALLOC_NORECLAIM) == 0;
191 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
192 pflags |= VM_ALLOC_NOWAIT;
193 for (tries = wait ? 3 : 1;; tries--) {
194 m = vm_page_alloc_contig_domain(object, pindex, domain, pflags,
195 npages, low, high, alignment, boundary, memattr);
196 if (m != NULL || tries == 0 || !reclaim)
199 VM_OBJECT_WUNLOCK(object);
200 if (!vm_page_reclaim_contig_domain(domain, pflags, npages,
201 low, high, alignment, boundary) && wait)
202 vm_wait_domain(domain);
203 VM_OBJECT_WLOCK(object);
209 * Allocates a region from the kernel address map and physical pages
210 * within the specified address range to the kernel object. Creates a
211 * wired mapping from this region to these pages, and returns the
212 * region's starting virtual address. The allocated pages are not
213 * necessarily physically contiguous. If M_ZERO is specified through the
214 * given flags, then the pages are zeroed before they are mapped.
217 kmem_alloc_attr_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
218 vm_paddr_t high, vm_memattr_t memattr)
222 vm_offset_t addr, i, offset;
228 object = kernel_object;
229 asize = round_page(size);
230 vmem = vm_dom[domain].vmd_kernel_arena;
231 if (vmem_alloc(vmem, asize, M_BESTFIT | flags, &addr))
233 offset = addr - VM_MIN_KERNEL_ADDRESS;
234 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
235 prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
236 VM_OBJECT_WLOCK(object);
237 for (i = 0; i < asize; i += PAGE_SIZE) {
238 m = kmem_alloc_contig_pages(object, atop(offset + i),
239 domain, pflags, 1, low, high, PAGE_SIZE, 0, memattr);
241 VM_OBJECT_WUNLOCK(object);
242 kmem_unback(object, addr, i);
243 vmem_free(vmem, addr, asize);
246 KASSERT(vm_page_domain(m) == domain,
247 ("kmem_alloc_attr_domain: Domain mismatch %d != %d",
248 vm_page_domain(m), domain));
249 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
252 pmap_enter(kernel_pmap, addr + i, m, prot,
253 prot | PMAP_ENTER_WIRED, 0);
255 VM_OBJECT_WUNLOCK(object);
256 kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
261 kmem_alloc_attr(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
262 vm_memattr_t memattr)
265 return (kmem_alloc_attr_domainset(DOMAINSET_RR(), size, flags, low,
270 kmem_alloc_attr_domainset(struct domainset *ds, vm_size_t size, int flags,
271 vm_paddr_t low, vm_paddr_t high, vm_memattr_t memattr)
273 struct vm_domainset_iter di;
277 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
279 addr = kmem_alloc_attr_domain(domain, size, flags, low, high,
283 } while (vm_domainset_iter_policy(&di, &domain) == 0);
289 * Allocates a region from the kernel address map and physically
290 * contiguous pages within the specified address range to the kernel
291 * object. Creates a wired mapping from this region to these pages, and
292 * returns the region's starting virtual address. If M_ZERO is specified
293 * through the given flags, then the pages are zeroed before they are
297 kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
298 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
299 vm_memattr_t memattr)
303 vm_offset_t addr, offset, tmp;
309 object = kernel_object;
310 asize = round_page(size);
311 vmem = vm_dom[domain].vmd_kernel_arena;
312 if (vmem_alloc(vmem, asize, flags | M_BESTFIT, &addr))
314 offset = addr - VM_MIN_KERNEL_ADDRESS;
315 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
316 npages = atop(asize);
317 VM_OBJECT_WLOCK(object);
318 m = kmem_alloc_contig_pages(object, atop(offset), domain,
319 pflags, npages, low, high, alignment, boundary, memattr);
321 VM_OBJECT_WUNLOCK(object);
322 vmem_free(vmem, addr, asize);
325 KASSERT(vm_page_domain(m) == domain,
326 ("kmem_alloc_contig_domain: Domain mismatch %d != %d",
327 vm_page_domain(m), domain));
330 for (; m < end_m; m++) {
331 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
334 pmap_enter(kernel_pmap, tmp, m, VM_PROT_RW,
335 VM_PROT_RW | PMAP_ENTER_WIRED, 0);
338 VM_OBJECT_WUNLOCK(object);
339 kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
344 kmem_alloc_contig(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
345 u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
348 return (kmem_alloc_contig_domainset(DOMAINSET_RR(), size, flags, low,
349 high, alignment, boundary, memattr));
353 kmem_alloc_contig_domainset(struct domainset *ds, vm_size_t size, int flags,
354 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
355 vm_memattr_t memattr)
357 struct vm_domainset_iter di;
361 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
363 addr = kmem_alloc_contig_domain(domain, size, flags, low, high,
364 alignment, boundary, memattr);
367 } while (vm_domainset_iter_policy(&di, &domain) == 0);
375 * Initializes a map to manage a subrange
376 * of the kernel virtual address space.
378 * Arguments are as follows:
380 * parent Map to take range from
381 * min, max Returned endpoints of map
382 * size Size of range to find
383 * superpage_align Request that min is superpage aligned
386 kmem_subinit(vm_map_t map, vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
387 vm_size_t size, bool superpage_align)
391 size = round_page(size);
393 *min = vm_map_min(parent);
394 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
395 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
397 if (ret != KERN_SUCCESS)
398 panic("kmem_subinit: bad status return of %d", ret);
400 vm_map_init(map, vm_map_pmap(parent), *min, *max);
401 if (vm_map_submap(parent, *min, *max, map) != KERN_SUCCESS)
402 panic("kmem_subinit: unable to change range to submap");
406 * kmem_malloc_domain:
408 * Allocate wired-down pages in the kernel's address space.
411 kmem_malloc_domain(int domain, vm_size_t size, int flags)
418 if (__predict_true((flags & M_EXEC) == 0))
419 arena = vm_dom[domain].vmd_kernel_arena;
421 arena = vm_dom[domain].vmd_kernel_rwx_arena;
422 asize = round_page(size);
423 if (vmem_alloc(arena, asize, flags | M_BESTFIT, &addr))
426 rv = kmem_back_domain(domain, kernel_object, addr, asize, flags);
427 if (rv != KERN_SUCCESS) {
428 vmem_free(arena, addr, asize);
431 kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
436 kmem_malloc(vm_size_t size, int flags)
439 return (kmem_malloc_domainset(DOMAINSET_RR(), size, flags));
443 kmem_malloc_domainset(struct domainset *ds, vm_size_t size, int flags)
445 struct vm_domainset_iter di;
449 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
451 addr = kmem_malloc_domain(domain, size, flags);
454 } while (vm_domainset_iter_policy(&di, &domain) == 0);
462 * Allocate physical pages from the specified domain for the specified
463 * virtual address range.
466 kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr,
467 vm_size_t size, int flags)
469 vm_offset_t offset, i;
474 KASSERT(object == kernel_object,
475 ("kmem_back_domain: only supports kernel object."));
477 offset = addr - VM_MIN_KERNEL_ADDRESS;
478 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
479 pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
480 if (flags & M_WAITOK)
481 pflags |= VM_ALLOC_WAITFAIL;
482 prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
485 VM_OBJECT_WLOCK(object);
487 mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
488 for (; i < size; i += PAGE_SIZE, mpred = m) {
489 m = vm_page_alloc_domain_after(object, atop(offset + i),
490 domain, pflags, mpred);
493 * Ran out of space, free everything up and return. Don't need
494 * to lock page queues here as we know that the pages we got
495 * aren't on any queues.
498 if ((flags & M_NOWAIT) == 0)
500 VM_OBJECT_WUNLOCK(object);
501 kmem_unback(object, addr, i);
502 return (KERN_NO_SPACE);
504 KASSERT(vm_page_domain(m) == domain,
505 ("kmem_back_domain: Domain mismatch %d != %d",
506 vm_page_domain(m), domain));
507 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
509 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
510 ("kmem_malloc: page %p is managed", m));
512 pmap_enter(kernel_pmap, addr + i, m, prot,
513 prot | PMAP_ENTER_WIRED, 0);
514 if (__predict_false((prot & VM_PROT_EXECUTE) != 0))
515 m->oflags |= VPO_KMEM_EXEC;
517 VM_OBJECT_WUNLOCK(object);
519 return (KERN_SUCCESS);
525 * Allocate physical pages for the specified virtual address range.
528 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
530 vm_offset_t end, next, start;
533 KASSERT(object == kernel_object,
534 ("kmem_back: only supports kernel object."));
536 for (start = addr, end = addr + size; addr < end; addr = next) {
538 * We must ensure that pages backing a given large virtual page
539 * all come from the same physical domain.
541 if (vm_ndomains > 1) {
542 domain = (addr >> KVA_QUANTUM_SHIFT) % vm_ndomains;
543 while (VM_DOMAIN_EMPTY(domain))
545 next = roundup2(addr + 1, KVA_QUANTUM);
546 if (next > end || next < start)
552 rv = kmem_back_domain(domain, object, addr, next - addr, flags);
553 if (rv != KERN_SUCCESS) {
554 kmem_unback(object, start, addr - start);
564 * Unmap and free the physical pages underlying the specified virtual
567 * A physical page must exist within the specified object at each index
568 * that is being unmapped.
571 _kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
575 vm_offset_t end, offset;
578 KASSERT(object == kernel_object,
579 ("kmem_unback: only supports kernel object."));
583 pmap_remove(kernel_pmap, addr, addr + size);
584 offset = addr - VM_MIN_KERNEL_ADDRESS;
586 VM_OBJECT_WLOCK(object);
587 m = vm_page_lookup(object, atop(offset));
588 domain = vm_page_domain(m);
589 if (__predict_true((m->oflags & VPO_KMEM_EXEC) == 0))
590 arena = vm_dom[domain].vmd_kernel_arena;
592 arena = vm_dom[domain].vmd_kernel_rwx_arena;
593 for (; offset < end; offset += PAGE_SIZE, m = next) {
594 next = vm_page_next(m);
595 vm_page_xbusy_claim(m);
596 vm_page_unwire_noq(m);
599 VM_OBJECT_WUNLOCK(object);
605 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
608 (void)_kmem_unback(object, addr, size);
614 * Free memory allocated with kmem_malloc. The size must match the
615 * original allocation.
618 kmem_free(vm_offset_t addr, vm_size_t size)
622 size = round_page(size);
623 kasan_mark((void *)addr, size, size, 0);
624 arena = _kmem_unback(kernel_object, addr, size);
626 vmem_free(arena, addr, size);
632 * Allocates pageable memory from a sub-map of the kernel. If the submap
633 * has no room, the caller sleeps waiting for more memory in the submap.
635 * This routine may block.
638 kmap_alloc_wait(vm_map_t map, vm_size_t size)
642 size = round_page(size);
643 if (!swap_reserve(size))
648 * To make this work for more than one map, use the map's lock
649 * to lock out sleepers/wakers.
652 addr = vm_map_findspace(map, vm_map_min(map), size);
653 if (addr + size <= vm_map_max(map))
655 /* no space now; see if we can ever get space */
656 if (vm_map_max(map) - vm_map_min(map) < size) {
661 map->needs_wakeup = TRUE;
662 vm_map_unlock_and_wait(map, 0);
664 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_RW, VM_PROT_RW,
673 * Returns memory to a submap of the kernel, and wakes up any processes
674 * waiting for memory in that map.
677 kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
681 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
682 if (map->needs_wakeup) {
683 map->needs_wakeup = FALSE;
690 kmem_init_zero_region(void)
696 * Map a single physical page of zeros to a larger virtual range.
697 * This requires less looping in places that want large amounts of
698 * zeros, while not using much more physical resources.
700 addr = kva_alloc(ZERO_REGION_SIZE);
701 m = vm_page_alloc_noobj(VM_ALLOC_WIRED | VM_ALLOC_ZERO);
702 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
703 pmap_qenter(addr + i, &m, 1);
704 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
706 zero_region = (const void *)addr;
710 * Import KVA from the kernel map into the kernel arena.
713 kva_import(void *unused, vmem_size_t size, int flags, vmem_addr_t *addrp)
718 KASSERT((size % KVA_QUANTUM) == 0,
719 ("kva_import: Size %jd is not a multiple of %d",
720 (intmax_t)size, (int)KVA_QUANTUM));
721 addr = vm_map_min(kernel_map);
722 result = vm_map_find(kernel_map, NULL, 0, &addr, size, 0,
723 VMFS_SUPER_SPACE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
724 if (result != KERN_SUCCESS)
733 * Import KVA from a parent arena into a per-domain arena. Imports must be
734 * KVA_QUANTUM-aligned and a multiple of KVA_QUANTUM in size.
737 kva_import_domain(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp)
740 KASSERT((size % KVA_QUANTUM) == 0,
741 ("kva_import_domain: Size %jd is not a multiple of %d",
742 (intmax_t)size, (int)KVA_QUANTUM));
743 return (vmem_xalloc(arena, size, KVA_QUANTUM, 0, 0, VMEM_ADDR_MIN,
744 VMEM_ADDR_MAX, flags, addrp));
750 * Create the kernel map; insert a mapping covering kernel text,
751 * data, bss, and all space allocated thus far (`boostrap' data). The
752 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
753 * `start' as allocated, and the range between `start' and `end' as free.
754 * Create the kernel vmem arena and its per-domain children.
757 kmem_init(vm_offset_t start, vm_offset_t end)
762 vm_map_init(kernel_map, kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
763 kernel_map->system_map = 1;
764 vm_map_lock(kernel_map);
765 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
766 (void)vm_map_insert(kernel_map, NULL, 0,
770 VM_MIN_KERNEL_ADDRESS,
772 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
773 /* ... and ending with the completion of the above `insert' */
777 * Mark KVA used for the page array as allocated. Other platforms
778 * that handle vm_page_array allocation can simply adjust virtual_avail
781 (void)vm_map_insert(kernel_map, NULL, 0, (vm_offset_t)vm_page_array,
782 (vm_offset_t)vm_page_array + round_2mpage(vm_page_array_size *
783 sizeof(struct vm_page)),
784 VM_PROT_RW, VM_PROT_RW, MAP_NOFAULT);
786 vm_map_unlock(kernel_map);
789 * Use a large import quantum on NUMA systems. This helps minimize
790 * interleaving of superpages, reducing internal fragmentation within
791 * the per-domain arenas.
793 if (vm_ndomains > 1 && PMAP_HAS_DMAP)
794 quantum = KVA_NUMA_IMPORT_QUANTUM;
796 quantum = KVA_QUANTUM;
799 * Initialize the kernel_arena. This can grow on demand.
801 vmem_init(kernel_arena, "kernel arena", 0, 0, PAGE_SIZE, 0, 0);
802 vmem_set_import(kernel_arena, kva_import, NULL, NULL, quantum);
804 for (domain = 0; domain < vm_ndomains; domain++) {
806 * Initialize the per-domain arenas. These are used to color
807 * the KVA space in a way that ensures that virtual large pages
808 * are backed by memory from the same physical domain,
809 * maximizing the potential for superpage promotion.
811 vm_dom[domain].vmd_kernel_arena = vmem_create(
812 "kernel arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
813 vmem_set_import(vm_dom[domain].vmd_kernel_arena,
814 kva_import_domain, NULL, kernel_arena, quantum);
817 * In architectures with superpages, maintain separate arenas
818 * for allocations with permissions that differ from the
819 * "standard" read/write permissions used for kernel memory,
820 * so as not to inhibit superpage promotion.
822 * Use the base import quantum since this arena is rarely used.
824 #if VM_NRESERVLEVEL > 0
825 vm_dom[domain].vmd_kernel_rwx_arena = vmem_create(
826 "kernel rwx arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
827 vmem_set_import(vm_dom[domain].vmd_kernel_rwx_arena,
828 kva_import_domain, (vmem_release_t *)vmem_xfree,
829 kernel_arena, KVA_QUANTUM);
831 vm_dom[domain].vmd_kernel_rwx_arena =
832 vm_dom[domain].vmd_kernel_arena;
837 * This must be the very first call so that the virtual address
838 * space used for early allocations is properly marked used in
845 * kmem_bootstrap_free:
847 * Free pages backing preloaded data (e.g., kernel modules) to the
848 * system. Currently only supported on platforms that create a
849 * vm_phys segment for preloaded data.
852 kmem_bootstrap_free(vm_offset_t start, vm_size_t size)
854 #if defined(__i386__) || defined(__amd64__)
855 struct vm_domain *vmd;
860 end = trunc_page(start + size);
861 start = round_page(start);
865 * Preloaded files do not have execute permissions by default on amd64.
866 * Restore the default permissions to ensure that the direct map alias
869 pmap_change_prot(start, end - start, VM_PROT_RW);
871 for (va = start; va < end; va += PAGE_SIZE) {
872 pa = pmap_kextract(va);
873 m = PHYS_TO_VM_PAGE(pa);
875 vmd = vm_pagequeue_domain(m);
876 vm_domain_free_lock(vmd);
877 vm_phys_free_pages(m, 0);
878 vm_domain_free_unlock(vmd);
880 vm_domain_freecnt_inc(vmd, 1);
881 vm_cnt.v_page_count++;
883 pmap_remove(kernel_pmap, start, end);
884 (void)vmem_add(kernel_arena, start, end - start, M_WAITOK);
889 * Allow userspace to directly trigger the VM drain routine for testing
893 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
898 error = sysctl_handle_int(oidp, &i, 0, req);
901 if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
904 EVENTHANDLER_INVOKE(vm_lowmem, i);
907 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem,
908 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_vm_lowmem, "I",
909 "set to trigger vm_lowmem event with given flags");
912 debug_uma_reclaim(SYSCTL_HANDLER_ARGS)
917 error = sysctl_handle_int(oidp, &i, 0, req);
918 if (error != 0 || req->newptr == NULL)
920 if (i != UMA_RECLAIM_TRIM && i != UMA_RECLAIM_DRAIN &&
921 i != UMA_RECLAIM_DRAIN_CPU)
926 SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim,
927 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_uma_reclaim, "I",
928 "set to generate request to reclaim uma caches");
931 debug_uma_reclaim_domain(SYSCTL_HANDLER_ARGS)
933 int domain, error, request;
936 error = sysctl_handle_int(oidp, &request, 0, req);
937 if (error != 0 || req->newptr == NULL)
940 domain = request >> 4;
942 if (request != UMA_RECLAIM_TRIM && request != UMA_RECLAIM_DRAIN &&
943 request != UMA_RECLAIM_DRAIN_CPU)
945 if (domain < 0 || domain >= vm_ndomains)
947 uma_reclaim_domain(request, domain);
950 SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim_domain,
951 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0,
952 debug_uma_reclaim_domain, "I",