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>
79 #include <vm/vm_param.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pageout.h>
85 #include <vm/vm_extern.h>
93 vm_map_t bio_transient_map;
95 const void *zero_region;
96 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
98 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
99 NULL, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
101 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
102 #if defined(__arm__) || defined(__sparc64__)
103 &vm_max_kernel_address, 0,
105 NULL, VM_MAX_KERNEL_ADDRESS,
107 "Max kernel address");
110 * kmem_alloc_nofault:
112 * Allocate a virtual address range with no underlying object and
113 * no initial mapping to physical memory. Any mapping from this
114 * range to physical memory must be explicitly created prior to
115 * its use, typically with pmap_qenter(). Any attempt to create
116 * a mapping on demand through vm_fault() will result in a panic.
119 kmem_alloc_nofault(map, size)
126 size = round_page(size);
127 addr = vm_map_min(map);
128 result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
129 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
130 if (result != KERN_SUCCESS) {
137 * kmem_alloc_nofault_space:
139 * Allocate a virtual address range with no underlying object and
140 * no initial mapping to physical memory within the specified
141 * address space. Any mapping from this range to physical memory
142 * must be explicitly created prior to its use, typically with
143 * pmap_qenter(). Any attempt to create a mapping on demand
144 * through vm_fault() will result in a panic.
147 kmem_alloc_nofault_space(map, size, find_space)
155 size = round_page(size);
156 addr = vm_map_min(map);
157 result = vm_map_find(map, NULL, 0, &addr, size, find_space,
158 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
159 if (result != KERN_SUCCESS) {
166 * Allocate wired-down memory in the kernel's address map
170 kmem_alloc(map, size)
177 size = round_page(size);
180 * Use the kernel object for wired-down kernel pages. Assume that no
181 * region of the kernel object is referenced more than once.
185 * Locate sufficient space in the map. This will give us the final
186 * virtual address for the new memory, and thus will tell us the
187 * offset within the kernel map.
190 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
194 offset = addr - VM_MIN_KERNEL_ADDRESS;
195 vm_object_reference(kernel_object);
196 vm_map_insert(map, kernel_object, offset, addr, addr + size,
197 VM_PROT_ALL, VM_PROT_ALL, 0);
201 * And finally, mark the data as non-pageable.
203 (void) vm_map_wire(map, addr, addr + size,
204 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
210 * Allocates a region from the kernel address map and physical pages
211 * within the specified address range to the kernel object. Creates a
212 * wired mapping from this region to these pages, and returns the
213 * region's starting virtual address. The allocated pages are not
214 * necessarily physically contiguous. If M_ZERO is specified through the
215 * given flags, then the pages are zeroed before they are mapped.
218 kmem_alloc_attr(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
219 vm_paddr_t high, vm_memattr_t memattr)
221 vm_object_t object = kernel_object;
223 vm_ooffset_t end_offset, offset;
227 size = round_page(size);
229 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
233 offset = addr - VM_MIN_KERNEL_ADDRESS;
234 vm_object_reference(object);
235 vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
237 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
238 VM_OBJECT_WLOCK(object);
239 end_offset = offset + size;
240 for (; offset < end_offset; offset += PAGE_SIZE) {
243 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags, 1,
244 low, high, PAGE_SIZE, 0, memattr);
246 VM_OBJECT_WUNLOCK(object);
247 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
249 vm_pageout_grow_cache(tries, low, high);
251 VM_OBJECT_WLOCK(object);
257 * Since the pages that were allocated by any previous
258 * iterations of this loop are not busy, they can be
259 * freed by vm_object_page_remove(), which is called
260 * by vm_map_delete().
262 vm_map_delete(map, addr, addr + size);
266 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
268 m->valid = VM_PAGE_BITS_ALL;
270 VM_OBJECT_WUNLOCK(object);
272 vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
273 VM_MAP_WIRE_NOHOLES);
278 * Allocates a region from the kernel address map and physically
279 * contiguous pages within the specified address range to the kernel
280 * object. Creates a wired mapping from this region to these pages, and
281 * returns the region's starting virtual address. If M_ZERO is specified
282 * through the given flags, then the pages are zeroed before they are
286 kmem_alloc_contig(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
287 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
288 vm_memattr_t memattr)
290 vm_object_t object = kernel_object;
296 size = round_page(size);
298 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
302 offset = addr - VM_MIN_KERNEL_ADDRESS;
303 vm_object_reference(object);
304 vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
306 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
307 VM_OBJECT_WLOCK(object);
310 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
311 atop(size), low, high, alignment, boundary, memattr);
313 VM_OBJECT_WUNLOCK(object);
314 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
316 vm_pageout_grow_cache(tries, low, high);
318 VM_OBJECT_WLOCK(object);
322 vm_map_delete(map, addr, addr + size);
326 end_m = m + atop(size);
327 for (; m < end_m; m++) {
328 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
330 m->valid = VM_PAGE_BITS_ALL;
332 VM_OBJECT_WUNLOCK(object);
334 vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
335 VM_MAP_WIRE_NOHOLES);
342 * Release a region of kernel virtual memory allocated
343 * with kmem_alloc, and return the physical pages
344 * associated with that region.
346 * This routine may not block on kernel maps.
349 kmem_free(map, addr, size)
355 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
361 * Allocates a map to manage a subrange
362 * of the kernel virtual address space.
364 * Arguments are as follows:
366 * parent Map to take range from
367 * min, max Returned endpoints of map
368 * size Size of range to find
369 * superpage_align Request that min is superpage aligned
372 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
373 vm_size_t size, boolean_t superpage_align)
378 size = round_page(size);
380 *min = vm_map_min(parent);
381 ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
382 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
384 if (ret != KERN_SUCCESS)
385 panic("kmem_suballoc: bad status return of %d", ret);
387 result = vm_map_create(vm_map_pmap(parent), *min, *max);
389 panic("kmem_suballoc: cannot create submap");
390 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
391 panic("kmem_suballoc: unable to change range to submap");
398 * Allocate wired-down memory in the kernel's address map for the higher
399 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
400 * kmem_alloc() because we may need to allocate memory at interrupt
401 * level where we cannot block (canwait == FALSE).
403 * This routine has its own private kernel submap (kmem_map) and object
404 * (kmem_object). This, combined with the fact that only malloc uses
405 * this routine, ensures that we will never block in map or object waits.
407 * We don't worry about expanding the map (adding entries) since entries
408 * for wired maps are statically allocated.
410 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
411 * which we never free.
414 kmem_malloc(map, size, flags)
422 size = round_page(size);
423 addr = vm_map_min(map);
426 * Locate sufficient space in the map. This will give us the final
427 * virtual address for the new memory, and thus will tell us the
428 * offset within the kernel map.
431 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
433 if ((flags & M_NOWAIT) == 0) {
434 for (i = 0; i < 8; i++) {
435 EVENTHANDLER_INVOKE(vm_lowmem, 0);
438 if (vm_map_findspace(map, vm_map_min(map),
443 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
446 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
447 (long)size, (long)map->size);
454 rv = kmem_back(map, addr, size, flags);
456 return (rv == KERN_SUCCESS ? addr : 0);
462 * Allocate physical pages for the specified virtual address range.
465 kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
467 vm_offset_t offset, i;
468 vm_map_entry_t entry;
473 KASSERT(vm_map_locked(map), ("kmem_back: map %p is not locked", map));
474 offset = addr - VM_MIN_KERNEL_ADDRESS;
475 vm_object_reference(kmem_object);
476 vm_map_insert(map, kmem_object, offset, addr, addr + size,
477 VM_PROT_ALL, VM_PROT_ALL, 0);
480 * Assert: vm_map_insert() will never be able to extend the
481 * previous entry so vm_map_lookup_entry() will find a new
482 * entry exactly corresponding to this address range and it
483 * will have wired_count == 0.
485 found = vm_map_lookup_entry(map, addr, &entry);
486 KASSERT(found && entry->start == addr && entry->end == addr + size &&
487 entry->wired_count == 0 && (entry->eflags & MAP_ENTRY_IN_TRANSITION)
488 == 0, ("kmem_back: entry not found or misaligned"));
490 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
492 VM_OBJECT_WLOCK(kmem_object);
493 for (i = 0; i < size; i += PAGE_SIZE) {
495 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
498 * Ran out of space, free everything up and return. Don't need
499 * to lock page queues here as we know that the pages we got
500 * aren't on any queues.
503 if ((flags & M_NOWAIT) == 0) {
504 VM_OBJECT_WUNLOCK(kmem_object);
505 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
510 (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_NEEDS_WAKEUP)) ==
511 MAP_ENTRY_IN_TRANSITION,
512 ("kmem_back: volatile entry"));
513 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
514 VM_OBJECT_WLOCK(kmem_object);
518 * Free the pages before removing the map entry.
519 * They are already marked busy. Calling
520 * vm_map_delete before the pages has been freed or
521 * unbusied will cause a deadlock.
525 m = vm_page_lookup(kmem_object,
526 OFF_TO_IDX(offset + i));
527 vm_page_unwire(m, 0);
530 VM_OBJECT_WUNLOCK(kmem_object);
531 vm_map_delete(map, addr, addr + size);
532 return (KERN_NO_SPACE);
534 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
536 m->valid = VM_PAGE_BITS_ALL;
537 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
538 ("kmem_malloc: page %p is managed", m));
540 VM_OBJECT_WUNLOCK(kmem_object);
543 * Mark map entry as non-pageable. Repeat the assert.
545 KASSERT(entry->start == addr && entry->end == addr + size &&
546 entry->wired_count == 0,
547 ("kmem_back: entry not found or misaligned after allocation"));
548 entry->wired_count = 1;
551 * At this point, the kmem_object must be unlocked because
552 * vm_map_simplify_entry() calls vm_object_deallocate(), which
553 * locks the kmem_object.
555 vm_map_simplify_entry(map, entry);
558 * Loop thru pages, entering them in the pmap.
560 VM_OBJECT_WLOCK(kmem_object);
561 for (i = 0; i < size; i += PAGE_SIZE) {
562 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
564 * Because this is kernel_pmap, this call will not block.
566 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
570 VM_OBJECT_WUNLOCK(kmem_object);
572 return (KERN_SUCCESS);
578 * Allocates pageable memory from a sub-map of the kernel. If the submap
579 * has no room, the caller sleeps waiting for more memory in the submap.
581 * This routine may block.
584 kmem_alloc_wait(map, size)
590 size = round_page(size);
591 if (!swap_reserve(size))
596 * To make this work for more than one map, use the map's lock
597 * to lock out sleepers/wakers.
600 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
602 /* no space now; see if we can ever get space */
603 if (vm_map_max(map) - vm_map_min(map) < size) {
608 map->needs_wakeup = TRUE;
609 vm_map_unlock_and_wait(map, 0);
611 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
612 VM_PROT_ALL, MAP_ACC_CHARGED);
620 * Returns memory to a submap of the kernel, and wakes up any processes
621 * waiting for memory in that map.
624 kmem_free_wakeup(map, addr, size)
631 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
632 if (map->needs_wakeup) {
633 map->needs_wakeup = FALSE;
640 kmem_init_zero_region(void)
647 * Map a single physical page of zeros to a larger virtual range.
648 * This requires less looping in places that want large amounts of
649 * zeros, while not using much more physical resources.
651 addr = kmem_alloc_nofault(kernel_map, ZERO_REGION_SIZE);
652 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
653 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
654 if ((m->flags & PG_ZERO) == 0)
656 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
657 pmap_qenter(addr + i, &m, 1);
658 error = vm_map_protect(kernel_map, addr, addr + ZERO_REGION_SIZE,
660 KASSERT(error == 0, ("error=%d", error));
662 zero_region = (const void *)addr;
668 * Create the kernel map; insert a mapping covering kernel text,
669 * data, bss, and all space allocated thus far (`boostrap' data). The
670 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
671 * `start' as allocated, and the range between `start' and `end' as free.
674 kmem_init(start, end)
675 vm_offset_t start, end;
679 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
682 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
684 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
688 VM_MIN_KERNEL_ADDRESS,
690 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
691 /* ... and ending with the completion of the above `insert' */
694 kmem_init_zero_region();
699 * Allow userspace to directly trigger the VM drain routine for testing
703 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
708 error = sysctl_handle_int(oidp, &i, 0, req);
712 EVENTHANDLER_INVOKE(vm_lowmem, 0);
716 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
717 debug_vm_lowmem, "I", "set to trigger vm_lowmem event");