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>
73 #include <sys/mutex.h>
75 #include <sys/malloc.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>
88 vm_map_t kernel_map=0;
92 vm_map_t buffer_map=0;
94 const void *zero_region;
95 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
97 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
98 NULL, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
100 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
101 #if defined(__arm__) || defined(__sparc64__)
102 &vm_max_kernel_address, 0,
104 NULL, VM_MAX_KERNEL_ADDRESS,
106 "Max kernel address");
109 * kmem_alloc_nofault:
111 * Allocate a virtual address range with no underlying object and
112 * no initial mapping to physical memory. Any mapping from this
113 * range to physical memory must be explicitly created prior to
114 * its use, typically with pmap_qenter(). Any attempt to create
115 * a mapping on demand through vm_fault() will result in a panic.
118 kmem_alloc_nofault(map, size)
125 size = round_page(size);
126 addr = vm_map_min(map);
127 result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
128 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
129 if (result != KERN_SUCCESS) {
136 * kmem_alloc_nofault_space:
138 * Allocate a virtual address range with no underlying object and
139 * no initial mapping to physical memory within the specified
140 * address space. Any mapping from this range to physical memory
141 * must be explicitly created prior to its use, typically with
142 * pmap_qenter(). Any attempt to create a mapping on demand
143 * through vm_fault() will result in a panic.
146 kmem_alloc_nofault_space(map, size, find_space)
154 size = round_page(size);
155 addr = vm_map_min(map);
156 result = vm_map_find(map, NULL, 0, &addr, size, find_space,
157 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
158 if (result != KERN_SUCCESS) {
165 * Allocate wired-down memory in the kernel's address map
169 kmem_alloc(map, size)
176 size = round_page(size);
179 * Use the kernel object for wired-down kernel pages. Assume that no
180 * region of the kernel object is referenced more than once.
184 * Locate sufficient space in the map. This will give us the final
185 * virtual address for the new memory, and thus will tell us the
186 * offset within the kernel map.
189 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
193 offset = addr - VM_MIN_KERNEL_ADDRESS;
194 vm_object_reference(kernel_object);
195 vm_map_insert(map, kernel_object, offset, addr, addr + size,
196 VM_PROT_ALL, VM_PROT_ALL, 0);
200 * And finally, mark the data as non-pageable.
202 (void) vm_map_wire(map, addr, addr + size,
203 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
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(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
218 vm_paddr_t high, vm_memattr_t memattr)
220 vm_object_t object = kernel_object;
222 vm_ooffset_t end_offset, offset;
226 size = round_page(size);
228 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
232 offset = addr - VM_MIN_KERNEL_ADDRESS;
233 vm_object_reference(object);
234 vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
236 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
237 VM_OBJECT_LOCK(object);
238 end_offset = offset + size;
239 for (; offset < end_offset; offset += PAGE_SIZE) {
242 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags, 1,
243 low, high, PAGE_SIZE, 0, memattr);
245 VM_OBJECT_UNLOCK(object);
246 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
248 vm_pageout_grow_cache(tries, low, high);
250 VM_OBJECT_LOCK(object);
256 * Since the pages that were allocated by any previous
257 * iterations of this loop are not busy, they can be
258 * freed by vm_object_page_remove(), which is called
259 * by vm_map_delete().
261 vm_map_delete(map, addr, addr + size);
265 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
267 m->valid = VM_PAGE_BITS_ALL;
269 VM_OBJECT_UNLOCK(object);
271 vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
272 VM_MAP_WIRE_NOHOLES);
277 * Allocates a region from the kernel address map and physically
278 * contiguous pages within the specified address range to the kernel
279 * object. Creates a wired mapping from this region to these pages, and
280 * returns the region's starting virtual address. If M_ZERO is specified
281 * through the given flags, then the pages are zeroed before they are
285 kmem_alloc_contig(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
286 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
287 vm_memattr_t memattr)
289 vm_object_t object = kernel_object;
295 size = round_page(size);
297 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
301 offset = addr - VM_MIN_KERNEL_ADDRESS;
302 vm_object_reference(object);
303 vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
305 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
306 VM_OBJECT_LOCK(object);
309 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
310 atop(size), low, high, alignment, boundary, memattr);
312 VM_OBJECT_UNLOCK(object);
313 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
315 vm_pageout_grow_cache(tries, low, high);
317 VM_OBJECT_LOCK(object);
321 vm_map_delete(map, addr, addr + size);
325 end_m = m + atop(size);
326 for (; m < end_m; m++) {
327 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
329 m->valid = VM_PAGE_BITS_ALL;
331 VM_OBJECT_UNLOCK(object);
333 vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
334 VM_MAP_WIRE_NOHOLES);
341 * Release a region of kernel virtual memory allocated
342 * with kmem_alloc, and return the physical pages
343 * associated with that region.
345 * This routine may not block on kernel maps.
348 kmem_free(map, addr, size)
354 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
360 * Allocates a map to manage a subrange
361 * of the kernel virtual address space.
363 * Arguments are as follows:
365 * parent Map to take range from
366 * min, max Returned endpoints of map
367 * size Size of range to find
368 * superpage_align Request that min is superpage aligned
371 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
372 vm_size_t size, boolean_t superpage_align)
377 size = round_page(size);
379 *min = vm_map_min(parent);
380 ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
381 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
383 if (ret != KERN_SUCCESS)
384 panic("kmem_suballoc: bad status return of %d", ret);
386 result = vm_map_create(vm_map_pmap(parent), *min, *max);
388 panic("kmem_suballoc: cannot create submap");
389 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
390 panic("kmem_suballoc: unable to change range to submap");
397 * Allocate wired-down memory in the kernel's address map for the higher
398 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
399 * kmem_alloc() because we may need to allocate memory at interrupt
400 * level where we cannot block (canwait == FALSE).
402 * This routine has its own private kernel submap (kmem_map) and object
403 * (kmem_object). This, combined with the fact that only malloc uses
404 * this routine, ensures that we will never block in map or object waits.
406 * We don't worry about expanding the map (adding entries) since entries
407 * for wired maps are statically allocated.
409 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
410 * which we never free.
413 kmem_malloc(map, size, flags)
421 size = round_page(size);
422 addr = vm_map_min(map);
425 * Locate sufficient space in the map. This will give us the final
426 * virtual address for the new memory, and thus will tell us the
427 * offset within the kernel map.
430 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
432 if ((flags & M_NOWAIT) == 0) {
433 for (i = 0; i < 8; i++) {
434 EVENTHANDLER_INVOKE(vm_lowmem, 0);
437 if (vm_map_findspace(map, vm_map_min(map),
442 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
445 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
446 (long)size, (long)map->size);
453 rv = kmem_back(map, addr, size, flags);
455 return (rv == KERN_SUCCESS ? addr : 0);
461 * Allocate physical pages for the specified virtual address range.
464 kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
466 vm_offset_t offset, i;
467 vm_map_entry_t entry;
472 KASSERT(vm_map_locked(map), ("kmem_back: map %p is not locked", map));
473 offset = addr - VM_MIN_KERNEL_ADDRESS;
474 vm_object_reference(kmem_object);
475 vm_map_insert(map, kmem_object, offset, addr, addr + size,
476 VM_PROT_ALL, VM_PROT_ALL, 0);
479 * Assert: vm_map_insert() will never be able to extend the
480 * previous entry so vm_map_lookup_entry() will find a new
481 * entry exactly corresponding to this address range and it
482 * will have wired_count == 0.
484 found = vm_map_lookup_entry(map, addr, &entry);
485 KASSERT(found && entry->start == addr && entry->end == addr + size &&
486 entry->wired_count == 0 && (entry->eflags & MAP_ENTRY_IN_TRANSITION)
487 == 0, ("kmem_back: entry not found or misaligned"));
489 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
491 VM_OBJECT_LOCK(kmem_object);
492 for (i = 0; i < size; i += PAGE_SIZE) {
494 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
497 * Ran out of space, free everything up and return. Don't need
498 * to lock page queues here as we know that the pages we got
499 * aren't on any queues.
502 if ((flags & M_NOWAIT) == 0) {
503 VM_OBJECT_UNLOCK(kmem_object);
504 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
509 (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_NEEDS_WAKEUP)) ==
510 MAP_ENTRY_IN_TRANSITION,
511 ("kmem_back: volatile entry"));
512 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
513 VM_OBJECT_LOCK(kmem_object);
517 * Free the pages before removing the map entry.
518 * They are already marked busy. Calling
519 * vm_map_delete before the pages has been freed or
520 * unbusied will cause a deadlock.
524 m = vm_page_lookup(kmem_object,
525 OFF_TO_IDX(offset + i));
526 vm_page_unwire(m, 0);
529 VM_OBJECT_UNLOCK(kmem_object);
530 vm_map_delete(map, addr, addr + size);
531 return (KERN_NO_SPACE);
533 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
535 m->valid = VM_PAGE_BITS_ALL;
536 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
537 ("kmem_malloc: page %p is managed", m));
539 VM_OBJECT_UNLOCK(kmem_object);
542 * Mark map entry as non-pageable. Repeat the assert.
544 KASSERT(entry->start == addr && entry->end == addr + size &&
545 entry->wired_count == 0,
546 ("kmem_back: entry not found or misaligned after allocation"));
547 entry->wired_count = 1;
550 * At this point, the kmem_object must be unlocked because
551 * vm_map_simplify_entry() calls vm_object_deallocate(), which
552 * locks the kmem_object.
554 vm_map_simplify_entry(map, entry);
557 * Loop thru pages, entering them in the pmap.
559 VM_OBJECT_LOCK(kmem_object);
560 for (i = 0; i < size; i += PAGE_SIZE) {
561 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
563 * Because this is kernel_pmap, this call will not block.
565 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
569 VM_OBJECT_UNLOCK(kmem_object);
571 return (KERN_SUCCESS);
577 * Allocates pageable memory from a sub-map of the kernel. If the submap
578 * has no room, the caller sleeps waiting for more memory in the submap.
580 * This routine may block.
583 kmem_alloc_wait(map, size)
589 size = round_page(size);
590 if (!swap_reserve(size))
595 * To make this work for more than one map, use the map's lock
596 * to lock out sleepers/wakers.
599 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
601 /* no space now; see if we can ever get space */
602 if (vm_map_max(map) - vm_map_min(map) < size) {
607 map->needs_wakeup = TRUE;
608 vm_map_unlock_and_wait(map, 0);
610 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
611 VM_PROT_ALL, MAP_ACC_CHARGED);
619 * Returns memory to a submap of the kernel, and wakes up any processes
620 * waiting for memory in that map.
623 kmem_free_wakeup(map, addr, size)
630 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
631 if (map->needs_wakeup) {
632 map->needs_wakeup = FALSE;
639 kmem_init_zero_region(void)
646 * Map a single physical page of zeros to a larger virtual range.
647 * This requires less looping in places that want large amounts of
648 * zeros, while not using much more physical resources.
650 addr = kmem_alloc_nofault(kernel_map, ZERO_REGION_SIZE);
651 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
652 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
653 if ((m->flags & PG_ZERO) == 0)
655 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
656 pmap_qenter(addr + i, &m, 1);
657 error = vm_map_protect(kernel_map, addr, addr + ZERO_REGION_SIZE,
659 KASSERT(error == 0, ("error=%d", error));
661 zero_region = (const void *)addr;
667 * Create the kernel map; insert a mapping covering kernel text,
668 * data, bss, and all space allocated thus far (`boostrap' data). The
669 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
670 * `start' as allocated, and the range between `start' and `end' as free.
673 kmem_init(start, end)
674 vm_offset_t start, end;
678 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
681 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
683 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
687 VM_MIN_KERNEL_ADDRESS,
689 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
690 /* ... and ending with the completion of the above `insert' */
693 kmem_init_zero_region();
698 * Allow userspace to directly trigger the VM drain routine for testing
702 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
707 error = sysctl_handle_int(oidp, &i, 0, req);
711 EVENTHANDLER_INVOKE(vm_lowmem, 0);
715 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
716 debug_vm_lowmem, "I", "set to trigger vm_lowmem event");