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);
100 * Allocate a virtual address range with no underlying object and
101 * no initial mapping to physical memory. Any mapping from this
102 * range to physical memory must be explicitly created prior to
103 * its use, typically with pmap_qenter(). Any attempt to create
104 * a mapping on demand through vm_fault() will result in a panic.
107 kmem_alloc_nofault(map, size)
114 size = round_page(size);
115 addr = vm_map_min(map);
116 result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
117 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
118 if (result != KERN_SUCCESS) {
125 * kmem_alloc_nofault_space:
127 * Allocate a virtual address range with no underlying object and
128 * no initial mapping to physical memory within the specified
129 * address space. Any mapping from this range to physical memory
130 * must be explicitly created prior to its use, typically with
131 * pmap_qenter(). Any attempt to create a mapping on demand
132 * through vm_fault() will result in a panic.
135 kmem_alloc_nofault_space(map, size, find_space)
143 size = round_page(size);
144 addr = vm_map_min(map);
145 result = vm_map_find(map, NULL, 0, &addr, size, find_space,
146 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
147 if (result != KERN_SUCCESS) {
154 * Allocate wired-down memory in the kernel's address map
158 kmem_alloc(map, size)
166 size = round_page(size);
169 * Use the kernel object for wired-down kernel pages. Assume that no
170 * region of the kernel object is referenced more than once.
174 * Locate sufficient space in the map. This will give us the final
175 * virtual address for the new memory, and thus will tell us the
176 * offset within the kernel map.
179 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
183 offset = addr - VM_MIN_KERNEL_ADDRESS;
184 vm_object_reference(kernel_object);
185 vm_map_insert(map, kernel_object, offset, addr, addr + size,
186 VM_PROT_ALL, VM_PROT_ALL, 0);
190 * Guarantee that there are pages already in this object before
191 * calling vm_map_wire. This is to prevent the following
194 * 1) Threads have swapped out, so that there is a pager for the
195 * kernel_object. 2) The kmsg zone is empty, and so we are
196 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
197 * there is no page, but there is a pager, so we call
198 * pager_data_request. But the kmsg zone is empty, so we must
199 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
200 * we get the data back from the pager, it will be (very stale)
201 * non-zero data. kmem_alloc is defined to return zero-filled memory.
203 * We're intentionally not activating the pages we allocate to prevent a
204 * race with page-out. vm_map_wire will wire the pages.
206 VM_OBJECT_LOCK(kernel_object);
207 for (i = 0; i < size; i += PAGE_SIZE) {
210 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
211 VM_ALLOC_NOBUSY | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
212 mem->valid = VM_PAGE_BITS_ALL;
213 KASSERT((mem->oflags & VPO_UNMANAGED) != 0,
214 ("kmem_alloc: page %p is managed", mem));
216 VM_OBJECT_UNLOCK(kernel_object);
219 * And finally, mark the data as non-pageable.
221 (void) vm_map_wire(map, addr, addr + size,
222 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
230 * Release a region of kernel virtual memory allocated
231 * with kmem_alloc, and return the physical pages
232 * associated with that region.
234 * This routine may not block on kernel maps.
237 kmem_free(map, addr, size)
243 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
249 * Allocates a map to manage a subrange
250 * of the kernel virtual address space.
252 * Arguments are as follows:
254 * parent Map to take range from
255 * min, max Returned endpoints of map
256 * size Size of range to find
257 * superpage_align Request that min is superpage aligned
260 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
261 vm_size_t size, boolean_t superpage_align)
266 size = round_page(size);
268 *min = vm_map_min(parent);
269 ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
270 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
272 if (ret != KERN_SUCCESS)
273 panic("kmem_suballoc: bad status return of %d", ret);
275 result = vm_map_create(vm_map_pmap(parent), *min, *max);
277 panic("kmem_suballoc: cannot create submap");
278 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
279 panic("kmem_suballoc: unable to change range to submap");
286 * Allocate wired-down memory in the kernel's address map for the higher
287 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
288 * kmem_alloc() because we may need to allocate memory at interrupt
289 * level where we cannot block (canwait == FALSE).
291 * This routine has its own private kernel submap (kmem_map) and object
292 * (kmem_object). This, combined with the fact that only malloc uses
293 * this routine, ensures that we will never block in map or object waits.
295 * We don't worry about expanding the map (adding entries) since entries
296 * for wired maps are statically allocated.
298 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
299 * which we never free.
302 kmem_malloc(map, size, flags)
310 size = round_page(size);
311 addr = vm_map_min(map);
314 * Locate sufficient space in the map. This will give us the final
315 * virtual address for the new memory, and thus will tell us the
316 * offset within the kernel map.
319 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
321 if ((flags & M_NOWAIT) == 0) {
322 for (i = 0; i < 8; i++) {
323 EVENTHANDLER_INVOKE(vm_lowmem, 0);
326 if (vm_map_findspace(map, vm_map_min(map),
331 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
334 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
335 (long)size, (long)map->size);
342 rv = kmem_back(map, addr, size, flags);
344 return (rv == KERN_SUCCESS ? addr : 0);
350 * Allocate physical pages for the specified virtual address range.
353 kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
355 vm_offset_t offset, i;
356 vm_map_entry_t entry;
361 KASSERT(vm_map_locked(map), ("kmem_back: map %p is not locked", map));
362 offset = addr - VM_MIN_KERNEL_ADDRESS;
363 vm_object_reference(kmem_object);
364 vm_map_insert(map, kmem_object, offset, addr, addr + size,
365 VM_PROT_ALL, VM_PROT_ALL, 0);
368 * Assert: vm_map_insert() will never be able to extend the
369 * previous entry so vm_map_lookup_entry() will find a new
370 * entry exactly corresponding to this address range and it
371 * will have wired_count == 0.
373 found = vm_map_lookup_entry(map, addr, &entry);
374 KASSERT(found && entry->start == addr && entry->end == addr + size &&
375 entry->wired_count == 0 && (entry->eflags & MAP_ENTRY_IN_TRANSITION)
376 == 0, ("kmem_back: entry not found or misaligned"));
378 if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
379 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
381 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
384 pflags |= VM_ALLOC_ZERO;
386 VM_OBJECT_LOCK(kmem_object);
387 for (i = 0; i < size; i += PAGE_SIZE) {
389 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
392 * Ran out of space, free everything up and return. Don't need
393 * to lock page queues here as we know that the pages we got
394 * aren't on any queues.
397 if ((flags & M_NOWAIT) == 0) {
398 VM_OBJECT_UNLOCK(kmem_object);
399 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
404 (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_NEEDS_WAKEUP)) ==
405 MAP_ENTRY_IN_TRANSITION,
406 ("kmem_back: volatile entry"));
407 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
408 VM_OBJECT_LOCK(kmem_object);
412 * Free the pages before removing the map entry.
413 * They are already marked busy. Calling
414 * vm_map_delete before the pages has been freed or
415 * unbusied will cause a deadlock.
419 m = vm_page_lookup(kmem_object,
420 OFF_TO_IDX(offset + i));
421 vm_page_unwire(m, 0);
424 VM_OBJECT_UNLOCK(kmem_object);
425 vm_map_delete(map, addr, addr + size);
426 return (KERN_NO_SPACE);
428 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
430 m->valid = VM_PAGE_BITS_ALL;
431 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
432 ("kmem_malloc: page %p is managed", m));
434 VM_OBJECT_UNLOCK(kmem_object);
437 * Mark map entry as non-pageable. Repeat the assert.
439 KASSERT(entry->start == addr && entry->end == addr + size &&
440 entry->wired_count == 0,
441 ("kmem_back: entry not found or misaligned after allocation"));
442 entry->wired_count = 1;
445 * At this point, the kmem_object must be unlocked because
446 * vm_map_simplify_entry() calls vm_object_deallocate(), which
447 * locks the kmem_object.
449 vm_map_simplify_entry(map, entry);
452 * Loop thru pages, entering them in the pmap.
454 VM_OBJECT_LOCK(kmem_object);
455 for (i = 0; i < size; i += PAGE_SIZE) {
456 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
458 * Because this is kernel_pmap, this call will not block.
460 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
464 VM_OBJECT_UNLOCK(kmem_object);
466 return (KERN_SUCCESS);
472 * Allocates pageable memory from a sub-map of the kernel. If the submap
473 * has no room, the caller sleeps waiting for more memory in the submap.
475 * This routine may block.
478 kmem_alloc_wait(map, size)
484 size = round_page(size);
485 if (!swap_reserve(size))
490 * To make this work for more than one map, use the map's lock
491 * to lock out sleepers/wakers.
494 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
496 /* no space now; see if we can ever get space */
497 if (vm_map_max(map) - vm_map_min(map) < size) {
502 map->needs_wakeup = TRUE;
503 vm_map_unlock_and_wait(map, 0);
505 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
506 VM_PROT_ALL, MAP_ACC_CHARGED);
514 * Returns memory to a submap of the kernel, and wakes up any processes
515 * waiting for memory in that map.
518 kmem_free_wakeup(map, addr, size)
525 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
526 if (map->needs_wakeup) {
527 map->needs_wakeup = FALSE;
534 kmem_init_zero_region(void)
541 * Map a single physical page of zeros to a larger virtual range.
542 * This requires less looping in places that want large amounts of
543 * zeros, while not using much more physical resources.
545 addr = kmem_alloc_nofault(kernel_map, ZERO_REGION_SIZE);
546 m = vm_page_alloc(NULL, OFF_TO_IDX(addr - VM_MIN_KERNEL_ADDRESS),
547 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
548 if ((m->flags & PG_ZERO) == 0)
550 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
551 pmap_qenter(addr + i, &m, 1);
552 error = vm_map_protect(kernel_map, addr, addr + ZERO_REGION_SIZE,
554 KASSERT(error == 0, ("error=%d", error));
556 zero_region = (const void *)addr;
562 * Create the kernel map; insert a mapping covering kernel text,
563 * data, bss, and all space allocated thus far (`boostrap' data). The
564 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
565 * `start' as allocated, and the range between `start' and `end' as free.
568 kmem_init(start, end)
569 vm_offset_t start, end;
573 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
576 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
578 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
582 VM_MIN_KERNEL_ADDRESS,
584 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
585 /* ... and ending with the completion of the above `insert' */
588 kmem_init_zero_region();
593 * Allow userspace to directly trigger the VM drain routine for testing
597 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
602 error = sysctl_handle_int(oidp, &i, 0, req);
606 EVENTHANDLER_INVOKE(vm_lowmem, 0);
610 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
611 debug_vm_lowmem, "I", "set to trigger vm_lowmem event");