vm_kern.c revision 248277
1169695Skan/*-
2169695Skan * Copyright (c) 1991, 1993
3169695Skan *	The Regents of the University of California.  All rights reserved.
4169695Skan *
5169695Skan * This code is derived from software contributed to Berkeley by
6169695Skan * The Mach Operating System project at Carnegie-Mellon University.
7169695Skan *
8169695Skan * Redistribution and use in source and binary forms, with or without
9169695Skan * modification, are permitted provided that the following conditions
10169695Skan * are met:
11169695Skan * 1. Redistributions of source code must retain the above copyright
12169695Skan *    notice, this list of conditions and the following disclaimer.
13169695Skan * 2. Redistributions in binary form must reproduce the above copyright
14169695Skan *    notice, this list of conditions and the following disclaimer in the
15169695Skan *    documentation and/or other materials provided with the distribution.
16169695Skan * 4. Neither the name of the University nor the names of its contributors
17169695Skan *    may be used to endorse or promote products derived from this software
18169695Skan *    without specific prior written permission.
19169695Skan *
20169695Skan * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21169695Skan * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22169695Skan * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23169695Skan * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24169695Skan * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25169695Skan * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26169695Skan * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27169695Skan * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28169695Skan * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29169695Skan * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30169695Skan * SUCH DAMAGE.
31169695Skan *
32169695Skan *	from: @(#)vm_kern.c	8.3 (Berkeley) 1/12/94
33169695Skan *
34169695Skan *
35169695Skan * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36169695Skan * All rights reserved.
37169695Skan *
38169695Skan * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39169695Skan *
40169695Skan * Permission to use, copy, modify and distribute this software and
41169695Skan * its documentation is hereby granted, provided that both the copyright
42169695Skan * notice and this permission notice appear in all copies of the
43169695Skan * software, derivative works or modified versions, and any portions
44169695Skan * thereof, and that both notices appear in supporting documentation.
45169695Skan *
46169695Skan * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47169695Skan * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48169695Skan * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49169695Skan *
50169695Skan * Carnegie Mellon requests users of this software to return to
51169695Skan *
52169695Skan *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
53169695Skan *  School of Computer Science
54169695Skan *  Carnegie Mellon University
55169695Skan *  Pittsburgh PA 15213-3890
56169695Skan *
57169695Skan * any improvements or extensions that they make and grant Carnegie the
58169695Skan * rights to redistribute these changes.
59169695Skan */
60169695Skan
61169695Skan/*
62169695Skan *	Kernel memory management.
63169695Skan */
64169695Skan
65169695Skan#include <sys/cdefs.h>
66169695Skan__FBSDID("$FreeBSD: head/sys/vm/vm_kern.c 248277 2013-03-14 19:50:09Z kib $");
67169695Skan
68169695Skan#include <sys/param.h>
69169695Skan#include <sys/systm.h>
70169695Skan#include <sys/kernel.h>		/* for ticks and hz */
71169695Skan#include <sys/eventhandler.h>
72169695Skan#include <sys/lock.h>
73169695Skan#include <sys/proc.h>
74169695Skan#include <sys/malloc.h>
75169695Skan#include <sys/rwlock.h>
76169695Skan#include <sys/sysctl.h>
77169695Skan
78169695Skan#include <vm/vm.h>
79169695Skan#include <vm/vm_param.h>
80169695Skan#include <vm/pmap.h>
81169695Skan#include <vm/vm_map.h>
82169695Skan#include <vm/vm_object.h>
83169695Skan#include <vm/vm_page.h>
84169695Skan#include <vm/vm_pageout.h>
85169695Skan#include <vm/vm_extern.h>
86169695Skan#include <vm/uma.h>
87169695Skan
88169695Skanvm_map_t kernel_map;
89169695Skanvm_map_t kmem_map;
90169695Skanvm_map_t exec_map;
91169695Skanvm_map_t pipe_map;
92169695Skanvm_map_t buffer_map;
93169695Skan
94169695Skanconst void *zero_region;
95169695SkanCTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
96169695Skan
97169695SkanSYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
98169695Skan    NULL, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
99169695Skan
100169695SkanSYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
101169695Skan#if defined(__arm__) || defined(__sparc64__)
102169695Skan    &vm_max_kernel_address, 0,
103169695Skan#else
104169695Skan    NULL, VM_MAX_KERNEL_ADDRESS,
105169695Skan#endif
106169695Skan    "Max kernel address");
107169695Skan
108169695Skan/*
109169695Skan *	kmem_alloc_nofault:
110169695Skan *
111169695Skan *	Allocate a virtual address range with no underlying object and
112169695Skan *	no initial mapping to physical memory.  Any mapping from this
113169695Skan *	range to physical memory must be explicitly created prior to
114169695Skan *	its use, typically with pmap_qenter().  Any attempt to create
115169695Skan *	a mapping on demand through vm_fault() will result in a panic.
116169695Skan */
117169695Skanvm_offset_t
118169695Skankmem_alloc_nofault(map, size)
119169695Skan	vm_map_t map;
120169695Skan	vm_size_t size;
121169695Skan{
122169695Skan	vm_offset_t addr;
123169695Skan	int result;
124169695Skan
125169695Skan	size = round_page(size);
126169695Skan	addr = vm_map_min(map);
127169695Skan	result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
128169695Skan	    VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
129169695Skan	if (result != KERN_SUCCESS) {
130169695Skan		return (0);
131169695Skan	}
132169695Skan	return (addr);
133169695Skan}
134169695Skan
135169695Skan/*
136169695Skan *	kmem_alloc_nofault_space:
137169695Skan *
138169695Skan *	Allocate a virtual address range with no underlying object and
139169695Skan *	no initial mapping to physical memory within the specified
140169695Skan *	address space.  Any mapping from this range to physical memory
141169695Skan *	must be explicitly created prior to its use, typically with
142169695Skan *	pmap_qenter().  Any attempt to create a mapping on demand
143169695Skan *	through vm_fault() will result in a panic.
144169695Skan */
145169695Skanvm_offset_t
146169695Skankmem_alloc_nofault_space(map, size, find_space)
147169695Skan	vm_map_t map;
148169695Skan	vm_size_t size;
149169695Skan	int find_space;
150169695Skan{
151169695Skan	vm_offset_t addr;
152169695Skan	int result;
153169695Skan
154169695Skan	size = round_page(size);
155169695Skan	addr = vm_map_min(map);
156169695Skan	result = vm_map_find(map, NULL, 0, &addr, size, find_space,
157169695Skan	    VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
158169695Skan	if (result != KERN_SUCCESS) {
159169695Skan		return (0);
160169695Skan	}
161169695Skan	return (addr);
162169695Skan}
163169695Skan
164169695Skan/*
165169695Skan *	Allocate wired-down memory in the kernel's address map
166169695Skan *	or a submap.
167169695Skan */
168169695Skanvm_offset_t
169169695Skankmem_alloc(map, size)
170169695Skan	vm_map_t map;
171169695Skan	vm_size_t size;
172169695Skan{
173169695Skan	vm_offset_t addr;
174169695Skan	vm_offset_t offset;
175169695Skan
176169695Skan	size = round_page(size);
177169695Skan
178169695Skan	/*
179169695Skan	 * Use the kernel object for wired-down kernel pages. Assume that no
180169695Skan	 * region of the kernel object is referenced more than once.
181169695Skan	 */
182169695Skan
183169695Skan	/*
184169695Skan	 * Locate sufficient space in the map.  This will give us the final
185169695Skan	 * virtual address for the new memory, and thus will tell us the
186169695Skan	 * offset within the kernel map.
187169695Skan	 */
188169695Skan	vm_map_lock(map);
189169695Skan	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
190169695Skan		vm_map_unlock(map);
191169695Skan		return (0);
192169695Skan	}
193169695Skan	offset = addr - VM_MIN_KERNEL_ADDRESS;
194169695Skan	vm_object_reference(kernel_object);
195169695Skan	vm_map_insert(map, kernel_object, offset, addr, addr + size,
196169695Skan		VM_PROT_ALL, VM_PROT_ALL, 0);
197169695Skan	vm_map_unlock(map);
198169695Skan
199169695Skan	/*
200169695Skan	 * And finally, mark the data as non-pageable.
201169695Skan	 */
202169695Skan	(void) vm_map_wire(map, addr, addr + size,
203169695Skan	    VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
204169695Skan
205169695Skan	return (addr);
206169695Skan}
207169695Skan
208169695Skan/*
209169695Skan *	Allocates a region from the kernel address map and physical pages
210169695Skan *	within the specified address range to the kernel object.  Creates a
211169695Skan *	wired mapping from this region to these pages, and returns the
212169695Skan *	region's starting virtual address.  The allocated pages are not
213169695Skan *	necessarily physically contiguous.  If M_ZERO is specified through the
214169695Skan *	given flags, then the pages are zeroed before they are mapped.
215169695Skan */
216169695Skanvm_offset_t
217169695Skankmem_alloc_attr(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
218169695Skan    vm_paddr_t high, vm_memattr_t memattr)
219169695Skan{
220169695Skan	vm_object_t object = kernel_object;
221169695Skan	vm_offset_t addr;
222169695Skan	vm_ooffset_t end_offset, offset;
223169695Skan	vm_page_t m;
224169695Skan	int pflags, tries;
225169695Skan
226169695Skan	size = round_page(size);
227169695Skan	vm_map_lock(map);
228169695Skan	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
229169695Skan		vm_map_unlock(map);
230169695Skan		return (0);
231169695Skan	}
232169695Skan	offset = addr - VM_MIN_KERNEL_ADDRESS;
233169695Skan	vm_object_reference(object);
234169695Skan	vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
235169695Skan	    VM_PROT_ALL, 0);
236169695Skan	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
237169695Skan	VM_OBJECT_WLOCK(object);
238169695Skan	end_offset = offset + size;
239169695Skan	for (; offset < end_offset; offset += PAGE_SIZE) {
240169695Skan		tries = 0;
241169695Skanretry:
242169695Skan		m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags, 1,
243169695Skan		    low, high, PAGE_SIZE, 0, memattr);
244169695Skan		if (m == NULL) {
245169695Skan			VM_OBJECT_WUNLOCK(object);
246169695Skan			if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
247169695Skan				vm_map_unlock(map);
248169695Skan				vm_pageout_grow_cache(tries, low, high);
249169695Skan				vm_map_lock(map);
250169695Skan				VM_OBJECT_WLOCK(object);
251169695Skan				tries++;
252169695Skan				goto retry;
253169695Skan			}
254169695Skan
255169695Skan			/*
256169695Skan			 * Since the pages that were allocated by any previous
257169695Skan			 * iterations of this loop are not busy, they can be
258169695Skan			 * freed by vm_object_page_remove(), which is called
259169695Skan			 * by vm_map_delete().
260169695Skan			 */
261169695Skan			vm_map_delete(map, addr, addr + size);
262169695Skan			vm_map_unlock(map);
263169695Skan			return (0);
264169695Skan		}
265169695Skan		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
266169695Skan			pmap_zero_page(m);
267169695Skan		m->valid = VM_PAGE_BITS_ALL;
268169695Skan	}
269169695Skan	VM_OBJECT_WUNLOCK(object);
270169695Skan	vm_map_unlock(map);
271169695Skan	vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
272169695Skan	    VM_MAP_WIRE_NOHOLES);
273169695Skan	return (addr);
274169695Skan}
275169695Skan
276169695Skan/*
277169695Skan *	Allocates a region from the kernel address map and physically
278169695Skan *	contiguous pages within the specified address range to the kernel
279169695Skan *	object.  Creates a wired mapping from this region to these pages, and
280169695Skan *	returns the region's starting virtual address.  If M_ZERO is specified
281169695Skan *	through the given flags, then the pages are zeroed before they are
282169695Skan *	mapped.
283169695Skan */
284169695Skanvm_offset_t
285169695Skankmem_alloc_contig(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
286169695Skan    vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
287169695Skan    vm_memattr_t memattr)
288169695Skan{
289169695Skan	vm_object_t object = kernel_object;
290169695Skan	vm_offset_t addr;
291169695Skan	vm_ooffset_t offset;
292169695Skan	vm_page_t end_m, m;
293169695Skan	int pflags, tries;
294169695Skan
295169695Skan	size = round_page(size);
296169695Skan	vm_map_lock(map);
297169695Skan	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
298169695Skan		vm_map_unlock(map);
299169695Skan		return (0);
300169695Skan	}
301169695Skan	offset = addr - VM_MIN_KERNEL_ADDRESS;
302169695Skan	vm_object_reference(object);
303169695Skan	vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
304169695Skan	    VM_PROT_ALL, 0);
305169695Skan	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
306169695Skan	VM_OBJECT_WLOCK(object);
307169695Skan	tries = 0;
308169695Skanretry:
309169695Skan	m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
310169695Skan	    atop(size), low, high, alignment, boundary, memattr);
311169695Skan	if (m == NULL) {
312169695Skan		VM_OBJECT_WUNLOCK(object);
313169695Skan		if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
314169695Skan			vm_map_unlock(map);
315169695Skan			vm_pageout_grow_cache(tries, low, high);
316169695Skan			vm_map_lock(map);
317169695Skan			VM_OBJECT_WLOCK(object);
318169695Skan			tries++;
319169695Skan			goto retry;
320169695Skan		}
321169695Skan		vm_map_delete(map, addr, addr + size);
322169695Skan		vm_map_unlock(map);
323169695Skan		return (0);
324169695Skan	}
325	end_m = m + atop(size);
326	for (; m < end_m; m++) {
327		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
328			pmap_zero_page(m);
329		m->valid = VM_PAGE_BITS_ALL;
330	}
331	VM_OBJECT_WUNLOCK(object);
332	vm_map_unlock(map);
333	vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
334	    VM_MAP_WIRE_NOHOLES);
335	return (addr);
336}
337
338/*
339 *	kmem_free:
340 *
341 *	Release a region of kernel virtual memory allocated
342 *	with kmem_alloc, and return the physical pages
343 *	associated with that region.
344 *
345 *	This routine may not block on kernel maps.
346 */
347void
348kmem_free(map, addr, size)
349	vm_map_t map;
350	vm_offset_t addr;
351	vm_size_t size;
352{
353
354	(void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
355}
356
357/*
358 *	kmem_suballoc:
359 *
360 *	Allocates a map to manage a subrange
361 *	of the kernel virtual address space.
362 *
363 *	Arguments are as follows:
364 *
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
369 */
370vm_map_t
371kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
372    vm_size_t size, boolean_t superpage_align)
373{
374	int ret;
375	vm_map_t result;
376
377	size = round_page(size);
378
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,
382	    MAP_ACC_NO_CHARGE);
383	if (ret != KERN_SUCCESS)
384		panic("kmem_suballoc: bad status return of %d", ret);
385	*max = *min + size;
386	result = vm_map_create(vm_map_pmap(parent), *min, *max);
387	if (result == NULL)
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");
391	return (result);
392}
393
394/*
395 *	kmem_malloc:
396 *
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).
401 *
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.
405 *
406 * 	We don't worry about expanding the map (adding entries) since entries
407 * 	for wired maps are statically allocated.
408 *
409 *	`map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
410 *	which we never free.
411 */
412vm_offset_t
413kmem_malloc(map, size, flags)
414	vm_map_t map;
415	vm_size_t size;
416	int flags;
417{
418	vm_offset_t addr;
419	int i, rv;
420
421	size = round_page(size);
422	addr = vm_map_min(map);
423
424	/*
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.
428	 */
429	vm_map_lock(map);
430	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
431		vm_map_unlock(map);
432                if ((flags & M_NOWAIT) == 0) {
433			for (i = 0; i < 8; i++) {
434				EVENTHANDLER_INVOKE(vm_lowmem, 0);
435				uma_reclaim();
436				vm_map_lock(map);
437				if (vm_map_findspace(map, vm_map_min(map),
438				    size, &addr) == 0) {
439					break;
440				}
441				vm_map_unlock(map);
442				tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
443			}
444			if (i == 8) {
445				panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
446				    (long)size, (long)map->size);
447			}
448		} else {
449			return (0);
450		}
451	}
452
453	rv = kmem_back(map, addr, size, flags);
454	vm_map_unlock(map);
455	return (rv == KERN_SUCCESS ? addr : 0);
456}
457
458/*
459 *	kmem_back:
460 *
461 *	Allocate physical pages for the specified virtual address range.
462 */
463int
464kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
465{
466	vm_offset_t offset, i;
467	vm_map_entry_t entry;
468	vm_page_t m;
469	int pflags;
470	boolean_t found;
471
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);
477
478	/*
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.
483	 */
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"));
488
489	pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
490
491	VM_OBJECT_WLOCK(kmem_object);
492	for (i = 0; i < size; i += PAGE_SIZE) {
493retry:
494		m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
495
496		/*
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.
500		 */
501		if (m == NULL) {
502			if ((flags & M_NOWAIT) == 0) {
503				VM_OBJECT_WUNLOCK(kmem_object);
504				entry->eflags |= MAP_ENTRY_IN_TRANSITION;
505				vm_map_unlock(map);
506				VM_WAIT;
507				vm_map_lock(map);
508				KASSERT(
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_WLOCK(kmem_object);
514				goto retry;
515			}
516			/*
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.
521			 */
522			while (i != 0) {
523				i -= PAGE_SIZE;
524				m = vm_page_lookup(kmem_object,
525						   OFF_TO_IDX(offset + i));
526				vm_page_unwire(m, 0);
527				vm_page_free(m);
528			}
529			VM_OBJECT_WUNLOCK(kmem_object);
530			vm_map_delete(map, addr, addr + size);
531			return (KERN_NO_SPACE);
532		}
533		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
534			pmap_zero_page(m);
535		m->valid = VM_PAGE_BITS_ALL;
536		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
537		    ("kmem_malloc: page %p is managed", m));
538	}
539	VM_OBJECT_WUNLOCK(kmem_object);
540
541	/*
542	 * Mark map entry as non-pageable.  Repeat the assert.
543	 */
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;
548
549	/*
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.
553	 */
554	vm_map_simplify_entry(map, entry);
555
556	/*
557	 * Loop thru pages, entering them in the pmap.
558	 */
559	VM_OBJECT_WLOCK(kmem_object);
560	for (i = 0; i < size; i += PAGE_SIZE) {
561		m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
562		/*
563		 * Because this is kernel_pmap, this call will not block.
564		 */
565		pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
566		    TRUE);
567		vm_page_wakeup(m);
568	}
569	VM_OBJECT_WUNLOCK(kmem_object);
570
571	return (KERN_SUCCESS);
572}
573
574/*
575 *	kmem_alloc_wait:
576 *
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.
579 *
580 *	This routine may block.
581 */
582vm_offset_t
583kmem_alloc_wait(map, size)
584	vm_map_t map;
585	vm_size_t size;
586{
587	vm_offset_t addr;
588
589	size = round_page(size);
590	if (!swap_reserve(size))
591		return (0);
592
593	for (;;) {
594		/*
595		 * To make this work for more than one map, use the map's lock
596		 * to lock out sleepers/wakers.
597		 */
598		vm_map_lock(map);
599		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
600			break;
601		/* no space now; see if we can ever get space */
602		if (vm_map_max(map) - vm_map_min(map) < size) {
603			vm_map_unlock(map);
604			swap_release(size);
605			return (0);
606		}
607		map->needs_wakeup = TRUE;
608		vm_map_unlock_and_wait(map, 0);
609	}
610	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
611	    VM_PROT_ALL, MAP_ACC_CHARGED);
612	vm_map_unlock(map);
613	return (addr);
614}
615
616/*
617 *	kmem_free_wakeup:
618 *
619 *	Returns memory to a submap of the kernel, and wakes up any processes
620 *	waiting for memory in that map.
621 */
622void
623kmem_free_wakeup(map, addr, size)
624	vm_map_t map;
625	vm_offset_t addr;
626	vm_size_t size;
627{
628
629	vm_map_lock(map);
630	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
631	if (map->needs_wakeup) {
632		map->needs_wakeup = FALSE;
633		vm_map_wakeup(map);
634	}
635	vm_map_unlock(map);
636}
637
638static void
639kmem_init_zero_region(void)
640{
641	vm_offset_t addr, i;
642	vm_page_t m;
643	int error;
644
645	/*
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.
649	 */
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)
654		pmap_zero_page(m);
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,
658	    VM_PROT_READ, TRUE);
659	KASSERT(error == 0, ("error=%d", error));
660
661	zero_region = (const void *)addr;
662}
663
664/*
665 * 	kmem_init:
666 *
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.
671 */
672void
673kmem_init(start, end)
674	vm_offset_t start, end;
675{
676	vm_map_t m;
677
678	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
679	m->system_map = 1;
680	vm_map_lock(m);
681	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
682	kernel_map = m;
683	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
684#ifdef __amd64__
685	    KERNBASE,
686#else
687	    VM_MIN_KERNEL_ADDRESS,
688#endif
689	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
690	/* ... and ending with the completion of the above `insert' */
691	vm_map_unlock(m);
692
693	kmem_init_zero_region();
694}
695
696#ifdef DIAGNOSTIC
697/*
698 * Allow userspace to directly trigger the VM drain routine for testing
699 * purposes.
700 */
701static int
702debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
703{
704	int error, i;
705
706	i = 0;
707	error = sysctl_handle_int(oidp, &i, 0, req);
708	if (error)
709		return (error);
710	if (i)
711		EVENTHANDLER_INVOKE(vm_lowmem, 0);
712	return (0);
713}
714
715SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
716    debug_vm_lowmem, "I", "set to trigger vm_lowmem event");
717#endif
718