vm_kern.c revision 252330
1/*-
2 * Copyright (c) 1991, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
19 *
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
30 * SUCH DAMAGE.
31 *
32 *	from: @(#)vm_kern.c	8.3 (Berkeley) 1/12/94
33 *
34 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
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.
45 *
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.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
53 *  School of Computer Science
54 *  Carnegie Mellon University
55 *  Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61/*
62 *	Kernel memory management.
63 */
64
65#include <sys/cdefs.h>
66__FBSDID("$FreeBSD: head/sys/vm/vm_kern.c 252330 2013-06-28 03:51:20Z jeff $");
67
68#include <sys/param.h>
69#include <sys/systm.h>
70#include <sys/kernel.h>		/* for ticks and hz */
71#include <sys/eventhandler.h>
72#include <sys/lock.h>
73#include <sys/proc.h>
74#include <sys/malloc.h>
75#include <sys/rwlock.h>
76#include <sys/sysctl.h>
77
78#include <vm/vm.h>
79#include <vm/vm_param.h>
80#include <vm/pmap.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>
86#include <vm/uma.h>
87
88vm_map_t kernel_map;
89vm_map_t kmem_map;
90vm_map_t exec_map;
91vm_map_t pipe_map;
92
93const void *zero_region;
94CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
95
96SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
97    NULL, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
98
99SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
100#if defined(__arm__) || defined(__sparc64__)
101    &vm_max_kernel_address, 0,
102#else
103    NULL, VM_MAX_KERNEL_ADDRESS,
104#endif
105    "Max kernel address");
106
107/*
108 *	kmem_alloc_nofault:
109 *
110 *	Allocate a virtual address range with no underlying object and
111 *	no initial mapping to physical memory.  Any mapping from this
112 *	range to physical memory must be explicitly created prior to
113 *	its use, typically with pmap_qenter().  Any attempt to create
114 *	a mapping on demand through vm_fault() will result in a panic.
115 */
116vm_offset_t
117kmem_alloc_nofault(map, size)
118	vm_map_t map;
119	vm_size_t size;
120{
121	vm_offset_t addr;
122	int result;
123
124	size = round_page(size);
125	addr = vm_map_min(map);
126	result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
127	    VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
128	if (result != KERN_SUCCESS) {
129		return (0);
130	}
131	return (addr);
132}
133
134/*
135 *	kmem_alloc_nofault_space:
136 *
137 *	Allocate a virtual address range with no underlying object and
138 *	no initial mapping to physical memory within the specified
139 *	address space.  Any mapping from this range to physical memory
140 *	must be explicitly created prior to its use, typically with
141 *	pmap_qenter().  Any attempt to create a mapping on demand
142 *	through vm_fault() will result in a panic.
143 */
144vm_offset_t
145kmem_alloc_nofault_space(map, size, find_space)
146	vm_map_t map;
147	vm_size_t size;
148	int find_space;
149{
150	vm_offset_t addr;
151	int result;
152
153	size = round_page(size);
154	addr = vm_map_min(map);
155	result = vm_map_find(map, NULL, 0, &addr, size, find_space,
156	    VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
157	if (result != KERN_SUCCESS) {
158		return (0);
159	}
160	return (addr);
161}
162
163/*
164 *	Allocate wired-down memory in the kernel's address map
165 *	or a submap.
166 */
167vm_offset_t
168kmem_alloc(map, size)
169	vm_map_t map;
170	vm_size_t size;
171{
172	vm_offset_t addr;
173	vm_offset_t offset;
174
175	size = round_page(size);
176
177	/*
178	 * Use the kernel object for wired-down kernel pages. Assume that no
179	 * region of the kernel object is referenced more than once.
180	 */
181
182	/*
183	 * Locate sufficient space in the map.  This will give us the final
184	 * virtual address for the new memory, and thus will tell us the
185	 * offset within the kernel map.
186	 */
187	vm_map_lock(map);
188	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
189		vm_map_unlock(map);
190		return (0);
191	}
192	offset = addr - VM_MIN_KERNEL_ADDRESS;
193	vm_object_reference(kernel_object);
194	vm_map_insert(map, kernel_object, offset, addr, addr + size,
195		VM_PROT_ALL, VM_PROT_ALL, 0);
196	vm_map_unlock(map);
197
198	/*
199	 * And finally, mark the data as non-pageable.
200	 */
201	(void) vm_map_wire(map, addr, addr + size,
202	    VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
203
204	return (addr);
205}
206
207/*
208 *	Allocates a region from the kernel address map and physical pages
209 *	within the specified address range to the kernel object.  Creates a
210 *	wired mapping from this region to these pages, and returns the
211 *	region's starting virtual address.  The allocated pages are not
212 *	necessarily physically contiguous.  If M_ZERO is specified through the
213 *	given flags, then the pages are zeroed before they are mapped.
214 */
215vm_offset_t
216kmem_alloc_attr(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
217    vm_paddr_t high, vm_memattr_t memattr)
218{
219	vm_object_t object = kernel_object;
220	vm_offset_t addr;
221	vm_ooffset_t end_offset, offset;
222	vm_page_t m;
223	int pflags, tries;
224
225	size = round_page(size);
226	vm_map_lock(map);
227	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
228		vm_map_unlock(map);
229		return (0);
230	}
231	offset = addr - VM_MIN_KERNEL_ADDRESS;
232	vm_object_reference(object);
233	vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
234	    VM_PROT_ALL, 0);
235	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
236	VM_OBJECT_WLOCK(object);
237	end_offset = offset + size;
238	for (; offset < end_offset; offset += PAGE_SIZE) {
239		tries = 0;
240retry:
241		m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags, 1,
242		    low, high, PAGE_SIZE, 0, memattr);
243		if (m == NULL) {
244			VM_OBJECT_WUNLOCK(object);
245			if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
246				vm_map_unlock(map);
247				vm_pageout_grow_cache(tries, low, high);
248				vm_map_lock(map);
249				VM_OBJECT_WLOCK(object);
250				tries++;
251				goto retry;
252			}
253
254			/*
255			 * Since the pages that were allocated by any previous
256			 * iterations of this loop are not busy, they can be
257			 * freed by vm_object_page_remove(), which is called
258			 * by vm_map_delete().
259			 */
260			vm_map_delete(map, addr, addr + size);
261			vm_map_unlock(map);
262			return (0);
263		}
264		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
265			pmap_zero_page(m);
266		m->valid = VM_PAGE_BITS_ALL;
267	}
268	VM_OBJECT_WUNLOCK(object);
269	vm_map_unlock(map);
270	vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
271	    VM_MAP_WIRE_NOHOLES);
272	return (addr);
273}
274
275/*
276 *	Allocates a region from the kernel address map and physically
277 *	contiguous pages within the specified address range to the kernel
278 *	object.  Creates a wired mapping from this region to these pages, and
279 *	returns the region's starting virtual address.  If M_ZERO is specified
280 *	through the given flags, then the pages are zeroed before they are
281 *	mapped.
282 */
283vm_offset_t
284kmem_alloc_contig(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
285    vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
286    vm_memattr_t memattr)
287{
288	vm_object_t object = kernel_object;
289	vm_offset_t addr;
290	vm_ooffset_t offset;
291	vm_page_t end_m, m;
292	int pflags, tries;
293
294	size = round_page(size);
295	vm_map_lock(map);
296	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
297		vm_map_unlock(map);
298		return (0);
299	}
300	offset = addr - VM_MIN_KERNEL_ADDRESS;
301	vm_object_reference(object);
302	vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
303	    VM_PROT_ALL, 0);
304	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
305	VM_OBJECT_WLOCK(object);
306	tries = 0;
307retry:
308	m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
309	    atop(size), low, high, alignment, boundary, memattr);
310	if (m == NULL) {
311		VM_OBJECT_WUNLOCK(object);
312		if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
313			vm_map_unlock(map);
314			vm_pageout_grow_cache(tries, low, high);
315			vm_map_lock(map);
316			VM_OBJECT_WLOCK(object);
317			tries++;
318			goto retry;
319		}
320		vm_map_delete(map, addr, addr + size);
321		vm_map_unlock(map);
322		return (0);
323	}
324	end_m = m + atop(size);
325	for (; m < end_m; m++) {
326		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
327			pmap_zero_page(m);
328		m->valid = VM_PAGE_BITS_ALL;
329	}
330	VM_OBJECT_WUNLOCK(object);
331	vm_map_unlock(map);
332	vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
333	    VM_MAP_WIRE_NOHOLES);
334	return (addr);
335}
336
337/*
338 *	kmem_free:
339 *
340 *	Release a region of kernel virtual memory allocated
341 *	with kmem_alloc, and return the physical pages
342 *	associated with that region.
343 *
344 *	This routine may not block on kernel maps.
345 */
346void
347kmem_free(map, addr, size)
348	vm_map_t map;
349	vm_offset_t addr;
350	vm_size_t size;
351{
352
353	(void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
354}
355
356/*
357 *	kmem_suballoc:
358 *
359 *	Allocates a map to manage a subrange
360 *	of the kernel virtual address space.
361 *
362 *	Arguments are as follows:
363 *
364 *	parent		Map to take range from
365 *	min, max	Returned endpoints of map
366 *	size		Size of range to find
367 *	superpage_align	Request that min is superpage aligned
368 */
369vm_map_t
370kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
371    vm_size_t size, boolean_t superpage_align)
372{
373	int ret;
374	vm_map_t result;
375
376	size = round_page(size);
377
378	*min = vm_map_min(parent);
379	ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
380	    VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
381	    MAP_ACC_NO_CHARGE);
382	if (ret != KERN_SUCCESS)
383		panic("kmem_suballoc: bad status return of %d", ret);
384	*max = *min + size;
385	result = vm_map_create(vm_map_pmap(parent), *min, *max);
386	if (result == NULL)
387		panic("kmem_suballoc: cannot create submap");
388	if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
389		panic("kmem_suballoc: unable to change range to submap");
390	return (result);
391}
392
393/*
394 *	kmem_malloc:
395 *
396 * 	Allocate wired-down memory in the kernel's address map for the higher
397 * 	level kernel memory allocator (kern/kern_malloc.c).  We cannot use
398 * 	kmem_alloc() because we may need to allocate memory at interrupt
399 * 	level where we cannot block (canwait == FALSE).
400 *
401 * 	This routine has its own private kernel submap (kmem_map) and object
402 * 	(kmem_object).  This, combined with the fact that only malloc uses
403 * 	this routine, ensures that we will never block in map or object waits.
404 *
405 * 	We don't worry about expanding the map (adding entries) since entries
406 * 	for wired maps are statically allocated.
407 *
408 *	`map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
409 *	which we never free.
410 */
411vm_offset_t
412kmem_malloc(map, size, flags)
413	vm_map_t map;
414	vm_size_t size;
415	int flags;
416{
417	vm_offset_t addr;
418	int i, rv;
419
420	size = round_page(size);
421	addr = vm_map_min(map);
422
423	/*
424	 * Locate sufficient space in the map.  This will give us the final
425	 * virtual address for the new memory, and thus will tell us the
426	 * offset within the kernel map.
427	 */
428	vm_map_lock(map);
429	if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
430		vm_map_unlock(map);
431                if ((flags & M_NOWAIT) == 0) {
432			for (i = 0; i < 8; i++) {
433				EVENTHANDLER_INVOKE(vm_lowmem, 0);
434				uma_reclaim();
435				vm_map_lock(map);
436				if (vm_map_findspace(map, vm_map_min(map),
437				    size, &addr) == 0) {
438					break;
439				}
440				vm_map_unlock(map);
441				tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
442			}
443			if (i == 8) {
444				panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
445				    (long)size, (long)map->size);
446			}
447		} else {
448			return (0);
449		}
450	}
451
452	rv = kmem_back(map, addr, size, flags);
453	vm_map_unlock(map);
454	return (rv == KERN_SUCCESS ? addr : 0);
455}
456
457/*
458 *	kmem_back:
459 *
460 *	Allocate physical pages for the specified virtual address range.
461 */
462int
463kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
464{
465	vm_offset_t offset, i;
466	vm_map_entry_t entry;
467	vm_page_t m;
468	int pflags;
469	boolean_t found;
470
471	KASSERT(vm_map_locked(map), ("kmem_back: map %p is not locked", map));
472	offset = addr - VM_MIN_KERNEL_ADDRESS;
473	vm_object_reference(kmem_object);
474	vm_map_insert(map, kmem_object, offset, addr, addr + size,
475	    VM_PROT_ALL, VM_PROT_ALL, 0);
476
477	/*
478	 * Assert: vm_map_insert() will never be able to extend the
479	 * previous entry so vm_map_lookup_entry() will find a new
480	 * entry exactly corresponding to this address range and it
481	 * will have wired_count == 0.
482	 */
483	found = vm_map_lookup_entry(map, addr, &entry);
484	KASSERT(found && entry->start == addr && entry->end == addr + size &&
485	    entry->wired_count == 0 && (entry->eflags & MAP_ENTRY_IN_TRANSITION)
486	    == 0, ("kmem_back: entry not found or misaligned"));
487
488	pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
489
490	VM_OBJECT_WLOCK(kmem_object);
491	for (i = 0; i < size; i += PAGE_SIZE) {
492retry:
493		m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
494
495		/*
496		 * Ran out of space, free everything up and return. Don't need
497		 * to lock page queues here as we know that the pages we got
498		 * aren't on any queues.
499		 */
500		if (m == NULL) {
501			if ((flags & M_NOWAIT) == 0) {
502				VM_OBJECT_WUNLOCK(kmem_object);
503				entry->eflags |= MAP_ENTRY_IN_TRANSITION;
504				vm_map_unlock(map);
505				VM_WAIT;
506				vm_map_lock(map);
507				KASSERT(
508(entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_NEEDS_WAKEUP)) ==
509				    MAP_ENTRY_IN_TRANSITION,
510				    ("kmem_back: volatile entry"));
511				entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
512				VM_OBJECT_WLOCK(kmem_object);
513				goto retry;
514			}
515			/*
516			 * Free the pages before removing the map entry.
517			 * They are already marked busy.  Calling
518			 * vm_map_delete before the pages has been freed or
519			 * unbusied will cause a deadlock.
520			 */
521			while (i != 0) {
522				i -= PAGE_SIZE;
523				m = vm_page_lookup(kmem_object,
524						   OFF_TO_IDX(offset + i));
525				vm_page_unwire(m, 0);
526				vm_page_free(m);
527			}
528			VM_OBJECT_WUNLOCK(kmem_object);
529			vm_map_delete(map, addr, addr + size);
530			return (KERN_NO_SPACE);
531		}
532		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
533			pmap_zero_page(m);
534		m->valid = VM_PAGE_BITS_ALL;
535		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
536		    ("kmem_malloc: page %p is managed", m));
537	}
538	VM_OBJECT_WUNLOCK(kmem_object);
539
540	/*
541	 * Mark map entry as non-pageable.  Repeat the assert.
542	 */
543	KASSERT(entry->start == addr && entry->end == addr + size &&
544	    entry->wired_count == 0,
545	    ("kmem_back: entry not found or misaligned after allocation"));
546	entry->wired_count = 1;
547
548	/*
549	 * At this point, the kmem_object must be unlocked because
550	 * vm_map_simplify_entry() calls vm_object_deallocate(), which
551	 * locks the kmem_object.
552	 */
553	vm_map_simplify_entry(map, entry);
554
555	/*
556	 * Loop thru pages, entering them in the pmap.
557	 */
558	VM_OBJECT_WLOCK(kmem_object);
559	for (i = 0; i < size; i += PAGE_SIZE) {
560		m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
561		/*
562		 * Because this is kernel_pmap, this call will not block.
563		 */
564		pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
565		    TRUE);
566		vm_page_wakeup(m);
567	}
568	VM_OBJECT_WUNLOCK(kmem_object);
569
570	return (KERN_SUCCESS);
571}
572
573/*
574 *	kmem_alloc_wait:
575 *
576 *	Allocates pageable memory from a sub-map of the kernel.  If the submap
577 *	has no room, the caller sleeps waiting for more memory in the submap.
578 *
579 *	This routine may block.
580 */
581vm_offset_t
582kmem_alloc_wait(map, size)
583	vm_map_t map;
584	vm_size_t size;
585{
586	vm_offset_t addr;
587
588	size = round_page(size);
589	if (!swap_reserve(size))
590		return (0);
591
592	for (;;) {
593		/*
594		 * To make this work for more than one map, use the map's lock
595		 * to lock out sleepers/wakers.
596		 */
597		vm_map_lock(map);
598		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
599			break;
600		/* no space now; see if we can ever get space */
601		if (vm_map_max(map) - vm_map_min(map) < size) {
602			vm_map_unlock(map);
603			swap_release(size);
604			return (0);
605		}
606		map->needs_wakeup = TRUE;
607		vm_map_unlock_and_wait(map, 0);
608	}
609	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
610	    VM_PROT_ALL, MAP_ACC_CHARGED);
611	vm_map_unlock(map);
612	return (addr);
613}
614
615/*
616 *	kmem_free_wakeup:
617 *
618 *	Returns memory to a submap of the kernel, and wakes up any processes
619 *	waiting for memory in that map.
620 */
621void
622kmem_free_wakeup(map, addr, size)
623	vm_map_t map;
624	vm_offset_t addr;
625	vm_size_t size;
626{
627
628	vm_map_lock(map);
629	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
630	if (map->needs_wakeup) {
631		map->needs_wakeup = FALSE;
632		vm_map_wakeup(map);
633	}
634	vm_map_unlock(map);
635}
636
637static void
638kmem_init_zero_region(void)
639{
640	vm_offset_t addr, i;
641	vm_page_t m;
642	int error;
643
644	/*
645	 * Map a single physical page of zeros to a larger virtual range.
646	 * This requires less looping in places that want large amounts of
647	 * zeros, while not using much more physical resources.
648	 */
649	addr = kmem_alloc_nofault(kernel_map, ZERO_REGION_SIZE);
650	m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
651	    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
652	if ((m->flags & PG_ZERO) == 0)
653		pmap_zero_page(m);
654	for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
655		pmap_qenter(addr + i, &m, 1);
656	error = vm_map_protect(kernel_map, addr, addr + ZERO_REGION_SIZE,
657	    VM_PROT_READ, TRUE);
658	KASSERT(error == 0, ("error=%d", error));
659
660	zero_region = (const void *)addr;
661}
662
663/*
664 * 	kmem_init:
665 *
666 *	Create the kernel map; insert a mapping covering kernel text,
667 *	data, bss, and all space allocated thus far (`boostrap' data).  The
668 *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
669 *	`start' as allocated, and the range between `start' and `end' as free.
670 */
671void
672kmem_init(start, end)
673	vm_offset_t start, end;
674{
675	vm_map_t m;
676
677	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
678	m->system_map = 1;
679	vm_map_lock(m);
680	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
681	kernel_map = m;
682	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
683#ifdef __amd64__
684	    KERNBASE,
685#else
686	    VM_MIN_KERNEL_ADDRESS,
687#endif
688	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
689	/* ... and ending with the completion of the above `insert' */
690	vm_map_unlock(m);
691
692	kmem_init_zero_region();
693}
694
695#ifdef DIAGNOSTIC
696/*
697 * Allow userspace to directly trigger the VM drain routine for testing
698 * purposes.
699 */
700static int
701debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
702{
703	int error, i;
704
705	i = 0;
706	error = sysctl_handle_int(oidp, &i, 0, req);
707	if (error)
708		return (error);
709	if (i)
710		EVENTHANDLER_INVOKE(vm_lowmem, 0);
711	return (0);
712}
713
714SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
715    debug_vm_lowmem, "I", "set to trigger vm_lowmem event");
716#endif
717