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>
| 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 128613 2004-04-24 20:53:55Z alc $");
| 66__FBSDID("$FreeBSD: head/sys/vm/vm_kern.c 129906 2004-05-31 21:46:06Z bmilekic $");
|
67
68#include <sys/param.h>
69#include <sys/systm.h>
70#include <sys/kernel.h> /* for ticks and hz */
71#include <sys/lock.h>
72#include <sys/mutex.h>
73#include <sys/proc.h>
74#include <sys/malloc.h>
75
76#include <vm/vm.h>
77#include <vm/vm_param.h>
78#include <vm/pmap.h>
79#include <vm/vm_map.h>
80#include <vm/vm_object.h>
81#include <vm/vm_page.h>
82#include <vm/vm_pageout.h>
83#include <vm/vm_extern.h>
84
85vm_map_t kernel_map=0;
86vm_map_t kmem_map=0;
87vm_map_t exec_map=0;
88vm_map_t pipe_map;
89vm_map_t buffer_map=0;
90
91/*
92 * kmem_alloc_pageable:
93 *
94 * Allocate pageable memory to the kernel's address map.
95 * "map" must be kernel_map or a submap of kernel_map.
96 */
97vm_offset_t
98kmem_alloc_pageable(map, size)
99 vm_map_t map;
100 vm_size_t size;
101{
102 vm_offset_t addr;
103 int result;
104
105 size = round_page(size);
106 addr = vm_map_min(map);
107 result = vm_map_find(map, NULL, 0,
108 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
109 if (result != KERN_SUCCESS) {
110 return (0);
111 }
112 return (addr);
113}
114
115/*
116 * kmem_alloc_nofault:
117 *
118 * Allocate a virtual address range with no underlying object and
119 * no initial mapping to physical memory. Any mapping from this
120 * range to physical memory must be explicitly created prior to
121 * its use, typically with pmap_qenter(). Any attempt to create
122 * a mapping on demand through vm_fault() will result in a panic.
123 */
124vm_offset_t
125kmem_alloc_nofault(map, size)
126 vm_map_t map;
127 vm_size_t size;
128{
129 vm_offset_t addr;
130 int result;
131
132 size = round_page(size);
133 addr = vm_map_min(map);
134 result = vm_map_find(map, NULL, 0,
135 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
136 if (result != KERN_SUCCESS) {
137 return (0);
138 }
139 return (addr);
140}
141
142/*
143 * Allocate wired-down memory in the kernel's address map
144 * or a submap.
145 */
146vm_offset_t
147kmem_alloc(map, size)
148 vm_map_t map;
149 vm_size_t size;
150{
151 vm_offset_t addr;
152 vm_offset_t offset;
153 vm_offset_t i;
154
155 size = round_page(size);
156
157 /*
158 * Use the kernel object for wired-down kernel pages. Assume that no
159 * region of the kernel object is referenced more than once.
160 */
161
162 /*
163 * Locate sufficient space in the map. This will give us the final
164 * virtual address for the new memory, and thus will tell us the
165 * offset within the kernel map.
166 */
167 vm_map_lock(map);
168 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
169 vm_map_unlock(map);
170 return (0);
171 }
172 offset = addr - VM_MIN_KERNEL_ADDRESS;
173 vm_object_reference(kernel_object);
174 vm_map_insert(map, kernel_object, offset, addr, addr + size,
175 VM_PROT_ALL, VM_PROT_ALL, 0);
176 vm_map_unlock(map);
177
178 /*
179 * Guarantee that there are pages already in this object before
180 * calling vm_map_wire. This is to prevent the following
181 * scenario:
182 *
183 * 1) Threads have swapped out, so that there is a pager for the
184 * kernel_object. 2) The kmsg zone is empty, and so we are
185 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
186 * there is no page, but there is a pager, so we call
187 * pager_data_request. But the kmsg zone is empty, so we must
188 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
189 * we get the data back from the pager, it will be (very stale)
190 * non-zero data. kmem_alloc is defined to return zero-filled memory.
191 *
192 * We're intentionally not activating the pages we allocate to prevent a
193 * race with page-out. vm_map_wire will wire the pages.
194 */
195 VM_OBJECT_LOCK(kernel_object);
196 for (i = 0; i < size; i += PAGE_SIZE) {
197 vm_page_t mem;
198
199 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
200 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
201 mem->valid = VM_PAGE_BITS_ALL;
202 vm_page_lock_queues();
203 vm_page_unmanage(mem);
204 vm_page_wakeup(mem);
205 vm_page_unlock_queues();
206 }
207 VM_OBJECT_UNLOCK(kernel_object);
208
209 /*
210 * And finally, mark the data as non-pageable.
211 */
212 (void) vm_map_wire(map, addr, addr + size,
213 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
214
215 return (addr);
216}
217
218/*
219 * kmem_free:
220 *
221 * Release a region of kernel virtual memory allocated
222 * with kmem_alloc, and return the physical pages
223 * associated with that region.
224 *
225 * This routine may not block on kernel maps.
226 */
227void
228kmem_free(map, addr, size)
229 vm_map_t map;
230 vm_offset_t addr;
231 vm_size_t size;
232{
233
234 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
235}
236
237/*
238 * kmem_suballoc:
239 *
240 * Allocates a map to manage a subrange
241 * of the kernel virtual address space.
242 *
243 * Arguments are as follows:
244 *
245 * parent Map to take range from
246 * min, max Returned endpoints of map
247 * size Size of range to find
248 */
249vm_map_t
250kmem_suballoc(parent, min, max, size)
251 vm_map_t parent;
252 vm_offset_t *min, *max;
253 vm_size_t size;
254{
255 int ret;
256 vm_map_t result;
257
258 size = round_page(size);
259
260 *min = (vm_offset_t) vm_map_min(parent);
261 ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
262 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
263 if (ret != KERN_SUCCESS) {
264 printf("kmem_suballoc: bad status return of %d.\n", ret);
265 panic("kmem_suballoc");
266 }
267 *max = *min + size;
268 result = vm_map_create(vm_map_pmap(parent), *min, *max);
269 if (result == NULL)
270 panic("kmem_suballoc: cannot create submap");
271 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
272 panic("kmem_suballoc: unable to change range to submap");
273 return (result);
274}
275
276/*
277 * kmem_malloc:
278 *
279 * Allocate wired-down memory in the kernel's address map for the higher
280 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
281 * kmem_alloc() because we may need to allocate memory at interrupt
282 * level where we cannot block (canwait == FALSE).
283 *
284 * This routine has its own private kernel submap (kmem_map) and object
285 * (kmem_object). This, combined with the fact that only malloc uses
286 * this routine, ensures that we will never block in map or object waits.
287 *
288 * Note that this still only works in a uni-processor environment and
289 * when called at splhigh().
290 *
291 * We don't worry about expanding the map (adding entries) since entries
292 * for wired maps are statically allocated.
293 *
294 * NOTE: This routine is not supposed to block if M_NOWAIT is set, but
295 * I have not verified that it actually does not block.
296 *
297 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
298 * which we never free.
299 */
300vm_offset_t
301kmem_malloc(map, size, flags)
302 vm_map_t map;
303 vm_size_t size;
304 int flags;
305{
306 vm_offset_t offset, i;
307 vm_map_entry_t entry;
308 vm_offset_t addr;
309 vm_page_t m;
310 int pflags;
311
312 size = round_page(size);
313 addr = vm_map_min(map);
314
315 /*
316 * Locate sufficient space in the map. This will give us the final
317 * virtual address for the new memory, and thus will tell us the
318 * offset within the kernel map.
319 */
320 vm_map_lock(map);
321 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
322 vm_map_unlock(map);
| 67
68#include <sys/param.h>
69#include <sys/systm.h>
70#include <sys/kernel.h> /* for ticks and hz */
71#include <sys/lock.h>
72#include <sys/mutex.h>
73#include <sys/proc.h>
74#include <sys/malloc.h>
75
76#include <vm/vm.h>
77#include <vm/vm_param.h>
78#include <vm/pmap.h>
79#include <vm/vm_map.h>
80#include <vm/vm_object.h>
81#include <vm/vm_page.h>
82#include <vm/vm_pageout.h>
83#include <vm/vm_extern.h>
84
85vm_map_t kernel_map=0;
86vm_map_t kmem_map=0;
87vm_map_t exec_map=0;
88vm_map_t pipe_map;
89vm_map_t buffer_map=0;
90
91/*
92 * kmem_alloc_pageable:
93 *
94 * Allocate pageable memory to the kernel's address map.
95 * "map" must be kernel_map or a submap of kernel_map.
96 */
97vm_offset_t
98kmem_alloc_pageable(map, size)
99 vm_map_t map;
100 vm_size_t size;
101{
102 vm_offset_t addr;
103 int result;
104
105 size = round_page(size);
106 addr = vm_map_min(map);
107 result = vm_map_find(map, NULL, 0,
108 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
109 if (result != KERN_SUCCESS) {
110 return (0);
111 }
112 return (addr);
113}
114
115/*
116 * kmem_alloc_nofault:
117 *
118 * Allocate a virtual address range with no underlying object and
119 * no initial mapping to physical memory. Any mapping from this
120 * range to physical memory must be explicitly created prior to
121 * its use, typically with pmap_qenter(). Any attempt to create
122 * a mapping on demand through vm_fault() will result in a panic.
123 */
124vm_offset_t
125kmem_alloc_nofault(map, size)
126 vm_map_t map;
127 vm_size_t size;
128{
129 vm_offset_t addr;
130 int result;
131
132 size = round_page(size);
133 addr = vm_map_min(map);
134 result = vm_map_find(map, NULL, 0,
135 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
136 if (result != KERN_SUCCESS) {
137 return (0);
138 }
139 return (addr);
140}
141
142/*
143 * Allocate wired-down memory in the kernel's address map
144 * or a submap.
145 */
146vm_offset_t
147kmem_alloc(map, size)
148 vm_map_t map;
149 vm_size_t size;
150{
151 vm_offset_t addr;
152 vm_offset_t offset;
153 vm_offset_t i;
154
155 size = round_page(size);
156
157 /*
158 * Use the kernel object for wired-down kernel pages. Assume that no
159 * region of the kernel object is referenced more than once.
160 */
161
162 /*
163 * Locate sufficient space in the map. This will give us the final
164 * virtual address for the new memory, and thus will tell us the
165 * offset within the kernel map.
166 */
167 vm_map_lock(map);
168 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
169 vm_map_unlock(map);
170 return (0);
171 }
172 offset = addr - VM_MIN_KERNEL_ADDRESS;
173 vm_object_reference(kernel_object);
174 vm_map_insert(map, kernel_object, offset, addr, addr + size,
175 VM_PROT_ALL, VM_PROT_ALL, 0);
176 vm_map_unlock(map);
177
178 /*
179 * Guarantee that there are pages already in this object before
180 * calling vm_map_wire. This is to prevent the following
181 * scenario:
182 *
183 * 1) Threads have swapped out, so that there is a pager for the
184 * kernel_object. 2) The kmsg zone is empty, and so we are
185 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
186 * there is no page, but there is a pager, so we call
187 * pager_data_request. But the kmsg zone is empty, so we must
188 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
189 * we get the data back from the pager, it will be (very stale)
190 * non-zero data. kmem_alloc is defined to return zero-filled memory.
191 *
192 * We're intentionally not activating the pages we allocate to prevent a
193 * race with page-out. vm_map_wire will wire the pages.
194 */
195 VM_OBJECT_LOCK(kernel_object);
196 for (i = 0; i < size; i += PAGE_SIZE) {
197 vm_page_t mem;
198
199 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
200 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
201 mem->valid = VM_PAGE_BITS_ALL;
202 vm_page_lock_queues();
203 vm_page_unmanage(mem);
204 vm_page_wakeup(mem);
205 vm_page_unlock_queues();
206 }
207 VM_OBJECT_UNLOCK(kernel_object);
208
209 /*
210 * And finally, mark the data as non-pageable.
211 */
212 (void) vm_map_wire(map, addr, addr + size,
213 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
214
215 return (addr);
216}
217
218/*
219 * kmem_free:
220 *
221 * Release a region of kernel virtual memory allocated
222 * with kmem_alloc, and return the physical pages
223 * associated with that region.
224 *
225 * This routine may not block on kernel maps.
226 */
227void
228kmem_free(map, addr, size)
229 vm_map_t map;
230 vm_offset_t addr;
231 vm_size_t size;
232{
233
234 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
235}
236
237/*
238 * kmem_suballoc:
239 *
240 * Allocates a map to manage a subrange
241 * of the kernel virtual address space.
242 *
243 * Arguments are as follows:
244 *
245 * parent Map to take range from
246 * min, max Returned endpoints of map
247 * size Size of range to find
248 */
249vm_map_t
250kmem_suballoc(parent, min, max, size)
251 vm_map_t parent;
252 vm_offset_t *min, *max;
253 vm_size_t size;
254{
255 int ret;
256 vm_map_t result;
257
258 size = round_page(size);
259
260 *min = (vm_offset_t) vm_map_min(parent);
261 ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
262 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
263 if (ret != KERN_SUCCESS) {
264 printf("kmem_suballoc: bad status return of %d.\n", ret);
265 panic("kmem_suballoc");
266 }
267 *max = *min + size;
268 result = vm_map_create(vm_map_pmap(parent), *min, *max);
269 if (result == NULL)
270 panic("kmem_suballoc: cannot create submap");
271 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
272 panic("kmem_suballoc: unable to change range to submap");
273 return (result);
274}
275
276/*
277 * kmem_malloc:
278 *
279 * Allocate wired-down memory in the kernel's address map for the higher
280 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
281 * kmem_alloc() because we may need to allocate memory at interrupt
282 * level where we cannot block (canwait == FALSE).
283 *
284 * This routine has its own private kernel submap (kmem_map) and object
285 * (kmem_object). This, combined with the fact that only malloc uses
286 * this routine, ensures that we will never block in map or object waits.
287 *
288 * Note that this still only works in a uni-processor environment and
289 * when called at splhigh().
290 *
291 * We don't worry about expanding the map (adding entries) since entries
292 * for wired maps are statically allocated.
293 *
294 * NOTE: This routine is not supposed to block if M_NOWAIT is set, but
295 * I have not verified that it actually does not block.
296 *
297 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
298 * which we never free.
299 */
300vm_offset_t
301kmem_malloc(map, size, flags)
302 vm_map_t map;
303 vm_size_t size;
304 int flags;
305{
306 vm_offset_t offset, i;
307 vm_map_entry_t entry;
308 vm_offset_t addr;
309 vm_page_t m;
310 int pflags;
311
312 size = round_page(size);
313 addr = vm_map_min(map);
314
315 /*
316 * Locate sufficient space in the map. This will give us the final
317 * virtual address for the new memory, and thus will tell us the
318 * offset within the kernel map.
319 */
320 vm_map_lock(map);
321 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
322 vm_map_unlock(map);
|
323 if (map != kmem_map) {
324 static int last_report; /* when we did it (in ticks) */
325 if (ticks < last_report ||
326 (ticks - last_report) >= hz) {
327 last_report = ticks;
328 printf("Out of mbuf address space!\n");
329 printf("Consider increasing NMBCLUSTERS\n");
330 }
331 return (0);
332 }
| |
333 if ((flags & M_NOWAIT) == 0)
334 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
335 (long)size, (long)map->size);
336 return (0);
337 }
338 offset = addr - VM_MIN_KERNEL_ADDRESS;
339 vm_object_reference(kmem_object);
340 vm_map_insert(map, kmem_object, offset, addr, addr + size,
341 VM_PROT_ALL, VM_PROT_ALL, 0);
342
343 /*
344 * Note: if M_NOWAIT specified alone, allocate from
345 * interrupt-safe queues only (just the free list). If
346 * M_USE_RESERVE is also specified, we can also
347 * allocate from the cache. Neither of the latter two
348 * flags may be specified from an interrupt since interrupts
349 * are not allowed to mess with the cache queue.
350 */
351
352 if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
353 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
354 else
355 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
356
357 if (flags & M_ZERO)
358 pflags |= VM_ALLOC_ZERO;
359
360 VM_OBJECT_LOCK(kmem_object);
361 for (i = 0; i < size; i += PAGE_SIZE) {
362retry:
363 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
364
365 /*
366 * Ran out of space, free everything up and return. Don't need
367 * to lock page queues here as we know that the pages we got
368 * aren't on any queues.
369 */
370 if (m == NULL) {
371 if ((flags & M_NOWAIT) == 0) {
372 VM_OBJECT_UNLOCK(kmem_object);
373 vm_map_unlock(map);
374 VM_WAIT;
375 vm_map_lock(map);
376 VM_OBJECT_LOCK(kmem_object);
377 goto retry;
378 }
379 /*
380 * Free the pages before removing the map entry.
381 * They are already marked busy. Calling
382 * vm_map_delete before the pages has been freed or
383 * unbusied will cause a deadlock.
384 */
385 while (i != 0) {
386 i -= PAGE_SIZE;
387 m = vm_page_lookup(kmem_object,
388 OFF_TO_IDX(offset + i));
389 vm_page_lock_queues();
390 vm_page_unwire(m, 0);
391 vm_page_free(m);
392 vm_page_unlock_queues();
393 }
394 VM_OBJECT_UNLOCK(kmem_object);
395 vm_map_delete(map, addr, addr + size);
396 vm_map_unlock(map);
397 return (0);
398 }
399 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
400 pmap_zero_page(m);
401 m->valid = VM_PAGE_BITS_ALL;
402 vm_page_lock_queues();
403 vm_page_unmanage(m);
404 vm_page_unlock_queues();
405 }
406 VM_OBJECT_UNLOCK(kmem_object);
407
408 /*
409 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
410 * be able to extend the previous entry so there will be a new entry
411 * exactly corresponding to this address range and it will have
412 * wired_count == 0.
413 */
414 if (!vm_map_lookup_entry(map, addr, &entry) ||
415 entry->start != addr || entry->end != addr + size ||
416 entry->wired_count != 0)
417 panic("kmem_malloc: entry not found or misaligned");
418 entry->wired_count = 1;
419
420 /*
421 * At this point, the kmem_object must be unlocked because
422 * vm_map_simplify_entry() calls vm_object_deallocate(), which
423 * locks the kmem_object.
424 */
425 vm_map_simplify_entry(map, entry);
426
427 /*
428 * Loop thru pages, entering them in the pmap. (We cannot add them to
429 * the wired count without wrapping the vm_page_queue_lock in
430 * splimp...)
431 */
432 VM_OBJECT_LOCK(kmem_object);
433 for (i = 0; i < size; i += PAGE_SIZE) {
434 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
435 /*
436 * Because this is kernel_pmap, this call will not block.
437 */
438 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
439 vm_page_lock_queues();
440 vm_page_flag_set(m, PG_WRITEABLE | PG_REFERENCED);
441 vm_page_wakeup(m);
442 vm_page_unlock_queues();
443 }
444 VM_OBJECT_UNLOCK(kmem_object);
445 vm_map_unlock(map);
446
447 return (addr);
448}
449
450/*
451 * kmem_alloc_wait:
452 *
453 * Allocates pageable memory from a sub-map of the kernel. If the submap
454 * has no room, the caller sleeps waiting for more memory in the submap.
455 *
456 * This routine may block.
457 */
458vm_offset_t
459kmem_alloc_wait(map, size)
460 vm_map_t map;
461 vm_size_t size;
462{
463 vm_offset_t addr;
464
465 size = round_page(size);
466
467 for (;;) {
468 /*
469 * To make this work for more than one map, use the map's lock
470 * to lock out sleepers/wakers.
471 */
472 vm_map_lock(map);
473 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
474 break;
475 /* no space now; see if we can ever get space */
476 if (vm_map_max(map) - vm_map_min(map) < size) {
477 vm_map_unlock(map);
478 return (0);
479 }
480 map->needs_wakeup = TRUE;
481 vm_map_unlock_and_wait(map, FALSE);
482 }
483 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
484 vm_map_unlock(map);
485 return (addr);
486}
487
488/*
489 * kmem_free_wakeup:
490 *
491 * Returns memory to a submap of the kernel, and wakes up any processes
492 * waiting for memory in that map.
493 */
494void
495kmem_free_wakeup(map, addr, size)
496 vm_map_t map;
497 vm_offset_t addr;
498 vm_size_t size;
499{
500
501 vm_map_lock(map);
502 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
503 if (map->needs_wakeup) {
504 map->needs_wakeup = FALSE;
505 vm_map_wakeup(map);
506 }
507 vm_map_unlock(map);
508}
509
510/*
511 * kmem_init:
512 *
513 * Create the kernel map; insert a mapping covering kernel text,
514 * data, bss, and all space allocated thus far (`boostrap' data). The
515 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
516 * `start' as allocated, and the range between `start' and `end' as free.
517 */
518void
519kmem_init(start, end)
520 vm_offset_t start, end;
521{
522 vm_map_t m;
523
524 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
525 m->system_map = 1;
526 vm_map_lock(m);
527 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
528 kernel_map = m;
529 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
530 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
531 /* ... and ending with the completion of the above `insert' */
532 vm_map_unlock(m);
533}
| 323 if ((flags & M_NOWAIT) == 0)
324 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
325 (long)size, (long)map->size);
326 return (0);
327 }
328 offset = addr - VM_MIN_KERNEL_ADDRESS;
329 vm_object_reference(kmem_object);
330 vm_map_insert(map, kmem_object, offset, addr, addr + size,
331 VM_PROT_ALL, VM_PROT_ALL, 0);
332
333 /*
334 * Note: if M_NOWAIT specified alone, allocate from
335 * interrupt-safe queues only (just the free list). If
336 * M_USE_RESERVE is also specified, we can also
337 * allocate from the cache. Neither of the latter two
338 * flags may be specified from an interrupt since interrupts
339 * are not allowed to mess with the cache queue.
340 */
341
342 if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
343 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
344 else
345 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
346
347 if (flags & M_ZERO)
348 pflags |= VM_ALLOC_ZERO;
349
350 VM_OBJECT_LOCK(kmem_object);
351 for (i = 0; i < size; i += PAGE_SIZE) {
352retry:
353 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
354
355 /*
356 * Ran out of space, free everything up and return. Don't need
357 * to lock page queues here as we know that the pages we got
358 * aren't on any queues.
359 */
360 if (m == NULL) {
361 if ((flags & M_NOWAIT) == 0) {
362 VM_OBJECT_UNLOCK(kmem_object);
363 vm_map_unlock(map);
364 VM_WAIT;
365 vm_map_lock(map);
366 VM_OBJECT_LOCK(kmem_object);
367 goto retry;
368 }
369 /*
370 * Free the pages before removing the map entry.
371 * They are already marked busy. Calling
372 * vm_map_delete before the pages has been freed or
373 * unbusied will cause a deadlock.
374 */
375 while (i != 0) {
376 i -= PAGE_SIZE;
377 m = vm_page_lookup(kmem_object,
378 OFF_TO_IDX(offset + i));
379 vm_page_lock_queues();
380 vm_page_unwire(m, 0);
381 vm_page_free(m);
382 vm_page_unlock_queues();
383 }
384 VM_OBJECT_UNLOCK(kmem_object);
385 vm_map_delete(map, addr, addr + size);
386 vm_map_unlock(map);
387 return (0);
388 }
389 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
390 pmap_zero_page(m);
391 m->valid = VM_PAGE_BITS_ALL;
392 vm_page_lock_queues();
393 vm_page_unmanage(m);
394 vm_page_unlock_queues();
395 }
396 VM_OBJECT_UNLOCK(kmem_object);
397
398 /*
399 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
400 * be able to extend the previous entry so there will be a new entry
401 * exactly corresponding to this address range and it will have
402 * wired_count == 0.
403 */
404 if (!vm_map_lookup_entry(map, addr, &entry) ||
405 entry->start != addr || entry->end != addr + size ||
406 entry->wired_count != 0)
407 panic("kmem_malloc: entry not found or misaligned");
408 entry->wired_count = 1;
409
410 /*
411 * At this point, the kmem_object must be unlocked because
412 * vm_map_simplify_entry() calls vm_object_deallocate(), which
413 * locks the kmem_object.
414 */
415 vm_map_simplify_entry(map, entry);
416
417 /*
418 * Loop thru pages, entering them in the pmap. (We cannot add them to
419 * the wired count without wrapping the vm_page_queue_lock in
420 * splimp...)
421 */
422 VM_OBJECT_LOCK(kmem_object);
423 for (i = 0; i < size; i += PAGE_SIZE) {
424 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
425 /*
426 * Because this is kernel_pmap, this call will not block.
427 */
428 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
429 vm_page_lock_queues();
430 vm_page_flag_set(m, PG_WRITEABLE | PG_REFERENCED);
431 vm_page_wakeup(m);
432 vm_page_unlock_queues();
433 }
434 VM_OBJECT_UNLOCK(kmem_object);
435 vm_map_unlock(map);
436
437 return (addr);
438}
439
440/*
441 * kmem_alloc_wait:
442 *
443 * Allocates pageable memory from a sub-map of the kernel. If the submap
444 * has no room, the caller sleeps waiting for more memory in the submap.
445 *
446 * This routine may block.
447 */
448vm_offset_t
449kmem_alloc_wait(map, size)
450 vm_map_t map;
451 vm_size_t size;
452{
453 vm_offset_t addr;
454
455 size = round_page(size);
456
457 for (;;) {
458 /*
459 * To make this work for more than one map, use the map's lock
460 * to lock out sleepers/wakers.
461 */
462 vm_map_lock(map);
463 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
464 break;
465 /* no space now; see if we can ever get space */
466 if (vm_map_max(map) - vm_map_min(map) < size) {
467 vm_map_unlock(map);
468 return (0);
469 }
470 map->needs_wakeup = TRUE;
471 vm_map_unlock_and_wait(map, FALSE);
472 }
473 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
474 vm_map_unlock(map);
475 return (addr);
476}
477
478/*
479 * kmem_free_wakeup:
480 *
481 * Returns memory to a submap of the kernel, and wakes up any processes
482 * waiting for memory in that map.
483 */
484void
485kmem_free_wakeup(map, addr, size)
486 vm_map_t map;
487 vm_offset_t addr;
488 vm_size_t size;
489{
490
491 vm_map_lock(map);
492 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
493 if (map->needs_wakeup) {
494 map->needs_wakeup = FALSE;
495 vm_map_wakeup(map);
496 }
497 vm_map_unlock(map);
498}
499
500/*
501 * kmem_init:
502 *
503 * Create the kernel map; insert a mapping covering kernel text,
504 * data, bss, and all space allocated thus far (`boostrap' data). The
505 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
506 * `start' as allocated, and the range between `start' and `end' as free.
507 */
508void
509kmem_init(start, end)
510 vm_offset_t start, end;
511{
512 vm_map_t m;
513
514 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
515 m->system_map = 1;
516 vm_map_lock(m);
517 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
518 kernel_map = m;
519 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
520 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
521 /* ... and ending with the completion of the above `insert' */
522 vm_map_unlock(m);
523}
|