xref: /freebsd/sys/vm/vm_kern.c (revision 3e0f6b97b257a96f7275e4442204263e44b16686)
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  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)vm_kern.c	8.3 (Berkeley) 1/12/94
37  *
38  *
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  *
64  * $FreeBSD$
65  */
66 
67 /*
68  *	Kernel memory management.
69  */
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/proc.h>
75 #include <sys/malloc.h>
76 #include <sys/syslog.h>
77 #include <sys/queue.h>
78 #include <sys/vmmeter.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_prot.h>
83 #include <sys/lock.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_extern.h>
91 
92 vm_map_t kernel_map=0;
93 vm_map_t kmem_map=0;
94 vm_map_t exec_map=0;
95 vm_map_t exech_map=0;
96 vm_map_t clean_map=0;
97 vm_map_t u_map=0;
98 vm_map_t buffer_map=0;
99 vm_map_t mb_map=0;
100 int mb_map_full=0;
101 vm_map_t io_map=0;
102 vm_map_t phys_map=0;
103 
104 /*
105  *	kmem_alloc_pageable:
106  *
107  *	Allocate pageable memory to the kernel's address map.
108  *	"map" must be kernel_map or a submap of kernel_map.
109  */
110 
111 vm_offset_t
112 kmem_alloc_pageable(map, size)
113 	vm_map_t map;
114 	register vm_size_t size;
115 {
116 	vm_offset_t addr;
117 	register int result;
118 
119 	size = round_page(size);
120 	addr = vm_map_min(map);
121 	result = vm_map_find(map, NULL, (vm_offset_t) 0,
122 	    &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
123 	if (result != KERN_SUCCESS) {
124 		return (0);
125 	}
126 	return (addr);
127 }
128 
129 /*
130  *	Allocate wired-down memory in the kernel's address map
131  *	or a submap.
132  */
133 vm_offset_t
134 kmem_alloc(map, size)
135 	register vm_map_t map;
136 	register vm_size_t size;
137 {
138 	vm_offset_t addr;
139 	register vm_offset_t offset;
140 	vm_offset_t i;
141 
142 	size = round_page(size);
143 
144 	/*
145 	 * Use the kernel object for wired-down kernel pages. Assume that no
146 	 * region of the kernel object is referenced more than once.
147 	 */
148 
149 	/*
150 	 * Locate sufficient space in the map.  This will give us the final
151 	 * virtual address for the new memory, and thus will tell us the
152 	 * offset within the kernel map.
153 	 */
154 	vm_map_lock(map);
155 	if (vm_map_findspace(map, 0, size, &addr)) {
156 		vm_map_unlock(map);
157 		return (0);
158 	}
159 	offset = addr - VM_MIN_KERNEL_ADDRESS;
160 	vm_object_reference(kernel_object);
161 	vm_map_insert(map, kernel_object, offset, addr, addr + size,
162 		VM_PROT_ALL, VM_PROT_ALL, 0);
163 	vm_map_unlock(map);
164 
165 	/*
166 	 * Guarantee that there are pages already in this object before
167 	 * calling vm_map_pageable.  This is to prevent the following
168 	 * scenario:
169 	 *
170 	 * 1) Threads have swapped out, so that there is a pager for the
171 	 * kernel_object. 2) The kmsg zone is empty, and so we are
172 	 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault;
173 	 * there is no page, but there is a pager, so we call
174 	 * pager_data_request.  But the kmsg zone is empty, so we must
175 	 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
176 	 * we get the data back from the pager, it will be (very stale)
177 	 * non-zero data.  kmem_alloc is defined to return zero-filled memory.
178 	 *
179 	 * We're intentionally not activating the pages we allocate to prevent a
180 	 * race with page-out.  vm_map_pageable will wire the pages.
181 	 */
182 
183 	for (i = 0; i < size; i += PAGE_SIZE) {
184 		vm_page_t mem;
185 
186 		while ((mem = vm_page_alloc(kernel_object,
187 			OFF_TO_IDX(offset + i), VM_ALLOC_ZERO)) == NULL) {
188 			VM_WAIT;
189 		}
190 		if ((mem->flags & PG_ZERO) == 0)
191 			vm_page_zero_fill(mem);
192 		mem->flags &= ~(PG_BUSY|PG_ZERO);
193 		mem->valid = VM_PAGE_BITS_ALL;
194 	}
195 
196 	/*
197 	 * And finally, mark the data as non-pageable.
198 	 */
199 
200 	(void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE);
201 
202 	return (addr);
203 }
204 
205 /*
206  *	kmem_free:
207  *
208  *	Release a region of kernel virtual memory allocated
209  *	with kmem_alloc, and return the physical pages
210  *	associated with that region.
211  */
212 void
213 kmem_free(map, addr, size)
214 	vm_map_t map;
215 	register vm_offset_t addr;
216 	vm_size_t size;
217 {
218 	(void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
219 }
220 
221 /*
222  *	kmem_suballoc:
223  *
224  *	Allocates a map to manage a subrange
225  *	of the kernel virtual address space.
226  *
227  *	Arguments are as follows:
228  *
229  *	parent		Map to take range from
230  *	size		Size of range to find
231  *	min, max	Returned endpoints of map
232  *	pageable	Can the region be paged
233  */
234 vm_map_t
235 kmem_suballoc(parent, min, max, size, pageable)
236 	register vm_map_t parent;
237 	vm_offset_t *min, *max;
238 	register vm_size_t size;
239 	boolean_t pageable;
240 {
241 	register int ret;
242 	vm_map_t result;
243 
244 	size = round_page(size);
245 
246 	*min = (vm_offset_t) vm_map_min(parent);
247 	ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
248 	    min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
249 	if (ret != KERN_SUCCESS) {
250 		printf("kmem_suballoc: bad status return of %d.\n", ret);
251 		panic("kmem_suballoc");
252 	}
253 	*max = *min + size;
254 	pmap_reference(vm_map_pmap(parent));
255 	result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable);
256 	if (result == NULL)
257 		panic("kmem_suballoc: cannot create submap");
258 	if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
259 		panic("kmem_suballoc: unable to change range to submap");
260 	return (result);
261 }
262 
263 /*
264  * Allocate wired-down memory in the kernel's address map for the higher
265  * level kernel memory allocator (kern/kern_malloc.c).  We cannot use
266  * kmem_alloc() because we may need to allocate memory at interrupt
267  * level where we cannot block (canwait == FALSE).
268  *
269  * This routine has its own private kernel submap (kmem_map) and object
270  * (kmem_object).  This, combined with the fact that only malloc uses
271  * this routine, ensures that we will never block in map or object waits.
272  *
273  * Note that this still only works in a uni-processor environment and
274  * when called at splhigh().
275  *
276  * We don't worry about expanding the map (adding entries) since entries
277  * for wired maps are statically allocated.
278  */
279 vm_offset_t
280 kmem_malloc(map, size, waitflag)
281 	register vm_map_t map;
282 	register vm_size_t size;
283 	boolean_t waitflag;
284 {
285 	register vm_offset_t offset, i;
286 	vm_map_entry_t entry;
287 	vm_offset_t addr;
288 	vm_page_t m;
289 
290 	if (map != kmem_map && map != mb_map)
291 		panic("kmem_malloc: map != {kmem,mb}_map");
292 
293 	size = round_page(size);
294 	addr = vm_map_min(map);
295 
296 	/*
297 	 * Locate sufficient space in the map.  This will give us the final
298 	 * virtual address for the new memory, and thus will tell us the
299 	 * offset within the kernel map.
300 	 */
301 	vm_map_lock(map);
302 	if (vm_map_findspace(map, 0, size, &addr)) {
303 		vm_map_unlock(map);
304 		if (map == mb_map) {
305 			mb_map_full = TRUE;
306 			log(LOG_ERR, "Out of mbuf clusters - increase maxusers!\n");
307 			return (0);
308 		}
309 		if (waitflag == M_WAITOK)
310 			panic("kmem_malloc: kmem_map too small");
311 		return (0);
312 	}
313 	offset = addr - VM_MIN_KERNEL_ADDRESS;
314 	vm_object_reference(kmem_object);
315 	vm_map_insert(map, kmem_object, offset, addr, addr + size,
316 		VM_PROT_ALL, VM_PROT_ALL, 0);
317 
318 	for (i = 0; i < size; i += PAGE_SIZE) {
319 retry:
320 		m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i),
321 			(waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM);
322 
323 		/*
324 		 * Ran out of space, free everything up and return. Don't need
325 		 * to lock page queues here as we know that the pages we got
326 		 * aren't on any queues.
327 		 */
328 		if (m == NULL) {
329 			if (waitflag == M_WAITOK) {
330 				VM_WAIT;
331 				goto retry;
332 			}
333 			while (i != 0) {
334 				i -= PAGE_SIZE;
335 				m = vm_page_lookup(kmem_object,
336 					OFF_TO_IDX(offset + i));
337 				PAGE_WAKEUP(m);
338 				vm_page_free(m);
339 			}
340 			vm_map_delete(map, addr, addr + size);
341 			vm_map_unlock(map);
342 			return (0);
343 		}
344 		m->flags &= ~PG_ZERO;
345 		m->valid = VM_PAGE_BITS_ALL;
346 	}
347 
348 	/*
349 	 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
350 	 * be able to extend the previous entry so there will be a new entry
351 	 * exactly corresponding to this address range and it will have
352 	 * wired_count == 0.
353 	 */
354 	if (!vm_map_lookup_entry(map, addr, &entry) ||
355 	    entry->start != addr || entry->end != addr + size ||
356 	    entry->wired_count)
357 		panic("kmem_malloc: entry not found or misaligned");
358 	entry->wired_count++;
359 
360 	vm_map_simplify_entry(map, entry);
361 
362 	/*
363 	 * Loop thru pages, entering them in the pmap. (We cannot add them to
364 	 * the wired count without wrapping the vm_page_queue_lock in
365 	 * splimp...)
366 	 */
367 	for (i = 0; i < size; i += PAGE_SIZE) {
368 		m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
369 		vm_page_wire(m);
370 		PAGE_WAKEUP(m);
371 		pmap_enter(kernel_pmap, addr + i, VM_PAGE_TO_PHYS(m),
372 			VM_PROT_ALL, 1);
373 		m->flags |= PG_MAPPED|PG_WRITEABLE;
374 	}
375 	vm_map_unlock(map);
376 
377 	return (addr);
378 }
379 
380 /*
381  *	kmem_alloc_wait
382  *
383  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
384  *	has no room, the caller sleeps waiting for more memory in the submap.
385  *
386  */
387 vm_offset_t
388 kmem_alloc_wait(map, size)
389 	vm_map_t map;
390 	vm_size_t size;
391 {
392 	vm_offset_t addr;
393 
394 	size = round_page(size);
395 
396 	for (;;) {
397 		/*
398 		 * To make this work for more than one map, use the map's lock
399 		 * to lock out sleepers/wakers.
400 		 */
401 		vm_map_lock(map);
402 		if (vm_map_findspace(map, 0, size, &addr) == 0)
403 			break;
404 		/* no space now; see if we can ever get space */
405 		if (vm_map_max(map) - vm_map_min(map) < size) {
406 			vm_map_unlock(map);
407 			return (0);
408 		}
409 		vm_map_unlock(map);
410 		tsleep(map, PVM, "kmaw", 0);
411 	}
412 	vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
413 	vm_map_unlock(map);
414 	return (addr);
415 }
416 
417 /*
418  *	kmem_free_wakeup
419  *
420  *	Returns memory to a submap of the kernel, and wakes up any processes
421  *	waiting for memory in that map.
422  */
423 void
424 kmem_free_wakeup(map, addr, size)
425 	vm_map_t map;
426 	vm_offset_t addr;
427 	vm_size_t size;
428 {
429 	vm_map_lock(map);
430 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
431 	wakeup(map);
432 	vm_map_unlock(map);
433 }
434 
435 /*
436  * Create the kernel map; insert a mapping covering kernel text, data, bss,
437  * and all space allocated thus far (`boostrap' data).  The new map will thus
438  * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and
439  * the range between `start' and `end' as free.
440  */
441 void
442 kmem_init(start, end)
443 	vm_offset_t start, end;
444 {
445 	register vm_map_t m;
446 
447 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE);
448 	vm_map_lock(m);
449 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
450 	kernel_map = m;
451 	(void) vm_map_insert(m, NULL, (vm_offset_t) 0,
452 	    VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
453 	/* ... and ending with the completion of the above `insert' */
454 	vm_map_unlock(m);
455 }
456