xref: /freebsd/sys/vm/vm_kern.c (revision 952d112864d8008aa87278a30a539d888a8493cd)
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  * $Id: vm_kern.c,v 1.33 1997/02/22 09:48:21 peter Exp $
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 clean_map=0;
96 vm_map_t u_map=0;
97 vm_map_t buffer_map=0;
98 vm_map_t mb_map=0;
99 int mb_map_full=0;
100 vm_map_t io_map=0;
101 vm_map_t phys_map=0;
102 
103 /*
104  *	kmem_alloc_pageable:
105  *
106  *	Allocate pageable memory to the kernel's address map.
107  *	"map" must be kernel_map or a submap of kernel_map.
108  */
109 
110 vm_offset_t
111 kmem_alloc_pageable(map, size)
112 	vm_map_t map;
113 	register vm_size_t size;
114 {
115 	vm_offset_t addr;
116 	register int result;
117 
118 	size = round_page(size);
119 	addr = vm_map_min(map);
120 	result = vm_map_find(map, NULL, (vm_offset_t) 0,
121 	    &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
122 	if (result != KERN_SUCCESS) {
123 		return (0);
124 	}
125 	return (addr);
126 }
127 
128 /*
129  *	Allocate wired-down memory in the kernel's address map
130  *	or a submap.
131  */
132 vm_offset_t
133 kmem_alloc(map, size)
134 	register vm_map_t map;
135 	register vm_size_t size;
136 {
137 	vm_offset_t addr;
138 	register vm_offset_t offset;
139 	vm_offset_t i;
140 
141 	size = round_page(size);
142 
143 	/*
144 	 * Use the kernel object for wired-down kernel pages. Assume that no
145 	 * region of the kernel object is referenced more than once.
146 	 */
147 
148 	/*
149 	 * Locate sufficient space in the map.  This will give us the final
150 	 * virtual address for the new memory, and thus will tell us the
151 	 * offset within the kernel map.
152 	 */
153 	vm_map_lock(map);
154 	if (vm_map_findspace(map, 0, size, &addr)) {
155 		vm_map_unlock(map);
156 		return (0);
157 	}
158 	offset = addr - VM_MIN_KERNEL_ADDRESS;
159 	vm_object_reference(kernel_object);
160 	vm_map_insert(map, kernel_object, offset, addr, addr + size,
161 		VM_PROT_ALL, VM_PROT_ALL, 0);
162 	vm_map_unlock(map);
163 
164 	/*
165 	 * Guarantee that there are pages already in this object before
166 	 * calling vm_map_pageable.  This is to prevent the following
167 	 * scenario:
168 	 *
169 	 * 1) Threads have swapped out, so that there is a pager for the
170 	 * kernel_object. 2) The kmsg zone is empty, and so we are
171 	 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault;
172 	 * there is no page, but there is a pager, so we call
173 	 * pager_data_request.  But the kmsg zone is empty, so we must
174 	 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
175 	 * we get the data back from the pager, it will be (very stale)
176 	 * non-zero data.  kmem_alloc is defined to return zero-filled memory.
177 	 *
178 	 * We're intentionally not activating the pages we allocate to prevent a
179 	 * race with page-out.  vm_map_pageable will wire the pages.
180 	 */
181 
182 	for (i = 0; i < size; i += PAGE_SIZE) {
183 		vm_page_t mem;
184 
185 		while ((mem = vm_page_alloc(kernel_object,
186 			OFF_TO_IDX(offset + i), VM_ALLOC_ZERO)) == NULL) {
187 			VM_WAIT;
188 		}
189 		if ((mem->flags & PG_ZERO) == 0)
190 			vm_page_zero_fill(mem);
191 		mem->flags &= ~(PG_BUSY|PG_ZERO);
192 		mem->valid = VM_PAGE_BITS_ALL;
193 	}
194 
195 	/*
196 	 * And finally, mark the data as non-pageable.
197 	 */
198 
199 	(void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE);
200 
201 	return (addr);
202 }
203 
204 /*
205  *	kmem_free:
206  *
207  *	Release a region of kernel virtual memory allocated
208  *	with kmem_alloc, and return the physical pages
209  *	associated with that region.
210  */
211 void
212 kmem_free(map, addr, size)
213 	vm_map_t map;
214 	register vm_offset_t addr;
215 	vm_size_t size;
216 {
217 	(void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
218 }
219 
220 /*
221  *	kmem_suballoc:
222  *
223  *	Allocates a map to manage a subrange
224  *	of the kernel virtual address space.
225  *
226  *	Arguments are as follows:
227  *
228  *	parent		Map to take range from
229  *	size		Size of range to find
230  *	min, max	Returned endpoints of map
231  *	pageable	Can the region be paged
232  */
233 vm_map_t
234 kmem_suballoc(parent, min, max, size, pageable)
235 	register vm_map_t parent;
236 	vm_offset_t *min, *max;
237 	register vm_size_t size;
238 	boolean_t pageable;
239 {
240 	register int ret;
241 	vm_map_t result;
242 
243 	size = round_page(size);
244 
245 	*min = (vm_offset_t) vm_map_min(parent);
246 	ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
247 	    min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
248 	if (ret != KERN_SUCCESS) {
249 		printf("kmem_suballoc: bad status return of %d.\n", ret);
250 		panic("kmem_suballoc");
251 	}
252 	*max = *min + size;
253 	pmap_reference(vm_map_pmap(parent));
254 	result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable);
255 	if (result == NULL)
256 		panic("kmem_suballoc: cannot create submap");
257 	if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
258 		panic("kmem_suballoc: unable to change range to submap");
259 	return (result);
260 }
261 
262 /*
263  * Allocate wired-down memory in the kernel's address map for the higher
264  * level kernel memory allocator (kern/kern_malloc.c).  We cannot use
265  * kmem_alloc() because we may need to allocate memory at interrupt
266  * level where we cannot block (canwait == FALSE).
267  *
268  * This routine has its own private kernel submap (kmem_map) and object
269  * (kmem_object).  This, combined with the fact that only malloc uses
270  * this routine, ensures that we will never block in map or object waits.
271  *
272  * Note that this still only works in a uni-processor environment and
273  * when called at splhigh().
274  *
275  * We don't worry about expanding the map (adding entries) since entries
276  * for wired maps are statically allocated.
277  */
278 vm_offset_t
279 kmem_malloc(map, size, waitflag)
280 	register vm_map_t map;
281 	register vm_size_t size;
282 	boolean_t waitflag;
283 {
284 	register vm_offset_t offset, i;
285 	vm_map_entry_t entry;
286 	vm_offset_t addr;
287 	vm_page_t m;
288 
289 	if (map != kmem_map && map != mb_map)
290 		panic("kmem_malloc: map != {kmem,mb}_map");
291 
292 	size = round_page(size);
293 	addr = vm_map_min(map);
294 
295 	/*
296 	 * Locate sufficient space in the map.  This will give us the final
297 	 * virtual address for the new memory, and thus will tell us the
298 	 * offset within the kernel map.
299 	 */
300 	vm_map_lock(map);
301 	if (vm_map_findspace(map, 0, size, &addr)) {
302 		vm_map_unlock(map);
303 		if (map == mb_map) {
304 			mb_map_full = TRUE;
305 			log(LOG_ERR, "Out of mbuf clusters - increase maxusers!\n");
306 			return (0);
307 		}
308 		if (waitflag == M_WAITOK)
309 			panic("kmem_malloc: kmem_map too small");
310 		return (0);
311 	}
312 	offset = addr - VM_MIN_KERNEL_ADDRESS;
313 	vm_object_reference(kmem_object);
314 	vm_map_insert(map, kmem_object, offset, addr, addr + size,
315 		VM_PROT_ALL, VM_PROT_ALL, 0);
316 
317 	for (i = 0; i < size; i += PAGE_SIZE) {
318 retry:
319 		m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i),
320 			(waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM);
321 
322 		/*
323 		 * Ran out of space, free everything up and return. Don't need
324 		 * to lock page queues here as we know that the pages we got
325 		 * aren't on any queues.
326 		 */
327 		if (m == NULL) {
328 			if (waitflag == M_WAITOK) {
329 				VM_WAIT;
330 				goto retry;
331 			}
332 			while (i != 0) {
333 				i -= PAGE_SIZE;
334 				m = vm_page_lookup(kmem_object,
335 					OFF_TO_IDX(offset + i));
336 				PAGE_WAKEUP(m);
337 				vm_page_free(m);
338 			}
339 			vm_map_delete(map, addr, addr + size);
340 			vm_map_unlock(map);
341 			return (0);
342 		}
343 		m->flags &= ~PG_ZERO;
344 		m->valid = VM_PAGE_BITS_ALL;
345 	}
346 
347 	/*
348 	 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
349 	 * be able to extend the previous entry so there will be a new entry
350 	 * exactly corresponding to this address range and it will have
351 	 * wired_count == 0.
352 	 */
353 	if (!vm_map_lookup_entry(map, addr, &entry) ||
354 	    entry->start != addr || entry->end != addr + size ||
355 	    entry->wired_count)
356 		panic("kmem_malloc: entry not found or misaligned");
357 	entry->wired_count++;
358 
359 	vm_map_simplify_entry(map, entry);
360 
361 	/*
362 	 * Loop thru pages, entering them in the pmap. (We cannot add them to
363 	 * the wired count without wrapping the vm_page_queue_lock in
364 	 * splimp...)
365 	 */
366 	for (i = 0; i < size; i += PAGE_SIZE) {
367 		m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
368 		vm_page_wire(m);
369 		PAGE_WAKEUP(m);
370 		pmap_enter(kernel_pmap, addr + i, VM_PAGE_TO_PHYS(m),
371 			VM_PROT_ALL, 1);
372 		m->flags |= PG_MAPPED|PG_WRITEABLE;
373 	}
374 	vm_map_unlock(map);
375 
376 	return (addr);
377 }
378 
379 /*
380  *	kmem_alloc_wait
381  *
382  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
383  *	has no room, the caller sleeps waiting for more memory in the submap.
384  *
385  */
386 vm_offset_t
387 kmem_alloc_wait(map, size)
388 	vm_map_t map;
389 	vm_size_t size;
390 {
391 	vm_offset_t addr;
392 
393 	size = round_page(size);
394 
395 	for (;;) {
396 		/*
397 		 * To make this work for more than one map, use the map's lock
398 		 * to lock out sleepers/wakers.
399 		 */
400 		vm_map_lock(map);
401 		if (vm_map_findspace(map, 0, size, &addr) == 0)
402 			break;
403 		/* no space now; see if we can ever get space */
404 		if (vm_map_max(map) - vm_map_min(map) < size) {
405 			vm_map_unlock(map);
406 			return (0);
407 		}
408 		vm_map_unlock(map);
409 		tsleep(map, PVM, "kmaw", 0);
410 	}
411 	vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
412 	vm_map_unlock(map);
413 	return (addr);
414 }
415 
416 /*
417  *	kmem_free_wakeup
418  *
419  *	Returns memory to a submap of the kernel, and wakes up any processes
420  *	waiting for memory in that map.
421  */
422 void
423 kmem_free_wakeup(map, addr, size)
424 	vm_map_t map;
425 	vm_offset_t addr;
426 	vm_size_t size;
427 {
428 	vm_map_lock(map);
429 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
430 	wakeup(map);
431 	vm_map_unlock(map);
432 }
433 
434 /*
435  * Create the kernel map; insert a mapping covering kernel text, data, bss,
436  * and all space allocated thus far (`boostrap' data).  The new map will thus
437  * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and
438  * the range between `start' and `end' as free.
439  */
440 void
441 kmem_init(start, end)
442 	vm_offset_t start, end;
443 {
444 	register vm_map_t m;
445 
446 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE);
447 	vm_map_lock(m);
448 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
449 	kernel_map = m;
450 	(void) vm_map_insert(m, NULL, (vm_offset_t) 0,
451 	    VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
452 	/* ... and ending with the completion of the above `insert' */
453 	vm_map_unlock(m);
454 }
455