xref: /freebsd/sys/vm/vm_kern.c (revision afe61c15161c324a7af299a9b8457aba5afc92db)
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  *	@(#)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 
65 /*
66  *	Kernel memory management.
67  */
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_kern.h>
76 
77 /*
78  *	kmem_alloc_pageable:
79  *
80  *	Allocate pageable memory to the kernel's address map.
81  *	map must be "kernel_map" below.
82  */
83 
84 vm_offset_t kmem_alloc_pageable(map, size)
85 	vm_map_t		map;
86 	register vm_size_t	size;
87 {
88 	vm_offset_t		addr;
89 	register int		result;
90 
91 #if	0
92 	if (map != kernel_map)
93 		panic("kmem_alloc_pageable: not called with kernel_map");
94 #endif
95 
96 	size = round_page(size);
97 
98 	addr = vm_map_min(map);
99 	result = vm_map_find(map, NULL, (vm_offset_t) 0,
100 				&addr, size, TRUE);
101 	if (result != KERN_SUCCESS) {
102 		return(0);
103 	}
104 
105 	return(addr);
106 }
107 
108 /*
109  *	Allocate wired-down memory in the kernel's address map
110  *	or a submap.
111  */
112 vm_offset_t kmem_alloc(map, size)
113 	register vm_map_t	map;
114 	register vm_size_t	size;
115 {
116 	vm_offset_t		addr;
117 	register vm_offset_t	offset;
118 	extern vm_object_t	kernel_object;
119 	vm_offset_t		i;
120 
121 	size = round_page(size);
122 
123 	/*
124 	 *	Use the kernel object for wired-down kernel pages.
125 	 *	Assume that no region of the kernel object is
126 	 *	referenced more than once.
127 	 */
128 
129 	/*
130 	 * Locate sufficient space in the map.  This will give us the
131 	 * final virtual address for the new memory, and thus will tell
132 	 * us the offset within the kernel map.
133 	 */
134 	vm_map_lock(map);
135 	if (vm_map_findspace(map, 0, size, &addr)) {
136 		vm_map_unlock(map);
137 		return (0);
138 	}
139 	offset = addr - VM_MIN_KERNEL_ADDRESS;
140 	vm_object_reference(kernel_object);
141 	vm_map_insert(map, kernel_object, offset, addr, addr + size);
142 	vm_map_unlock(map);
143 
144 	/*
145 	 *	Guarantee that there are pages already in this object
146 	 *	before calling vm_map_pageable.  This is to prevent the
147 	 *	following scenario:
148 	 *
149 	 *		1) Threads have swapped out, so that there is a
150 	 *		   pager for the kernel_object.
151 	 *		2) The kmsg zone is empty, and so we are kmem_allocing
152 	 *		   a new page for it.
153 	 *		3) vm_map_pageable calls vm_fault; there is no page,
154 	 *		   but there is a pager, so we call
155 	 *		   pager_data_request.  But the kmsg zone is empty,
156 	 *		   so we must kmem_alloc.
157 	 *		4) goto 1
158 	 *		5) Even if the kmsg zone is not empty: when we get
159 	 *		   the data back from the pager, it will be (very
160 	 *		   stale) non-zero data.  kmem_alloc is defined to
161 	 *		   return zero-filled memory.
162 	 *
163 	 *	We're intentionally not activating the pages we allocate
164 	 *	to prevent a race with page-out.  vm_map_pageable will wire
165 	 *	the pages.
166 	 */
167 
168 	vm_object_lock(kernel_object);
169 	for (i = 0 ; i < size; i+= PAGE_SIZE) {
170 		vm_page_t	mem;
171 
172 		while ((mem = vm_page_alloc(kernel_object, offset+i)) == NULL) {
173 			vm_object_unlock(kernel_object);
174 			VM_WAIT;
175 			vm_object_lock(kernel_object);
176 		}
177 		vm_page_zero_fill(mem);
178 		mem->flags &= ~PG_BUSY;
179 	}
180 	vm_object_unlock(kernel_object);
181 
182 	/*
183 	 *	And finally, mark the data as non-pageable.
184 	 */
185 
186 	(void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE);
187 
188 	/*
189 	 *	Try to coalesce the map
190 	 */
191 
192 	vm_map_simplify(map, addr);
193 
194 	return(addr);
195 }
196 
197 /*
198  *	kmem_free:
199  *
200  *	Release a region of kernel virtual memory allocated
201  *	with kmem_alloc, and return the physical pages
202  *	associated with that region.
203  */
204 void kmem_free(map, addr, size)
205 	vm_map_t		map;
206 	register vm_offset_t	addr;
207 	vm_size_t		size;
208 {
209 	(void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
210 }
211 
212 /*
213  *	kmem_suballoc:
214  *
215  *	Allocates a map to manage a subrange
216  *	of the kernel virtual address space.
217  *
218  *	Arguments are as follows:
219  *
220  *	parent		Map to take range from
221  *	size		Size of range to find
222  *	min, max	Returned endpoints of map
223  *	pageable	Can the region be paged
224  */
225 vm_map_t kmem_suballoc(parent, min, max, size, pageable)
226 	register vm_map_t	parent;
227 	vm_offset_t		*min, *max;
228 	register vm_size_t	size;
229 	boolean_t		pageable;
230 {
231 	register int	ret;
232 	vm_map_t	result;
233 
234 	size = round_page(size);
235 
236 	*min = (vm_offset_t) vm_map_min(parent);
237 	ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
238 				min, size, TRUE);
239 	if (ret != KERN_SUCCESS) {
240 		printf("kmem_suballoc: bad status return of %d.\n", ret);
241 		panic("kmem_suballoc");
242 	}
243 	*max = *min + size;
244 	pmap_reference(vm_map_pmap(parent));
245 	result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable);
246 	if (result == NULL)
247 		panic("kmem_suballoc: cannot create submap");
248 	if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
249 		panic("kmem_suballoc: unable to change range to submap");
250 	return(result);
251 }
252 
253 /*
254  * Allocate wired-down memory in the kernel's address map for the higher
255  * level kernel memory allocator (kern/kern_malloc.c).  We cannot use
256  * kmem_alloc() because we may need to allocate memory at interrupt
257  * level where we cannot block (canwait == FALSE).
258  *
259  * This routine has its own private kernel submap (kmem_map) and object
260  * (kmem_object).  This, combined with the fact that only malloc uses
261  * this routine, ensures that we will never block in map or object waits.
262  *
263  * Note that this still only works in a uni-processor environment and
264  * when called at splhigh().
265  *
266  * We don't worry about expanding the map (adding entries) since entries
267  * for wired maps are statically allocated.
268  */
269 vm_offset_t
270 kmem_malloc(map, size, canwait)
271 	register vm_map_t	map;
272 	register vm_size_t	size;
273 	boolean_t		canwait;
274 {
275 	register vm_offset_t	offset, i;
276 	vm_map_entry_t		entry;
277 	vm_offset_t		addr;
278 	vm_page_t		m;
279 	extern vm_object_t	kmem_object;
280 
281 	if (map != kmem_map && map != mb_map)
282 		panic("kern_malloc_alloc: map != {kmem,mb}_map");
283 
284 	size = round_page(size);
285 	addr = vm_map_min(map);
286 
287 	/*
288 	 * Locate sufficient space in the map.  This will give us the
289 	 * final virtual address for the new memory, and thus will tell
290 	 * us the offset within the kernel map.
291 	 */
292 	vm_map_lock(map);
293 	if (vm_map_findspace(map, 0, size, &addr)) {
294 		vm_map_unlock(map);
295 #if 0
296 		if (canwait)		/* XXX  should wait */
297 			panic("kmem_malloc: %s too small",
298 			    map == kmem_map ? "kmem_map" : "mb_map");
299 #endif
300 		if (canwait)
301 			panic("kmem_malloc: map too small");
302 		return (0);
303 	}
304 	offset = addr - vm_map_min(kmem_map);
305 	vm_object_reference(kmem_object);
306 	vm_map_insert(map, kmem_object, offset, addr, addr + size);
307 
308 	/*
309 	 * If we can wait, just mark the range as wired
310 	 * (will fault pages as necessary).
311 	 */
312 	if (canwait) {
313 		vm_map_unlock(map);
314 		(void) vm_map_pageable(map, (vm_offset_t) addr, addr + size,
315 				       FALSE);
316 		vm_map_simplify(map, addr);
317 		return(addr);
318 	}
319 
320 	/*
321 	 * If we cannot wait then we must allocate all memory up front,
322 	 * pulling it off the active queue to prevent pageout.
323 	 */
324 	vm_object_lock(kmem_object);
325 	for (i = 0; i < size; i += PAGE_SIZE) {
326 		m = vm_page_alloc(kmem_object, offset + i);
327 
328 		/*
329 		 * Ran out of space, free everything up and return.
330 		 * Don't need to lock page queues here as we know
331 		 * that the pages we got aren't on any queues.
332 		 */
333 		if (m == NULL) {
334 			while (i != 0) {
335 				i -= PAGE_SIZE;
336 				m = vm_page_lookup(kmem_object, offset + i);
337 				vm_page_free(m);
338 			}
339 			vm_object_unlock(kmem_object);
340 			vm_map_delete(map, addr, addr + size);
341 			vm_map_unlock(map);
342 			return(0);
343 		}
344 #if 0
345 		vm_page_zero_fill(m);
346 #endif
347 		m->flags &= ~PG_BUSY;
348 	}
349 	vm_object_unlock(kmem_object);
350 
351 	/*
352 	 * Mark map entry as non-pageable.
353 	 * Assert: vm_map_insert() will never be able to extend the previous
354 	 * entry so there will be a new entry exactly corresponding to this
355 	 * address range and it will have wired_count == 0.
356 	 */
357 	if (!vm_map_lookup_entry(map, addr, &entry) ||
358 	    entry->start != addr || entry->end != addr + size ||
359 	    entry->wired_count)
360 		panic("kmem_malloc: entry not found or misaligned");
361 	entry->wired_count++;
362 
363 	/*
364 	 * Loop thru pages, entering them in the pmap.
365 	 * (We cannot add them to the wired count without
366 	 * wrapping the vm_page_queue_lock in splimp...)
367 	 */
368 	for (i = 0; i < size; i += PAGE_SIZE) {
369 		vm_object_lock(kmem_object);
370 		m = vm_page_lookup(kmem_object, offset + i);
371 		vm_object_unlock(kmem_object);
372 		pmap_enter(map->pmap, addr + i, VM_PAGE_TO_PHYS(m),
373 			   VM_PROT_DEFAULT, TRUE);
374 	}
375 	vm_map_unlock(map);
376 
377 	vm_map_simplify(map, addr);
378 	return(addr);
379 }
380 
381 /*
382  *	kmem_alloc_wait
383  *
384  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
385  *	has no room, the caller sleeps waiting for more memory in the submap.
386  *
387  */
388 vm_offset_t 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,
399 		 * use the map's lock 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 		assert_wait((int)map, TRUE);
410 		vm_map_unlock(map);
411 		thread_block("kmaw");
412 	}
413 	vm_map_insert(map, NULL, (vm_offset_t)0, addr, addr + size);
414 	vm_map_unlock(map);
415 	return (addr);
416 }
417 
418 /*
419  *	kmem_free_wakeup
420  *
421  *	Returns memory to a submap of the kernel, and wakes up any threads
422  *	waiting for memory in that map.
423  */
424 void	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 	thread_wakeup((int)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 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);
452 	/* ... and ending with the completion of the above `insert' */
453 	vm_map_unlock(m);
454 }
455