xref: /dragonfly/sys/vm/vm_kern.c (revision aa8d5dcb) 
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: src/sys/vm/vm_kern.c,v 1.61.2.2 2002/03/12 18:25:26 tegge Exp $
65  * $DragonFly: src/sys/vm/vm_kern.c,v 1.15 2004/03/01 06:33:24 dillon Exp $
66  */
67 
68 /*
69  *	Kernel memory management.
70  */
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/proc.h>
75 #include <sys/malloc.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <sys/lock.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_kern.h>
86 #include <vm/vm_extern.h>
87 
88 vm_map_t kernel_map=0;
89 vm_map_t exec_map=0;
90 vm_map_t clean_map=0;
91 vm_map_t buffer_map=0;
92 vm_map_t mb_map=0;
93 int mb_map_full=0;
94 
95 /*
96  *	kmem_alloc_pageable:
97  *
98  *	Allocate pageable memory to the kernel's address map.
99  *	"map" must be kernel_map or a submap of kernel_map.
100  */
101 vm_offset_t
102 kmem_alloc_pageable(vm_map_t map, vm_size_t size)
103 {
104 	vm_offset_t addr;
105 	int result;
106 
107 	size = round_page(size);
108 	addr = vm_map_min(map);
109 	result = vm_map_find(map, NULL, (vm_offset_t) 0,
110 	    &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
111 	if (result != KERN_SUCCESS) {
112 		return (0);
113 	}
114 	return (addr);
115 }
116 
117 /*
118  *	kmem_alloc_nofault:
119  *
120  *	Same as kmem_alloc_pageable, except that it create a nofault entry.
121  */
122 vm_offset_t
123 kmem_alloc_nofault(vm_map_t map, vm_size_t size)
124 {
125 	vm_offset_t addr;
126 	int result;
127 
128 	size = round_page(size);
129 	addr = vm_map_min(map);
130 	result = vm_map_find(map, NULL, (vm_offset_t) 0,
131 	    &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
132 	if (result != KERN_SUCCESS) {
133 		return (0);
134 	}
135 	return (addr);
136 }
137 
138 /*
139  *	Allocate wired-down memory in the kernel's address map
140  *	or a submap.
141  */
142 vm_offset_t
143 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags)
144 {
145 	vm_offset_t addr;
146 	vm_offset_t offset;
147 	vm_offset_t i;
148 	int count;
149 
150 	size = round_page(size);
151 
152 	if (kmflags & KM_KRESERVE)
153 		count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
154 	else
155 		count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
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 	 * Locate sufficient space in the map.  This will give us the final
162 	 * virtual address for the new memory, and thus will tell us the
163 	 * offset within the kernel map.
164 	 */
165 	vm_map_lock(map);
166 	if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
167 		vm_map_unlock(map);
168 		if (kmflags & KM_KRESERVE)
169 			vm_map_entry_krelease(count);
170 		else
171 			vm_map_entry_release(count);
172 		return (0);
173 	}
174 	offset = addr - VM_MIN_KERNEL_ADDRESS;
175 	vm_object_reference(kernel_object);
176 	vm_map_insert(map, &count,
177 		kernel_object, offset, addr, addr + size,
178 		VM_PROT_ALL, VM_PROT_ALL, 0);
179 	vm_map_unlock(map);
180 	if (kmflags & KM_KRESERVE)
181 		vm_map_entry_krelease(count);
182 	else
183 		vm_map_entry_release(count);
184 
185 	/*
186 	 * Guarantee that there are pages already in this object before
187 	 * calling vm_map_wire.  This is to prevent the following
188 	 * scenario:
189 	 *
190 	 * 1) Threads have swapped out, so that there is a pager for the
191 	 * kernel_object. 2) The kmsg zone is empty, and so we are
192 	 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
193 	 * there is no page, but there is a pager, so we call
194 	 * pager_data_request.  But the kmsg zone is empty, so we must
195 	 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
196 	 * we get the data back from the pager, it will be (very stale)
197 	 * non-zero data.  kmem_alloc is defined to return zero-filled memory.
198 	 *
199 	 * We're intentionally not activating the pages we allocate to prevent a
200 	 * race with page-out.  vm_map_wire will wire the pages.
201 	 */
202 
203 	for (i = 0; i < size; i += PAGE_SIZE) {
204 		vm_page_t mem;
205 
206 		mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
207 			    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
208 		if ((mem->flags & PG_ZERO) == 0)
209 			vm_page_zero_fill(mem);
210 		mem->valid = VM_PAGE_BITS_ALL;
211 		vm_page_flag_clear(mem, PG_ZERO);
212 		vm_page_wakeup(mem);
213 	}
214 
215 	/*
216 	 * And finally, mark the data as non-pageable.
217 	 */
218 
219 	(void) vm_map_wire(map, (vm_offset_t) addr, addr + size, kmflags);
220 
221 	return (addr);
222 }
223 
224 /*
225  *	kmem_free:
226  *
227  *	Release a region of kernel virtual memory allocated
228  *	with kmem_alloc, and return the physical pages
229  *	associated with that region.
230  *
231  *	This routine may not block on kernel maps.
232  */
233 void
234 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size)
235 {
236 	(void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
237 }
238 
239 /*
240  *	kmem_suballoc:
241  *
242  *	Allocates a map to manage a subrange
243  *	of the kernel virtual address space.
244  *
245  *	Arguments are as follows:
246  *
247  *	parent		Map to take range from
248  *	size		Size of range to find
249  *	min, max	Returned endpoints of map
250  *	pageable	Can the region be paged
251  */
252 vm_map_t
253 kmem_suballoc(parent, min, max, size)
254 	vm_map_t parent;
255 	vm_offset_t *min, *max;
256 	vm_size_t size;
257 {
258 	int ret;
259 	vm_map_t result;
260 
261 	size = round_page(size);
262 
263 	*min = (vm_offset_t) vm_map_min(parent);
264 	ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
265 	    min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
266 	if (ret != KERN_SUCCESS) {
267 		printf("kmem_suballoc: bad status return of %d.\n", ret);
268 		panic("kmem_suballoc");
269 	}
270 	*max = *min + size;
271 	pmap_reference(vm_map_pmap(parent));
272 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
273 	if (result == NULL)
274 		panic("kmem_suballoc: cannot create submap");
275 	if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
276 		panic("kmem_suballoc: unable to change range to submap");
277 	return (result);
278 }
279 
280 /*
281  *	kmem_malloc:
282  *
283  * 	Allocate wired-down memory in the kernel's address map for the higher
284  * 	level kernel memory allocator (kern/kern_malloc.c).  We cannot use
285  * 	kmem_alloc() because we may need to allocate memory at interrupt
286  * 	level where we cannot block (canwait == FALSE).
287  *
288  * 	We don't worry about expanding the map (adding entries) since entries
289  * 	for wired maps are statically allocated.
290  *
291  *	NOTE:  Please see kmem_slab_alloc() for a better explanation of the
292  *	M_* flags.
293  */
294 vm_offset_t
295 kmem_malloc(vm_map_t map, vm_size_t size, int flags)
296 {
297 	vm_offset_t offset, i;
298 	vm_map_entry_t entry;
299 	vm_offset_t addr;
300 	vm_page_t m;
301 	int count;
302 	thread_t td;
303 	int wanted_reserve;
304 
305 	if (map != kernel_map && map != mb_map)
306 		panic("kmem_malloc: map != {kmem,mb}_map");
307 
308 	size = round_page(size);
309 	addr = vm_map_min(map);
310 
311 	/*
312 	 * Locate sufficient space in the map.  This will give us the final
313 	 * virtual address for the new memory, and thus will tell us the
314 	 * offset within the kernel map.
315 	 */
316 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
317 	vm_map_lock(map);
318 	if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
319 		vm_map_unlock(map);
320 		vm_map_entry_release(count);
321 		if (map == mb_map) {
322 			mb_map_full = TRUE;
323 			printf("Out of mbuf clusters - adjust NMBCLUSTERS or increase maxusers!\n");
324 			return (0);
325 		}
326 		if ((flags & (M_RNOWAIT|M_NULLOK)) == 0 ||
327 		    (flags & (M_FAILSAFE|M_NULLOK)) == M_FAILSAFE
328 		) {
329 			panic("kmem_malloc(%ld): kernel_map too small: "
330 				"%ld total allocated",
331 				(long)size, (long)map->size);
332 		}
333 		return (0);
334 	}
335 	offset = addr - VM_MIN_KERNEL_ADDRESS;
336 	vm_object_reference(kmem_object);
337 	vm_map_insert(map, &count,
338 		kmem_object, offset, addr, addr + size,
339 		VM_PROT_ALL, VM_PROT_ALL, 0);
340 
341 	td = curthread;
342 	wanted_reserve = 0;
343 
344 	for (i = 0; i < size; i += PAGE_SIZE) {
345 		int vmflags;
346 
347 		vmflags = VM_ALLOC_SYSTEM;	/* XXX M_USE_RESERVE? */
348 		if ((flags & (M_WAITOK|M_RNOWAIT)) == 0)
349 			printf("kmem_malloc: bad flags %08x (%p)\n", flags, ((int **)&map)[-1]);
350 		if (flags & M_USE_INTERRUPT_RESERVE)
351 			vmflags |= VM_ALLOC_INTERRUPT;
352 		if (flags & (M_FAILSAFE|M_WAITOK)) {
353 			if (td->td_preempted) {
354 				wanted_reserve = 1;
355 			} else {
356 				vmflags |= VM_ALLOC_NORMAL;
357 				wanted_reserve = 0;
358 			}
359 		}
360 
361 		m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), vmflags);
362 
363 		/*
364 		 * Ran out of space, free everything up and return. Don't need
365 		 * to lock page queues here as we know that the pages we got
366 		 * aren't on any queues.
367 		 *
368 		 * If M_WAITOK or M_FAILSAFE is set we can yield or block.
369 		 */
370 		if (m == NULL) {
371 			if (flags & (M_FAILSAFE|M_WAITOK)) {
372 				if (wanted_reserve) {
373 					if (flags & M_FAILSAFE)
374 						printf("kmem_malloc: no memory, try failsafe\n");
375 					vm_map_unlock(map);
376 					lwkt_yield();
377 					vm_map_lock(map);
378 				} else {
379 					if (flags & M_FAILSAFE)
380 						printf("kmem_malloc: no memory, block even though we shouldn't\n");
381 					vm_map_unlock(map);
382 					VM_WAIT;
383 					vm_map_lock(map);
384 				}
385 				i -= PAGE_SIZE;	/* retry */
386 				continue;
387 			}
388 			/*
389 			 * Free the pages before removing the map entry.
390 			 * They are already marked busy.  Calling
391 			 * vm_map_delete before the pages has been freed or
392 			 * unbusied will cause a deadlock.
393 			 */
394 			while (i != 0) {
395 				i -= PAGE_SIZE;
396 				m = vm_page_lookup(kmem_object,
397 						   OFF_TO_IDX(offset + i));
398 				vm_page_free(m);
399 			}
400 			vm_map_delete(map, addr, addr + size, &count);
401 			vm_map_unlock(map);
402 			vm_map_entry_release(count);
403 			return (0);
404 		}
405 		vm_page_flag_clear(m, PG_ZERO);
406 		m->valid = VM_PAGE_BITS_ALL;
407 	}
408 
409 	/*
410 	 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
411 	 * be able to extend the previous entry so there will be a new entry
412 	 * exactly corresponding to this address range and it will have
413 	 * wired_count == 0.
414 	 */
415 	if (!vm_map_lookup_entry(map, addr, &entry) ||
416 	    entry->start != addr || entry->end != addr + size ||
417 	    entry->wired_count != 0)
418 		panic("kmem_malloc: entry not found or misaligned");
419 	entry->wired_count = 1;
420 
421 	vm_map_simplify_entry(map, entry, &count);
422 
423 	/*
424 	 * Loop thru pages, entering them in the pmap. (We cannot add them to
425 	 * the wired count without wrapping the vm_page_queue_lock in
426 	 * splimp...)
427 	 */
428 	for (i = 0; i < size; i += PAGE_SIZE) {
429 		m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
430 		vm_page_wire(m);
431 		vm_page_wakeup(m);
432 		/*
433 		 * Because this is kernel_pmap, this call will not block.
434 		 */
435 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
436 		vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED);
437 	}
438 	vm_map_unlock(map);
439 	vm_map_entry_release(count);
440 
441 	return (addr);
442 }
443 
444 /*
445  *	kmem_alloc_wait:
446  *
447  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
448  *	has no room, the caller sleeps waiting for more memory in the submap.
449  *
450  *	This routine may block.
451  */
452 
453 vm_offset_t
454 kmem_alloc_wait(vm_map_t map, vm_size_t size)
455 {
456 	vm_offset_t addr;
457 	int count;
458 
459 	size = round_page(size);
460 
461 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
462 
463 	for (;;) {
464 		/*
465 		 * To make this work for more than one map, use the map's lock
466 		 * to lock out sleepers/wakers.
467 		 */
468 		vm_map_lock(map);
469 		if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) == 0)
470 			break;
471 		/* no space now; see if we can ever get space */
472 		if (vm_map_max(map) - vm_map_min(map) < size) {
473 			vm_map_entry_release(count);
474 			vm_map_unlock(map);
475 			return (0);
476 		}
477 		vm_map_unlock(map);
478 		tsleep(map, 0, "kmaw", 0);
479 	}
480 	vm_map_insert(map, &count,
481 		    NULL, (vm_offset_t) 0,
482 		    addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
483 	vm_map_unlock(map);
484 	vm_map_entry_release(count);
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  */
494 void
495 kmem_free_wakeup(map, addr, size)
496 	vm_map_t map;
497 	vm_offset_t addr;
498 	vm_size_t size;
499 {
500 	int count;
501 
502 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
503 	vm_map_lock(map);
504 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count);
505 	wakeup(map);
506 	vm_map_unlock(map);
507 	vm_map_entry_release(count);
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  *
518  *	Depend on the zalloc bootstrap cache to get our vm_map_entry_t.
519  */
520 void
521 kmem_init(vm_offset_t start, vm_offset_t end)
522 {
523 	vm_map_t m;
524 	int count;
525 
526 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
527 	vm_map_lock(m);
528 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
529 	kernel_map = m;
530 	kernel_map->system_map = 1;
531 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
532 	(void) vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
533 	    VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
534 	/* ... and ending with the completion of the above `insert' */
535 	vm_map_unlock(m);
536 	vm_map_entry_release(count);
537 }
538 
539