xref: /dflybsd-src/sys/vm/vm_kern.c (revision 1f7ab7c9fc18f47a2f16dc45b13dee254c603ce7)
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 1.23 2006/09/11 20:25:31 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 
93 /*
94  *	kmem_alloc_pageable:
95  *
96  *	Allocate pageable memory to the kernel's address map.
97  *	"map" must be kernel_map or a submap of kernel_map.
98  */
99 vm_offset_t
100 kmem_alloc_pageable(vm_map_t map, 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, (vm_offset_t) 0,
108 			     &addr, size,
109 			     TRUE,
110 			     VM_MAPTYPE_NORMAL,
111 			     VM_PROT_ALL, VM_PROT_ALL,
112 			     0);
113 	if (result != KERN_SUCCESS) {
114 		return (0);
115 	}
116 	return (addr);
117 }
118 
119 /*
120  *	kmem_alloc_nofault:
121  *
122  *	Same as kmem_alloc_pageable, except that it create a nofault entry.
123  */
124 vm_offset_t
125 kmem_alloc_nofault(vm_map_t map, vm_size_t size)
126 {
127 	vm_offset_t addr;
128 	int result;
129 
130 	size = round_page(size);
131 	addr = vm_map_min(map);
132 	result = vm_map_find(map, NULL, (vm_offset_t) 0,
133 			     &addr, size,
134 			     TRUE,
135 			     VM_MAPTYPE_NORMAL,
136 			     VM_PROT_ALL, VM_PROT_ALL,
137 			     MAP_NOFAULT);
138 	if (result != KERN_SUCCESS) {
139 		return (0);
140 	}
141 	return (addr);
142 }
143 
144 /*
145  *	Allocate wired-down memory in the kernel's address map
146  *	or a submap.
147  */
148 vm_offset_t
149 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags)
150 {
151 	vm_offset_t addr;
152 	vm_offset_t offset;
153 	vm_offset_t i;
154 	int count;
155 
156 	size = round_page(size);
157 
158 	if (kmflags & KM_KRESERVE)
159 		count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
160 	else
161 		count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
162 
163 	/*
164 	 * Use the kernel object for wired-down kernel pages. Assume that no
165 	 * region of the kernel object is referenced more than once.
166 	 *
167 	 * Locate sufficient space in the map.  This will give us the final
168 	 * virtual address for the new memory, and thus will tell us the
169 	 * offset within the kernel map.
170 	 */
171 	vm_map_lock(map);
172 	if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
173 		vm_map_unlock(map);
174 		if (kmflags & KM_KRESERVE)
175 			vm_map_entry_krelease(count);
176 		else
177 			vm_map_entry_release(count);
178 		return (0);
179 	}
180 	offset = addr - VM_MIN_KERNEL_ADDRESS;
181 	vm_object_reference(kernel_object);
182 	vm_map_insert(map, &count,
183 		      kernel_object, offset, addr, addr + size,
184 		      VM_MAPTYPE_NORMAL,
185 		      VM_PROT_ALL, VM_PROT_ALL,
186 		      0);
187 	vm_map_unlock(map);
188 	if (kmflags & KM_KRESERVE)
189 		vm_map_entry_krelease(count);
190 	else
191 		vm_map_entry_release(count);
192 
193 	/*
194 	 * Guarantee that there are pages already in this object before
195 	 * calling vm_map_wire.  This is to prevent the following
196 	 * scenario:
197 	 *
198 	 * 1) Threads have swapped out, so that there is a pager for the
199 	 * kernel_object. 2) The kmsg zone is empty, and so we are
200 	 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
201 	 * there is no page, but there is a pager, so we call
202 	 * pager_data_request.  But the kmsg zone is empty, so we must
203 	 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
204 	 * we get the data back from the pager, it will be (very stale)
205 	 * non-zero data.  kmem_alloc is defined to return zero-filled memory.
206 	 *
207 	 * We're intentionally not activating the pages we allocate to prevent a
208 	 * race with page-out.  vm_map_wire will wire the pages.
209 	 */
210 
211 	for (i = 0; i < size; i += PAGE_SIZE) {
212 		vm_page_t mem;
213 
214 		mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
215 			    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
216 		if ((mem->flags & PG_ZERO) == 0)
217 			vm_page_zero_fill(mem);
218 		mem->valid = VM_PAGE_BITS_ALL;
219 		vm_page_flag_clear(mem, PG_ZERO);
220 		vm_page_wakeup(mem);
221 	}
222 
223 	/*
224 	 * And finally, mark the data as non-pageable.
225 	 */
226 
227 	vm_map_wire(map, (vm_offset_t) addr, addr + size, kmflags);
228 
229 	return (addr);
230 }
231 
232 /*
233  *	kmem_free:
234  *
235  *	Release a region of kernel virtual memory allocated
236  *	with kmem_alloc, and return the physical pages
237  *	associated with that region.
238  *
239  *	This routine may not block on kernel maps.
240  */
241 void
242 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size)
243 {
244 	vm_map_remove(map, trunc_page(addr), round_page(addr + size));
245 }
246 
247 /*
248  *	kmem_suballoc:
249  *
250  *	Allocates a map to manage a subrange
251  *	of the kernel virtual address space.
252  *
253  *	Arguments are as follows:
254  *
255  *	parent		Map to take range from
256  *	size		Size of range to find
257  *	min, max	Returned endpoints of map
258  *	pageable	Can the region be paged
259  */
260 vm_map_t
261 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
262     vm_size_t size)
263 {
264 	int ret;
265 	vm_map_t result;
266 
267 	size = round_page(size);
268 
269 	*min = (vm_offset_t) vm_map_min(parent);
270 	ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
271 			  min, size,
272 			  TRUE,
273 			  VM_MAPTYPE_UNSPECIFIED,
274 			  VM_PROT_ALL, VM_PROT_ALL,
275 			  0);
276 	if (ret != KERN_SUCCESS) {
277 		printf("kmem_suballoc: bad status return of %d.\n", ret);
278 		panic("kmem_suballoc");
279 	}
280 	*max = *min + size;
281 	pmap_reference(vm_map_pmap(parent));
282 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
283 	if (result == NULL)
284 		panic("kmem_suballoc: cannot create submap");
285 	if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
286 		panic("kmem_suballoc: unable to change range to submap");
287 	return (result);
288 }
289 
290 /*
291  *	kmem_malloc:
292  *
293  * 	Allocate wired-down memory in the kernel's address map for the higher
294  * 	level kernel memory allocator (kern/kern_malloc.c).  We cannot use
295  * 	kmem_alloc() because we may need to allocate memory at interrupt
296  * 	level where we cannot block (canwait == FALSE).
297  *
298  * 	We don't worry about expanding the map (adding entries) since entries
299  * 	for wired maps are statically allocated.
300  *
301  *	NOTE:  Please see kmem_slab_alloc() for a better explanation of the
302  *	M_* flags.
303  */
304 vm_offset_t
305 kmem_malloc(vm_map_t map, vm_size_t size, int flags)
306 {
307 	vm_offset_t offset, i;
308 	vm_map_entry_t entry;
309 	vm_offset_t addr;
310 	vm_page_t m;
311 	int count, vmflags, wanted_reserve;
312 	thread_t td;
313 
314 	if (map != kernel_map)
315 		panic("kmem_malloc: map != kernel_map");
316 
317 	size = round_page(size);
318 	addr = vm_map_min(map);
319 
320 	/*
321 	 * Locate sufficient space in the map.  This will give us the final
322 	 * virtual address for the new memory, and thus will tell us the
323 	 * offset within the kernel map.  If we are unable to allocate space
324 	 * and neither RNOWAIT or NULLOK is set, we panic.
325 	 */
326 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
327 	vm_map_lock(map);
328 	if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
329 		vm_map_unlock(map);
330 		vm_map_entry_release(count);
331 		if ((flags & M_NULLOK) == 0) {
332 			panic("kmem_malloc(%ld): kernel_map too small: "
333 				"%ld total allocated",
334 				(long)size, (long)map->size);
335 		}
336 		return (0);
337 	}
338 	offset = addr - VM_MIN_KERNEL_ADDRESS;
339 	vm_object_reference(kmem_object);
340 	vm_map_insert(map, &count,
341 		      kmem_object, offset, addr, addr + size,
342 		      VM_MAPTYPE_NORMAL,
343 		      VM_PROT_ALL, VM_PROT_ALL,
344 		      0);
345 
346 	td = curthread;
347 	wanted_reserve = 0;
348 
349 	vmflags = VM_ALLOC_SYSTEM;	/* XXX M_USE_RESERVE? */
350 	if ((flags & (M_WAITOK|M_RNOWAIT)) == 0)
351 		panic("kmem_malloc: bad flags %08x (%p)\n", flags, ((int **)&map)[-1]);
352 	if (flags & M_USE_INTERRUPT_RESERVE)
353 		vmflags |= VM_ALLOC_INTERRUPT;
354 
355 	for (i = 0; i < size; i += PAGE_SIZE) {
356 		/*
357 		 * Only allocate PQ_CACHE pages for M_WAITOK requests and
358 		 * then only if we are not preempting.
359 		 */
360 		if (flags & M_WAITOK) {
361 			if (td->td_preempted) {
362 				vmflags &= ~VM_ALLOC_NORMAL;
363 				wanted_reserve = 1;
364 			} else {
365 				vmflags |= VM_ALLOC_NORMAL;
366 				wanted_reserve = 0;
367 			}
368 		}
369 
370 		m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), vmflags);
371 
372 		/*
373 		 * Ran out of space, free everything up and return.  Don't need
374 		 * to lock page queues here as we know that the pages we got
375 		 * aren't on any queues.
376 		 *
377 		 * If M_WAITOK is set we can yield or block.
378 		 */
379 		if (m == NULL) {
380 			if (flags & M_WAITOK) {
381 				if (wanted_reserve) {
382 					vm_map_unlock(map);
383 					lwkt_yield();
384 					vm_map_lock(map);
385 				} else {
386 					vm_map_unlock(map);
387 					vm_wait();
388 					vm_map_lock(map);
389 				}
390 				i -= PAGE_SIZE;	/* retry */
391 				continue;
392 			}
393 			/*
394 			 * Free the pages before removing the map entry.
395 			 * They are already marked busy.  Calling
396 			 * vm_map_delete before the pages has been freed or
397 			 * unbusied will cause a deadlock.
398 			 */
399 			while (i != 0) {
400 				i -= PAGE_SIZE;
401 				m = vm_page_lookup(kmem_object,
402 						   OFF_TO_IDX(offset + i));
403 				vm_page_free(m);
404 			}
405 			vm_map_delete(map, addr, addr + size, &count);
406 			vm_map_unlock(map);
407 			vm_map_entry_release(count);
408 			return (0);
409 		}
410 		vm_page_flag_clear(m, PG_ZERO);
411 		m->valid = VM_PAGE_BITS_ALL;
412 	}
413 
414 	/*
415 	 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
416 	 * be able to extend the previous entry so there will be a new entry
417 	 * exactly corresponding to this address range and it will have
418 	 * wired_count == 0.
419 	 */
420 	if (!vm_map_lookup_entry(map, addr, &entry) ||
421 	    entry->start != addr || entry->end != addr + size ||
422 	    entry->wired_count != 0)
423 		panic("kmem_malloc: entry not found or misaligned");
424 	entry->wired_count = 1;
425 
426 	vm_map_simplify_entry(map, entry, &count);
427 
428 	/*
429 	 * Loop thru pages, entering them in the pmap. (We cannot add them to
430 	 * the wired count without wrapping the vm_page_queue_lock in
431 	 * splimp...)
432 	 */
433 	for (i = 0; i < size; i += PAGE_SIZE) {
434 		m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
435 		vm_page_wire(m);
436 		vm_page_wakeup(m);
437 		/*
438 		 * Because this is kernel_pmap, this call will not block.
439 		 */
440 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
441 		vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED);
442 	}
443 	vm_map_unlock(map);
444 	vm_map_entry_release(count);
445 
446 	return (addr);
447 }
448 
449 /*
450  *	kmem_alloc_wait:
451  *
452  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
453  *	has no room, the caller sleeps waiting for more memory in the submap.
454  *
455  *	This routine may block.
456  */
457 
458 vm_offset_t
459 kmem_alloc_wait(vm_map_t map, vm_size_t size)
460 {
461 	vm_offset_t addr;
462 	int count;
463 
464 	size = round_page(size);
465 
466 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
467 
468 	for (;;) {
469 		/*
470 		 * To make this work for more than one map, use the map's lock
471 		 * to lock out sleepers/wakers.
472 		 */
473 		vm_map_lock(map);
474 		if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) == 0)
475 			break;
476 		/* no space now; see if we can ever get space */
477 		if (vm_map_max(map) - vm_map_min(map) < size) {
478 			vm_map_entry_release(count);
479 			vm_map_unlock(map);
480 			return (0);
481 		}
482 		vm_map_unlock(map);
483 		tsleep(map, 0, "kmaw", 0);
484 	}
485 	vm_map_insert(map, &count,
486 		      NULL, (vm_offset_t) 0,
487 		      addr, addr + size,
488 		      VM_MAPTYPE_NORMAL,
489 		      VM_PROT_ALL, VM_PROT_ALL,
490 		      0);
491 	vm_map_unlock(map);
492 	vm_map_entry_release(count);
493 	return (addr);
494 }
495 
496 /*
497  *	kmem_free_wakeup:
498  *
499  *	Returns memory to a submap of the kernel, and wakes up any processes
500  *	waiting for memory in that map.
501  */
502 void
503 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
504 {
505 	int count;
506 
507 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
508 	vm_map_lock(map);
509 	vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count);
510 	wakeup(map);
511 	vm_map_unlock(map);
512 	vm_map_entry_release(count);
513 }
514 
515 /*
516  * 	kmem_init:
517  *
518  *	Create the kernel map; insert a mapping covering kernel text,
519  *	data, bss, and all space allocated thus far (`boostrap' data).  The
520  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
521  *	`start' as allocated, and the range between `start' and `end' as free.
522  *
523  *	Depend on the zalloc bootstrap cache to get our vm_map_entry_t.
524  */
525 void
526 kmem_init(vm_offset_t start, vm_offset_t end)
527 {
528 	vm_map_t m;
529 	int count;
530 
531 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
532 	vm_map_lock(m);
533 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
534 	kernel_map = m;
535 	kernel_map->system_map = 1;
536 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
537 	vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
538 		      VM_MIN_KERNEL_ADDRESS, start,
539 		      VM_MAPTYPE_NORMAL,
540 		      VM_PROT_ALL, VM_PROT_ALL,
541 		      0);
542 	/* ... and ending with the completion of the above `insert' */
543 	vm_map_unlock(m);
544 	vm_map_entry_release(count);
545 }
546