xref: /dragonfly/sys/vm/vm_kern.c (revision 1bf4b486) 
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 1.21 2005/04/02 15:58:16 joerg 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, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
109 	if (result != KERN_SUCCESS) {
110 		return (0);
111 	}
112 	return (addr);
113 }
114 
115 /*
116  *	kmem_alloc_nofault:
117  *
118  *	Same as kmem_alloc_pageable, except that it create a nofault entry.
119  */
120 vm_offset_t
121 kmem_alloc_nofault(vm_map_t map, vm_size_t size)
122 {
123 	vm_offset_t addr;
124 	int result;
125 
126 	size = round_page(size);
127 	addr = vm_map_min(map);
128 	result = vm_map_find(map, NULL, (vm_offset_t) 0,
129 	    &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
130 	if (result != KERN_SUCCESS) {
131 		return (0);
132 	}
133 	return (addr);
134 }
135 
136 /*
137  *	Allocate wired-down memory in the kernel's address map
138  *	or a submap.
139  */
140 vm_offset_t
141 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags)
142 {
143 	vm_offset_t addr;
144 	vm_offset_t offset;
145 	vm_offset_t i;
146 	int count;
147 
148 	size = round_page(size);
149 
150 	if (kmflags & KM_KRESERVE)
151 		count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
152 	else
153 		count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
154 
155 	/*
156 	 * Use the kernel object for wired-down kernel pages. Assume that no
157 	 * region of the kernel object is referenced more than once.
158 	 *
159 	 * Locate sufficient space in the map.  This will give us the final
160 	 * virtual address for the new memory, and thus will tell us the
161 	 * offset within the kernel map.
162 	 */
163 	vm_map_lock(map);
164 	if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
165 		vm_map_unlock(map);
166 		if (kmflags & KM_KRESERVE)
167 			vm_map_entry_krelease(count);
168 		else
169 			vm_map_entry_release(count);
170 		return (0);
171 	}
172 	offset = addr - VM_MIN_KERNEL_ADDRESS;
173 	vm_object_reference(kernel_object);
174 	vm_map_insert(map, &count,
175 		kernel_object, offset, addr, addr + size,
176 		VM_PROT_ALL, VM_PROT_ALL, 0);
177 	vm_map_unlock(map);
178 	if (kmflags & KM_KRESERVE)
179 		vm_map_entry_krelease(count);
180 	else
181 		vm_map_entry_release(count);
182 
183 	/*
184 	 * Guarantee that there are pages already in this object before
185 	 * calling vm_map_wire.  This is to prevent the following
186 	 * scenario:
187 	 *
188 	 * 1) Threads have swapped out, so that there is a pager for the
189 	 * kernel_object. 2) The kmsg zone is empty, and so we are
190 	 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
191 	 * there is no page, but there is a pager, so we call
192 	 * pager_data_request.  But the kmsg zone is empty, so we must
193 	 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
194 	 * we get the data back from the pager, it will be (very stale)
195 	 * non-zero data.  kmem_alloc is defined to return zero-filled memory.
196 	 *
197 	 * We're intentionally not activating the pages we allocate to prevent a
198 	 * race with page-out.  vm_map_wire will wire the pages.
199 	 */
200 
201 	for (i = 0; i < size; i += PAGE_SIZE) {
202 		vm_page_t mem;
203 
204 		mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
205 			    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
206 		if ((mem->flags & PG_ZERO) == 0)
207 			vm_page_zero_fill(mem);
208 		mem->valid = VM_PAGE_BITS_ALL;
209 		vm_page_flag_clear(mem, PG_ZERO);
210 		vm_page_wakeup(mem);
211 	}
212 
213 	/*
214 	 * And finally, mark the data as non-pageable.
215 	 */
216 
217 	(void) vm_map_wire(map, (vm_offset_t) addr, addr + size, kmflags);
218 
219 	return (addr);
220 }
221 
222 /*
223  *	kmem_free:
224  *
225  *	Release a region of kernel virtual memory allocated
226  *	with kmem_alloc, and return the physical pages
227  *	associated with that region.
228  *
229  *	This routine may not block on kernel maps.
230  */
231 void
232 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size)
233 {
234 	(void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
235 }
236 
237 /*
238  *	kmem_suballoc:
239  *
240  *	Allocates a map to manage a subrange
241  *	of the kernel virtual address space.
242  *
243  *	Arguments are as follows:
244  *
245  *	parent		Map to take range from
246  *	size		Size of range to find
247  *	min, max	Returned endpoints of map
248  *	pageable	Can the region be paged
249  */
250 vm_map_t
251 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
252     vm_size_t size)
253 {
254 	int ret;
255 	vm_map_t result;
256 
257 	size = round_page(size);
258 
259 	*min = (vm_offset_t) vm_map_min(parent);
260 	ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
261 	    min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
262 	if (ret != KERN_SUCCESS) {
263 		printf("kmem_suballoc: bad status return of %d.\n", ret);
264 		panic("kmem_suballoc");
265 	}
266 	*max = *min + size;
267 	pmap_reference(vm_map_pmap(parent));
268 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
269 	if (result == NULL)
270 		panic("kmem_suballoc: cannot create submap");
271 	if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
272 		panic("kmem_suballoc: unable to change range to submap");
273 	return (result);
274 }
275 
276 /*
277  *	kmem_malloc:
278  *
279  * 	Allocate wired-down memory in the kernel's address map for the higher
280  * 	level kernel memory allocator (kern/kern_malloc.c).  We cannot use
281  * 	kmem_alloc() because we may need to allocate memory at interrupt
282  * 	level where we cannot block (canwait == FALSE).
283  *
284  * 	We don't worry about expanding the map (adding entries) since entries
285  * 	for wired maps are statically allocated.
286  *
287  *	NOTE:  Please see kmem_slab_alloc() for a better explanation of the
288  *	M_* flags.
289  */
290 vm_offset_t
291 kmem_malloc(vm_map_t map, vm_size_t size, int flags)
292 {
293 	vm_offset_t offset, i;
294 	vm_map_entry_t entry;
295 	vm_offset_t addr;
296 	vm_page_t m;
297 	int count, vmflags, wanted_reserve;
298 	thread_t td;
299 
300 	if (map != kernel_map)
301 		panic("kmem_malloc: map != kernel_map");
302 
303 	size = round_page(size);
304 	addr = vm_map_min(map);
305 
306 	/*
307 	 * Locate sufficient space in the map.  This will give us the final
308 	 * virtual address for the new memory, and thus will tell us the
309 	 * offset within the kernel map.  If we are unable to allocate space
310 	 * and neither RNOWAIT or NULLOK is set, we panic.
311 	 */
312 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
313 	vm_map_lock(map);
314 	if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) {
315 		vm_map_unlock(map);
316 		vm_map_entry_release(count);
317 		if ((flags & M_NULLOK) == 0) {
318 			panic("kmem_malloc(%ld): kernel_map too small: "
319 				"%ld total allocated",
320 				(long)size, (long)map->size);
321 		}
322 		return (0);
323 	}
324 	offset = addr - VM_MIN_KERNEL_ADDRESS;
325 	vm_object_reference(kmem_object);
326 	vm_map_insert(map, &count,
327 		kmem_object, offset, addr, addr + size,
328 		VM_PROT_ALL, VM_PROT_ALL, 0);
329 
330 	td = curthread;
331 	wanted_reserve = 0;
332 
333 	vmflags = VM_ALLOC_SYSTEM;	/* XXX M_USE_RESERVE? */
334 	if ((flags & (M_WAITOK|M_RNOWAIT)) == 0)
335 		panic("kmem_malloc: bad flags %08x (%p)\n", flags, ((int **)&map)[-1]);
336 	if (flags & M_USE_INTERRUPT_RESERVE)
337 		vmflags |= VM_ALLOC_INTERRUPT;
338 
339 	for (i = 0; i < size; i += PAGE_SIZE) {
340 		/*
341 		 * Only allocate PQ_CACHE pages for M_WAITOK requests and
342 		 * then only if we are not preempting.
343 		 */
344 		if (flags & M_WAITOK) {
345 			if (td->td_preempted) {
346 				vmflags &= ~VM_ALLOC_NORMAL;
347 				wanted_reserve = 1;
348 			} else {
349 				vmflags |= VM_ALLOC_NORMAL;
350 				wanted_reserve = 0;
351 			}
352 		}
353 
354 		m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), vmflags);
355 
356 		/*
357 		 * Ran out of space, free everything up and return.  Don't need
358 		 * to lock page queues here as we know that the pages we got
359 		 * aren't on any queues.
360 		 *
361 		 * If M_WAITOK is set we can yield or block.
362 		 */
363 		if (m == NULL) {
364 			if (flags & M_WAITOK) {
365 				if (wanted_reserve) {
366 					vm_map_unlock(map);
367 					lwkt_yield();
368 					vm_map_lock(map);
369 				} else {
370 					vm_map_unlock(map);
371 					vm_wait();
372 					vm_map_lock(map);
373 				}
374 				i -= PAGE_SIZE;	/* retry */
375 				continue;
376 			}
377 			/*
378 			 * Free the pages before removing the map entry.
379 			 * They are already marked busy.  Calling
380 			 * vm_map_delete before the pages has been freed or
381 			 * unbusied will cause a deadlock.
382 			 */
383 			while (i != 0) {
384 				i -= PAGE_SIZE;
385 				m = vm_page_lookup(kmem_object,
386 						   OFF_TO_IDX(offset + i));
387 				vm_page_free(m);
388 			}
389 			vm_map_delete(map, addr, addr + size, &count);
390 			vm_map_unlock(map);
391 			vm_map_entry_release(count);
392 			return (0);
393 		}
394 		vm_page_flag_clear(m, PG_ZERO);
395 		m->valid = VM_PAGE_BITS_ALL;
396 	}
397 
398 	/*
399 	 * Mark map entry as non-pageable. Assert: vm_map_insert() will never
400 	 * be able to extend the previous entry so there will be a new entry
401 	 * exactly corresponding to this address range and it will have
402 	 * wired_count == 0.
403 	 */
404 	if (!vm_map_lookup_entry(map, addr, &entry) ||
405 	    entry->start != addr || entry->end != addr + size ||
406 	    entry->wired_count != 0)
407 		panic("kmem_malloc: entry not found or misaligned");
408 	entry->wired_count = 1;
409 
410 	vm_map_simplify_entry(map, entry, &count);
411 
412 	/*
413 	 * Loop thru pages, entering them in the pmap. (We cannot add them to
414 	 * the wired count without wrapping the vm_page_queue_lock in
415 	 * splimp...)
416 	 */
417 	for (i = 0; i < size; i += PAGE_SIZE) {
418 		m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
419 		vm_page_wire(m);
420 		vm_page_wakeup(m);
421 		/*
422 		 * Because this is kernel_pmap, this call will not block.
423 		 */
424 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
425 		vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED);
426 	}
427 	vm_map_unlock(map);
428 	vm_map_entry_release(count);
429 
430 	return (addr);
431 }
432 
433 /*
434  *	kmem_alloc_wait:
435  *
436  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
437  *	has no room, the caller sleeps waiting for more memory in the submap.
438  *
439  *	This routine may block.
440  */
441 
442 vm_offset_t
443 kmem_alloc_wait(vm_map_t map, vm_size_t size)
444 {
445 	vm_offset_t addr;
446 	int count;
447 
448 	size = round_page(size);
449 
450 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
451 
452 	for (;;) {
453 		/*
454 		 * To make this work for more than one map, use the map's lock
455 		 * to lock out sleepers/wakers.
456 		 */
457 		vm_map_lock(map);
458 		if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) == 0)
459 			break;
460 		/* no space now; see if we can ever get space */
461 		if (vm_map_max(map) - vm_map_min(map) < size) {
462 			vm_map_entry_release(count);
463 			vm_map_unlock(map);
464 			return (0);
465 		}
466 		vm_map_unlock(map);
467 		tsleep(map, 0, "kmaw", 0);
468 	}
469 	vm_map_insert(map, &count,
470 		    NULL, (vm_offset_t) 0,
471 		    addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
472 	vm_map_unlock(map);
473 	vm_map_entry_release(count);
474 	return (addr);
475 }
476 
477 /*
478  *	kmem_free_wakeup:
479  *
480  *	Returns memory to a submap of the kernel, and wakes up any processes
481  *	waiting for memory in that map.
482  */
483 void
484 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
485 {
486 	int count;
487 
488 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
489 	vm_map_lock(map);
490 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count);
491 	wakeup(map);
492 	vm_map_unlock(map);
493 	vm_map_entry_release(count);
494 }
495 
496 /*
497  * 	kmem_init:
498  *
499  *	Create the kernel map; insert a mapping covering kernel text,
500  *	data, bss, and all space allocated thus far (`boostrap' data).  The
501  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
502  *	`start' as allocated, and the range between `start' and `end' as free.
503  *
504  *	Depend on the zalloc bootstrap cache to get our vm_map_entry_t.
505  */
506 void
507 kmem_init(vm_offset_t start, vm_offset_t end)
508 {
509 	vm_map_t m;
510 	int count;
511 
512 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
513 	vm_map_lock(m);
514 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
515 	kernel_map = m;
516 	kernel_map->system_map = 1;
517 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
518 	(void) vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
519 	    VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0);
520 	/* ... and ending with the completion of the above `insert' */
521 	vm_map_unlock(m);
522 	vm_map_entry_release(count);
523 }
524