xref: /freebsd/sys/vm/vm_kern.c (revision 41059135ce931c0f1014a999ffabc6bc470ce856)
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. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from: @(#)vm_kern.c	8.3 (Berkeley) 1/12/94
33  *
34  *
35  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36  * All rights reserved.
37  *
38  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39  *
40  * Permission to use, copy, modify and distribute this software and
41  * its documentation is hereby granted, provided that both the copyright
42  * notice and this permission notice appear in all copies of the
43  * software, derivative works or modified versions, and any portions
44  * thereof, and that both notices appear in supporting documentation.
45  *
46  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49  *
50  * Carnegie Mellon requests users of this software to return to
51  *
52  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
53  *  School of Computer Science
54  *  Carnegie Mellon University
55  *  Pittsburgh PA 15213-3890
56  *
57  * any improvements or extensions that they make and grant Carnegie the
58  * rights to redistribute these changes.
59  */
60 
61 /*
62  *	Kernel memory management.
63  */
64 
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>		/* for ticks and hz */
71 #include <sys/eventhandler.h>
72 #include <sys/lock.h>
73 #include <sys/proc.h>
74 #include <sys/malloc.h>
75 #include <sys/rwlock.h>
76 #include <sys/sysctl.h>
77 #include <sys/vmem.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/vm_kern.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pageout.h>
87 #include <vm/vm_extern.h>
88 #include <vm/uma.h>
89 
90 vm_map_t kernel_map;
91 vm_map_t exec_map;
92 vm_map_t pipe_map;
93 
94 const void *zero_region;
95 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
96 
97 /* NB: Used by kernel debuggers. */
98 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
99 
100 u_int exec_map_entry_size;
101 u_int exec_map_entries;
102 
103 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
104     SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
105 
106 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
107 #if defined(__arm__) || defined(__sparc64__)
108     &vm_max_kernel_address, 0,
109 #else
110     SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
111 #endif
112     "Max kernel address");
113 
114 /*
115  *	kva_alloc:
116  *
117  *	Allocate a virtual address range with no underlying object and
118  *	no initial mapping to physical memory.  Any mapping from this
119  *	range to physical memory must be explicitly created prior to
120  *	its use, typically with pmap_qenter().  Any attempt to create
121  *	a mapping on demand through vm_fault() will result in a panic.
122  */
123 vm_offset_t
124 kva_alloc(size)
125 	vm_size_t size;
126 {
127 	vm_offset_t addr;
128 
129 	size = round_page(size);
130 	if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
131 		return (0);
132 
133 	return (addr);
134 }
135 
136 /*
137  *	kva_free:
138  *
139  *	Release a region of kernel virtual memory allocated
140  *	with kva_alloc, and return the physical pages
141  *	associated with that region.
142  *
143  *	This routine may not block on kernel maps.
144  */
145 void
146 kva_free(addr, size)
147 	vm_offset_t addr;
148 	vm_size_t size;
149 {
150 
151 	size = round_page(size);
152 	vmem_free(kernel_arena, addr, size);
153 }
154 
155 /*
156  *	Allocates a region from the kernel address map and physical pages
157  *	within the specified address range to the kernel object.  Creates a
158  *	wired mapping from this region to these pages, and returns the
159  *	region's starting virtual address.  The allocated pages are not
160  *	necessarily physically contiguous.  If M_ZERO is specified through the
161  *	given flags, then the pages are zeroed before they are mapped.
162  */
163 vm_offset_t
164 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
165     vm_paddr_t high, vm_memattr_t memattr)
166 {
167 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
168 	vm_offset_t addr, i, offset;
169 	vm_page_t m;
170 	int pflags, tries;
171 
172 	size = round_page(size);
173 	if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
174 		return (0);
175 	offset = addr - VM_MIN_KERNEL_ADDRESS;
176 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
177 	VM_OBJECT_WLOCK(object);
178 	for (i = 0; i < size; i += PAGE_SIZE) {
179 		tries = 0;
180 retry:
181 		m = vm_page_alloc_contig(object, atop(offset + i),
182 		    pflags, 1, low, high, PAGE_SIZE, 0, memattr);
183 		if (m == NULL) {
184 			VM_OBJECT_WUNLOCK(object);
185 			if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
186 				if (!vm_page_reclaim_contig(pflags, 1,
187 				    low, high, PAGE_SIZE, 0) &&
188 				    (flags & M_WAITOK) != 0)
189 					VM_WAIT;
190 				VM_OBJECT_WLOCK(object);
191 				tries++;
192 				goto retry;
193 			}
194 			kmem_unback(object, addr, i);
195 			vmem_free(vmem, addr, size);
196 			return (0);
197 		}
198 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
199 			pmap_zero_page(m);
200 		m->valid = VM_PAGE_BITS_ALL;
201 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
202 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
203 	}
204 	VM_OBJECT_WUNLOCK(object);
205 	return (addr);
206 }
207 
208 /*
209  *	Allocates a region from the kernel address map and physically
210  *	contiguous pages within the specified address range to the kernel
211  *	object.  Creates a wired mapping from this region to these pages, and
212  *	returns the region's starting virtual address.  If M_ZERO is specified
213  *	through the given flags, then the pages are zeroed before they are
214  *	mapped.
215  */
216 vm_offset_t
217 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
218     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
219     vm_memattr_t memattr)
220 {
221 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
222 	vm_offset_t addr, offset, tmp;
223 	vm_page_t end_m, m;
224 	u_long npages;
225 	int pflags, tries;
226 
227 	size = round_page(size);
228 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
229 		return (0);
230 	offset = addr - VM_MIN_KERNEL_ADDRESS;
231 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
232 	npages = atop(size);
233 	VM_OBJECT_WLOCK(object);
234 	tries = 0;
235 retry:
236 	m = vm_page_alloc_contig(object, atop(offset), pflags,
237 	    npages, low, high, alignment, boundary, memattr);
238 	if (m == NULL) {
239 		VM_OBJECT_WUNLOCK(object);
240 		if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
241 			if (!vm_page_reclaim_contig(pflags, npages, low, high,
242 			    alignment, boundary) && (flags & M_WAITOK) != 0)
243 				VM_WAIT;
244 			VM_OBJECT_WLOCK(object);
245 			tries++;
246 			goto retry;
247 		}
248 		vmem_free(vmem, addr, size);
249 		return (0);
250 	}
251 	end_m = m + npages;
252 	tmp = addr;
253 	for (; m < end_m; m++) {
254 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
255 			pmap_zero_page(m);
256 		m->valid = VM_PAGE_BITS_ALL;
257 		pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
258 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
259 		tmp += PAGE_SIZE;
260 	}
261 	VM_OBJECT_WUNLOCK(object);
262 	return (addr);
263 }
264 
265 /*
266  *	kmem_suballoc:
267  *
268  *	Allocates a map to manage a subrange
269  *	of the kernel virtual address space.
270  *
271  *	Arguments are as follows:
272  *
273  *	parent		Map to take range from
274  *	min, max	Returned endpoints of map
275  *	size		Size of range to find
276  *	superpage_align	Request that min is superpage aligned
277  */
278 vm_map_t
279 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
280     vm_size_t size, boolean_t superpage_align)
281 {
282 	int ret;
283 	vm_map_t result;
284 
285 	size = round_page(size);
286 
287 	*min = vm_map_min(parent);
288 	ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
289 	    VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
290 	    MAP_ACC_NO_CHARGE);
291 	if (ret != KERN_SUCCESS)
292 		panic("kmem_suballoc: bad status return of %d", ret);
293 	*max = *min + size;
294 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
295 	if (result == NULL)
296 		panic("kmem_suballoc: cannot create submap");
297 	if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
298 		panic("kmem_suballoc: unable to change range to submap");
299 	return (result);
300 }
301 
302 /*
303  *	kmem_malloc:
304  *
305  *	Allocate wired-down pages in the kernel's address space.
306  */
307 vm_offset_t
308 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
309 {
310 	vm_offset_t addr;
311 	int rv;
312 
313 	size = round_page(size);
314 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
315 		return (0);
316 
317 	rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
318 	    addr, size, flags);
319 	if (rv != KERN_SUCCESS) {
320 		vmem_free(vmem, addr, size);
321 		return (0);
322 	}
323 	return (addr);
324 }
325 
326 /*
327  *	kmem_back:
328  *
329  *	Allocate physical pages for the specified virtual address range.
330  */
331 int
332 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
333 {
334 	vm_offset_t offset, i;
335 	vm_page_t m;
336 	int pflags;
337 
338 	KASSERT(object == kmem_object || object == kernel_object,
339 	    ("kmem_back: only supports kernel objects."));
340 
341 	offset = addr - VM_MIN_KERNEL_ADDRESS;
342 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
343 
344 	VM_OBJECT_WLOCK(object);
345 	for (i = 0; i < size; i += PAGE_SIZE) {
346 retry:
347 		m = vm_page_alloc(object, atop(offset + i), pflags);
348 
349 		/*
350 		 * Ran out of space, free everything up and return. Don't need
351 		 * to lock page queues here as we know that the pages we got
352 		 * aren't on any queues.
353 		 */
354 		if (m == NULL) {
355 			VM_OBJECT_WUNLOCK(object);
356 			if ((flags & M_NOWAIT) == 0) {
357 				VM_WAIT;
358 				VM_OBJECT_WLOCK(object);
359 				goto retry;
360 			}
361 			kmem_unback(object, addr, i);
362 			return (KERN_NO_SPACE);
363 		}
364 		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
365 			pmap_zero_page(m);
366 		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
367 		    ("kmem_malloc: page %p is managed", m));
368 		m->valid = VM_PAGE_BITS_ALL;
369 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
370 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
371 	}
372 	VM_OBJECT_WUNLOCK(object);
373 
374 	return (KERN_SUCCESS);
375 }
376 
377 /*
378  *	kmem_unback:
379  *
380  *	Unmap and free the physical pages underlying the specified virtual
381  *	address range.
382  *
383  *	A physical page must exist within the specified object at each index
384  *	that is being unmapped.
385  */
386 void
387 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
388 {
389 	vm_page_t m;
390 	vm_offset_t i, offset;
391 
392 	KASSERT(object == kmem_object || object == kernel_object,
393 	    ("kmem_unback: only supports kernel objects."));
394 
395 	pmap_remove(kernel_pmap, addr, addr + size);
396 	offset = addr - VM_MIN_KERNEL_ADDRESS;
397 	VM_OBJECT_WLOCK(object);
398 	for (i = 0; i < size; i += PAGE_SIZE) {
399 		m = vm_page_lookup(object, atop(offset + i));
400 		vm_page_unwire(m, PQ_NONE);
401 		vm_page_free(m);
402 	}
403 	VM_OBJECT_WUNLOCK(object);
404 }
405 
406 /*
407  *	kmem_free:
408  *
409  *	Free memory allocated with kmem_malloc.  The size must match the
410  *	original allocation.
411  */
412 void
413 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
414 {
415 
416 	size = round_page(size);
417 	kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
418 	    addr, size);
419 	vmem_free(vmem, addr, size);
420 }
421 
422 /*
423  *	kmap_alloc_wait:
424  *
425  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
426  *	has no room, the caller sleeps waiting for more memory in the submap.
427  *
428  *	This routine may block.
429  */
430 vm_offset_t
431 kmap_alloc_wait(map, size)
432 	vm_map_t map;
433 	vm_size_t size;
434 {
435 	vm_offset_t addr;
436 
437 	size = round_page(size);
438 	if (!swap_reserve(size))
439 		return (0);
440 
441 	for (;;) {
442 		/*
443 		 * To make this work for more than one map, use the map's lock
444 		 * to lock out sleepers/wakers.
445 		 */
446 		vm_map_lock(map);
447 		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
448 			break;
449 		/* no space now; see if we can ever get space */
450 		if (vm_map_max(map) - vm_map_min(map) < size) {
451 			vm_map_unlock(map);
452 			swap_release(size);
453 			return (0);
454 		}
455 		map->needs_wakeup = TRUE;
456 		vm_map_unlock_and_wait(map, 0);
457 	}
458 	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
459 	    VM_PROT_ALL, MAP_ACC_CHARGED);
460 	vm_map_unlock(map);
461 	return (addr);
462 }
463 
464 /*
465  *	kmap_free_wakeup:
466  *
467  *	Returns memory to a submap of the kernel, and wakes up any processes
468  *	waiting for memory in that map.
469  */
470 void
471 kmap_free_wakeup(map, addr, size)
472 	vm_map_t map;
473 	vm_offset_t addr;
474 	vm_size_t size;
475 {
476 
477 	vm_map_lock(map);
478 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
479 	if (map->needs_wakeup) {
480 		map->needs_wakeup = FALSE;
481 		vm_map_wakeup(map);
482 	}
483 	vm_map_unlock(map);
484 }
485 
486 void
487 kmem_init_zero_region(void)
488 {
489 	vm_offset_t addr, i;
490 	vm_page_t m;
491 
492 	/*
493 	 * Map a single physical page of zeros to a larger virtual range.
494 	 * This requires less looping in places that want large amounts of
495 	 * zeros, while not using much more physical resources.
496 	 */
497 	addr = kva_alloc(ZERO_REGION_SIZE);
498 	m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
499 	    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
500 	if ((m->flags & PG_ZERO) == 0)
501 		pmap_zero_page(m);
502 	for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
503 		pmap_qenter(addr + i, &m, 1);
504 	pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
505 
506 	zero_region = (const void *)addr;
507 }
508 
509 /*
510  * 	kmem_init:
511  *
512  *	Create the kernel map; insert a mapping covering kernel text,
513  *	data, bss, and all space allocated thus far (`boostrap' data).  The
514  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
515  *	`start' as allocated, and the range between `start' and `end' as free.
516  */
517 void
518 kmem_init(start, end)
519 	vm_offset_t start, end;
520 {
521 	vm_map_t m;
522 
523 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
524 	m->system_map = 1;
525 	vm_map_lock(m);
526 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
527 	kernel_map = m;
528 	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
529 #ifdef __amd64__
530 	    KERNBASE,
531 #else
532 	    VM_MIN_KERNEL_ADDRESS,
533 #endif
534 	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
535 	/* ... and ending with the completion of the above `insert' */
536 	vm_map_unlock(m);
537 }
538 
539 #ifdef DIAGNOSTIC
540 /*
541  * Allow userspace to directly trigger the VM drain routine for testing
542  * purposes.
543  */
544 static int
545 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
546 {
547 	int error, i;
548 
549 	i = 0;
550 	error = sysctl_handle_int(oidp, &i, 0, req);
551 	if (error)
552 		return (error);
553 	if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
554 		return (EINVAL);
555 	if (i != 0)
556 		EVENTHANDLER_INVOKE(vm_lowmem, i);
557 	return (0);
558 }
559 
560 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
561     debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");
562 #endif
563