xref: /freebsd/sys/vm/vm_kern.c (revision ec65e4f8d0654361df5e97d4de3518edebf76b46)
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  * 4. 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 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
101     SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
102 
103 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
104 #if defined(__arm__) || defined(__sparc64__)
105     &vm_max_kernel_address, 0,
106 #else
107     SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
108 #endif
109     "Max kernel address");
110 
111 /*
112  *	kva_alloc:
113  *
114  *	Allocate a virtual address range with no underlying object and
115  *	no initial mapping to physical memory.  Any mapping from this
116  *	range to physical memory must be explicitly created prior to
117  *	its use, typically with pmap_qenter().  Any attempt to create
118  *	a mapping on demand through vm_fault() will result in a panic.
119  */
120 vm_offset_t
121 kva_alloc(size)
122 	vm_size_t size;
123 {
124 	vm_offset_t addr;
125 
126 	size = round_page(size);
127 	if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
128 		return (0);
129 
130 	return (addr);
131 }
132 
133 /*
134  *	kva_free:
135  *
136  *	Release a region of kernel virtual memory allocated
137  *	with kva_alloc, and return the physical pages
138  *	associated with that region.
139  *
140  *	This routine may not block on kernel maps.
141  */
142 void
143 kva_free(addr, size)
144 	vm_offset_t addr;
145 	vm_size_t size;
146 {
147 
148 	size = round_page(size);
149 	vmem_free(kernel_arena, addr, size);
150 }
151 
152 /*
153  *	Allocates a region from the kernel address map and physical pages
154  *	within the specified address range to the kernel object.  Creates a
155  *	wired mapping from this region to these pages, and returns the
156  *	region's starting virtual address.  The allocated pages are not
157  *	necessarily physically contiguous.  If M_ZERO is specified through the
158  *	given flags, then the pages are zeroed before they are mapped.
159  */
160 vm_offset_t
161 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
162     vm_paddr_t high, vm_memattr_t memattr)
163 {
164 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
165 	vm_offset_t addr, i;
166 	vm_ooffset_t offset;
167 	vm_page_t m;
168 	int pflags, tries;
169 
170 	size = round_page(size);
171 	if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
172 		return (0);
173 	offset = addr - VM_MIN_KERNEL_ADDRESS;
174 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
175 	VM_OBJECT_WLOCK(object);
176 	for (i = 0; i < size; i += PAGE_SIZE) {
177 		tries = 0;
178 retry:
179 		m = vm_page_alloc_contig(object, OFF_TO_IDX(offset + i),
180 		    pflags, 1, low, high, PAGE_SIZE, 0, memattr);
181 		if (m == NULL) {
182 			VM_OBJECT_WUNLOCK(object);
183 			if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
184 				if (!vm_page_reclaim_contig(pflags, 1,
185 				    low, high, PAGE_SIZE, 0) &&
186 				    (flags & M_WAITOK) != 0)
187 					VM_WAIT;
188 				VM_OBJECT_WLOCK(object);
189 				tries++;
190 				goto retry;
191 			}
192 			kmem_unback(object, addr, i);
193 			vmem_free(vmem, addr, size);
194 			return (0);
195 		}
196 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
197 			pmap_zero_page(m);
198 		m->valid = VM_PAGE_BITS_ALL;
199 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
200 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
201 	}
202 	VM_OBJECT_WUNLOCK(object);
203 	return (addr);
204 }
205 
206 /*
207  *	Allocates a region from the kernel address map and physically
208  *	contiguous pages within the specified address range to the kernel
209  *	object.  Creates a wired mapping from this region to these pages, and
210  *	returns the region's starting virtual address.  If M_ZERO is specified
211  *	through the given flags, then the pages are zeroed before they are
212  *	mapped.
213  */
214 vm_offset_t
215 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
216     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
217     vm_memattr_t memattr)
218 {
219 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
220 	vm_offset_t addr, tmp;
221 	vm_ooffset_t offset;
222 	vm_page_t end_m, m;
223 	u_long npages;
224 	int pflags, tries;
225 
226 	size = round_page(size);
227 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
228 		return (0);
229 	offset = addr - VM_MIN_KERNEL_ADDRESS;
230 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
231 	npages = atop(size);
232 	VM_OBJECT_WLOCK(object);
233 	tries = 0;
234 retry:
235 	m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
236 	    npages, low, high, alignment, boundary, memattr);
237 	if (m == NULL) {
238 		VM_OBJECT_WUNLOCK(object);
239 		if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
240 			if (!vm_page_reclaim_contig(pflags, npages, low, high,
241 			    alignment, boundary) && (flags & M_WAITOK) != 0)
242 				VM_WAIT;
243 			VM_OBJECT_WLOCK(object);
244 			tries++;
245 			goto retry;
246 		}
247 		vmem_free(vmem, addr, size);
248 		return (0);
249 	}
250 	end_m = m + npages;
251 	tmp = addr;
252 	for (; m < end_m; m++) {
253 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
254 			pmap_zero_page(m);
255 		m->valid = VM_PAGE_BITS_ALL;
256 		pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
257 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
258 		tmp += PAGE_SIZE;
259 	}
260 	VM_OBJECT_WUNLOCK(object);
261 	return (addr);
262 }
263 
264 /*
265  *	kmem_suballoc:
266  *
267  *	Allocates a map to manage a subrange
268  *	of the kernel virtual address space.
269  *
270  *	Arguments are as follows:
271  *
272  *	parent		Map to take range from
273  *	min, max	Returned endpoints of map
274  *	size		Size of range to find
275  *	superpage_align	Request that min is superpage aligned
276  */
277 vm_map_t
278 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
279     vm_size_t size, boolean_t superpage_align)
280 {
281 	int ret;
282 	vm_map_t result;
283 
284 	size = round_page(size);
285 
286 	*min = vm_map_min(parent);
287 	ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
288 	    VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
289 	    MAP_ACC_NO_CHARGE);
290 	if (ret != KERN_SUCCESS)
291 		panic("kmem_suballoc: bad status return of %d", ret);
292 	*max = *min + size;
293 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
294 	if (result == NULL)
295 		panic("kmem_suballoc: cannot create submap");
296 	if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
297 		panic("kmem_suballoc: unable to change range to submap");
298 	return (result);
299 }
300 
301 /*
302  *	kmem_malloc:
303  *
304  *	Allocate wired-down pages in the kernel's address space.
305  */
306 vm_offset_t
307 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
308 {
309 	vm_offset_t addr;
310 	int rv;
311 
312 	size = round_page(size);
313 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
314 		return (0);
315 
316 	rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
317 	    addr, size, flags);
318 	if (rv != KERN_SUCCESS) {
319 		vmem_free(vmem, addr, size);
320 		return (0);
321 	}
322 	return (addr);
323 }
324 
325 /*
326  *	kmem_back:
327  *
328  *	Allocate physical pages for the specified virtual address range.
329  */
330 int
331 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
332 {
333 	vm_offset_t offset, i;
334 	vm_page_t m;
335 	int pflags;
336 
337 	KASSERT(object == kmem_object || object == kernel_object,
338 	    ("kmem_back: only supports kernel objects."));
339 
340 	offset = addr - VM_MIN_KERNEL_ADDRESS;
341 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
342 
343 	VM_OBJECT_WLOCK(object);
344 	for (i = 0; i < size; i += PAGE_SIZE) {
345 retry:
346 		m = vm_page_alloc(object, OFF_TO_IDX(offset + i), pflags);
347 
348 		/*
349 		 * Ran out of space, free everything up and return. Don't need
350 		 * to lock page queues here as we know that the pages we got
351 		 * aren't on any queues.
352 		 */
353 		if (m == NULL) {
354 			VM_OBJECT_WUNLOCK(object);
355 			if ((flags & M_NOWAIT) == 0) {
356 				VM_WAIT;
357 				VM_OBJECT_WLOCK(object);
358 				goto retry;
359 			}
360 			kmem_unback(object, addr, i);
361 			return (KERN_NO_SPACE);
362 		}
363 		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
364 			pmap_zero_page(m);
365 		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
366 		    ("kmem_malloc: page %p is managed", m));
367 		m->valid = VM_PAGE_BITS_ALL;
368 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
369 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
370 	}
371 	VM_OBJECT_WUNLOCK(object);
372 
373 	return (KERN_SUCCESS);
374 }
375 
376 /*
377  *	kmem_unback:
378  *
379  *	Unmap and free the physical pages underlying the specified virtual
380  *	address range.
381  *
382  *	A physical page must exist within the specified object at each index
383  *	that is being unmapped.
384  */
385 void
386 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
387 {
388 	vm_page_t m;
389 	vm_offset_t i, offset;
390 
391 	KASSERT(object == kmem_object || object == kernel_object,
392 	    ("kmem_unback: only supports kernel objects."));
393 
394 	pmap_remove(kernel_pmap, addr, addr + size);
395 	offset = addr - VM_MIN_KERNEL_ADDRESS;
396 	VM_OBJECT_WLOCK(object);
397 	for (i = 0; i < size; i += PAGE_SIZE) {
398 		m = vm_page_lookup(object, OFF_TO_IDX(offset + i));
399 		vm_page_unwire(m, PQ_NONE);
400 		vm_page_free(m);
401 	}
402 	VM_OBJECT_WUNLOCK(object);
403 }
404 
405 /*
406  *	kmem_free:
407  *
408  *	Free memory allocated with kmem_malloc.  The size must match the
409  *	original allocation.
410  */
411 void
412 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
413 {
414 
415 	size = round_page(size);
416 	kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
417 	    addr, size);
418 	vmem_free(vmem, addr, size);
419 }
420 
421 /*
422  *	kmap_alloc_wait:
423  *
424  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
425  *	has no room, the caller sleeps waiting for more memory in the submap.
426  *
427  *	This routine may block.
428  */
429 vm_offset_t
430 kmap_alloc_wait(map, size)
431 	vm_map_t map;
432 	vm_size_t size;
433 {
434 	vm_offset_t addr;
435 
436 	size = round_page(size);
437 	if (!swap_reserve(size))
438 		return (0);
439 
440 	for (;;) {
441 		/*
442 		 * To make this work for more than one map, use the map's lock
443 		 * to lock out sleepers/wakers.
444 		 */
445 		vm_map_lock(map);
446 		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
447 			break;
448 		/* no space now; see if we can ever get space */
449 		if (vm_map_max(map) - vm_map_min(map) < size) {
450 			vm_map_unlock(map);
451 			swap_release(size);
452 			return (0);
453 		}
454 		map->needs_wakeup = TRUE;
455 		vm_map_unlock_and_wait(map, 0);
456 	}
457 	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
458 	    VM_PROT_ALL, MAP_ACC_CHARGED);
459 	vm_map_unlock(map);
460 	return (addr);
461 }
462 
463 /*
464  *	kmap_free_wakeup:
465  *
466  *	Returns memory to a submap of the kernel, and wakes up any processes
467  *	waiting for memory in that map.
468  */
469 void
470 kmap_free_wakeup(map, addr, size)
471 	vm_map_t map;
472 	vm_offset_t addr;
473 	vm_size_t size;
474 {
475 
476 	vm_map_lock(map);
477 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
478 	if (map->needs_wakeup) {
479 		map->needs_wakeup = FALSE;
480 		vm_map_wakeup(map);
481 	}
482 	vm_map_unlock(map);
483 }
484 
485 void
486 kmem_init_zero_region(void)
487 {
488 	vm_offset_t addr, i;
489 	vm_page_t m;
490 
491 	/*
492 	 * Map a single physical page of zeros to a larger virtual range.
493 	 * This requires less looping in places that want large amounts of
494 	 * zeros, while not using much more physical resources.
495 	 */
496 	addr = kva_alloc(ZERO_REGION_SIZE);
497 	m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
498 	    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
499 	if ((m->flags & PG_ZERO) == 0)
500 		pmap_zero_page(m);
501 	for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
502 		pmap_qenter(addr + i, &m, 1);
503 	pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
504 
505 	zero_region = (const void *)addr;
506 }
507 
508 /*
509  * 	kmem_init:
510  *
511  *	Create the kernel map; insert a mapping covering kernel text,
512  *	data, bss, and all space allocated thus far (`boostrap' data).  The
513  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
514  *	`start' as allocated, and the range between `start' and `end' as free.
515  */
516 void
517 kmem_init(start, end)
518 	vm_offset_t start, end;
519 {
520 	vm_map_t m;
521 
522 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
523 	m->system_map = 1;
524 	vm_map_lock(m);
525 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
526 	kernel_map = m;
527 	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
528 #ifdef __amd64__
529 	    KERNBASE,
530 #else
531 	    VM_MIN_KERNEL_ADDRESS,
532 #endif
533 	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
534 	/* ... and ending with the completion of the above `insert' */
535 	vm_map_unlock(m);
536 }
537 
538 #ifdef DIAGNOSTIC
539 /*
540  * Allow userspace to directly trigger the VM drain routine for testing
541  * purposes.
542  */
543 static int
544 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
545 {
546 	int error, i;
547 
548 	i = 0;
549 	error = sysctl_handle_int(oidp, &i, 0, req);
550 	if (error)
551 		return (error);
552 	if (i)
553 		EVENTHANDLER_INVOKE(vm_lowmem, 0);
554 	return (0);
555 }
556 
557 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
558     debug_vm_lowmem, "I", "set to trigger vm_lowmem event");
559 #endif
560