xref: /freebsd/sys/vm/vm_kern.c (revision ca987d4641cdcd7f27e153db17c5bf064934faf5)
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_radix.h>
88 #include <vm/vm_extern.h>
89 #include <vm/uma.h>
90 
91 vm_map_t kernel_map;
92 vm_map_t exec_map;
93 vm_map_t pipe_map;
94 
95 const void *zero_region;
96 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
97 
98 /* NB: Used by kernel debuggers. */
99 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
100 
101 u_int exec_map_entry_size;
102 u_int exec_map_entries;
103 
104 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
105     SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
106 
107 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
108 #if defined(__arm__) || defined(__sparc64__)
109     &vm_max_kernel_address, 0,
110 #else
111     SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
112 #endif
113     "Max kernel address");
114 
115 /*
116  *	kva_alloc:
117  *
118  *	Allocate a virtual address range with no underlying object and
119  *	no initial mapping to physical memory.  Any mapping from this
120  *	range to physical memory must be explicitly created prior to
121  *	its use, typically with pmap_qenter().  Any attempt to create
122  *	a mapping on demand through vm_fault() will result in a panic.
123  */
124 vm_offset_t
125 kva_alloc(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(vm_offset_t addr, vm_size_t size)
147 {
148 
149 	size = round_page(size);
150 	vmem_free(kernel_arena, addr, size);
151 }
152 
153 /*
154  *	Allocates a region from the kernel address map and physical pages
155  *	within the specified address range to the kernel object.  Creates a
156  *	wired mapping from this region to these pages, and returns the
157  *	region's starting virtual address.  The allocated pages are not
158  *	necessarily physically contiguous.  If M_ZERO is specified through the
159  *	given flags, then the pages are zeroed before they are mapped.
160  */
161 vm_offset_t
162 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
163     vm_paddr_t high, vm_memattr_t memattr)
164 {
165 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
166 	vm_offset_t addr, i, 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 	pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
176 	pflags |= VM_ALLOC_NOWAIT;
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 	pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
233 	pflags |= VM_ALLOC_NOWAIT;
234 	npages = atop(size);
235 	VM_OBJECT_WLOCK(object);
236 	tries = 0;
237 retry:
238 	m = vm_page_alloc_contig(object, atop(offset), pflags,
239 	    npages, low, high, alignment, boundary, memattr);
240 	if (m == NULL) {
241 		VM_OBJECT_WUNLOCK(object);
242 		if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
243 			if (!vm_page_reclaim_contig(pflags, npages, low, high,
244 			    alignment, boundary) && (flags & M_WAITOK) != 0)
245 				VM_WAIT;
246 			VM_OBJECT_WLOCK(object);
247 			tries++;
248 			goto retry;
249 		}
250 		vmem_free(vmem, addr, size);
251 		return (0);
252 	}
253 	end_m = m + npages;
254 	tmp = addr;
255 	for (; m < end_m; m++) {
256 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
257 			pmap_zero_page(m);
258 		m->valid = VM_PAGE_BITS_ALL;
259 		pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
260 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
261 		tmp += PAGE_SIZE;
262 	}
263 	VM_OBJECT_WUNLOCK(object);
264 	return (addr);
265 }
266 
267 /*
268  *	kmem_suballoc:
269  *
270  *	Allocates a map to manage a subrange
271  *	of the kernel virtual address space.
272  *
273  *	Arguments are as follows:
274  *
275  *	parent		Map to take range from
276  *	min, max	Returned endpoints of map
277  *	size		Size of range to find
278  *	superpage_align	Request that min is superpage aligned
279  */
280 vm_map_t
281 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
282     vm_size_t size, boolean_t superpage_align)
283 {
284 	int ret;
285 	vm_map_t result;
286 
287 	size = round_page(size);
288 
289 	*min = vm_map_min(parent);
290 	ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
291 	    VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
292 	    MAP_ACC_NO_CHARGE);
293 	if (ret != KERN_SUCCESS)
294 		panic("kmem_suballoc: bad status return of %d", ret);
295 	*max = *min + size;
296 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
297 	if (result == NULL)
298 		panic("kmem_suballoc: cannot create submap");
299 	if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
300 		panic("kmem_suballoc: unable to change range to submap");
301 	return (result);
302 }
303 
304 /*
305  *	kmem_malloc:
306  *
307  *	Allocate wired-down pages in the kernel's address space.
308  */
309 vm_offset_t
310 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
311 {
312 	vm_offset_t addr;
313 	int rv;
314 
315 	size = round_page(size);
316 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
317 		return (0);
318 
319 	rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
320 	    addr, size, flags);
321 	if (rv != KERN_SUCCESS) {
322 		vmem_free(vmem, addr, size);
323 		return (0);
324 	}
325 	return (addr);
326 }
327 
328 /*
329  *	kmem_back:
330  *
331  *	Allocate physical pages for the specified virtual address range.
332  */
333 int
334 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
335 {
336 	vm_offset_t offset, i;
337 	vm_page_t m, mpred;
338 	int pflags;
339 
340 	KASSERT(object == kmem_object || object == kernel_object,
341 	    ("kmem_back: only supports kernel objects."));
342 
343 	offset = addr - VM_MIN_KERNEL_ADDRESS;
344 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
345 	pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
346 	if (flags & M_WAITOK)
347 		pflags |= VM_ALLOC_WAITFAIL;
348 
349 	i = 0;
350 	VM_OBJECT_WLOCK(object);
351 retry:
352 	mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
353 	for (; i < size; i += PAGE_SIZE, mpred = m) {
354 		m = vm_page_alloc_after(object, atop(offset + i), pflags,
355 		    mpred);
356 
357 		/*
358 		 * Ran out of space, free everything up and return. Don't need
359 		 * to lock page queues here as we know that the pages we got
360 		 * aren't on any queues.
361 		 */
362 		if (m == NULL) {
363 			if ((flags & M_NOWAIT) == 0)
364 				goto retry;
365 			VM_OBJECT_WUNLOCK(object);
366 			kmem_unback(object, addr, i);
367 			return (KERN_NO_SPACE);
368 		}
369 		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
370 			pmap_zero_page(m);
371 		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
372 		    ("kmem_malloc: page %p is managed", m));
373 		m->valid = VM_PAGE_BITS_ALL;
374 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
375 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
376 	}
377 	VM_OBJECT_WUNLOCK(object);
378 
379 	return (KERN_SUCCESS);
380 }
381 
382 /*
383  *	kmem_unback:
384  *
385  *	Unmap and free the physical pages underlying the specified virtual
386  *	address range.
387  *
388  *	A physical page must exist within the specified object at each index
389  *	that is being unmapped.
390  */
391 void
392 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
393 {
394 	vm_page_t m, next;
395 	vm_offset_t end, offset;
396 
397 	KASSERT(object == kmem_object || object == kernel_object,
398 	    ("kmem_unback: only supports kernel objects."));
399 
400 	pmap_remove(kernel_pmap, addr, addr + size);
401 	offset = addr - VM_MIN_KERNEL_ADDRESS;
402 	end = offset + size;
403 	VM_OBJECT_WLOCK(object);
404 	for (m = vm_page_lookup(object, atop(offset)); offset < end;
405 	    offset += PAGE_SIZE, m = next) {
406 		next = vm_page_next(m);
407 		vm_page_unwire(m, PQ_NONE);
408 		vm_page_free(m);
409 	}
410 	VM_OBJECT_WUNLOCK(object);
411 }
412 
413 /*
414  *	kmem_free:
415  *
416  *	Free memory allocated with kmem_malloc.  The size must match the
417  *	original allocation.
418  */
419 void
420 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
421 {
422 
423 	size = round_page(size);
424 	kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
425 	    addr, size);
426 	vmem_free(vmem, addr, size);
427 }
428 
429 /*
430  *	kmap_alloc_wait:
431  *
432  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
433  *	has no room, the caller sleeps waiting for more memory in the submap.
434  *
435  *	This routine may block.
436  */
437 vm_offset_t
438 kmap_alloc_wait(vm_map_t map, vm_size_t size)
439 {
440 	vm_offset_t addr;
441 
442 	size = round_page(size);
443 	if (!swap_reserve(size))
444 		return (0);
445 
446 	for (;;) {
447 		/*
448 		 * To make this work for more than one map, use the map's lock
449 		 * to lock out sleepers/wakers.
450 		 */
451 		vm_map_lock(map);
452 		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
453 			break;
454 		/* no space now; see if we can ever get space */
455 		if (vm_map_max(map) - vm_map_min(map) < size) {
456 			vm_map_unlock(map);
457 			swap_release(size);
458 			return (0);
459 		}
460 		map->needs_wakeup = TRUE;
461 		vm_map_unlock_and_wait(map, 0);
462 	}
463 	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
464 	    VM_PROT_ALL, MAP_ACC_CHARGED);
465 	vm_map_unlock(map);
466 	return (addr);
467 }
468 
469 /*
470  *	kmap_free_wakeup:
471  *
472  *	Returns memory to a submap of the kernel, and wakes up any processes
473  *	waiting for memory in that map.
474  */
475 void
476 kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
477 {
478 
479 	vm_map_lock(map);
480 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
481 	if (map->needs_wakeup) {
482 		map->needs_wakeup = FALSE;
483 		vm_map_wakeup(map);
484 	}
485 	vm_map_unlock(map);
486 }
487 
488 void
489 kmem_init_zero_region(void)
490 {
491 	vm_offset_t addr, i;
492 	vm_page_t m;
493 
494 	/*
495 	 * Map a single physical page of zeros to a larger virtual range.
496 	 * This requires less looping in places that want large amounts of
497 	 * zeros, while not using much more physical resources.
498 	 */
499 	addr = kva_alloc(ZERO_REGION_SIZE);
500 	m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
501 	    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
502 	if ((m->flags & PG_ZERO) == 0)
503 		pmap_zero_page(m);
504 	for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
505 		pmap_qenter(addr + i, &m, 1);
506 	pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
507 
508 	zero_region = (const void *)addr;
509 }
510 
511 /*
512  * 	kmem_init:
513  *
514  *	Create the kernel map; insert a mapping covering kernel text,
515  *	data, bss, and all space allocated thus far (`boostrap' data).  The
516  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
517  *	`start' as allocated, and the range between `start' and `end' as free.
518  */
519 void
520 kmem_init(vm_offset_t start, vm_offset_t end)
521 {
522 	vm_map_t m;
523 
524 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
525 	m->system_map = 1;
526 	vm_map_lock(m);
527 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
528 	kernel_map = m;
529 	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
530 #ifdef __amd64__
531 	    KERNBASE,
532 #else
533 	    VM_MIN_KERNEL_ADDRESS,
534 #endif
535 	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
536 	/* ... and ending with the completion of the above `insert' */
537 	vm_map_unlock(m);
538 }
539 
540 #ifdef DIAGNOSTIC
541 /*
542  * Allow userspace to directly trigger the VM drain routine for testing
543  * purposes.
544  */
545 static int
546 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
547 {
548 	int error, i;
549 
550 	i = 0;
551 	error = sysctl_handle_int(oidp, &i, 0, req);
552 	if (error)
553 		return (error);
554 	if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
555 		return (EINVAL);
556 	if (i != 0)
557 		EVENTHANDLER_INVOKE(vm_lowmem, i);
558 	return (0);
559 }
560 
561 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
562     debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");
563 #endif
564