xref: /freebsd/sys/vm/vm_kern.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
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;
169 	vm_ooffset_t offset;
170 	vm_page_t m;
171 	int pflags, tries;
172 
173 	size = round_page(size);
174 	if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
175 		return (0);
176 	offset = addr - VM_MIN_KERNEL_ADDRESS;
177 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
178 	VM_OBJECT_WLOCK(object);
179 	for (i = 0; i < size; i += PAGE_SIZE) {
180 		tries = 0;
181 retry:
182 		m = vm_page_alloc_contig(object, OFF_TO_IDX(offset + i),
183 		    pflags, 1, low, high, PAGE_SIZE, 0, memattr);
184 		if (m == NULL) {
185 			VM_OBJECT_WUNLOCK(object);
186 			if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
187 				if (!vm_page_reclaim_contig(pflags, 1,
188 				    low, high, PAGE_SIZE, 0) &&
189 				    (flags & M_WAITOK) != 0)
190 					VM_WAIT;
191 				VM_OBJECT_WLOCK(object);
192 				tries++;
193 				goto retry;
194 			}
195 			kmem_unback(object, addr, i);
196 			vmem_free(vmem, addr, size);
197 			return (0);
198 		}
199 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
200 			pmap_zero_page(m);
201 		m->valid = VM_PAGE_BITS_ALL;
202 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
203 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
204 	}
205 	VM_OBJECT_WUNLOCK(object);
206 	return (addr);
207 }
208 
209 /*
210  *	Allocates a region from the kernel address map and physically
211  *	contiguous pages within the specified address range to the kernel
212  *	object.  Creates a wired mapping from this region to these pages, and
213  *	returns the region's starting virtual address.  If M_ZERO is specified
214  *	through the given flags, then the pages are zeroed before they are
215  *	mapped.
216  */
217 vm_offset_t
218 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
219     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
220     vm_memattr_t memattr)
221 {
222 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
223 	vm_offset_t addr, tmp;
224 	vm_ooffset_t offset;
225 	vm_page_t end_m, m;
226 	u_long npages;
227 	int pflags, tries;
228 
229 	size = round_page(size);
230 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
231 		return (0);
232 	offset = addr - VM_MIN_KERNEL_ADDRESS;
233 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
234 	npages = atop(size);
235 	VM_OBJECT_WLOCK(object);
236 	tries = 0;
237 retry:
238 	m = vm_page_alloc_contig(object, OFF_TO_IDX(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;
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 
346 	VM_OBJECT_WLOCK(object);
347 	for (i = 0; i < size; i += PAGE_SIZE) {
348 retry:
349 		m = vm_page_alloc(object, OFF_TO_IDX(offset + i), pflags);
350 
351 		/*
352 		 * Ran out of space, free everything up and return. Don't need
353 		 * to lock page queues here as we know that the pages we got
354 		 * aren't on any queues.
355 		 */
356 		if (m == NULL) {
357 			VM_OBJECT_WUNLOCK(object);
358 			if ((flags & M_NOWAIT) == 0) {
359 				VM_WAIT;
360 				VM_OBJECT_WLOCK(object);
361 				goto retry;
362 			}
363 			kmem_unback(object, addr, i);
364 			return (KERN_NO_SPACE);
365 		}
366 		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
367 			pmap_zero_page(m);
368 		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
369 		    ("kmem_malloc: page %p is managed", m));
370 		m->valid = VM_PAGE_BITS_ALL;
371 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
372 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
373 	}
374 	VM_OBJECT_WUNLOCK(object);
375 
376 	return (KERN_SUCCESS);
377 }
378 
379 /*
380  *	kmem_unback:
381  *
382  *	Unmap and free the physical pages underlying the specified virtual
383  *	address range.
384  *
385  *	A physical page must exist within the specified object at each index
386  *	that is being unmapped.
387  */
388 void
389 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
390 {
391 	vm_page_t m;
392 	vm_offset_t i, offset;
393 
394 	KASSERT(object == kmem_object || object == kernel_object,
395 	    ("kmem_unback: only supports kernel objects."));
396 
397 	pmap_remove(kernel_pmap, addr, addr + size);
398 	offset = addr - VM_MIN_KERNEL_ADDRESS;
399 	VM_OBJECT_WLOCK(object);
400 	for (i = 0; i < size; i += PAGE_SIZE) {
401 		m = vm_page_lookup(object, OFF_TO_IDX(offset + i));
402 		vm_page_unwire(m, PQ_NONE);
403 		vm_page_free(m);
404 	}
405 	VM_OBJECT_WUNLOCK(object);
406 }
407 
408 /*
409  *	kmem_free:
410  *
411  *	Free memory allocated with kmem_malloc.  The size must match the
412  *	original allocation.
413  */
414 void
415 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
416 {
417 
418 	size = round_page(size);
419 	kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
420 	    addr, size);
421 	vmem_free(vmem, addr, size);
422 }
423 
424 /*
425  *	kmap_alloc_wait:
426  *
427  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
428  *	has no room, the caller sleeps waiting for more memory in the submap.
429  *
430  *	This routine may block.
431  */
432 vm_offset_t
433 kmap_alloc_wait(map, size)
434 	vm_map_t map;
435 	vm_size_t size;
436 {
437 	vm_offset_t addr;
438 
439 	size = round_page(size);
440 	if (!swap_reserve(size))
441 		return (0);
442 
443 	for (;;) {
444 		/*
445 		 * To make this work for more than one map, use the map's lock
446 		 * to lock out sleepers/wakers.
447 		 */
448 		vm_map_lock(map);
449 		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
450 			break;
451 		/* no space now; see if we can ever get space */
452 		if (vm_map_max(map) - vm_map_min(map) < size) {
453 			vm_map_unlock(map);
454 			swap_release(size);
455 			return (0);
456 		}
457 		map->needs_wakeup = TRUE;
458 		vm_map_unlock_and_wait(map, 0);
459 	}
460 	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
461 	    VM_PROT_ALL, MAP_ACC_CHARGED);
462 	vm_map_unlock(map);
463 	return (addr);
464 }
465 
466 /*
467  *	kmap_free_wakeup:
468  *
469  *	Returns memory to a submap of the kernel, and wakes up any processes
470  *	waiting for memory in that map.
471  */
472 void
473 kmap_free_wakeup(map, addr, size)
474 	vm_map_t map;
475 	vm_offset_t addr;
476 	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(start, end)
521 	vm_offset_t start, end;
522 {
523 	vm_map_t m;
524 
525 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
526 	m->system_map = 1;
527 	vm_map_lock(m);
528 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
529 	kernel_map = m;
530 	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
531 #ifdef __amd64__
532 	    KERNBASE,
533 #else
534 	    VM_MIN_KERNEL_ADDRESS,
535 #endif
536 	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
537 	/* ... and ending with the completion of the above `insert' */
538 	vm_map_unlock(m);
539 }
540 
541 #ifdef DIAGNOSTIC
542 /*
543  * Allow userspace to directly trigger the VM drain routine for testing
544  * purposes.
545  */
546 static int
547 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
548 {
549 	int error, i;
550 
551 	i = 0;
552 	error = sysctl_handle_int(oidp, &i, 0, req);
553 	if (error)
554 		return (error);
555 	if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
556 		return (EINVAL);
557 	if (i != 0)
558 		EVENTHANDLER_INVOKE(vm_lowmem, i);
559 	return (0);
560 }
561 
562 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
563     debug_vm_lowmem, "I", "set to trigger vm_lowmem event with given flags");
564 #endif
565