xref: /freebsd/sys/vm/vm_kern.c (revision 96cdb0ab9d6d89d47c14f239933391a1f7d465c7)
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 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
98     SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
99 
100 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
101 #if defined(__arm__) || defined(__sparc64__)
102     &vm_max_kernel_address, 0,
103 #else
104     SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
105 #endif
106     "Max kernel address");
107 
108 /*
109  *	kva_alloc:
110  *
111  *	Allocate a virtual address range with no underlying object and
112  *	no initial mapping to physical memory.  Any mapping from this
113  *	range to physical memory must be explicitly created prior to
114  *	its use, typically with pmap_qenter().  Any attempt to create
115  *	a mapping on demand through vm_fault() will result in a panic.
116  */
117 vm_offset_t
118 kva_alloc(size)
119 	vm_size_t size;
120 {
121 	vm_offset_t addr;
122 
123 	size = round_page(size);
124 	if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
125 		return (0);
126 
127 	return (addr);
128 }
129 
130 /*
131  *	kva_free:
132  *
133  *	Release a region of kernel virtual memory allocated
134  *	with kva_alloc, and return the physical pages
135  *	associated with that region.
136  *
137  *	This routine may not block on kernel maps.
138  */
139 void
140 kva_free(addr, size)
141 	vm_offset_t addr;
142 	vm_size_t size;
143 {
144 
145 	size = round_page(size);
146 	vmem_free(kernel_arena, addr, size);
147 }
148 
149 /*
150  *	Allocates a region from the kernel address map and physical pages
151  *	within the specified address range to the kernel object.  Creates a
152  *	wired mapping from this region to these pages, and returns the
153  *	region's starting virtual address.  The allocated pages are not
154  *	necessarily physically contiguous.  If M_ZERO is specified through the
155  *	given flags, then the pages are zeroed before they are mapped.
156  */
157 vm_offset_t
158 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
159     vm_paddr_t high, vm_memattr_t memattr)
160 {
161 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
162 	vm_offset_t addr, i;
163 	vm_ooffset_t offset;
164 	vm_page_t m;
165 	int pflags, tries;
166 
167 	size = round_page(size);
168 	if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
169 		return (0);
170 	offset = addr - VM_MIN_KERNEL_ADDRESS;
171 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
172 	VM_OBJECT_WLOCK(object);
173 	for (i = 0; i < size; i += PAGE_SIZE) {
174 		tries = 0;
175 retry:
176 		m = vm_page_alloc_contig(object, OFF_TO_IDX(offset + i),
177 		    pflags, 1, low, high, PAGE_SIZE, 0, memattr);
178 		if (m == NULL) {
179 			VM_OBJECT_WUNLOCK(object);
180 			if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
181 				vm_pageout_grow_cache(tries, low, high);
182 				VM_OBJECT_WLOCK(object);
183 				tries++;
184 				goto retry;
185 			}
186 			kmem_unback(object, addr, i);
187 			vmem_free(vmem, addr, size);
188 			return (0);
189 		}
190 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
191 			pmap_zero_page(m);
192 		m->valid = VM_PAGE_BITS_ALL;
193 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
194 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
195 	}
196 	VM_OBJECT_WUNLOCK(object);
197 	return (addr);
198 }
199 
200 /*
201  *	Allocates a region from the kernel address map and physically
202  *	contiguous pages within the specified address range to the kernel
203  *	object.  Creates a wired mapping from this region to these pages, and
204  *	returns the region's starting virtual address.  If M_ZERO is specified
205  *	through the given flags, then the pages are zeroed before they are
206  *	mapped.
207  */
208 vm_offset_t
209 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
210     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
211     vm_memattr_t memattr)
212 {
213 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
214 	vm_offset_t addr, tmp;
215 	vm_ooffset_t offset;
216 	vm_page_t end_m, m;
217 	int pflags, tries;
218 
219 	size = round_page(size);
220 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
221 		return (0);
222 	offset = addr - VM_MIN_KERNEL_ADDRESS;
223 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
224 	VM_OBJECT_WLOCK(object);
225 	tries = 0;
226 retry:
227 	m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
228 	    atop(size), low, high, alignment, boundary, memattr);
229 	if (m == NULL) {
230 		VM_OBJECT_WUNLOCK(object);
231 		if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
232 			vm_pageout_grow_cache(tries, low, high);
233 			VM_OBJECT_WLOCK(object);
234 			tries++;
235 			goto retry;
236 		}
237 		vmem_free(vmem, addr, size);
238 		return (0);
239 	}
240 	end_m = m + atop(size);
241 	tmp = addr;
242 	for (; m < end_m; m++) {
243 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
244 			pmap_zero_page(m);
245 		m->valid = VM_PAGE_BITS_ALL;
246 		pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
247 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
248 		tmp += PAGE_SIZE;
249 	}
250 	VM_OBJECT_WUNLOCK(object);
251 	return (addr);
252 }
253 
254 /*
255  *	kmem_suballoc:
256  *
257  *	Allocates a map to manage a subrange
258  *	of the kernel virtual address space.
259  *
260  *	Arguments are as follows:
261  *
262  *	parent		Map to take range from
263  *	min, max	Returned endpoints of map
264  *	size		Size of range to find
265  *	superpage_align	Request that min is superpage aligned
266  */
267 vm_map_t
268 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
269     vm_size_t size, boolean_t superpage_align)
270 {
271 	int ret;
272 	vm_map_t result;
273 
274 	size = round_page(size);
275 
276 	*min = vm_map_min(parent);
277 	ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
278 	    VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
279 	    MAP_ACC_NO_CHARGE);
280 	if (ret != KERN_SUCCESS)
281 		panic("kmem_suballoc: bad status return of %d", ret);
282 	*max = *min + size;
283 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
284 	if (result == NULL)
285 		panic("kmem_suballoc: cannot create submap");
286 	if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
287 		panic("kmem_suballoc: unable to change range to submap");
288 	return (result);
289 }
290 
291 /*
292  *	kmem_malloc:
293  *
294  *	Allocate wired-down pages in the kernel's address space.
295  */
296 vm_offset_t
297 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
298 {
299 	vm_offset_t addr;
300 	int rv;
301 
302 	size = round_page(size);
303 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
304 		return (0);
305 
306 	rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
307 	    addr, size, flags);
308 	if (rv != KERN_SUCCESS) {
309 		vmem_free(vmem, addr, size);
310 		return (0);
311 	}
312 	return (addr);
313 }
314 
315 /*
316  *	kmem_back:
317  *
318  *	Allocate physical pages for the specified virtual address range.
319  */
320 int
321 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
322 {
323 	vm_offset_t offset, i;
324 	vm_page_t m;
325 	int pflags;
326 
327 	KASSERT(object == kmem_object || object == kernel_object,
328 	    ("kmem_back: only supports kernel objects."));
329 
330 	offset = addr - VM_MIN_KERNEL_ADDRESS;
331 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
332 
333 	VM_OBJECT_WLOCK(object);
334 	for (i = 0; i < size; i += PAGE_SIZE) {
335 retry:
336 		m = vm_page_alloc(object, OFF_TO_IDX(offset + i), pflags);
337 
338 		/*
339 		 * Ran out of space, free everything up and return. Don't need
340 		 * to lock page queues here as we know that the pages we got
341 		 * aren't on any queues.
342 		 */
343 		if (m == NULL) {
344 			VM_OBJECT_WUNLOCK(object);
345 			if ((flags & M_NOWAIT) == 0) {
346 				VM_WAIT;
347 				VM_OBJECT_WLOCK(object);
348 				goto retry;
349 			}
350 			kmem_unback(object, addr, i);
351 			return (KERN_NO_SPACE);
352 		}
353 		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
354 			pmap_zero_page(m);
355 		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
356 		    ("kmem_malloc: page %p is managed", m));
357 		m->valid = VM_PAGE_BITS_ALL;
358 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
359 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
360 	}
361 	VM_OBJECT_WUNLOCK(object);
362 
363 	return (KERN_SUCCESS);
364 }
365 
366 /*
367  *	kmem_unback:
368  *
369  *	Unmap and free the physical pages underlying the specified virtual
370  *	address range.
371  *
372  *	A physical page must exist within the specified object at each index
373  *	that is being unmapped.
374  */
375 void
376 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
377 {
378 	vm_page_t m;
379 	vm_offset_t i, offset;
380 
381 	KASSERT(object == kmem_object || object == kernel_object,
382 	    ("kmem_unback: only supports kernel objects."));
383 
384 	pmap_remove(kernel_pmap, addr, addr + size);
385 	offset = addr - VM_MIN_KERNEL_ADDRESS;
386 	VM_OBJECT_WLOCK(object);
387 	for (i = 0; i < size; i += PAGE_SIZE) {
388 		m = vm_page_lookup(object, OFF_TO_IDX(offset + i));
389 		vm_page_unwire(m, PQ_INACTIVE);
390 		vm_page_free(m);
391 	}
392 	VM_OBJECT_WUNLOCK(object);
393 }
394 
395 /*
396  *	kmem_free:
397  *
398  *	Free memory allocated with kmem_malloc.  The size must match the
399  *	original allocation.
400  */
401 void
402 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
403 {
404 
405 	size = round_page(size);
406 	kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
407 	    addr, size);
408 	vmem_free(vmem, addr, size);
409 }
410 
411 /*
412  *	kmap_alloc_wait:
413  *
414  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
415  *	has no room, the caller sleeps waiting for more memory in the submap.
416  *
417  *	This routine may block.
418  */
419 vm_offset_t
420 kmap_alloc_wait(map, size)
421 	vm_map_t map;
422 	vm_size_t size;
423 {
424 	vm_offset_t addr;
425 
426 	size = round_page(size);
427 	if (!swap_reserve(size))
428 		return (0);
429 
430 	for (;;) {
431 		/*
432 		 * To make this work for more than one map, use the map's lock
433 		 * to lock out sleepers/wakers.
434 		 */
435 		vm_map_lock(map);
436 		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
437 			break;
438 		/* no space now; see if we can ever get space */
439 		if (vm_map_max(map) - vm_map_min(map) < size) {
440 			vm_map_unlock(map);
441 			swap_release(size);
442 			return (0);
443 		}
444 		map->needs_wakeup = TRUE;
445 		vm_map_unlock_and_wait(map, 0);
446 	}
447 	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
448 	    VM_PROT_ALL, MAP_ACC_CHARGED);
449 	vm_map_unlock(map);
450 	return (addr);
451 }
452 
453 /*
454  *	kmap_free_wakeup:
455  *
456  *	Returns memory to a submap of the kernel, and wakes up any processes
457  *	waiting for memory in that map.
458  */
459 void
460 kmap_free_wakeup(map, addr, size)
461 	vm_map_t map;
462 	vm_offset_t addr;
463 	vm_size_t size;
464 {
465 
466 	vm_map_lock(map);
467 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
468 	if (map->needs_wakeup) {
469 		map->needs_wakeup = FALSE;
470 		vm_map_wakeup(map);
471 	}
472 	vm_map_unlock(map);
473 }
474 
475 void
476 kmem_init_zero_region(void)
477 {
478 	vm_offset_t addr, i;
479 	vm_page_t m;
480 
481 	/*
482 	 * Map a single physical page of zeros to a larger virtual range.
483 	 * This requires less looping in places that want large amounts of
484 	 * zeros, while not using much more physical resources.
485 	 */
486 	addr = kva_alloc(ZERO_REGION_SIZE);
487 	m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
488 	    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
489 	if ((m->flags & PG_ZERO) == 0)
490 		pmap_zero_page(m);
491 	for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
492 		pmap_qenter(addr + i, &m, 1);
493 	pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
494 
495 	zero_region = (const void *)addr;
496 }
497 
498 /*
499  * 	kmem_init:
500  *
501  *	Create the kernel map; insert a mapping covering kernel text,
502  *	data, bss, and all space allocated thus far (`boostrap' data).  The
503  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
504  *	`start' as allocated, and the range between `start' and `end' as free.
505  */
506 void
507 kmem_init(start, end)
508 	vm_offset_t start, end;
509 {
510 	vm_map_t m;
511 
512 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
513 	m->system_map = 1;
514 	vm_map_lock(m);
515 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
516 	kernel_map = m;
517 	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
518 #ifdef __amd64__
519 	    KERNBASE,
520 #else
521 	    VM_MIN_KERNEL_ADDRESS,
522 #endif
523 	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
524 	/* ... and ending with the completion of the above `insert' */
525 	vm_map_unlock(m);
526 }
527 
528 #ifdef DIAGNOSTIC
529 /*
530  * Allow userspace to directly trigger the VM drain routine for testing
531  * purposes.
532  */
533 static int
534 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
535 {
536 	int error, i;
537 
538 	i = 0;
539 	error = sysctl_handle_int(oidp, &i, 0, req);
540 	if (error)
541 		return (error);
542 	if (i)
543 		EVENTHANDLER_INVOKE(vm_lowmem, 0);
544 	return (0);
545 }
546 
547 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
548     debug_vm_lowmem, "I", "set to trigger vm_lowmem event");
549 #endif
550