xref: /NextBSD/sys/vm/vm_kern.c (revision 4557fabb34e865d7f40be64b39c9e34fa41dbb60)
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
kva_alloc(size)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
kva_free(addr,size)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
kmem_alloc_attr(vmem_t * vmem,vm_size_t size,int flags,vm_paddr_t low,vm_paddr_t high,vm_memattr_t memattr)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, level;
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 		level = 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 (level < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
184 				vm_pageout_reclaim_contig(1, low, high,
185 				    PAGE_SIZE, 0, level);
186 				VM_OBJECT_WLOCK(object);
187 				level++;
188 				goto retry;
189 			}
190 			kmem_unback(object, addr, i);
191 			vmem_free(vmem, addr, size);
192 			return (0);
193 		}
194 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
195 			pmap_zero_page(m);
196 		m->valid = VM_PAGE_BITS_ALL;
197 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
198 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
199 	}
200 	VM_OBJECT_WUNLOCK(object);
201 	return (addr);
202 }
203 
204 /*
205  *	Allocates a region from the kernel address map and physically
206  *	contiguous pages within the specified address range to the kernel
207  *	object.  Creates a wired mapping from this region to these pages, and
208  *	returns the region's starting virtual address.  If M_ZERO is specified
209  *	through the given flags, then the pages are zeroed before they are
210  *	mapped.
211  */
212 vm_offset_t
kmem_alloc_contig(struct vmem * vmem,vm_size_t size,int flags,vm_paddr_t low,vm_paddr_t high,u_long alignment,vm_paddr_t boundary,vm_memattr_t memattr)213 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
214     vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
215     vm_memattr_t memattr)
216 {
217 	vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
218 	vm_offset_t addr, tmp;
219 	vm_ooffset_t offset;
220 	vm_page_t end_m, m;
221 	int pflags, level;
222 	u_long npages;
223 
224 	size = round_page(size);
225 	npages = atop(size);
226 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
227 		return (0);
228 	offset = addr - VM_MIN_KERNEL_ADDRESS;
229 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
230 	VM_OBJECT_WLOCK(object);
231 	level = 0;
232 retry:
233 	m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
234 	    npages, low, high, alignment, boundary, memattr);
235 	if (m == NULL) {
236 		VM_OBJECT_WUNLOCK(object);
237 		if (level < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
238 			vm_pageout_reclaim_contig(atop(size), low, high,
239 			    alignment, boundary, level);
240 			VM_OBJECT_WLOCK(object);
241 			level++;
242 			goto retry;
243 		}
244 		vmem_free(vmem, addr, size);
245 		return (0);
246 	}
247 	end_m = m + npages;
248 	tmp = addr;
249 	for (; m < end_m; m++) {
250 		if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
251 			pmap_zero_page(m);
252 		m->valid = VM_PAGE_BITS_ALL;
253 		pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
254 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
255 		tmp += PAGE_SIZE;
256 	}
257 	VM_OBJECT_WUNLOCK(object);
258 	return (addr);
259 }
260 
261 /*
262  *	kmem_suballoc:
263  *
264  *	Allocates a map to manage a subrange
265  *	of the kernel virtual address space.
266  *
267  *	Arguments are as follows:
268  *
269  *	parent		Map to take range from
270  *	min, max	Returned endpoints of map
271  *	size		Size of range to find
272  *	superpage_align	Request that min is superpage aligned
273  */
274 vm_map_t
kmem_suballoc(vm_map_t parent,vm_offset_t * min,vm_offset_t * max,vm_size_t size,boolean_t superpage_align)275 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
276     vm_size_t size, boolean_t superpage_align)
277 {
278 	int ret;
279 	vm_map_t result;
280 
281 	size = round_page(size);
282 
283 	*min = vm_map_min(parent);
284 	ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
285 	    VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
286 	    MAP_ACC_NO_CHARGE);
287 	if (ret != KERN_SUCCESS)
288 		panic("kmem_suballoc: bad status return of %d", ret);
289 	*max = *min + size;
290 	result = vm_map_create(vm_map_pmap(parent), *min, *max);
291 	if (result == NULL)
292 		panic("kmem_suballoc: cannot create submap");
293 	if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
294 		panic("kmem_suballoc: unable to change range to submap");
295 	return (result);
296 }
297 
298 /*
299  *	kmem_malloc:
300  *
301  *	Allocate wired-down pages in the kernel's address space.
302  */
303 vm_offset_t
kmem_malloc(struct vmem * vmem,vm_size_t size,int flags)304 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
305 {
306 	vm_offset_t addr;
307 	int rv;
308 
309 	size = round_page(size);
310 	if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
311 		return (0);
312 
313 	rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
314 	    addr, size, flags);
315 	if (rv != KERN_SUCCESS) {
316 		vmem_free(vmem, addr, size);
317 		return (0);
318 	}
319 	return (addr);
320 }
321 
322 /*
323  *	kmem_back:
324  *
325  *	Allocate physical pages for the specified virtual address range.
326  */
327 int
kmem_back(vm_object_t object,vm_offset_t addr,vm_size_t size,int flags)328 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
329 {
330 	vm_offset_t offset, i;
331 	vm_page_t m;
332 	int pflags;
333 
334 	KASSERT(object == kmem_object || object == kernel_object,
335 	    ("kmem_back: only supports kernel objects."));
336 
337 	offset = addr - VM_MIN_KERNEL_ADDRESS;
338 	pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
339 
340 	VM_OBJECT_WLOCK(object);
341 	for (i = 0; i < size; i += PAGE_SIZE) {
342 retry:
343 		m = vm_page_alloc(object, OFF_TO_IDX(offset + i), pflags);
344 
345 		/*
346 		 * Ran out of space, free everything up and return. Don't need
347 		 * to lock page queues here as we know that the pages we got
348 		 * aren't on any queues.
349 		 */
350 		if (m == NULL) {
351 			VM_OBJECT_WUNLOCK(object);
352 			if ((flags & M_NOWAIT) == 0) {
353 				VM_WAIT;
354 				VM_OBJECT_WLOCK(object);
355 				goto retry;
356 			}
357 			kmem_unback(object, addr, i);
358 			return (KERN_NO_SPACE);
359 		}
360 		if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
361 			pmap_zero_page(m);
362 		KASSERT((m->oflags & VPO_UNMANAGED) != 0,
363 		    ("kmem_malloc: page %p is managed", m));
364 		m->valid = VM_PAGE_BITS_ALL;
365 		pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
366 		    VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
367 	}
368 	VM_OBJECT_WUNLOCK(object);
369 
370 	return (KERN_SUCCESS);
371 }
372 
373 /*
374  *	kmem_unback:
375  *
376  *	Unmap and free the physical pages underlying the specified virtual
377  *	address range.
378  *
379  *	A physical page must exist within the specified object at each index
380  *	that is being unmapped.
381  */
382 void
kmem_unback(vm_object_t object,vm_offset_t addr,vm_size_t size)383 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
384 {
385 	vm_page_t m;
386 	vm_offset_t i, offset;
387 
388 	KASSERT(object == kmem_object || object == kernel_object,
389 	    ("kmem_unback: only supports kernel objects."));
390 
391 	pmap_remove(kernel_pmap, addr, addr + size);
392 	offset = addr - VM_MIN_KERNEL_ADDRESS;
393 	VM_OBJECT_WLOCK(object);
394 	for (i = 0; i < size; i += PAGE_SIZE) {
395 		m = vm_page_lookup(object, OFF_TO_IDX(offset + i));
396 		vm_page_unwire(m, PQ_NONE);
397 		vm_page_free(m);
398 	}
399 	VM_OBJECT_WUNLOCK(object);
400 }
401 
402 /*
403  *	kmem_free:
404  *
405  *	Free memory allocated with kmem_malloc.  The size must match the
406  *	original allocation.
407  */
408 void
kmem_free(struct vmem * vmem,vm_offset_t addr,vm_size_t size)409 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
410 {
411 
412 	size = round_page(size);
413 	kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
414 	    addr, size);
415 	vmem_free(vmem, addr, size);
416 }
417 
418 /*
419  *	kmap_alloc_wait:
420  *
421  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
422  *	has no room, the caller sleeps waiting for more memory in the submap.
423  *
424  *	This routine may block.
425  */
426 vm_offset_t
kmap_alloc_wait(map,size)427 kmap_alloc_wait(map, size)
428 	vm_map_t map;
429 	vm_size_t size;
430 {
431 	vm_offset_t addr;
432 
433 	size = round_page(size);
434 	if (!swap_reserve(size))
435 		return (0);
436 
437 	for (;;) {
438 		/*
439 		 * To make this work for more than one map, use the map's lock
440 		 * to lock out sleepers/wakers.
441 		 */
442 		vm_map_lock(map);
443 		if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
444 			break;
445 		/* no space now; see if we can ever get space */
446 		if (vm_map_max(map) - vm_map_min(map) < size) {
447 			vm_map_unlock(map);
448 			swap_release(size);
449 			return (0);
450 		}
451 		map->needs_wakeup = TRUE;
452 		vm_map_unlock_and_wait(map, 0);
453 	}
454 	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
455 	    VM_PROT_ALL, MAP_ACC_CHARGED);
456 	vm_map_unlock(map);
457 	return (addr);
458 }
459 
460 /*
461  *	kmap_free_wakeup:
462  *
463  *	Returns memory to a submap of the kernel, and wakes up any processes
464  *	waiting for memory in that map.
465  */
466 void
kmap_free_wakeup(map,addr,size)467 kmap_free_wakeup(map, addr, size)
468 	vm_map_t map;
469 	vm_offset_t addr;
470 	vm_size_t size;
471 {
472 
473 	vm_map_lock(map);
474 	(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
475 	if (map->needs_wakeup) {
476 		map->needs_wakeup = FALSE;
477 		vm_map_wakeup(map);
478 	}
479 	vm_map_unlock(map);
480 }
481 
482 void
kmem_init_zero_region(void)483 kmem_init_zero_region(void)
484 {
485 	vm_offset_t addr, i;
486 	vm_page_t m;
487 
488 	/*
489 	 * Map a single physical page of zeros to a larger virtual range.
490 	 * This requires less looping in places that want large amounts of
491 	 * zeros, while not using much more physical resources.
492 	 */
493 	addr = kva_alloc(ZERO_REGION_SIZE);
494 	m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
495 	    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
496 	if ((m->flags & PG_ZERO) == 0)
497 		pmap_zero_page(m);
498 	for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
499 		pmap_qenter(addr + i, &m, 1);
500 	pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
501 
502 	zero_region = (const void *)addr;
503 }
504 
505 /*
506  * 	kmem_init:
507  *
508  *	Create the kernel map; insert a mapping covering kernel text,
509  *	data, bss, and all space allocated thus far (`boostrap' data).  The
510  *	new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
511  *	`start' as allocated, and the range between `start' and `end' as free.
512  */
513 void
kmem_init(start,end)514 kmem_init(start, end)
515 	vm_offset_t start, end;
516 {
517 	vm_map_t m;
518 
519 	m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
520 	m->system_map = 1;
521 	vm_map_lock(m);
522 	/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
523 	kernel_map = m;
524 	(void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
525 #ifdef __amd64__
526 	    KERNBASE,
527 #else
528 	    VM_MIN_KERNEL_ADDRESS,
529 #endif
530 	    start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
531 	/* ... and ending with the completion of the above `insert' */
532 	vm_map_unlock(m);
533 }
534 
535 #ifdef DIAGNOSTIC
536 /*
537  * Allow userspace to directly trigger the VM drain routine for testing
538  * purposes.
539  */
540 static int
debug_vm_lowmem(SYSCTL_HANDLER_ARGS)541 debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
542 {
543 	int error, i;
544 
545 	i = 0;
546 	error = sysctl_handle_int(oidp, &i, 0, req);
547 	if (error)
548 		return (error);
549 	if (i)
550 		EVENTHANDLER_INVOKE(vm_lowmem, 0);
551 	return (0);
552 }
553 
554 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0,
555     debug_vm_lowmem, "I", "set to trigger vm_lowmem event");
556 #endif
557