Deleted Added
sdiff udiff text old ( 252330 ) new ( 254025 )
full compact
vm_kern.c (252330) vm_kern.c (254025)
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

--- 49 unchanged lines hidden (view full) ---

58 * rights to redistribute these changes.
59 */
60
61/*
62 * Kernel memory management.
63 */
64
65#include <sys/cdefs.h>
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

--- 49 unchanged lines hidden (view full) ---

58 * rights to redistribute these changes.
59 */
60
61/*
62 * Kernel memory management.
63 */
64
65#include <sys/cdefs.h>
66__FBSDID("$FreeBSD: head/sys/vm/vm_kern.c 252330 2013-06-28 03:51:20Z jeff $");
66__FBSDID("$FreeBSD: head/sys/vm/vm_kern.c 254025 2013-08-07 06:21:20Z jeff $");
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>
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>
77
78#include <vm/vm.h>
79#include <vm/vm_param.h>
78
79#include <vm/vm.h>
80#include <vm/vm_param.h>
81#include <vm/vm_kern.h>
80#include <vm/pmap.h>
81#include <vm/vm_map.h>
82#include <vm/vm_object.h>
83#include <vm/vm_page.h>
84#include <vm/vm_pageout.h>
85#include <vm/vm_extern.h>
86#include <vm/uma.h>
87
88vm_map_t kernel_map;
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
90vm_map_t kernel_map;
89vm_map_t kmem_map;
90vm_map_t exec_map;
91vm_map_t pipe_map;
92
93const void *zero_region;
94CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
95
96SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
97 NULL, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
98
99SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
100#if defined(__arm__) || defined(__sparc64__)
101 &vm_max_kernel_address, 0,
102#else
103 NULL, VM_MAX_KERNEL_ADDRESS,
104#endif
105 "Max kernel address");
106
107/*
91vm_map_t exec_map;
92vm_map_t pipe_map;
93
94const void *zero_region;
95CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
96
97SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
98 NULL, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
99
100SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
101#if defined(__arm__) || defined(__sparc64__)
102 &vm_max_kernel_address, 0,
103#else
104 NULL, VM_MAX_KERNEL_ADDRESS,
105#endif
106 "Max kernel address");
107
108/*
108 * kmem_alloc_nofault:
109 * kva_alloc:
109 *
110 * Allocate a virtual address range with no underlying object and
111 * no initial mapping to physical memory. Any mapping from this
112 * range to physical memory must be explicitly created prior to
113 * its use, typically with pmap_qenter(). Any attempt to create
114 * a mapping on demand through vm_fault() will result in a panic.
115 */
116vm_offset_t
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 */
117vm_offset_t
117kmem_alloc_nofault(map, size)
118 vm_map_t map;
118kva_alloc(size)
119 vm_size_t size;
120{
121 vm_offset_t addr;
119 vm_size_t size;
120{
121 vm_offset_t addr;
122 int result;
123
124 size = round_page(size);
122
123 size = round_page(size);
125 addr = vm_map_min(map);
126 result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
127 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
128 if (result != KERN_SUCCESS) {
124 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr))
129 return (0);
125 return (0);
130 }
126
131 return (addr);
132}
133
134/*
127 return (addr);
128}
129
130/*
135 * kmem_alloc_nofault_space:
131 * kva_free:
136 *
132 *
137 * Allocate a virtual address range with no underlying object and
138 * no initial mapping to physical memory within the specified
139 * address space. Any mapping from this range to physical memory
140 * must be explicitly created prior to its use, typically with
141 * pmap_qenter(). Any attempt to create a mapping on demand
142 * through vm_fault() will result in a panic.
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.
143 */
138 */
144vm_offset_t
145kmem_alloc_nofault_space(map, size, find_space)
146 vm_map_t map;
147 vm_size_t size;
148 int find_space;
149{
139void
140kva_free(addr, size)
150 vm_offset_t addr;
141 vm_offset_t addr;
151 int result;
152
153 size = round_page(size);
154 addr = vm_map_min(map);
155 result = vm_map_find(map, NULL, 0, &addr, size, find_space,
156 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
157 if (result != KERN_SUCCESS) {
158 return (0);
159 }
160 return (addr);
161}
162
163/*
164 * Allocate wired-down memory in the kernel's address map
165 * or a submap.
166 */
167vm_offset_t
168kmem_alloc(map, size)
169 vm_map_t map;
170 vm_size_t size;
171{
142 vm_size_t size;
143{
172 vm_offset_t addr;
173 vm_offset_t offset;
174
175 size = round_page(size);
144
145 size = round_page(size);
176
177 /*
178 * Use the kernel object for wired-down kernel pages. Assume that no
179 * region of the kernel object is referenced more than once.
180 */
181
182 /*
183 * Locate sufficient space in the map. This will give us the final
184 * virtual address for the new memory, and thus will tell us the
185 * offset within the kernel map.
186 */
187 vm_map_lock(map);
188 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
189 vm_map_unlock(map);
190 return (0);
191 }
192 offset = addr - VM_MIN_KERNEL_ADDRESS;
193 vm_object_reference(kernel_object);
194 vm_map_insert(map, kernel_object, offset, addr, addr + size,
195 VM_PROT_ALL, VM_PROT_ALL, 0);
196 vm_map_unlock(map);
197
198 /*
199 * And finally, mark the data as non-pageable.
200 */
201 (void) vm_map_wire(map, addr, addr + size,
202 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
203
204 return (addr);
146 vmem_free(kernel_arena, addr, size);
205}
206
207/*
208 * Allocates a region from the kernel address map and physical pages
209 * within the specified address range to the kernel object. Creates a
210 * wired mapping from this region to these pages, and returns the
211 * region's starting virtual address. The allocated pages are not
212 * necessarily physically contiguous. If M_ZERO is specified through the
213 * given flags, then the pages are zeroed before they are mapped.
214 */
215vm_offset_t
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 */
157vm_offset_t
216kmem_alloc_attr(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
158kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low,
217 vm_paddr_t high, vm_memattr_t memattr)
218{
159 vm_paddr_t high, vm_memattr_t memattr)
160{
219 vm_object_t object = kernel_object;
161 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
220 vm_offset_t addr;
162 vm_offset_t addr;
221 vm_ooffset_t end_offset, offset;
163 vm_ooffset_t offset;
222 vm_page_t m;
223 int pflags, tries;
164 vm_page_t m;
165 int pflags, tries;
166 int i;
224
225 size = round_page(size);
167
168 size = round_page(size);
226 vm_map_lock(map);
227 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
228 vm_map_unlock(map);
169 if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
229 return (0);
170 return (0);
230 }
231 offset = addr - VM_MIN_KERNEL_ADDRESS;
171 offset = addr - VM_MIN_KERNEL_ADDRESS;
232 vm_object_reference(object);
233 vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
234 VM_PROT_ALL, 0);
235 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
172 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
236 VM_OBJECT_WLOCK(object);
173 VM_OBJECT_WLOCK(object);
237 end_offset = offset + size;
238 for (; offset < end_offset; offset += PAGE_SIZE) {
174 for (i = 0; i < size; i += PAGE_SIZE) {
239 tries = 0;
240retry:
175 tries = 0;
176retry:
241 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags, 1,
242 low, high, PAGE_SIZE, 0, memattr);
177 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset + i),
178 pflags, 1, low, high, PAGE_SIZE, 0, memattr);
243 if (m == NULL) {
244 VM_OBJECT_WUNLOCK(object);
245 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
179 if (m == NULL) {
180 VM_OBJECT_WUNLOCK(object);
181 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
246 vm_map_unlock(map);
247 vm_pageout_grow_cache(tries, low, high);
182 vm_pageout_grow_cache(tries, low, high);
248 vm_map_lock(map);
249 VM_OBJECT_WLOCK(object);
250 tries++;
251 goto retry;
252 }
183 VM_OBJECT_WLOCK(object);
184 tries++;
185 goto retry;
186 }
253
254 /*
255 * Since the pages that were allocated by any previous
256 * iterations of this loop are not busy, they can be
257 * freed by vm_object_page_remove(), which is called
258 * by vm_map_delete().
187 /*
188 * Unmap and free the pages.
259 */
189 */
260 vm_map_delete(map, addr, addr + size);
261 vm_map_unlock(map);
190 if (i != 0)
191 pmap_remove(kernel_pmap, addr, addr + i);
192 while (i != 0) {
193 i -= PAGE_SIZE;
194 m = vm_page_lookup(object,
195 OFF_TO_IDX(offset + i));
196 vm_page_unwire(m, 0);
197 vm_page_free(m);
198 }
199 vmem_free(vmem, addr, size);
262 return (0);
263 }
264 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
265 pmap_zero_page(m);
266 m->valid = VM_PAGE_BITS_ALL;
200 return (0);
201 }
202 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
203 pmap_zero_page(m);
204 m->valid = VM_PAGE_BITS_ALL;
205 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
206 TRUE);
267 }
268 VM_OBJECT_WUNLOCK(object);
207 }
208 VM_OBJECT_WUNLOCK(object);
269 vm_map_unlock(map);
270 vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
271 VM_MAP_WIRE_NOHOLES);
272 return (addr);
273}
274
275/*
276 * Allocates a region from the kernel address map and physically
277 * contiguous pages within the specified address range to the kernel
278 * object. Creates a wired mapping from this region to these pages, and
279 * returns the region's starting virtual address. If M_ZERO is specified
280 * through the given flags, then the pages are zeroed before they are
281 * mapped.
282 */
283vm_offset_t
209 return (addr);
210}
211
212/*
213 * Allocates a region from the kernel address map and physically
214 * contiguous pages within the specified address range to the kernel
215 * object. Creates a wired mapping from this region to these pages, and
216 * returns the region's starting virtual address. If M_ZERO is specified
217 * through the given flags, then the pages are zeroed before they are
218 * mapped.
219 */
220vm_offset_t
284kmem_alloc_contig(vm_map_t map, vm_size_t size, int flags, vm_paddr_t low,
221kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low,
285 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
286 vm_memattr_t memattr)
287{
222 vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
223 vm_memattr_t memattr)
224{
288 vm_object_t object = kernel_object;
289 vm_offset_t addr;
225 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object;
226 vm_offset_t addr, tmp;
290 vm_ooffset_t offset;
291 vm_page_t end_m, m;
292 int pflags, tries;
293
294 size = round_page(size);
227 vm_ooffset_t offset;
228 vm_page_t end_m, m;
229 int pflags, tries;
230
231 size = round_page(size);
295 vm_map_lock(map);
296 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
297 vm_map_unlock(map);
232 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
298 return (0);
233 return (0);
299 }
300 offset = addr - VM_MIN_KERNEL_ADDRESS;
234 offset = addr - VM_MIN_KERNEL_ADDRESS;
301 vm_object_reference(object);
302 vm_map_insert(map, object, offset, addr, addr + size, VM_PROT_ALL,
303 VM_PROT_ALL, 0);
304 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY;
235 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
305 VM_OBJECT_WLOCK(object);
306 tries = 0;
307retry:
308 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
309 atop(size), low, high, alignment, boundary, memattr);
310 if (m == NULL) {
311 VM_OBJECT_WUNLOCK(object);
312 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
236 VM_OBJECT_WLOCK(object);
237 tries = 0;
238retry:
239 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags,
240 atop(size), low, high, alignment, boundary, memattr);
241 if (m == NULL) {
242 VM_OBJECT_WUNLOCK(object);
243 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
313 vm_map_unlock(map);
314 vm_pageout_grow_cache(tries, low, high);
244 vm_pageout_grow_cache(tries, low, high);
315 vm_map_lock(map);
316 VM_OBJECT_WLOCK(object);
317 tries++;
318 goto retry;
319 }
245 VM_OBJECT_WLOCK(object);
246 tries++;
247 goto retry;
248 }
320 vm_map_delete(map, addr, addr + size);
321 vm_map_unlock(map);
249 vmem_free(vmem, addr, size);
322 return (0);
323 }
324 end_m = m + atop(size);
250 return (0);
251 }
252 end_m = m + atop(size);
253 tmp = addr;
325 for (; m < end_m; m++) {
326 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
327 pmap_zero_page(m);
328 m->valid = VM_PAGE_BITS_ALL;
254 for (; m < end_m; m++) {
255 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
256 pmap_zero_page(m);
257 m->valid = VM_PAGE_BITS_ALL;
258 pmap_enter(kernel_pmap, tmp, VM_PROT_ALL, m, VM_PROT_ALL, true);
259 tmp += PAGE_SIZE;
329 }
330 VM_OBJECT_WUNLOCK(object);
260 }
261 VM_OBJECT_WUNLOCK(object);
331 vm_map_unlock(map);
332 vm_map_wire(map, addr, addr + size, VM_MAP_WIRE_SYSTEM |
333 VM_MAP_WIRE_NOHOLES);
334 return (addr);
335}
336
337/*
262 return (addr);
263}
264
265/*
338 * kmem_free:
339 *
340 * Release a region of kernel virtual memory allocated
341 * with kmem_alloc, and return the physical pages
342 * associated with that region.
343 *
344 * This routine may not block on kernel maps.
345 */
346void
347kmem_free(map, addr, size)
348 vm_map_t map;
349 vm_offset_t addr;
350 vm_size_t size;
351{
352
353 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
354}
355
356/*
357 * kmem_suballoc:
358 *
359 * Allocates a map to manage a subrange
360 * of the kernel virtual address space.
361 *
362 * Arguments are as follows:
363 *
364 * parent Map to take range from

--- 23 unchanged lines hidden (view full) ---

388 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
389 panic("kmem_suballoc: unable to change range to submap");
390 return (result);
391}
392
393/*
394 * kmem_malloc:
395 *
266 * kmem_suballoc:
267 *
268 * Allocates a map to manage a subrange
269 * of the kernel virtual address space.
270 *
271 * Arguments are as follows:
272 *
273 * parent Map to take range from

--- 23 unchanged lines hidden (view full) ---

297 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
298 panic("kmem_suballoc: unable to change range to submap");
299 return (result);
300}
301
302/*
303 * kmem_malloc:
304 *
396 * Allocate wired-down memory in the kernel's address map for the higher
397 * level kernel memory allocator (kern/kern_malloc.c). We cannot use
398 * kmem_alloc() because we may need to allocate memory at interrupt
399 * level where we cannot block (canwait == FALSE).
400 *
401 * This routine has its own private kernel submap (kmem_map) and object
402 * (kmem_object). This, combined with the fact that only malloc uses
403 * this routine, ensures that we will never block in map or object waits.
404 *
405 * We don't worry about expanding the map (adding entries) since entries
406 * for wired maps are statically allocated.
407 *
408 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
409 * which we never free.
305 * Allocate wired-down pages in the kernel's address space.
410 */
411vm_offset_t
306 */
307vm_offset_t
412kmem_malloc(map, size, flags)
413 vm_map_t map;
414 vm_size_t size;
415 int flags;
308kmem_malloc(struct vmem *vmem, vm_size_t size, int flags)
416{
417 vm_offset_t addr;
309{
310 vm_offset_t addr;
418 int i, rv;
311 int rv;
419
420 size = round_page(size);
312
313 size = round_page(size);
421 addr = vm_map_min(map);
314 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
315 return (0);
422
316
423 /*
424 * Locate sufficient space in the map. This will give us the final
425 * virtual address for the new memory, and thus will tell us the
426 * offset within the kernel map.
427 */
428 vm_map_lock(map);
429 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
430 vm_map_unlock(map);
431 if ((flags & M_NOWAIT) == 0) {
432 for (i = 0; i < 8; i++) {
433 EVENTHANDLER_INVOKE(vm_lowmem, 0);
434 uma_reclaim();
435 vm_map_lock(map);
436 if (vm_map_findspace(map, vm_map_min(map),
437 size, &addr) == 0) {
438 break;
439 }
440 vm_map_unlock(map);
441 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
442 }
443 if (i == 8) {
444 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
445 (long)size, (long)map->size);
446 }
447 } else {
448 return (0);
449 }
317 rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object,
318 addr, size, flags);
319 if (rv != KERN_SUCCESS) {
320 vmem_free(vmem, addr, size);
321 return (0);
450 }
322 }
451
452 rv = kmem_back(map, addr, size, flags);
453 vm_map_unlock(map);
454 return (rv == KERN_SUCCESS ? addr : 0);
323 return (addr);
455}
456
457/*
458 * kmem_back:
459 *
460 * Allocate physical pages for the specified virtual address range.
461 */
462int
324}
325
326/*
327 * kmem_back:
328 *
329 * Allocate physical pages for the specified virtual address range.
330 */
331int
463kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
332kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
464{
465 vm_offset_t offset, i;
333{
334 vm_offset_t offset, i;
466 vm_map_entry_t entry;
467 vm_page_t m;
468 int pflags;
335 vm_page_t m;
336 int pflags;
469 boolean_t found;
470
337
471 KASSERT(vm_map_locked(map), ("kmem_back: map %p is not locked", map));
338 KASSERT(object == kmem_object || object == kernel_object,
339 ("kmem_back: only supports kernel objects."));
340
472 offset = addr - VM_MIN_KERNEL_ADDRESS;
341 offset = addr - VM_MIN_KERNEL_ADDRESS;
473 vm_object_reference(kmem_object);
474 vm_map_insert(map, kmem_object, offset, addr, addr + size,
475 VM_PROT_ALL, VM_PROT_ALL, 0);
342 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
476
343
477 /*
478 * Assert: vm_map_insert() will never be able to extend the
479 * previous entry so vm_map_lookup_entry() will find a new
480 * entry exactly corresponding to this address range and it
481 * will have wired_count == 0.
482 */
483 found = vm_map_lookup_entry(map, addr, &entry);
484 KASSERT(found && entry->start == addr && entry->end == addr + size &&
485 entry->wired_count == 0 && (entry->eflags & MAP_ENTRY_IN_TRANSITION)
486 == 0, ("kmem_back: entry not found or misaligned"));
487
488 pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
489
490 VM_OBJECT_WLOCK(kmem_object);
344 VM_OBJECT_WLOCK(object);
491 for (i = 0; i < size; i += PAGE_SIZE) {
492retry:
345 for (i = 0; i < size; i += PAGE_SIZE) {
346retry:
493 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
347 m = vm_page_alloc(object, OFF_TO_IDX(offset + i), pflags);
494
495 /*
496 * Ran out of space, free everything up and return. Don't need
497 * to lock page queues here as we know that the pages we got
498 * aren't on any queues.
499 */
500 if (m == NULL) {
501 if ((flags & M_NOWAIT) == 0) {
348
349 /*
350 * Ran out of space, free everything up and return. Don't need
351 * to lock page queues here as we know that the pages we got
352 * aren't on any queues.
353 */
354 if (m == NULL) {
355 if ((flags & M_NOWAIT) == 0) {
502 VM_OBJECT_WUNLOCK(kmem_object);
503 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
504 vm_map_unlock(map);
356 VM_OBJECT_WUNLOCK(object);
505 VM_WAIT;
357 VM_WAIT;
506 vm_map_lock(map);
507 KASSERT(
508(entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_NEEDS_WAKEUP)) ==
509 MAP_ENTRY_IN_TRANSITION,
510 ("kmem_back: volatile entry"));
511 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
512 VM_OBJECT_WLOCK(kmem_object);
358 VM_OBJECT_WLOCK(object);
513 goto retry;
514 }
515 /*
359 goto retry;
360 }
361 /*
516 * Free the pages before removing the map entry.
517 * They are already marked busy. Calling
518 * vm_map_delete before the pages has been freed or
519 * unbusied will cause a deadlock.
362 * Unmap and free the pages.
520 */
363 */
364 if (i != 0)
365 pmap_remove(kernel_pmap, addr, addr + i);
521 while (i != 0) {
522 i -= PAGE_SIZE;
366 while (i != 0) {
367 i -= PAGE_SIZE;
523 m = vm_page_lookup(kmem_object,
368 m = vm_page_lookup(object,
524 OFF_TO_IDX(offset + i));
525 vm_page_unwire(m, 0);
526 vm_page_free(m);
527 }
369 OFF_TO_IDX(offset + i));
370 vm_page_unwire(m, 0);
371 vm_page_free(m);
372 }
528 VM_OBJECT_WUNLOCK(kmem_object);
529 vm_map_delete(map, addr, addr + size);
373 VM_OBJECT_WUNLOCK(object);
530 return (KERN_NO_SPACE);
531 }
532 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
533 pmap_zero_page(m);
374 return (KERN_NO_SPACE);
375 }
376 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
377 pmap_zero_page(m);
534 m->valid = VM_PAGE_BITS_ALL;
535 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
536 ("kmem_malloc: page %p is managed", m));
378 KASSERT((m->oflags & VPO_UNMANAGED) != 0,
379 ("kmem_malloc: page %p is managed", m));
380 m->valid = VM_PAGE_BITS_ALL;
381 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
382 TRUE);
537 }
383 }
538 VM_OBJECT_WUNLOCK(kmem_object);
384 VM_OBJECT_WUNLOCK(object);
539
385
540 /*
541 * Mark map entry as non-pageable. Repeat the assert.
542 */
543 KASSERT(entry->start == addr && entry->end == addr + size &&
544 entry->wired_count == 0,
545 ("kmem_back: entry not found or misaligned after allocation"));
546 entry->wired_count = 1;
386 return (KERN_SUCCESS);
387}
547
388
548 /*
549 * At this point, the kmem_object must be unlocked because
550 * vm_map_simplify_entry() calls vm_object_deallocate(), which
551 * locks the kmem_object.
552 */
553 vm_map_simplify_entry(map, entry);
389void
390kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
391{
392 vm_page_t m;
393 vm_offset_t offset;
394 int i;
554
395
555 /*
556 * Loop thru pages, entering them in the pmap.
557 */
558 VM_OBJECT_WLOCK(kmem_object);
396 KASSERT(object == kmem_object || object == kernel_object,
397 ("kmem_unback: only supports kernel objects."));
398
399 offset = addr - VM_MIN_KERNEL_ADDRESS;
400 VM_OBJECT_WLOCK(object);
401 pmap_remove(kernel_pmap, addr, addr + size);
559 for (i = 0; i < size; i += PAGE_SIZE) {
402 for (i = 0; i < size; i += PAGE_SIZE) {
560 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
561 /*
562 * Because this is kernel_pmap, this call will not block.
563 */
564 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
565 TRUE);
566 vm_page_wakeup(m);
403 m = vm_page_lookup(object, OFF_TO_IDX(offset + i));
404 vm_page_unwire(m, 0);
405 vm_page_free(m);
567 }
406 }
568 VM_OBJECT_WUNLOCK(kmem_object);
407 VM_OBJECT_WUNLOCK(object);
408}
569
409
570 return (KERN_SUCCESS);
410/*
411 * kmem_free:
412 *
413 * Free memory allocated with kmem_malloc. The size must match the
414 * original allocation.
415 */
416void
417kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size)
418{
419
420 size = round_page(size);
421 kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object,
422 addr, size);
423 vmem_free(vmem, addr, size);
571}
572
573/*
424}
425
426/*
574 * kmem_alloc_wait:
427 * kmap_alloc_wait:
575 *
576 * Allocates pageable memory from a sub-map of the kernel. If the submap
577 * has no room, the caller sleeps waiting for more memory in the submap.
578 *
579 * This routine may block.
580 */
581vm_offset_t
428 *
429 * Allocates pageable memory from a sub-map of the kernel. If the submap
430 * has no room, the caller sleeps waiting for more memory in the submap.
431 *
432 * This routine may block.
433 */
434vm_offset_t
582kmem_alloc_wait(map, size)
435kmap_alloc_wait(map, size)
583 vm_map_t map;
584 vm_size_t size;
585{
586 vm_offset_t addr;
587
588 size = round_page(size);
589 if (!swap_reserve(size))
590 return (0);

--- 17 unchanged lines hidden (view full) ---

608 }
609 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
610 VM_PROT_ALL, MAP_ACC_CHARGED);
611 vm_map_unlock(map);
612 return (addr);
613}
614
615/*
436 vm_map_t map;
437 vm_size_t size;
438{
439 vm_offset_t addr;
440
441 size = round_page(size);
442 if (!swap_reserve(size))
443 return (0);

--- 17 unchanged lines hidden (view full) ---

461 }
462 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL,
463 VM_PROT_ALL, MAP_ACC_CHARGED);
464 vm_map_unlock(map);
465 return (addr);
466}
467
468/*
616 * kmem_free_wakeup:
469 * kmap_free_wakeup:
617 *
618 * Returns memory to a submap of the kernel, and wakes up any processes
619 * waiting for memory in that map.
620 */
621void
470 *
471 * Returns memory to a submap of the kernel, and wakes up any processes
472 * waiting for memory in that map.
473 */
474void
622kmem_free_wakeup(map, addr, size)
475kmap_free_wakeup(map, addr, size)
623 vm_map_t map;
624 vm_offset_t addr;
625 vm_size_t size;
626{
627
628 vm_map_lock(map);
629 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
630 if (map->needs_wakeup) {
631 map->needs_wakeup = FALSE;
632 vm_map_wakeup(map);
633 }
634 vm_map_unlock(map);
635}
636
476 vm_map_t map;
477 vm_offset_t addr;
478 vm_size_t size;
479{
480
481 vm_map_lock(map);
482 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
483 if (map->needs_wakeup) {
484 map->needs_wakeup = FALSE;
485 vm_map_wakeup(map);
486 }
487 vm_map_unlock(map);
488}
489
637static void
490void
638kmem_init_zero_region(void)
639{
640 vm_offset_t addr, i;
641 vm_page_t m;
491kmem_init_zero_region(void)
492{
493 vm_offset_t addr, i;
494 vm_page_t m;
642 int error;
643
644 /*
645 * Map a single physical page of zeros to a larger virtual range.
646 * This requires less looping in places that want large amounts of
647 * zeros, while not using much more physical resources.
648 */
495
496 /*
497 * Map a single physical page of zeros to a larger virtual range.
498 * This requires less looping in places that want large amounts of
499 * zeros, while not using much more physical resources.
500 */
649 addr = kmem_alloc_nofault(kernel_map, ZERO_REGION_SIZE);
501 addr = kva_alloc(ZERO_REGION_SIZE);
650 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
651 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
652 if ((m->flags & PG_ZERO) == 0)
653 pmap_zero_page(m);
654 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
655 pmap_qenter(addr + i, &m, 1);
502 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL |
503 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO);
504 if ((m->flags & PG_ZERO) == 0)
505 pmap_zero_page(m);
506 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
507 pmap_qenter(addr + i, &m, 1);
656 error = vm_map_protect(kernel_map, addr, addr + ZERO_REGION_SIZE,
657 VM_PROT_READ, TRUE);
658 KASSERT(error == 0, ("error=%d", error));
508 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
659
660 zero_region = (const void *)addr;
661}
662
663/*
664 * kmem_init:
665 *
666 * Create the kernel map; insert a mapping covering kernel text,

--- 16 unchanged lines hidden (view full) ---

683#ifdef __amd64__
684 KERNBASE,
685#else
686 VM_MIN_KERNEL_ADDRESS,
687#endif
688 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
689 /* ... and ending with the completion of the above `insert' */
690 vm_map_unlock(m);
509
510 zero_region = (const void *)addr;
511}
512
513/*
514 * kmem_init:
515 *
516 * Create the kernel map; insert a mapping covering kernel text,

--- 16 unchanged lines hidden (view full) ---

533#ifdef __amd64__
534 KERNBASE,
535#else
536 VM_MIN_KERNEL_ADDRESS,
537#endif
538 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
539 /* ... and ending with the completion of the above `insert' */
540 vm_map_unlock(m);
691
692 kmem_init_zero_region();
693}
694
695#ifdef DIAGNOSTIC
696/*
697 * Allow userspace to directly trigger the VM drain routine for testing
698 * purposes.
699 */
700static int

--- 16 unchanged lines hidden --- 		541}
542
543#ifdef DIAGNOSTIC
544/*
545 * Allow userspace to directly trigger the VM drain routine for testing
546 * purposes.
547 */
548static int

--- 16 unchanged lines hidden ---