]> git.proxmox.com Git - mirror_qemu.git/blob - backends/hostmem.c
event_notifier: prevent accidental use after close
[mirror_qemu.git] / backends / hostmem.c
1 /*
2 * QEMU Host Memory Backend
3 *
4 * Copyright (C) 2013-2014 Red Hat Inc
5 *
6 * Authors:
7 * Igor Mammedov <imammedo@redhat.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
11 */
12 #include "qemu/osdep.h"
13 #include "sysemu/hostmem.h"
14 #include "hw/boards.h"
15 #include "qapi/error.h"
16 #include "qapi/visitor.h"
17 #include "qapi-types.h"
18 #include "qapi-visit.h"
19 #include "qemu/config-file.h"
20 #include "qom/object_interfaces.h"
21
22 #ifdef CONFIG_NUMA
23 #include <numaif.h>
24 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT != MPOL_DEFAULT);
25 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED != MPOL_PREFERRED);
26 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND != MPOL_BIND);
27 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE != MPOL_INTERLEAVE);
28 #endif
29
30 static void
31 host_memory_backend_get_size(Object *obj, Visitor *v, const char *name,
32 void *opaque, Error **errp)
33 {
34 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
35 uint64_t value = backend->size;
36
37 visit_type_size(v, name, &value, errp);
38 }
39
40 static void
41 host_memory_backend_set_size(Object *obj, Visitor *v, const char *name,
42 void *opaque, Error **errp)
43 {
44 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
45 Error *local_err = NULL;
46 uint64_t value;
47
48 if (memory_region_size(&backend->mr)) {
49 error_setg(&local_err, "cannot change property value");
50 goto out;
51 }
52
53 visit_type_size(v, name, &value, &local_err);
54 if (local_err) {
55 goto out;
56 }
57 if (!value) {
58 error_setg(&local_err, "Property '%s.%s' doesn't take value '%"
59 PRIu64 "'", object_get_typename(obj), name, value);
60 goto out;
61 }
62 backend->size = value;
63 out:
64 error_propagate(errp, local_err);
65 }
66
67 static void
68 host_memory_backend_get_host_nodes(Object *obj, Visitor *v, const char *name,
69 void *opaque, Error **errp)
70 {
71 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
72 uint16List *host_nodes = NULL;
73 uint16List **node = &host_nodes;
74 unsigned long value;
75
76 value = find_first_bit(backend->host_nodes, MAX_NODES);
77 if (value == MAX_NODES) {
78 return;
79 }
80
81 *node = g_malloc0(sizeof(**node));
82 (*node)->value = value;
83 node = &(*node)->next;
84
85 do {
86 value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
87 if (value == MAX_NODES) {
88 break;
89 }
90
91 *node = g_malloc0(sizeof(**node));
92 (*node)->value = value;
93 node = &(*node)->next;
94 } while (true);
95
96 visit_type_uint16List(v, name, &host_nodes, errp);
97 }
98
99 static void
100 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
101 void *opaque, Error **errp)
102 {
103 #ifdef CONFIG_NUMA
104 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
105 uint16List *l = NULL;
106
107 visit_type_uint16List(v, name, &l, errp);
108
109 while (l) {
110 bitmap_set(backend->host_nodes, l->value, 1);
111 l = l->next;
112 }
113 #else
114 error_setg(errp, "NUMA node binding are not supported by this QEMU");
115 #endif
116 }
117
118 static int
119 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
120 {
121 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
122 return backend->policy;
123 }
124
125 static void
126 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
127 {
128 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
129 backend->policy = policy;
130
131 #ifndef CONFIG_NUMA
132 if (policy != HOST_MEM_POLICY_DEFAULT) {
133 error_setg(errp, "NUMA policies are not supported by this QEMU");
134 }
135 #endif
136 }
137
138 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
139 {
140 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
141
142 return backend->merge;
143 }
144
145 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
146 {
147 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
148
149 if (!memory_region_size(&backend->mr)) {
150 backend->merge = value;
151 return;
152 }
153
154 if (value != backend->merge) {
155 void *ptr = memory_region_get_ram_ptr(&backend->mr);
156 uint64_t sz = memory_region_size(&backend->mr);
157
158 qemu_madvise(ptr, sz,
159 value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
160 backend->merge = value;
161 }
162 }
163
164 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
165 {
166 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
167
168 return backend->dump;
169 }
170
171 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
172 {
173 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
174
175 if (!memory_region_size(&backend->mr)) {
176 backend->dump = value;
177 return;
178 }
179
180 if (value != backend->dump) {
181 void *ptr = memory_region_get_ram_ptr(&backend->mr);
182 uint64_t sz = memory_region_size(&backend->mr);
183
184 qemu_madvise(ptr, sz,
185 value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
186 backend->dump = value;
187 }
188 }
189
190 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
191 {
192 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
193
194 return backend->prealloc || backend->force_prealloc;
195 }
196
197 static void host_memory_backend_set_prealloc(Object *obj, bool value,
198 Error **errp)
199 {
200 Error *local_err = NULL;
201 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
202
203 if (backend->force_prealloc) {
204 if (value) {
205 error_setg(errp,
206 "remove -mem-prealloc to use the prealloc property");
207 return;
208 }
209 }
210
211 if (!memory_region_size(&backend->mr)) {
212 backend->prealloc = value;
213 return;
214 }
215
216 if (value && !backend->prealloc) {
217 int fd = memory_region_get_fd(&backend->mr);
218 void *ptr = memory_region_get_ram_ptr(&backend->mr);
219 uint64_t sz = memory_region_size(&backend->mr);
220
221 os_mem_prealloc(fd, ptr, sz, smp_cpus, &local_err);
222 if (local_err) {
223 error_propagate(errp, local_err);
224 return;
225 }
226 backend->prealloc = true;
227 }
228 }
229
230 static void host_memory_backend_init(Object *obj)
231 {
232 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
233 MachineState *machine = MACHINE(qdev_get_machine());
234
235 backend->merge = machine_mem_merge(machine);
236 backend->dump = machine_dump_guest_core(machine);
237 backend->prealloc = mem_prealloc;
238 }
239
240 MemoryRegion *
241 host_memory_backend_get_memory(HostMemoryBackend *backend, Error **errp)
242 {
243 return memory_region_size(&backend->mr) ? &backend->mr : NULL;
244 }
245
246 void host_memory_backend_set_mapped(HostMemoryBackend *backend, bool mapped)
247 {
248 backend->is_mapped = mapped;
249 }
250
251 bool host_memory_backend_is_mapped(HostMemoryBackend *backend)
252 {
253 return backend->is_mapped;
254 }
255
256 static void
257 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
258 {
259 HostMemoryBackend *backend = MEMORY_BACKEND(uc);
260 HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
261 Error *local_err = NULL;
262 void *ptr;
263 uint64_t sz;
264
265 if (bc->alloc) {
266 bc->alloc(backend, &local_err);
267 if (local_err) {
268 goto out;
269 }
270
271 ptr = memory_region_get_ram_ptr(&backend->mr);
272 sz = memory_region_size(&backend->mr);
273
274 if (backend->merge) {
275 qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
276 }
277 if (!backend->dump) {
278 qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
279 }
280 #ifdef CONFIG_NUMA
281 unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
282 /* lastbit == MAX_NODES means maxnode = 0 */
283 unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
284 /* ensure policy won't be ignored in case memory is preallocated
285 * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
286 * this doesn't catch hugepage case. */
287 unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
288
289 /* check for invalid host-nodes and policies and give more verbose
290 * error messages than mbind(). */
291 if (maxnode && backend->policy == MPOL_DEFAULT) {
292 error_setg(errp, "host-nodes must be empty for policy default,"
293 " or you should explicitly specify a policy other"
294 " than default");
295 return;
296 } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
297 error_setg(errp, "host-nodes must be set for policy %s",
298 HostMemPolicy_lookup[backend->policy]);
299 return;
300 }
301
302 /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
303 * as argument to mbind() due to an old Linux bug (feature?) which
304 * cuts off the last specified node. This means backend->host_nodes
305 * must have MAX_NODES+1 bits available.
306 */
307 assert(sizeof(backend->host_nodes) >=
308 BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
309 assert(maxnode <= MAX_NODES);
310 if (mbind(ptr, sz, backend->policy,
311 maxnode ? backend->host_nodes : NULL, maxnode + 1, flags)) {
312 if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
313 error_setg_errno(errp, errno,
314 "cannot bind memory to host NUMA nodes");
315 return;
316 }
317 }
318 #endif
319 /* Preallocate memory after the NUMA policy has been instantiated.
320 * This is necessary to guarantee memory is allocated with
321 * specified NUMA policy in place.
322 */
323 if (backend->prealloc) {
324 os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz,
325 smp_cpus, &local_err);
326 if (local_err) {
327 goto out;
328 }
329 }
330 }
331 out:
332 error_propagate(errp, local_err);
333 }
334
335 static bool
336 host_memory_backend_can_be_deleted(UserCreatable *uc, Error **errp)
337 {
338 if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
339 return false;
340 } else {
341 return true;
342 }
343 }
344
345 static char *get_id(Object *o, Error **errp)
346 {
347 HostMemoryBackend *backend = MEMORY_BACKEND(o);
348
349 return g_strdup(backend->id);
350 }
351
352 static void set_id(Object *o, const char *str, Error **errp)
353 {
354 HostMemoryBackend *backend = MEMORY_BACKEND(o);
355
356 if (backend->id) {
357 error_setg(errp, "cannot change property value");
358 return;
359 }
360 backend->id = g_strdup(str);
361 }
362
363 static void
364 host_memory_backend_class_init(ObjectClass *oc, void *data)
365 {
366 UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
367
368 ucc->complete = host_memory_backend_memory_complete;
369 ucc->can_be_deleted = host_memory_backend_can_be_deleted;
370
371 object_class_property_add_bool(oc, "merge",
372 host_memory_backend_get_merge,
373 host_memory_backend_set_merge, &error_abort);
374 object_class_property_add_bool(oc, "dump",
375 host_memory_backend_get_dump,
376 host_memory_backend_set_dump, &error_abort);
377 object_class_property_add_bool(oc, "prealloc",
378 host_memory_backend_get_prealloc,
379 host_memory_backend_set_prealloc, &error_abort);
380 object_class_property_add(oc, "size", "int",
381 host_memory_backend_get_size,
382 host_memory_backend_set_size,
383 NULL, NULL, &error_abort);
384 object_class_property_add(oc, "host-nodes", "int",
385 host_memory_backend_get_host_nodes,
386 host_memory_backend_set_host_nodes,
387 NULL, NULL, &error_abort);
388 object_class_property_add_enum(oc, "policy", "HostMemPolicy",
389 HostMemPolicy_lookup,
390 host_memory_backend_get_policy,
391 host_memory_backend_set_policy, &error_abort);
392 object_class_property_add_str(oc, "id", get_id, set_id, &error_abort);
393 }
394
395 static void host_memory_backend_finalize(Object *o)
396 {
397 HostMemoryBackend *backend = MEMORY_BACKEND(o);
398 g_free(backend->id);
399 }
400
401 static const TypeInfo host_memory_backend_info = {
402 .name = TYPE_MEMORY_BACKEND,
403 .parent = TYPE_OBJECT,
404 .abstract = true,
405 .class_size = sizeof(HostMemoryBackendClass),
406 .class_init = host_memory_backend_class_init,
407 .instance_size = sizeof(HostMemoryBackend),
408 .instance_init = host_memory_backend_init,
409 .instance_finalize = host_memory_backend_finalize,
410 .interfaces = (InterfaceInfo[]) {
411 { TYPE_USER_CREATABLE },
412 { }
413 }
414 };
415
416 static void register_types(void)
417 {
418 type_register_static(&host_memory_backend_info);
419 }
420
421 type_init(register_types);