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
13#include "qemu/osdep.h"
14#include "sysemu/hostmem.h"
15#include "sysemu/sysemu.h"
16#include "hw/boards.h"
17#include "qapi/error.h"
18#include "qapi/qapi-builtin-visit.h"
19#include "qapi/visitor.h"
20#include "qemu/config-file.h"
21#include "qom/object_interfaces.h"
22#include "qemu/mmap-alloc.h"
23
24#ifdef CONFIG_NUMA
25#include <numaif.h>
26QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT != MPOL_DEFAULT);
27QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED != MPOL_PREFERRED);
28QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND != MPOL_BIND);
29QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE != MPOL_INTERLEAVE);
30#endif
31
32char *
33host_memory_backend_get_name(HostMemoryBackend *backend)
34{
35 if (!backend->use_canonical_path) {
36 return object_get_canonical_path_component(OBJECT(backend));
37 }
38
39 return object_get_canonical_path(OBJECT(backend));
40}
41
42static void
43host_memory_backend_get_size(Object *obj, Visitor *v, const char *name,
44 void *opaque, Error **errp)
45{
46 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
47 uint64_t value = backend->size;
48
49 visit_type_size(v, name, &value, errp);
50}
51
52static void
53host_memory_backend_set_size(Object *obj, Visitor *v, const char *name,
54 void *opaque, Error **errp)
55{
56 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
57 Error *local_err = NULL;
58 uint64_t value;
59
60 if (host_memory_backend_mr_inited(backend)) {
61 error_setg(&local_err, "cannot change property %s of %s ",
62 name, object_get_typename(obj));
63 goto out;
64 }
65
66 visit_type_size(v, name, &value, &local_err);
67 if (local_err) {
68 goto out;
69 }
70 if (!value) {
71 error_setg(&local_err,
72 "property '%s' of %s doesn't take value '%" PRIu64 "'",
73 name, object_get_typename(obj), value);
74 goto out;
75 }
76 backend->size = value;
77out:
78 error_propagate(errp, local_err);
79}
80
81static void
82host_memory_backend_get_host_nodes(Object *obj, Visitor *v, const char *name,
83 void *opaque, Error **errp)
84{
85 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
86 uint16List *host_nodes = NULL;
87 uint16List **node = &host_nodes;
88 unsigned long value;
89
90 value = find_first_bit(backend->host_nodes, MAX_NODES);
91 if (value == MAX_NODES) {
92 goto ret;
93 }
94
95 *node = g_malloc0(sizeof(**node));
96 (*node)->value = value;
97 node = &(*node)->next;
98
99 do {
100 value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
101 if (value == MAX_NODES) {
102 break;
103 }
104
105 *node = g_malloc0(sizeof(**node));
106 (*node)->value = value;
107 node = &(*node)->next;
108 } while (true);
109
110ret:
111 visit_type_uint16List(v, name, &host_nodes, errp);
112}
113
114static void
115host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
116 void *opaque, Error **errp)
117{
118#ifdef CONFIG_NUMA
119 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
120 uint16List *l, *host_nodes = NULL;
121
122 visit_type_uint16List(v, name, &host_nodes, errp);
123
124 for (l = host_nodes; l; l = l->next) {
125 if (l->value >= MAX_NODES) {
126 error_setg(errp, "Invalid host-nodes value: %d", l->value);
127 goto out;
128 }
129 }
130
131 for (l = host_nodes; l; l = l->next) {
132 bitmap_set(backend->host_nodes, l->value, 1);
133 }
134
135out:
136 qapi_free_uint16List(host_nodes);
137#else
138 error_setg(errp, "NUMA node binding are not supported by this QEMU");
139#endif
140}
141
142static int
143host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
144{
145 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
146 return backend->policy;
147}
148
149static void
150host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
151{
152 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
153 backend->policy = policy;
154
155#ifndef CONFIG_NUMA
156 if (policy != HOST_MEM_POLICY_DEFAULT) {
157 error_setg(errp, "NUMA policies are not supported by this QEMU");
158 }
159#endif
160}
161
162static bool host_memory_backend_get_merge(Object *obj, Error **errp)
163{
164 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
165
166 return backend->merge;
167}
168
169static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
170{
171 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
172
173 if (!host_memory_backend_mr_inited(backend)) {
174 backend->merge = value;
175 return;
176 }
177
178 if (value != backend->merge) {
179 void *ptr = memory_region_get_ram_ptr(&backend->mr);
180 uint64_t sz = memory_region_size(&backend->mr);
181
182 qemu_madvise(ptr, sz,
183 value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
184 backend->merge = value;
185 }
186}
187
188static bool host_memory_backend_get_dump(Object *obj, Error **errp)
189{
190 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
191
192 return backend->dump;
193}
194
195static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
196{
197 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
198
199 if (!host_memory_backend_mr_inited(backend)) {
200 backend->dump = value;
201 return;
202 }
203
204 if (value != backend->dump) {
205 void *ptr = memory_region_get_ram_ptr(&backend->mr);
206 uint64_t sz = memory_region_size(&backend->mr);
207
208 qemu_madvise(ptr, sz,
209 value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
210 backend->dump = value;
211 }
212}
213
214static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
215{
216 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
217
218 return backend->prealloc || backend->force_prealloc;
219}
220
221static void host_memory_backend_set_prealloc(Object *obj, bool value,
222 Error **errp)
223{
224 Error *local_err = NULL;
225 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
226 MachineState *ms = MACHINE(qdev_get_machine());
227
228 if (backend->force_prealloc) {
229 if (value) {
230 error_setg(errp,
231 "remove -mem-prealloc to use the prealloc property");
232 return;
233 }
234 }
235
236 if (!host_memory_backend_mr_inited(backend)) {
237 backend->prealloc = value;
238 return;
239 }
240
241 if (value && !backend->prealloc) {
242 int fd = memory_region_get_fd(&backend->mr);
243 void *ptr = memory_region_get_ram_ptr(&backend->mr);
244 uint64_t sz = memory_region_size(&backend->mr);
245
246 os_mem_prealloc(fd, ptr, sz, ms->smp.cpus, &local_err);
247 if (local_err) {
248 error_propagate(errp, local_err);
249 return;
250 }
251 backend->prealloc = true;
252 }
253}
254
255static void host_memory_backend_init(Object *obj)
256{
257 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
258 MachineState *machine = MACHINE(qdev_get_machine());
259
260 backend->merge = machine_mem_merge(machine);
261 backend->dump = machine_dump_guest_core(machine);
262 backend->prealloc = mem_prealloc;
263}
264
265static void host_memory_backend_post_init(Object *obj)
266{
267 object_apply_compat_props(obj);
268}
269
270bool host_memory_backend_mr_inited(HostMemoryBackend *backend)
271{
272 /*
273 * NOTE: We forbid zero-length memory backend, so here zero means
274 * "we haven't inited the backend memory region yet".
275 */
276 return memory_region_size(&backend->mr) != 0;
277}
278
279MemoryRegion *host_memory_backend_get_memory(HostMemoryBackend *backend)
280{
281 return host_memory_backend_mr_inited(backend) ? &backend->mr : NULL;
282}
283
284void host_memory_backend_set_mapped(HostMemoryBackend *backend, bool mapped)
285{
286 backend->is_mapped = mapped;
287}
288
289bool host_memory_backend_is_mapped(HostMemoryBackend *backend)
290{
291 return backend->is_mapped;
292}
293
294#ifdef __linux__
295size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
296{
297 Object *obj = OBJECT(memdev);
298 char *path = object_property_get_str(obj, "mem-path", NULL);
299 size_t pagesize = qemu_mempath_getpagesize(path);
300
301 g_free(path);
302 return pagesize;
303}
304#else
305size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
306{
307 return getpagesize();
308}
309#endif
310
311static void
312host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
313{
314 HostMemoryBackend *backend = MEMORY_BACKEND(uc);
315 HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
316 MachineState *ms = MACHINE(qdev_get_machine());
317 Error *local_err = NULL;
318 void *ptr;
319 uint64_t sz;
320
321 if (bc->alloc) {
322 bc->alloc(backend, &local_err);
323 if (local_err) {
324 goto out;
325 }
326
327 ptr = memory_region_get_ram_ptr(&backend->mr);
328 sz = memory_region_size(&backend->mr);
329
330 if (backend->merge) {
331 qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
332 }
333 if (!backend->dump) {
334 qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
335 }
336#ifdef CONFIG_NUMA
337 unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
338 /* lastbit == MAX_NODES means maxnode = 0 */
339 unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
340 /* ensure policy won't be ignored in case memory is preallocated
341 * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
342 * this doesn't catch hugepage case. */
343 unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
344
345 /* check for invalid host-nodes and policies and give more verbose
346 * error messages than mbind(). */
347 if (maxnode && backend->policy == MPOL_DEFAULT) {
348 error_setg(errp, "host-nodes must be empty for policy default,"
349 " or you should explicitly specify a policy other"
350 " than default");
351 return;
352 } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
353 error_setg(errp, "host-nodes must be set for policy %s",
354 HostMemPolicy_str(backend->policy));
355 return;
356 }
357
358 /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
359 * as argument to mbind() due to an old Linux bug (feature?) which
360 * cuts off the last specified node. This means backend->host_nodes
361 * must have MAX_NODES+1 bits available.
362 */
363 assert(sizeof(backend->host_nodes) >=
364 BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
365 assert(maxnode <= MAX_NODES);
366 if (mbind(ptr, sz, backend->policy,
367 maxnode ? backend->host_nodes : NULL, maxnode + 1, flags)) {
368 if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
369 error_setg_errno(errp, errno,
370 "cannot bind memory to host NUMA nodes");
371 return;
372 }
373 }
374#endif
375 /* Preallocate memory after the NUMA policy has been instantiated.
376 * This is necessary to guarantee memory is allocated with
377 * specified NUMA policy in place.
378 */
379 if (backend->prealloc) {
380 os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz,
381 ms->smp.cpus, &local_err);
382 if (local_err) {
383 goto out;
384 }
385 }
386 }
387out:
388 error_propagate(errp, local_err);
389}
390
391static bool
392host_memory_backend_can_be_deleted(UserCreatable *uc)
393{
394 if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
395 return false;
396 } else {
397 return true;
398 }
399}
400
401static bool host_memory_backend_get_share(Object *o, Error **errp)
402{
403 HostMemoryBackend *backend = MEMORY_BACKEND(o);
404
405 return backend->share;
406}
407
408static void host_memory_backend_set_share(Object *o, bool value, Error **errp)
409{
410 HostMemoryBackend *backend = MEMORY_BACKEND(o);
411
412 if (host_memory_backend_mr_inited(backend)) {
413 error_setg(errp, "cannot change property value");
414 return;
415 }
416 backend->share = value;
417}
418
419static bool
420host_memory_backend_get_use_canonical_path(Object *obj, Error **errp)
421{
422 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
423
424 return backend->use_canonical_path;
425}
426
427static void
428host_memory_backend_set_use_canonical_path(Object *obj, bool value,
429 Error **errp)
430{
431 HostMemoryBackend *backend = MEMORY_BACKEND(obj);
432
433 backend->use_canonical_path = value;
434}
435
436static void
437host_memory_backend_class_init(ObjectClass *oc, void *data)
438{
439 UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
440
441 ucc->complete = host_memory_backend_memory_complete;
442 ucc->can_be_deleted = host_memory_backend_can_be_deleted;
443
444 object_class_property_add_bool(oc, "merge",
445 host_memory_backend_get_merge,
446 host_memory_backend_set_merge, &error_abort);
447 object_class_property_set_description(oc, "merge",
448 "Mark memory as mergeable", &error_abort);
449 object_class_property_add_bool(oc, "dump",
450 host_memory_backend_get_dump,
451 host_memory_backend_set_dump, &error_abort);
452 object_class_property_set_description(oc, "dump",
453 "Set to 'off' to exclude from core dump", &error_abort);
454 object_class_property_add_bool(oc, "prealloc",
455 host_memory_backend_get_prealloc,
456 host_memory_backend_set_prealloc, &error_abort);
457 object_class_property_set_description(oc, "prealloc",
458 "Preallocate memory", &error_abort);
459 object_class_property_add(oc, "size", "int",
460 host_memory_backend_get_size,
461 host_memory_backend_set_size,
462 NULL, NULL, &error_abort);
463 object_class_property_set_description(oc, "size",
464 "Size of the memory region (ex: 500M)", &error_abort);
465 object_class_property_add(oc, "host-nodes", "int",
466 host_memory_backend_get_host_nodes,
467 host_memory_backend_set_host_nodes,
468 NULL, NULL, &error_abort);
469 object_class_property_set_description(oc, "host-nodes",
470 "Binds memory to the list of NUMA host nodes", &error_abort);
471 object_class_property_add_enum(oc, "policy", "HostMemPolicy",
472 &HostMemPolicy_lookup,
473 host_memory_backend_get_policy,
474 host_memory_backend_set_policy, &error_abort);
475 object_class_property_set_description(oc, "policy",
476 "Set the NUMA policy", &error_abort);
477 object_class_property_add_bool(oc, "share",
478 host_memory_backend_get_share, host_memory_backend_set_share,
479 &error_abort);
480 object_class_property_set_description(oc, "share",
481 "Mark the memory as private to QEMU or shared", &error_abort);
482 object_class_property_add_bool(oc, "x-use-canonical-path-for-ramblock-id",
483 host_memory_backend_get_use_canonical_path,
484 host_memory_backend_set_use_canonical_path, &error_abort);
485}
486
487static const TypeInfo host_memory_backend_info = {
488 .name = TYPE_MEMORY_BACKEND,
489 .parent = TYPE_OBJECT,
490 .abstract = true,
491 .class_size = sizeof(HostMemoryBackendClass),
492 .class_init = host_memory_backend_class_init,
493 .instance_size = sizeof(HostMemoryBackend),
494 .instance_init = host_memory_backend_init,
495 .instance_post_init = host_memory_backend_post_init,
496 .interfaces = (InterfaceInfo[]) {
497 { TYPE_USER_CREATABLE },
498 { }
499 }
500};
501
502static void register_types(void)
503{
504 type_register_static(&host_memory_backend_info);
505}
506
507type_init(register_types);
508