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