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 g_strdup(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     uint64_t value;
  57
  58     if (host_memory_backend_mr_inited(backend)) {
  59         error_setg(errp, "cannot change property %s of %s ", name,
  60                    object_get_typename(obj));
  61         return;
  62     }
  63
  64     if (!visit_type_size(v, name, &value, errp)) {
  65         return;
  66     }
  67     if (!value) {
  68         error_setg(errp,
  69                    "property '%s' of %s doesn't take value '%" PRIu64 "'",
  70                    name, object_get_typename(obj), value);
  71         return;
  72     }
  73     backend->size = value;
  74 }
  75
  76 static void
  77 host_memory_backend_get_host_nodes(Object *obj, Visitor *v, const char *name,
  78                                    void *opaque, Error **errp)
  79 {
  80     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  81     uint16List *host_nodes = NULL;
  82     uint16List **tail = &host_nodes;
  83     unsigned long value;
  84
  85     value = find_first_bit(backend->host_nodes, MAX_NODES);
  86     if (value == MAX_NODES) {
  87         goto ret;
  88     }
  89
  90     QAPI_LIST_APPEND(tail, value);
  91
  92     do {
  93         value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
  94         if (value == MAX_NODES) {
  95             break;
  96         }
  97
  98         QAPI_LIST_APPEND(tail, value);
  99     } while (true);
 100
 101 ret:
 102     visit_type_uint16List(v, name, &host_nodes, errp);
 103     qapi_free_uint16List(host_nodes);
 104 }
 105
 106 static void
 107 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
 108                                    void *opaque, Error **errp)
 109 {
 110 #ifdef CONFIG_NUMA
 111     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 112     uint16List *l, *host_nodes = NULL;
 113
 114     visit_type_uint16List(v, name, &host_nodes, errp);
 115
 116     for (l = host_nodes; l; l = l->next) {
 117         if (l->value >= MAX_NODES) {
 118             error_setg(errp, "Invalid host-nodes value: %d", l->value);
 119             goto out;
 120         }
 121     }
 122
 123     for (l = host_nodes; l; l = l->next) {
 124         bitmap_set(backend->host_nodes, l->value, 1);
 125     }
 126
 127 out:
 128     qapi_free_uint16List(host_nodes);
 129 #else
 130     error_setg(errp, "NUMA node binding are not supported by this QEMU");
 131 #endif
 132 }
 133
 134 static int
 135 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
 136 {
 137     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 138     return backend->policy;
 139 }
 140
 141 static void
 142 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
 143 {
 144     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 145     backend->policy = policy;
 146
 147 #ifndef CONFIG_NUMA
 148     if (policy != HOST_MEM_POLICY_DEFAULT) {
 149         error_setg(errp, "NUMA policies are not supported by this QEMU");
 150     }
 151 #endif
 152 }
 153
 154 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
 155 {
 156     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 157
 158     return backend->merge;
 159 }
 160
 161 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
 162 {
 163     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 164
 165     if (!host_memory_backend_mr_inited(backend)) {
 166         backend->merge = value;
 167         return;
 168     }
 169
 170     if (value != backend->merge) {
 171         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 172         uint64_t sz = memory_region_size(&backend->mr);
 173
 174         qemu_madvise(ptr, sz,
 175                      value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
 176         backend->merge = value;
 177     }
 178 }
 179
 180 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
 181 {
 182     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 183
 184     return backend->dump;
 185 }
 186
 187 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
 188 {
 189     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 190
 191     if (!host_memory_backend_mr_inited(backend)) {
 192         backend->dump = value;
 193         return;
 194     }
 195
 196     if (value != backend->dump) {
 197         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 198         uint64_t sz = memory_region_size(&backend->mr);
 199
 200         qemu_madvise(ptr, sz,
 201                      value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
 202         backend->dump = value;
 203     }
 204 }
 205
 206 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
 207 {
 208     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 209
 210     return backend->prealloc;
 211 }
 212
 213 static void host_memory_backend_set_prealloc(Object *obj, bool value,
 214                                              Error **errp)
 215 {
 216     Error *local_err = NULL;
 217     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 218
 219     if (!backend->reserve && value) {
 220         error_setg(errp, "'prealloc=on' and 'reserve=off' are incompatible");
 221         return;
 222     }
 223
 224     if (!host_memory_backend_mr_inited(backend)) {
 225         backend->prealloc = value;
 226         return;
 227     }
 228
 229     if (value && !backend->prealloc) {
 230         int fd = memory_region_get_fd(&backend->mr);
 231         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 232         uint64_t sz = memory_region_size(&backend->mr);
 233
 234         os_mem_prealloc(fd, ptr, sz, backend->prealloc_threads, &local_err);
 235         if (local_err) {
 236             error_propagate(errp, local_err);
 237             return;
 238         }
 239         backend->prealloc = true;
 240     }
 241 }
 242
 243 static void host_memory_backend_get_prealloc_threads(Object *obj, Visitor *v,
 244     const char *name, void *opaque, Error **errp)
 245 {
 246     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 247     visit_type_uint32(v, name, &backend->prealloc_threads, errp);
 248 }
 249
 250 static void host_memory_backend_set_prealloc_threads(Object *obj, Visitor *v,
 251     const char *name, void *opaque, Error **errp)
 252 {
 253     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 254     uint32_t value;
 255
 256     if (!visit_type_uint32(v, name, &value, errp)) {
 257         return;
 258     }
 259     if (value <= 0) {
 260         error_setg(errp, "property '%s' of %s doesn't take value '%d'", name,
 261                    object_get_typename(obj), value);
 262         return;
 263     }
 264     backend->prealloc_threads = value;
 265 }
 266
 267 static void host_memory_backend_init(Object *obj)
 268 {
 269     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 270     MachineState *machine = MACHINE(qdev_get_machine());
 271
 272     /* TODO: convert access to globals to compat properties */
 273     backend->merge = machine_mem_merge(machine);
 274     backend->dump = machine_dump_guest_core(machine);
 275     backend->reserve = true;
 276     backend->prealloc_threads = 1;
 277 }
 278
 279 static void host_memory_backend_post_init(Object *obj)
 280 {
 281     object_apply_compat_props(obj);
 282 }
 283
 284 bool host_memory_backend_mr_inited(HostMemoryBackend *backend)
 285 {
 286     /*
 287      * NOTE: We forbid zero-length memory backend, so here zero means
 288      * "we haven't inited the backend memory region yet".
 289      */
 290     return memory_region_size(&backend->mr) != 0;
 291 }
 292
 293 MemoryRegion *host_memory_backend_get_memory(HostMemoryBackend *backend)
 294 {
 295     return host_memory_backend_mr_inited(backend) ? &backend->mr : NULL;
 296 }
 297
 298 void host_memory_backend_set_mapped(HostMemoryBackend *backend, bool mapped)
 299 {
 300     backend->is_mapped = mapped;
 301 }
 302
 303 bool host_memory_backend_is_mapped(HostMemoryBackend *backend)
 304 {
 305     return backend->is_mapped;
 306 }
 307
 308 #ifdef __linux__
 309 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 310 {
 311     Object *obj = OBJECT(memdev);
 312     char *path = object_property_get_str(obj, "mem-path", NULL);
 313     size_t pagesize = qemu_mempath_getpagesize(path);
 314
 315     g_free(path);
 316     return pagesize;
 317 }
 318 #else
 319 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 320 {
 321     return qemu_real_host_page_size;
 322 }
 323 #endif
 324
 325 static void
 326 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 327 {
 328     HostMemoryBackend *backend = MEMORY_BACKEND(uc);
 329     HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
 330     Error *local_err = NULL;
 331     void *ptr;
 332     uint64_t sz;
 333
 334     if (bc->alloc) {
 335         bc->alloc(backend, &local_err);
 336         if (local_err) {
 337             goto out;
 338         }
 339
 340         ptr = memory_region_get_ram_ptr(&backend->mr);
 341         sz = memory_region_size(&backend->mr);
 342
 343         if (backend->merge) {
 344             qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
 345         }
 346         if (!backend->dump) {
 347             qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
 348         }
 349 #ifdef CONFIG_NUMA
 350         unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
 351         /* lastbit == MAX_NODES means maxnode = 0 */
 352         unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
 353         /* ensure policy won't be ignored in case memory is preallocated
 354          * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
 355          * this doesn't catch hugepage case. */
 356         unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
 357
 358         /* check for invalid host-nodes and policies and give more verbose
 359          * error messages than mbind(). */
 360         if (maxnode && backend->policy == MPOL_DEFAULT) {
 361             error_setg(errp, "host-nodes must be empty for policy default,"
 362                        " or you should explicitly specify a policy other"
 363                        " than default");
 364             return;
 365         } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
 366             error_setg(errp, "host-nodes must be set for policy %s",
 367                        HostMemPolicy_str(backend->policy));
 368             return;
 369         }
 370
 371         /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
 372          * as argument to mbind() due to an old Linux bug (feature?) which
 373          * cuts off the last specified node. This means backend->host_nodes
 374          * must have MAX_NODES+1 bits available.
 375          */
 376         assert(sizeof(backend->host_nodes) >=
 377                BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
 378         assert(maxnode <= MAX_NODES);
 379
 380         if (maxnode &&
 381             mbind(ptr, sz, backend->policy, backend->host_nodes, maxnode + 1,
 382                   flags)) {
 383             if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
 384                 error_setg_errno(errp, errno,
 385                                  "cannot bind memory to host NUMA nodes");
 386                 return;
 387             }
 388         }
 389 #endif
 390         /* Preallocate memory after the NUMA policy has been instantiated.
 391          * This is necessary to guarantee memory is allocated with
 392          * specified NUMA policy in place.
 393          */
 394         if (backend->prealloc) {
 395             os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz,
 396                             backend->prealloc_threads, &local_err);
 397             if (local_err) {
 398                 goto out;
 399             }
 400         }
 401     }
 402 out:
 403     error_propagate(errp, local_err);
 404 }
 405
 406 static bool
 407 host_memory_backend_can_be_deleted(UserCreatable *uc)
 408 {
 409     if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
 410         return false;
 411     } else {
 412         return true;
 413     }
 414 }
 415
 416 static bool host_memory_backend_get_share(Object *o, Error **errp)
 417 {
 418     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 419
 420     return backend->share;
 421 }
 422
 423 static void host_memory_backend_set_share(Object *o, bool value, Error **errp)
 424 {
 425     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 426
 427     if (host_memory_backend_mr_inited(backend)) {
 428         error_setg(errp, "cannot change property value");
 429         return;
 430     }
 431     backend->share = value;
 432 }
 433
 434 #ifdef CONFIG_LINUX
 435 static bool host_memory_backend_get_reserve(Object *o, Error **errp)
 436 {
 437     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 438
 439     return backend->reserve;
 440 }
 441
 442 static void host_memory_backend_set_reserve(Object *o, bool value, Error **errp)
 443 {
 444     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 445
 446     if (host_memory_backend_mr_inited(backend)) {
 447         error_setg(errp, "cannot change property value");
 448         return;
 449     }
 450     if (backend->prealloc && !value) {
 451         error_setg(errp, "'prealloc=on' and 'reserve=off' are incompatible");
 452         return;
 453     }
 454     backend->reserve = value;
 455 }
 456 #endif /* CONFIG_LINUX */
 457
 458 static bool
 459 host_memory_backend_get_use_canonical_path(Object *obj, Error **errp)
 460 {
 461     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 462
 463     return backend->use_canonical_path;
 464 }
 465
 466 static void
 467 host_memory_backend_set_use_canonical_path(Object *obj, bool value,
 468                                            Error **errp)
 469 {
 470     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 471
 472     backend->use_canonical_path = value;
 473 }
 474
 475 static void
 476 host_memory_backend_class_init(ObjectClass *oc, void *data)
 477 {
 478     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 479
 480     ucc->complete = host_memory_backend_memory_complete;
 481     ucc->can_be_deleted = host_memory_backend_can_be_deleted;
 482
 483     object_class_property_add_bool(oc, "merge",
 484         host_memory_backend_get_merge,
 485         host_memory_backend_set_merge);
 486     object_class_property_set_description(oc, "merge",
 487         "Mark memory as mergeable");
 488     object_class_property_add_bool(oc, "dump",
 489         host_memory_backend_get_dump,
 490         host_memory_backend_set_dump);
 491     object_class_property_set_description(oc, "dump",
 492         "Set to 'off' to exclude from core dump");
 493     object_class_property_add_bool(oc, "prealloc",
 494         host_memory_backend_get_prealloc,
 495         host_memory_backend_set_prealloc);
 496     object_class_property_set_description(oc, "prealloc",
 497         "Preallocate memory");
 498     object_class_property_add(oc, "prealloc-threads", "int",
 499         host_memory_backend_get_prealloc_threads,
 500         host_memory_backend_set_prealloc_threads,
 501         NULL, NULL);
 502     object_class_property_set_description(oc, "prealloc-threads",
 503         "Number of CPU threads to use for prealloc");
 504     object_class_property_add(oc, "size", "int",
 505         host_memory_backend_get_size,
 506         host_memory_backend_set_size,
 507         NULL, NULL);
 508     object_class_property_set_description(oc, "size",
 509         "Size of the memory region (ex: 500M)");
 510     object_class_property_add(oc, "host-nodes", "int",
 511         host_memory_backend_get_host_nodes,
 512         host_memory_backend_set_host_nodes,
 513         NULL, NULL);
 514     object_class_property_set_description(oc, "host-nodes",
 515         "Binds memory to the list of NUMA host nodes");
 516     object_class_property_add_enum(oc, "policy", "HostMemPolicy",
 517         &HostMemPolicy_lookup,
 518         host_memory_backend_get_policy,
 519         host_memory_backend_set_policy);
 520     object_class_property_set_description(oc, "policy",
 521         "Set the NUMA policy");
 522     object_class_property_add_bool(oc, "share",
 523         host_memory_backend_get_share, host_memory_backend_set_share);
 524     object_class_property_set_description(oc, "share",
 525         "Mark the memory as private to QEMU or shared");
 526 #ifdef CONFIG_LINUX
 527     object_class_property_add_bool(oc, "reserve",
 528         host_memory_backend_get_reserve, host_memory_backend_set_reserve);
 529     object_class_property_set_description(oc, "reserve",
 530         "Reserve swap space (or huge pages) if applicable");
 531 #endif /* CONFIG_LINUX */
 532     /*
 533      * Do not delete/rename option. This option must be considered stable
 534      * (as if it didn't have the 'x-' prefix including deprecation period) as
 535      * long as 4.0 and older machine types exists.
 536      * Option will be used by upper layers to override (disable) canonical path
 537      * for ramblock-id set by compat properties on old machine types ( <= 4.0),
 538      * to keep migration working when backend is used for main RAM with
 539      * -machine memory-backend= option (main RAM historically used prefix-less
 540      * ramblock-id).
 541      */
 542     object_class_property_add_bool(oc, "x-use-canonical-path-for-ramblock-id",
 543         host_memory_backend_get_use_canonical_path,
 544         host_memory_backend_set_use_canonical_path);
 545 }
 546
 547 static const TypeInfo host_memory_backend_info = {
 548     .name = TYPE_MEMORY_BACKEND,
 549     .parent = TYPE_OBJECT,
 550     .abstract = true,
 551     .class_size = sizeof(HostMemoryBackendClass),
 552     .class_init = host_memory_backend_class_init,
 553     .instance_size = sizeof(HostMemoryBackend),
 554     .instance_init = host_memory_backend_init,
 555     .instance_post_init = host_memory_backend_post_init,
 556     .interfaces = (InterfaceInfo[]) {
 557         { TYPE_USER_CREATABLE },
 558         { }
 559     }
 560 };
 561
 562 static void register_types(void)
 563 {
 564     type_register_static(&host_memory_backend_info);
 565 }
 566
 567 type_init(register_types);