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 "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>
  26 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT != MPOL_DEFAULT);
  27 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED != MPOL_PREFERRED);
  28 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND != MPOL_BIND);
  29 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE != MPOL_INTERLEAVE);
  30 #endif
  31
  32 char *
  33 host_memory_backend_get_name(HostMemoryBackend *backend)
  34 {
  35     if (!backend->use_canonical_path) {
  36         return g_strdup(object_get_canonical_path_component(OBJECT(backend)));
  37     }
  38
  39     return object_get_canonical_path(OBJECT(backend));
  40 }
  41
  42 static void
  43 host_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
  52 static void
  53 host_memory_backend_set_size(Object *obj, Visitor *v, const char *name,
  54                              void *opaque, Error **errp)
  55 {
  56     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  57     uint64_t value;
  58
  59     if (host_memory_backend_mr_inited(backend)) {
  60         error_setg(errp, "cannot change property %s of %s ", name,
  61                    object_get_typename(obj));
  62         return;
  63     }
  64
  65     if (!visit_type_size(v, name, &value, errp)) {
  66         return;
  67     }
  68     if (!value) {
  69         error_setg(errp,
  70                    "property '%s' of %s doesn't take value '%" PRIu64 "'",
  71                    name, object_get_typename(obj), value);
  72         return;
  73     }
  74     backend->size = value;
  75 }
  76
  77 static void
  78 host_memory_backend_get_host_nodes(Object *obj, Visitor *v, const char *name,
  79                                    void *opaque, Error **errp)
  80 {
  81     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  82     uint16List *host_nodes = NULL;
  83     uint16List **node = &host_nodes;
  84     unsigned long value;
  85
  86     value = find_first_bit(backend->host_nodes, MAX_NODES);
  87     if (value == MAX_NODES) {
  88         goto ret;
  89     }
  90
  91     *node = g_malloc0(sizeof(**node));
  92     (*node)->value = value;
  93     node = &(*node)->next;
  94
  95     do {
  96         value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
  97         if (value == MAX_NODES) {
  98             break;
  99         }
 100
 101         *node = g_malloc0(sizeof(**node));
 102         (*node)->value = value;
 103         node = &(*node)->next;
 104     } while (true);
 105
 106 ret:
 107     visit_type_uint16List(v, name, &host_nodes, errp);
 108     qapi_free_uint16List(host_nodes);
 109 }
 110
 111 static void
 112 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
 113                                    void *opaque, Error **errp)
 114 {
 115 #ifdef CONFIG_NUMA
 116     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 117     uint16List *l, *host_nodes = NULL;
 118
 119     visit_type_uint16List(v, name, &host_nodes, errp);
 120
 121     for (l = host_nodes; l; l = l->next) {
 122         if (l->value >= MAX_NODES) {
 123             error_setg(errp, "Invalid host-nodes value: %d", l->value);
 124             goto out;
 125         }
 126     }
 127
 128     for (l = host_nodes; l; l = l->next) {
 129         bitmap_set(backend->host_nodes, l->value, 1);
 130     }
 131
 132 out:
 133     qapi_free_uint16List(host_nodes);
 134 #else
 135     error_setg(errp, "NUMA node binding are not supported by this QEMU");
 136 #endif
 137 }
 138
 139 static int
 140 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
 141 {
 142     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 143     return backend->policy;
 144 }
 145
 146 static void
 147 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
 148 {
 149     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 150     backend->policy = policy;
 151
 152 #ifndef CONFIG_NUMA
 153     if (policy != HOST_MEM_POLICY_DEFAULT) {
 154         error_setg(errp, "NUMA policies are not supported by this QEMU");
 155     }
 156 #endif
 157 }
 158
 159 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
 160 {
 161     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 162
 163     return backend->merge;
 164 }
 165
 166 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
 167 {
 168     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 169
 170     if (!host_memory_backend_mr_inited(backend)) {
 171         backend->merge = value;
 172         return;
 173     }
 174
 175     if (value != backend->merge) {
 176         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 177         uint64_t sz = memory_region_size(&backend->mr);
 178
 179         qemu_madvise(ptr, sz,
 180                      value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
 181         backend->merge = value;
 182     }
 183 }
 184
 185 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
 186 {
 187     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 188
 189     return backend->dump;
 190 }
 191
 192 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
 193 {
 194     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 195
 196     if (!host_memory_backend_mr_inited(backend)) {
 197         backend->dump = value;
 198         return;
 199     }
 200
 201     if (value != backend->dump) {
 202         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 203         uint64_t sz = memory_region_size(&backend->mr);
 204
 205         qemu_madvise(ptr, sz,
 206                      value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
 207         backend->dump = value;
 208     }
 209 }
 210
 211 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
 212 {
 213     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 214
 215     return backend->prealloc;
 216 }
 217
 218 static void host_memory_backend_set_prealloc(Object *obj, bool value,
 219                                              Error **errp)
 220 {
 221     Error *local_err = NULL;
 222     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 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->prealloc_threads = 1;
 276 }
 277
 278 static void host_memory_backend_post_init(Object *obj)
 279 {
 280     object_apply_compat_props(obj);
 281 }
 282
 283 bool host_memory_backend_mr_inited(HostMemoryBackend *backend)
 284 {
 285     /*
 286      * NOTE: We forbid zero-length memory backend, so here zero means
 287      * "we haven't inited the backend memory region yet".
 288      */
 289     return memory_region_size(&backend->mr) != 0;
 290 }
 291
 292 MemoryRegion *host_memory_backend_get_memory(HostMemoryBackend *backend)
 293 {
 294     return host_memory_backend_mr_inited(backend) ? &backend->mr : NULL;
 295 }
 296
 297 void host_memory_backend_set_mapped(HostMemoryBackend *backend, bool mapped)
 298 {
 299     backend->is_mapped = mapped;
 300 }
 301
 302 bool host_memory_backend_is_mapped(HostMemoryBackend *backend)
 303 {
 304     return backend->is_mapped;
 305 }
 306
 307 #ifdef __linux__
 308 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 309 {
 310     Object *obj = OBJECT(memdev);
 311     char *path = object_property_get_str(obj, "mem-path", NULL);
 312     size_t pagesize = qemu_mempath_getpagesize(path);
 313
 314     g_free(path);
 315     return pagesize;
 316 }
 317 #else
 318 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 319 {
 320     return qemu_real_host_page_size;
 321 }
 322 #endif
 323
 324 static void
 325 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 326 {
 327     HostMemoryBackend *backend = MEMORY_BACKEND(uc);
 328     HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
 329     Error *local_err = NULL;
 330     void *ptr;
 331     uint64_t sz;
 332
 333     if (bc->alloc) {
 334         bc->alloc(backend, &local_err);
 335         if (local_err) {
 336             goto out;
 337         }
 338
 339         ptr = memory_region_get_ram_ptr(&backend->mr);
 340         sz = memory_region_size(&backend->mr);
 341
 342         if (backend->merge) {
 343             qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
 344         }
 345         if (!backend->dump) {
 346             qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
 347         }
 348 #ifdef CONFIG_NUMA
 349         unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
 350         /* lastbit == MAX_NODES means maxnode = 0 */
 351         unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
 352         /* ensure policy won't be ignored in case memory is preallocated
 353          * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
 354          * this doesn't catch hugepage case. */
 355         unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
 356
 357         /* check for invalid host-nodes and policies and give more verbose
 358          * error messages than mbind(). */
 359         if (maxnode && backend->policy == MPOL_DEFAULT) {
 360             error_setg(errp, "host-nodes must be empty for policy default,"
 361                        " or you should explicitly specify a policy other"
 362                        " than default");
 363             return;
 364         } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
 365             error_setg(errp, "host-nodes must be set for policy %s",
 366                        HostMemPolicy_str(backend->policy));
 367             return;
 368         }
 369
 370         /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
 371          * as argument to mbind() due to an old Linux bug (feature?) which
 372          * cuts off the last specified node. This means backend->host_nodes
 373          * must have MAX_NODES+1 bits available.
 374          */
 375         assert(sizeof(backend->host_nodes) >=
 376                BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
 377         assert(maxnode <= MAX_NODES);
 378
 379         if (maxnode &&
 380             mbind(ptr, sz, backend->policy, backend->host_nodes, maxnode + 1,
 381                   flags)) {
 382             if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
 383                 error_setg_errno(errp, errno,
 384                                  "cannot bind memory to host NUMA nodes");
 385                 return;
 386             }
 387         }
 388 #endif
 389         /* Preallocate memory after the NUMA policy has been instantiated.
 390          * This is necessary to guarantee memory is allocated with
 391          * specified NUMA policy in place.
 392          */
 393         if (backend->prealloc) {
 394             os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz,
 395                             backend->prealloc_threads, &local_err);
 396             if (local_err) {
 397                 goto out;
 398             }
 399         }
 400     }
 401 out:
 402     error_propagate(errp, local_err);
 403 }
 404
 405 static bool
 406 host_memory_backend_can_be_deleted(UserCreatable *uc)
 407 {
 408     if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
 409         return false;
 410     } else {
 411         return true;
 412     }
 413 }
 414
 415 static bool host_memory_backend_get_share(Object *o, Error **errp)
 416 {
 417     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 418
 419     return backend->share;
 420 }
 421
 422 static void host_memory_backend_set_share(Object *o, bool value, Error **errp)
 423 {
 424     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 425
 426     if (host_memory_backend_mr_inited(backend)) {
 427         error_setg(errp, "cannot change property value");
 428         return;
 429     }
 430     backend->share = value;
 431 }
 432
 433 static bool
 434 host_memory_backend_get_use_canonical_path(Object *obj, Error **errp)
 435 {
 436     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 437
 438     return backend->use_canonical_path;
 439 }
 440
 441 static void
 442 host_memory_backend_set_use_canonical_path(Object *obj, bool value,
 443                                            Error **errp)
 444 {
 445     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 446
 447     backend->use_canonical_path = value;
 448 }
 449
 450 static void
 451 host_memory_backend_class_init(ObjectClass *oc, void *data)
 452 {
 453     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 454
 455     ucc->complete = host_memory_backend_memory_complete;
 456     ucc->can_be_deleted = host_memory_backend_can_be_deleted;
 457
 458     object_class_property_add_bool(oc, "merge",
 459         host_memory_backend_get_merge,
 460         host_memory_backend_set_merge);
 461     object_class_property_set_description(oc, "merge",
 462         "Mark memory as mergeable");
 463     object_class_property_add_bool(oc, "dump",
 464         host_memory_backend_get_dump,
 465         host_memory_backend_set_dump);
 466     object_class_property_set_description(oc, "dump",
 467         "Set to 'off' to exclude from core dump");
 468     object_class_property_add_bool(oc, "prealloc",
 469         host_memory_backend_get_prealloc,
 470         host_memory_backend_set_prealloc);
 471     object_class_property_set_description(oc, "prealloc",
 472         "Preallocate memory");
 473     object_class_property_add(oc, "prealloc-threads", "int",
 474         host_memory_backend_get_prealloc_threads,
 475         host_memory_backend_set_prealloc_threads,
 476         NULL, NULL);
 477     object_class_property_set_description(oc, "prealloc-threads",
 478         "Number of CPU threads to use for prealloc");
 479     object_class_property_add(oc, "size", "int",
 480         host_memory_backend_get_size,
 481         host_memory_backend_set_size,
 482         NULL, NULL);
 483     object_class_property_set_description(oc, "size",
 484         "Size of the memory region (ex: 500M)");
 485     object_class_property_add(oc, "host-nodes", "int",
 486         host_memory_backend_get_host_nodes,
 487         host_memory_backend_set_host_nodes,
 488         NULL, NULL);
 489     object_class_property_set_description(oc, "host-nodes",
 490         "Binds memory to the list of NUMA host nodes");
 491     object_class_property_add_enum(oc, "policy", "HostMemPolicy",
 492         &HostMemPolicy_lookup,
 493         host_memory_backend_get_policy,
 494         host_memory_backend_set_policy);
 495     object_class_property_set_description(oc, "policy",
 496         "Set the NUMA policy");
 497     object_class_property_add_bool(oc, "share",
 498         host_memory_backend_get_share, host_memory_backend_set_share);
 499     object_class_property_set_description(oc, "share",
 500         "Mark the memory as private to QEMU or shared");
 501     object_class_property_add_bool(oc, "x-use-canonical-path-for-ramblock-id",
 502         host_memory_backend_get_use_canonical_path,
 503         host_memory_backend_set_use_canonical_path);
 504 }
 505
 506 static const TypeInfo host_memory_backend_info = {
 507     .name = TYPE_MEMORY_BACKEND,
 508     .parent = TYPE_OBJECT,
 509     .abstract = true,
 510     .class_size = sizeof(HostMemoryBackendClass),
 511     .class_init = host_memory_backend_class_init,
 512     .instance_size = sizeof(HostMemoryBackend),
 513     .instance_init = host_memory_backend_init,
 514     .instance_post_init = host_memory_backend_post_init,
 515     .interfaces = (InterfaceInfo[]) {
 516         { TYPE_USER_CREATABLE },
 517         { }
 518     }
 519 };
 520
 521 static void register_types(void)
 522 {
 523     type_register_static(&host_memory_backend_info);
 524 }
 525
 526 type_init(register_types);