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