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