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