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