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 **tail = &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     QAPI_LIST_APPEND(tail, value);
  92
  93     do {
  94         value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
  95         if (value == MAX_NODES) {
  96             break;
  97         }
  98
  99         QAPI_LIST_APPEND(tail, value);
 100     } while (true);
 101
 102 ret:
 103     visit_type_uint16List(v, name, &host_nodes, errp);
 104     qapi_free_uint16List(host_nodes);
 105 }
 106
 107 static void
 108 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
 109                                    void *opaque, Error **errp)
 110 {
 111 #ifdef CONFIG_NUMA
 112     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 113     uint16List *l, *host_nodes = NULL;
 114
 115     visit_type_uint16List(v, name, &host_nodes, errp);
 116
 117     for (l = host_nodes; l; l = l->next) {
 118         if (l->value >= MAX_NODES) {
 119             error_setg(errp, "Invalid host-nodes value: %d", l->value);
 120             goto out;
 121         }
 122     }
 123
 124     for (l = host_nodes; l; l = l->next) {
 125         bitmap_set(backend->host_nodes, l->value, 1);
 126     }
 127
 128 out:
 129     qapi_free_uint16List(host_nodes);
 130 #else
 131     error_setg(errp, "NUMA node binding are not supported by this QEMU");
 132 #endif
 133 }
 134
 135 static int
 136 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
 137 {
 138     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 139     return backend->policy;
 140 }
 141
 142 static void
 143 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
 144 {
 145     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 146     backend->policy = policy;
 147
 148 #ifndef CONFIG_NUMA
 149     if (policy != HOST_MEM_POLICY_DEFAULT) {
 150         error_setg(errp, "NUMA policies are not supported by this QEMU");
 151     }
 152 #endif
 153 }
 154
 155 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
 156 {
 157     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 158
 159     return backend->merge;
 160 }
 161
 162 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
 163 {
 164     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 165
 166     if (!host_memory_backend_mr_inited(backend)) {
 167         backend->merge = value;
 168         return;
 169     }
 170
 171     if (value != backend->merge) {
 172         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 173         uint64_t sz = memory_region_size(&backend->mr);
 174
 175         qemu_madvise(ptr, sz,
 176                      value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
 177         backend->merge = value;
 178     }
 179 }
 180
 181 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
 182 {
 183     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 184
 185     return backend->dump;
 186 }
 187
 188 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
 189 {
 190     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 191
 192     if (!host_memory_backend_mr_inited(backend)) {
 193         backend->dump = value;
 194         return;
 195     }
 196
 197     if (value != backend->dump) {
 198         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 199         uint64_t sz = memory_region_size(&backend->mr);
 200
 201         qemu_madvise(ptr, sz,
 202                      value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
 203         backend->dump = value;
 204     }
 205 }
 206
 207 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
 208 {
 209     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 210
 211     return backend->prealloc;
 212 }
 213
 214 static void host_memory_backend_set_prealloc(Object *obj, bool value,
 215                                              Error **errp)
 216 {
 217     Error *local_err = NULL;
 218     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 219
 220     if (!host_memory_backend_mr_inited(backend)) {
 221         backend->prealloc = value;
 222         return;
 223     }
 224
 225     if (value && !backend->prealloc) {
 226         int fd = memory_region_get_fd(&backend->mr);
 227         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 228         uint64_t sz = memory_region_size(&backend->mr);
 229
 230         os_mem_prealloc(fd, ptr, sz, backend->prealloc_threads, &local_err);
 231         if (local_err) {
 232             error_propagate(errp, local_err);
 233             return;
 234         }
 235         backend->prealloc = true;
 236     }
 237 }
 238
 239 static void host_memory_backend_get_prealloc_threads(Object *obj, Visitor *v,
 240     const char *name, void *opaque, Error **errp)
 241 {
 242     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 243     visit_type_uint32(v, name, &backend->prealloc_threads, errp);
 244 }
 245
 246 static void host_memory_backend_set_prealloc_threads(Object *obj, Visitor *v,
 247     const char *name, void *opaque, Error **errp)
 248 {
 249     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 250     uint32_t value;
 251
 252     if (!visit_type_uint32(v, name, &value, errp)) {
 253         return;
 254     }
 255     if (value <= 0) {
 256         error_setg(errp, "property '%s' of %s doesn't take value '%d'", name,
 257                    object_get_typename(obj), value);
 258         return;
 259     }
 260     backend->prealloc_threads = value;
 261 }
 262
 263 static void host_memory_backend_init(Object *obj)
 264 {
 265     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 266     MachineState *machine = MACHINE(qdev_get_machine());
 267
 268     /* TODO: convert access to globals to compat properties */
 269     backend->merge = machine_mem_merge(machine);
 270     backend->dump = machine_dump_guest_core(machine);
 271     backend->prealloc_threads = 1;
 272 }
 273
 274 static void host_memory_backend_post_init(Object *obj)
 275 {
 276     object_apply_compat_props(obj);
 277 }
 278
 279 bool host_memory_backend_mr_inited(HostMemoryBackend *backend)
 280 {
 281     /*
 282      * NOTE: We forbid zero-length memory backend, so here zero means
 283      * "we haven't inited the backend memory region yet".
 284      */
 285     return memory_region_size(&backend->mr) != 0;
 286 }
 287
 288 MemoryRegion *host_memory_backend_get_memory(HostMemoryBackend *backend)
 289 {
 290     return host_memory_backend_mr_inited(backend) ? &backend->mr : NULL;
 291 }
 292
 293 void host_memory_backend_set_mapped(HostMemoryBackend *backend, bool mapped)
 294 {
 295     backend->is_mapped = mapped;
 296 }
 297
 298 bool host_memory_backend_is_mapped(HostMemoryBackend *backend)
 299 {
 300     return backend->is_mapped;
 301 }
 302
 303 #ifdef __linux__
 304 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 305 {
 306     Object *obj = OBJECT(memdev);
 307     char *path = object_property_get_str(obj, "mem-path", NULL);
 308     size_t pagesize = qemu_mempath_getpagesize(path);
 309
 310     g_free(path);
 311     return pagesize;
 312 }
 313 #else
 314 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 315 {
 316     return qemu_real_host_page_size;
 317 }
 318 #endif
 319
 320 static void
 321 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 322 {
 323     HostMemoryBackend *backend = MEMORY_BACKEND(uc);
 324     HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
 325     Error *local_err = NULL;
 326     void *ptr;
 327     uint64_t sz;
 328
 329     if (bc->alloc) {
 330         bc->alloc(backend, &local_err);
 331         if (local_err) {
 332             goto out;
 333         }
 334
 335         ptr = memory_region_get_ram_ptr(&backend->mr);
 336         sz = memory_region_size(&backend->mr);
 337
 338         if (backend->merge) {
 339             qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
 340         }
 341         if (!backend->dump) {
 342             qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
 343         }
 344 #ifdef CONFIG_NUMA
 345         unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
 346         /* lastbit == MAX_NODES means maxnode = 0 */
 347         unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
 348         /* ensure policy won't be ignored in case memory is preallocated
 349          * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
 350          * this doesn't catch hugepage case. */
 351         unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
 352
 353         /* check for invalid host-nodes and policies and give more verbose
 354          * error messages than mbind(). */
 355         if (maxnode && backend->policy == MPOL_DEFAULT) {
 356             error_setg(errp, "host-nodes must be empty for policy default,"
 357                        " or you should explicitly specify a policy other"
 358                        " than default");
 359             return;
 360         } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
 361             error_setg(errp, "host-nodes must be set for policy %s",
 362                        HostMemPolicy_str(backend->policy));
 363             return;
 364         }
 365
 366         /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
 367          * as argument to mbind() due to an old Linux bug (feature?) which
 368          * cuts off the last specified node. This means backend->host_nodes
 369          * must have MAX_NODES+1 bits available.
 370          */
 371         assert(sizeof(backend->host_nodes) >=
 372                BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
 373         assert(maxnode <= MAX_NODES);
 374
 375         if (maxnode &&
 376             mbind(ptr, sz, backend->policy, backend->host_nodes, maxnode + 1,
 377                   flags)) {
 378             if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
 379                 error_setg_errno(errp, errno,
 380                                  "cannot bind memory to host NUMA nodes");
 381                 return;
 382             }
 383         }
 384 #endif
 385         /* Preallocate memory after the NUMA policy has been instantiated.
 386          * This is necessary to guarantee memory is allocated with
 387          * specified NUMA policy in place.
 388          */
 389         if (backend->prealloc) {
 390             os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz,
 391                             backend->prealloc_threads, &local_err);
 392             if (local_err) {
 393                 goto out;
 394             }
 395         }
 396     }
 397 out:
 398     error_propagate(errp, local_err);
 399 }
 400
 401 static bool
 402 host_memory_backend_can_be_deleted(UserCreatable *uc)
 403 {
 404     if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
 405         return false;
 406     } else {
 407         return true;
 408     }
 409 }
 410
 411 static bool host_memory_backend_get_share(Object *o, Error **errp)
 412 {
 413     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 414
 415     return backend->share;
 416 }
 417
 418 static void host_memory_backend_set_share(Object *o, bool value, Error **errp)
 419 {
 420     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 421
 422     if (host_memory_backend_mr_inited(backend)) {
 423         error_setg(errp, "cannot change property value");
 424         return;
 425     }
 426     backend->share = value;
 427 }
 428
 429 static bool
 430 host_memory_backend_get_use_canonical_path(Object *obj, Error **errp)
 431 {
 432     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 433
 434     return backend->use_canonical_path;
 435 }
 436
 437 static void
 438 host_memory_backend_set_use_canonical_path(Object *obj, bool value,
 439                                            Error **errp)
 440 {
 441     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 442
 443     backend->use_canonical_path = value;
 444 }
 445
 446 static void
 447 host_memory_backend_class_init(ObjectClass *oc, void *data)
 448 {
 449     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 450
 451     ucc->complete = host_memory_backend_memory_complete;
 452     ucc->can_be_deleted = host_memory_backend_can_be_deleted;
 453
 454     object_class_property_add_bool(oc, "merge",
 455         host_memory_backend_get_merge,
 456         host_memory_backend_set_merge);
 457     object_class_property_set_description(oc, "merge",
 458         "Mark memory as mergeable");
 459     object_class_property_add_bool(oc, "dump",
 460         host_memory_backend_get_dump,
 461         host_memory_backend_set_dump);
 462     object_class_property_set_description(oc, "dump",
 463         "Set to 'off' to exclude from core dump");
 464     object_class_property_add_bool(oc, "prealloc",
 465         host_memory_backend_get_prealloc,
 466         host_memory_backend_set_prealloc);
 467     object_class_property_set_description(oc, "prealloc",
 468         "Preallocate memory");
 469     object_class_property_add(oc, "prealloc-threads", "int",
 470         host_memory_backend_get_prealloc_threads,
 471         host_memory_backend_set_prealloc_threads,
 472         NULL, NULL);
 473     object_class_property_set_description(oc, "prealloc-threads",
 474         "Number of CPU threads to use for prealloc");
 475     object_class_property_add(oc, "size", "int",
 476         host_memory_backend_get_size,
 477         host_memory_backend_set_size,
 478         NULL, NULL);
 479     object_class_property_set_description(oc, "size",
 480         "Size of the memory region (ex: 500M)");
 481     object_class_property_add(oc, "host-nodes", "int",
 482         host_memory_backend_get_host_nodes,
 483         host_memory_backend_set_host_nodes,
 484         NULL, NULL);
 485     object_class_property_set_description(oc, "host-nodes",
 486         "Binds memory to the list of NUMA host nodes");
 487     object_class_property_add_enum(oc, "policy", "HostMemPolicy",
 488         &HostMemPolicy_lookup,
 489         host_memory_backend_get_policy,
 490         host_memory_backend_set_policy);
 491     object_class_property_set_description(oc, "policy",
 492         "Set the NUMA policy");
 493     object_class_property_add_bool(oc, "share",
 494         host_memory_backend_get_share, host_memory_backend_set_share);
 495     object_class_property_set_description(oc, "share",
 496         "Mark the memory as private to QEMU or shared");
 497     /*
 498      * Do not delete/rename option. This option must be considered stable
 499      * (as if it didn't have the 'x-' prefix including deprecation period) as
 500      * long as 4.0 and older machine types exists.
 501      * Option will be used by upper layers to override (disable) canonical path
 502      * for ramblock-id set by compat properties on old machine types ( <= 4.0),
 503      * to keep migration working when backend is used for main RAM with
 504      * -machine memory-backend= option (main RAM historically used prefix-less
 505      * ramblock-id).
 506      */
 507     object_class_property_add_bool(oc, "x-use-canonical-path-for-ramblock-id",
 508         host_memory_backend_get_use_canonical_path,
 509         host_memory_backend_set_use_canonical_path);
 510 }
 511
 512 static const TypeInfo host_memory_backend_info = {
 513     .name = TYPE_MEMORY_BACKEND,
 514     .parent = TYPE_OBJECT,
 515     .abstract = true,
 516     .class_size = sizeof(HostMemoryBackendClass),
 517     .class_init = host_memory_backend_class_init,
 518     .instance_size = sizeof(HostMemoryBackend),
 519     .instance_init = host_memory_backend_init,
 520     .instance_post_init = host_memory_backend_post_init,
 521     .interfaces = (InterfaceInfo[]) {
 522         { TYPE_USER_CREATABLE },
 523         { }
 524     }
 525 };
 526
 527 static void register_types(void)
 528 {
 529     type_register_static(&host_memory_backend_info);
 530 }
 531
 532 type_init(register_types);