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 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     Error *local_err = NULL;
  58     uint64_t value;
  59
  60     if (host_memory_backend_mr_inited(backend)) {
  61         error_setg(&local_err, "cannot change property %s of %s ",
  62                    name, object_get_typename(obj));
  63         goto out;
  64     }
  65
  66     visit_type_size(v, name, &value, &local_err);
  67     if (local_err) {
  68         goto out;
  69     }
  70     if (!value) {
  71         error_setg(&local_err,
  72                    "property '%s' of %s doesn't take value '%" PRIu64 "'",
  73                    name, object_get_typename(obj), value);
  74         goto out;
  75     }
  76     backend->size = value;
  77 out:
  78     error_propagate(errp, local_err);
  79 }
  80
  81 static void
  82 host_memory_backend_get_host_nodes(Object *obj, Visitor *v, const char *name,
  83                                    void *opaque, Error **errp)
  84 {
  85     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  86     uint16List *host_nodes = NULL;
  87     uint16List **node = &host_nodes;
  88     unsigned long value;
  89
  90     value = find_first_bit(backend->host_nodes, MAX_NODES);
  91     if (value == MAX_NODES) {
  92         goto ret;
  93     }
  94
  95     *node = g_malloc0(sizeof(**node));
  96     (*node)->value = value;
  97     node = &(*node)->next;
  98
  99     do {
 100         value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
 101         if (value == MAX_NODES) {
 102             break;
 103         }
 104
 105         *node = g_malloc0(sizeof(**node));
 106         (*node)->value = value;
 107         node = &(*node)->next;
 108     } while (true);
 109
 110 ret:
 111     visit_type_uint16List(v, name, &host_nodes, errp);
 112 }
 113
 114 static void
 115 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
 116                                    void *opaque, Error **errp)
 117 {
 118 #ifdef CONFIG_NUMA
 119     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 120     uint16List *l, *host_nodes = NULL;
 121
 122     visit_type_uint16List(v, name, &host_nodes, errp);
 123
 124     for (l = host_nodes; l; l = l->next) {
 125         if (l->value >= MAX_NODES) {
 126             error_setg(errp, "Invalid host-nodes value: %d", l->value);
 127             goto out;
 128         }
 129     }
 130
 131     for (l = host_nodes; l; l = l->next) {
 132         bitmap_set(backend->host_nodes, l->value, 1);
 133     }
 134
 135 out:
 136     qapi_free_uint16List(host_nodes);
 137 #else
 138     error_setg(errp, "NUMA node binding are not supported by this QEMU");
 139 #endif
 140 }
 141
 142 static int
 143 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
 144 {
 145     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 146     return backend->policy;
 147 }
 148
 149 static void
 150 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
 151 {
 152     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 153     backend->policy = policy;
 154
 155 #ifndef CONFIG_NUMA
 156     if (policy != HOST_MEM_POLICY_DEFAULT) {
 157         error_setg(errp, "NUMA policies are not supported by this QEMU");
 158     }
 159 #endif
 160 }
 161
 162 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
 163 {
 164     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 165
 166     return backend->merge;
 167 }
 168
 169 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
 170 {
 171     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 172
 173     if (!host_memory_backend_mr_inited(backend)) {
 174         backend->merge = value;
 175         return;
 176     }
 177
 178     if (value != backend->merge) {
 179         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 180         uint64_t sz = memory_region_size(&backend->mr);
 181
 182         qemu_madvise(ptr, sz,
 183                      value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
 184         backend->merge = value;
 185     }
 186 }
 187
 188 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
 189 {
 190     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 191
 192     return backend->dump;
 193 }
 194
 195 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
 196 {
 197     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 198
 199     if (!host_memory_backend_mr_inited(backend)) {
 200         backend->dump = value;
 201         return;
 202     }
 203
 204     if (value != backend->dump) {
 205         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 206         uint64_t sz = memory_region_size(&backend->mr);
 207
 208         qemu_madvise(ptr, sz,
 209                      value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
 210         backend->dump = value;
 211     }
 212 }
 213
 214 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
 215 {
 216     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 217
 218     return backend->prealloc;
 219 }
 220
 221 static void host_memory_backend_set_prealloc(Object *obj, bool value,
 222                                              Error **errp)
 223 {
 224     Error *local_err = NULL;
 225     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 226
 227     if (!host_memory_backend_mr_inited(backend)) {
 228         backend->prealloc = value;
 229         return;
 230     }
 231
 232     if (value && !backend->prealloc) {
 233         int fd = memory_region_get_fd(&backend->mr);
 234         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 235         uint64_t sz = memory_region_size(&backend->mr);
 236
 237         os_mem_prealloc(fd, ptr, sz, backend->prealloc_threads, &local_err);
 238         if (local_err) {
 239             error_propagate(errp, local_err);
 240             return;
 241         }
 242         backend->prealloc = true;
 243     }
 244 }
 245
 246 static void host_memory_backend_get_prealloc_threads(Object *obj, Visitor *v,
 247     const char *name, void *opaque, Error **errp)
 248 {
 249     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 250     visit_type_uint32(v, name, &backend->prealloc_threads, errp);
 251 }
 252
 253 static void host_memory_backend_set_prealloc_threads(Object *obj, Visitor *v,
 254     const char *name, void *opaque, Error **errp)
 255 {
 256     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 257     Error *local_err = NULL;
 258     uint32_t value;
 259
 260     visit_type_uint32(v, name, &value, &local_err);
 261     if (local_err) {
 262         goto out;
 263     }
 264     if (value <= 0) {
 265         error_setg(&local_err,
 266                    "property '%s' of %s doesn't take value '%d'",
 267                    name, object_get_typename(obj), value);
 268         goto out;
 269     }
 270     backend->prealloc_threads = value;
 271 out:
 272     error_propagate(errp, local_err);
 273 }
 274
 275 static void host_memory_backend_init(Object *obj)
 276 {
 277     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 278     MachineState *machine = MACHINE(qdev_get_machine());
 279
 280     /* TODO: convert access to globals to compat properties */
 281     backend->merge = machine_mem_merge(machine);
 282     backend->dump = machine_dump_guest_core(machine);
 283     backend->prealloc_threads = 1;
 284 }
 285
 286 static void host_memory_backend_post_init(Object *obj)
 287 {
 288     object_apply_compat_props(obj);
 289 }
 290
 291 bool host_memory_backend_mr_inited(HostMemoryBackend *backend)
 292 {
 293     /*
 294      * NOTE: We forbid zero-length memory backend, so here zero means
 295      * "we haven't inited the backend memory region yet".
 296      */
 297     return memory_region_size(&backend->mr) != 0;
 298 }
 299
 300 MemoryRegion *host_memory_backend_get_memory(HostMemoryBackend *backend)
 301 {
 302     return host_memory_backend_mr_inited(backend) ? &backend->mr : NULL;
 303 }
 304
 305 void host_memory_backend_set_mapped(HostMemoryBackend *backend, bool mapped)
 306 {
 307     backend->is_mapped = mapped;
 308 }
 309
 310 bool host_memory_backend_is_mapped(HostMemoryBackend *backend)
 311 {
 312     return backend->is_mapped;
 313 }
 314
 315 #ifdef __linux__
 316 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 317 {
 318     Object *obj = OBJECT(memdev);
 319     char *path = object_property_get_str(obj, "mem-path", NULL);
 320     size_t pagesize = qemu_mempath_getpagesize(path);
 321
 322     g_free(path);
 323     return pagesize;
 324 }
 325 #else
 326 size_t host_memory_backend_pagesize(HostMemoryBackend *memdev)
 327 {
 328     return qemu_real_host_page_size;
 329 }
 330 #endif
 331
 332 static void
 333 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 334 {
 335     HostMemoryBackend *backend = MEMORY_BACKEND(uc);
 336     HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
 337     Error *local_err = NULL;
 338     void *ptr;
 339     uint64_t sz;
 340
 341     if (bc->alloc) {
 342         bc->alloc(backend, &local_err);
 343         if (local_err) {
 344             goto out;
 345         }
 346
 347         ptr = memory_region_get_ram_ptr(&backend->mr);
 348         sz = memory_region_size(&backend->mr);
 349
 350         if (backend->merge) {
 351             qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
 352         }
 353         if (!backend->dump) {
 354             qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
 355         }
 356 #ifdef CONFIG_NUMA
 357         unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
 358         /* lastbit == MAX_NODES means maxnode = 0 */
 359         unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
 360         /* ensure policy won't be ignored in case memory is preallocated
 361          * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
 362          * this doesn't catch hugepage case. */
 363         unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
 364
 365         /* check for invalid host-nodes and policies and give more verbose
 366          * error messages than mbind(). */
 367         if (maxnode && backend->policy == MPOL_DEFAULT) {
 368             error_setg(errp, "host-nodes must be empty for policy default,"
 369                        " or you should explicitly specify a policy other"
 370                        " than default");
 371             return;
 372         } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
 373             error_setg(errp, "host-nodes must be set for policy %s",
 374                        HostMemPolicy_str(backend->policy));
 375             return;
 376         }
 377
 378         /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
 379          * as argument to mbind() due to an old Linux bug (feature?) which
 380          * cuts off the last specified node. This means backend->host_nodes
 381          * must have MAX_NODES+1 bits available.
 382          */
 383         assert(sizeof(backend->host_nodes) >=
 384                BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
 385         assert(maxnode <= MAX_NODES);
 386
 387         if (maxnode &&
 388             mbind(ptr, sz, backend->policy, backend->host_nodes, maxnode + 1,
 389                   flags)) {
 390             if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
 391                 error_setg_errno(errp, errno,
 392                                  "cannot bind memory to host NUMA nodes");
 393                 return;
 394             }
 395         }
 396 #endif
 397         /* Preallocate memory after the NUMA policy has been instantiated.
 398          * This is necessary to guarantee memory is allocated with
 399          * specified NUMA policy in place.
 400          */
 401         if (backend->prealloc) {
 402             os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz,
 403                             backend->prealloc_threads, &local_err);
 404             if (local_err) {
 405                 goto out;
 406             }
 407         }
 408     }
 409 out:
 410     error_propagate(errp, local_err);
 411 }
 412
 413 static bool
 414 host_memory_backend_can_be_deleted(UserCreatable *uc)
 415 {
 416     if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
 417         return false;
 418     } else {
 419         return true;
 420     }
 421 }
 422
 423 static bool host_memory_backend_get_share(Object *o, Error **errp)
 424 {
 425     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 426
 427     return backend->share;
 428 }
 429
 430 static void host_memory_backend_set_share(Object *o, bool value, Error **errp)
 431 {
 432     HostMemoryBackend *backend = MEMORY_BACKEND(o);
 433
 434     if (host_memory_backend_mr_inited(backend)) {
 435         error_setg(errp, "cannot change property value");
 436         return;
 437     }
 438     backend->share = value;
 439 }
 440
 441 static bool
 442 host_memory_backend_get_use_canonical_path(Object *obj, Error **errp)
 443 {
 444     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 445
 446     return backend->use_canonical_path;
 447 }
 448
 449 static void
 450 host_memory_backend_set_use_canonical_path(Object *obj, bool value,
 451                                            Error **errp)
 452 {
 453     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 454
 455     backend->use_canonical_path = value;
 456 }
 457
 458 static void
 459 host_memory_backend_class_init(ObjectClass *oc, void *data)
 460 {
 461     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 462
 463     ucc->complete = host_memory_backend_memory_complete;
 464     ucc->can_be_deleted = host_memory_backend_can_be_deleted;
 465
 466     object_class_property_add_bool(oc, "merge",
 467         host_memory_backend_get_merge,
 468         host_memory_backend_set_merge);
 469     object_class_property_set_description(oc, "merge",
 470         "Mark memory as mergeable");
 471     object_class_property_add_bool(oc, "dump",
 472         host_memory_backend_get_dump,
 473         host_memory_backend_set_dump);
 474     object_class_property_set_description(oc, "dump",
 475         "Set to 'off' to exclude from core dump");
 476     object_class_property_add_bool(oc, "prealloc",
 477         host_memory_backend_get_prealloc,
 478         host_memory_backend_set_prealloc);
 479     object_class_property_set_description(oc, "prealloc",
 480         "Preallocate memory");
 481     object_class_property_add(oc, "prealloc-threads", "int",
 482         host_memory_backend_get_prealloc_threads,
 483         host_memory_backend_set_prealloc_threads,
 484         NULL, NULL);
 485     object_class_property_set_description(oc, "prealloc-threads",
 486         "Number of CPU threads to use for prealloc");
 487     object_class_property_add(oc, "size", "int",
 488         host_memory_backend_get_size,
 489         host_memory_backend_set_size,
 490         NULL, NULL);
 491     object_class_property_set_description(oc, "size",
 492         "Size of the memory region (ex: 500M)");
 493     object_class_property_add(oc, "host-nodes", "int",
 494         host_memory_backend_get_host_nodes,
 495         host_memory_backend_set_host_nodes,
 496         NULL, NULL);
 497     object_class_property_set_description(oc, "host-nodes",
 498         "Binds memory to the list of NUMA host nodes");
 499     object_class_property_add_enum(oc, "policy", "HostMemPolicy",
 500         &HostMemPolicy_lookup,
 501         host_memory_backend_get_policy,
 502         host_memory_backend_set_policy);
 503     object_class_property_set_description(oc, "policy",
 504         "Set the NUMA policy");
 505     object_class_property_add_bool(oc, "share",
 506         host_memory_backend_get_share, host_memory_backend_set_share);
 507     object_class_property_set_description(oc, "share",
 508         "Mark the memory as private to QEMU or shared");
 509     object_class_property_add_bool(oc, "x-use-canonical-path-for-ramblock-id",
 510         host_memory_backend_get_use_canonical_path,
 511         host_memory_backend_set_use_canonical_path);
 512 }
 513
 514 static const TypeInfo host_memory_backend_info = {
 515     .name = TYPE_MEMORY_BACKEND,
 516     .parent = TYPE_OBJECT,
 517     .abstract = true,
 518     .class_size = sizeof(HostMemoryBackendClass),
 519     .class_init = host_memory_backend_class_init,
 520     .instance_size = sizeof(HostMemoryBackend),
 521     .instance_init = host_memory_backend_init,
 522     .instance_post_init = host_memory_backend_post_init,
 523     .interfaces = (InterfaceInfo[]) {
 524         { TYPE_USER_CREATABLE },
 525         { }
 526     }
 527 };
 528
 529 static void register_types(void)
 530 {
 531     type_register_static(&host_memory_backend_info);
 532 }
 533
 534 type_init(register_types);