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 #include "sysemu/hostmem.h"
  13 #include "hw/boards.h"
  14 #include "qapi/visitor.h"
  15 #include "qapi-types.h"
  16 #include "qapi-visit.h"
  17 #include "qemu/config-file.h"
  18 #include "qom/object_interfaces.h"
  19
  20 #ifdef CONFIG_NUMA
  21 #include <numaif.h>
  22 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT != MPOL_DEFAULT);
  23 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED != MPOL_PREFERRED);
  24 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND != MPOL_BIND);
  25 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE != MPOL_INTERLEAVE);
  26 #endif
  27
  28 static void
  29 host_memory_backend_get_size(Object *obj, Visitor *v, void *opaque,
  30                              const char *name, Error **errp)
  31 {
  32     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  33     uint64_t value = backend->size;
  34
  35     visit_type_size(v, &value, name, errp);
  36 }
  37
  38 static void
  39 host_memory_backend_set_size(Object *obj, Visitor *v, void *opaque,
  40                              const char *name, Error **errp)
  41 {
  42     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  43     Error *local_err = NULL;
  44     uint64_t value;
  45
  46     if (memory_region_size(&backend->mr)) {
  47         error_setg(&local_err, "cannot change property value");
  48         goto out;
  49     }
  50
  51     visit_type_size(v, &value, name, &local_err);
  52     if (local_err) {
  53         goto out;
  54     }
  55     if (!value) {
  56         error_setg(&local_err, "Property '%s.%s' doesn't take value '%"
  57                    PRIu64 "'", object_get_typename(obj), name, value);
  58         goto out;
  59     }
  60     backend->size = value;
  61 out:
  62     error_propagate(errp, local_err);
  63 }
  64
  65 static void
  66 host_memory_backend_get_host_nodes(Object *obj, Visitor *v, void *opaque,
  67                                    const char *name, Error **errp)
  68 {
  69     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  70     uint16List *host_nodes = NULL;
  71     uint16List **node = &host_nodes;
  72     unsigned long value;
  73
  74     value = find_first_bit(backend->host_nodes, MAX_NODES);
  75     if (value == MAX_NODES) {
  76         return;
  77     }
  78
  79     *node = g_malloc0(sizeof(**node));
  80     (*node)->value = value;
  81     node = &(*node)->next;
  82
  83     do {
  84         value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
  85         if (value == MAX_NODES) {
  86             break;
  87         }
  88
  89         *node = g_malloc0(sizeof(**node));
  90         (*node)->value = value;
  91         node = &(*node)->next;
  92     } while (true);
  93
  94     visit_type_uint16List(v, &host_nodes, name, errp);
  95 }
  96
  97 static void
  98 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, void *opaque,
  99                                    const char *name, Error **errp)
 100 {
 101 #ifdef CONFIG_NUMA
 102     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 103     uint16List *l = NULL;
 104
 105     visit_type_uint16List(v, &l, name, errp);
 106
 107     while (l) {
 108         bitmap_set(backend->host_nodes, l->value, 1);
 109         l = l->next;
 110     }
 111 #else
 112     error_setg(errp, "NUMA node binding are not supported by this QEMU");
 113 #endif
 114 }
 115
 116 static int
 117 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
 118 {
 119     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 120     return backend->policy;
 121 }
 122
 123 static void
 124 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
 125 {
 126     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 127     backend->policy = policy;
 128
 129 #ifndef CONFIG_NUMA
 130     if (policy != HOST_MEM_POLICY_DEFAULT) {
 131         error_setg(errp, "NUMA policies are not supported by this QEMU");
 132     }
 133 #endif
 134 }
 135
 136 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
 137 {
 138     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 139
 140     return backend->merge;
 141 }
 142
 143 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
 144 {
 145     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 146
 147     if (!memory_region_size(&backend->mr)) {
 148         backend->merge = value;
 149         return;
 150     }
 151
 152     if (value != backend->merge) {
 153         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 154         uint64_t sz = memory_region_size(&backend->mr);
 155
 156         qemu_madvise(ptr, sz,
 157                      value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
 158         backend->merge = value;
 159     }
 160 }
 161
 162 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
 163 {
 164     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 165
 166     return backend->dump;
 167 }
 168
 169 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
 170 {
 171     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 172
 173     if (!memory_region_size(&backend->mr)) {
 174         backend->dump = value;
 175         return;
 176     }
 177
 178     if (value != backend->dump) {
 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_DODUMP : QEMU_MADV_DONTDUMP);
 184         backend->dump = value;
 185     }
 186 }
 187
 188 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
 189 {
 190     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 191
 192     return backend->prealloc || backend->force_prealloc;
 193 }
 194
 195 static void host_memory_backend_set_prealloc(Object *obj, bool value,
 196                                              Error **errp)
 197 {
 198     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 199
 200     if (backend->force_prealloc) {
 201         if (value) {
 202             error_setg(errp,
 203                        "remove -mem-prealloc to use the prealloc property");
 204             return;
 205         }
 206     }
 207
 208     if (!memory_region_size(&backend->mr)) {
 209         backend->prealloc = value;
 210         return;
 211     }
 212
 213     if (value && !backend->prealloc) {
 214         int fd = memory_region_get_fd(&backend->mr);
 215         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 216         uint64_t sz = memory_region_size(&backend->mr);
 217
 218         os_mem_prealloc(fd, ptr, sz);
 219         backend->prealloc = true;
 220     }
 221 }
 222
 223 static void host_memory_backend_init(Object *obj)
 224 {
 225     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 226     MachineState *machine = MACHINE(qdev_get_machine());
 227
 228     backend->merge = machine_mem_merge(machine);
 229     backend->dump = machine_dump_guest_core(machine);
 230     backend->prealloc = mem_prealloc;
 231
 232     object_property_add_bool(obj, "merge",
 233                         host_memory_backend_get_merge,
 234                         host_memory_backend_set_merge, NULL);
 235     object_property_add_bool(obj, "dump",
 236                         host_memory_backend_get_dump,
 237                         host_memory_backend_set_dump, NULL);
 238     object_property_add_bool(obj, "prealloc",
 239                         host_memory_backend_get_prealloc,
 240                         host_memory_backend_set_prealloc, NULL);
 241     object_property_add(obj, "size", "int",
 242                         host_memory_backend_get_size,
 243                         host_memory_backend_set_size, NULL, NULL, NULL);
 244     object_property_add(obj, "host-nodes", "int",
 245                         host_memory_backend_get_host_nodes,
 246                         host_memory_backend_set_host_nodes, NULL, NULL, NULL);
 247     object_property_add_enum(obj, "policy", "HostMemPolicy",
 248                              HostMemPolicy_lookup,
 249                              host_memory_backend_get_policy,
 250                              host_memory_backend_set_policy, NULL);
 251 }
 252
 253 MemoryRegion *
 254 host_memory_backend_get_memory(HostMemoryBackend *backend, Error **errp)
 255 {
 256     return memory_region_size(&backend->mr) ? &backend->mr : NULL;
 257 }
 258
 259 static void
 260 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 261 {
 262     HostMemoryBackend *backend = MEMORY_BACKEND(uc);
 263     HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
 264     Error *local_err = NULL;
 265     void *ptr;
 266     uint64_t sz;
 267
 268     if (bc->alloc) {
 269         bc->alloc(backend, &local_err);
 270         if (local_err) {
 271             error_propagate(errp, local_err);
 272             return;
 273         }
 274
 275         ptr = memory_region_get_ram_ptr(&backend->mr);
 276         sz = memory_region_size(&backend->mr);
 277
 278         if (backend->merge) {
 279             qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
 280         }
 281         if (!backend->dump) {
 282             qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
 283         }
 284 #ifdef CONFIG_NUMA
 285         unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
 286         /* lastbit == MAX_NODES means maxnode = 0 */
 287         unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
 288         /* ensure policy won't be ignored in case memory is preallocated
 289          * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
 290          * this doesn't catch hugepage case. */
 291         unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
 292
 293         /* check for invalid host-nodes and policies and give more verbose
 294          * error messages than mbind(). */
 295         if (maxnode && backend->policy == MPOL_DEFAULT) {
 296             error_setg(errp, "host-nodes must be empty for policy default,"
 297                        " or you should explicitly specify a policy other"
 298                        " than default");
 299             return;
 300         } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
 301             error_setg(errp, "host-nodes must be set for policy %s",
 302                        HostMemPolicy_lookup[backend->policy]);
 303             return;
 304         }
 305
 306         /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
 307          * as argument to mbind() due to an old Linux bug (feature?) which
 308          * cuts off the last specified node. This means backend->host_nodes
 309          * must have MAX_NODES+1 bits available.
 310          */
 311         assert(sizeof(backend->host_nodes) >=
 312                BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
 313         assert(maxnode <= MAX_NODES);
 314         if (mbind(ptr, sz, backend->policy,
 315                   maxnode ? backend->host_nodes : NULL, maxnode + 1, flags)) {
 316             if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
 317                 error_setg_errno(errp, errno,
 318                                  "cannot bind memory to host NUMA nodes");
 319                 return;
 320             }
 321         }
 322 #endif
 323         /* Preallocate memory after the NUMA policy has been instantiated.
 324          * This is necessary to guarantee memory is allocated with
 325          * specified NUMA policy in place.
 326          */
 327         if (backend->prealloc) {
 328             os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz);
 329         }
 330     }
 331 }
 332
 333 static bool
 334 host_memory_backend_can_be_deleted(UserCreatable *uc, Error **errp)
 335 {
 336     MemoryRegion *mr;
 337
 338     mr = host_memory_backend_get_memory(MEMORY_BACKEND(uc), errp);
 339     if (memory_region_is_mapped(mr)) {
 340         return false;
 341     } else {
 342         return true;
 343     }
 344 }
 345
 346 static void
 347 host_memory_backend_class_init(ObjectClass *oc, void *data)
 348 {
 349     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 350
 351     ucc->complete = host_memory_backend_memory_complete;
 352     ucc->can_be_deleted = host_memory_backend_can_be_deleted;
 353 }
 354
 355 static const TypeInfo host_memory_backend_info = {
 356     .name = TYPE_MEMORY_BACKEND,
 357     .parent = TYPE_OBJECT,
 358     .abstract = true,
 359     .class_size = sizeof(HostMemoryBackendClass),
 360     .class_init = host_memory_backend_class_init,
 361     .instance_size = sizeof(HostMemoryBackend),
 362     .instance_init = host_memory_backend_init,
 363     .interfaces = (InterfaceInfo[]) {
 364         { TYPE_USER_CREATABLE },
 365         { }
 366     }
 367 };
 368
 369 static void register_types(void)
 370 {
 371     type_register_static(&host_memory_backend_info);
 372 }
 373
 374 type_init(register_types);