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 "qemu/osdep.h"
  13 #include "sysemu/hostmem.h"
  14 #include "hw/boards.h"
  15 #include "qapi/error.h"
  16 #include "qapi/visitor.h"
  17 #include "qapi-types.h"
  18 #include "qapi-visit.h"
  19 #include "qemu/config-file.h"
  20 #include "qom/object_interfaces.h"
  21
  22 #ifdef CONFIG_NUMA
  23 #include <numaif.h>
  24 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT != MPOL_DEFAULT);
  25 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED != MPOL_PREFERRED);
  26 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND != MPOL_BIND);
  27 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE != MPOL_INTERLEAVE);
  28 #endif
  29
  30 static void
  31 host_memory_backend_get_size(Object *obj, Visitor *v, const char *name,
  32                              void *opaque, Error **errp)
  33 {
  34     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  35     uint64_t value = backend->size;
  36
  37     visit_type_size(v, name, &value, errp);
  38 }
  39
  40 static void
  41 host_memory_backend_set_size(Object *obj, Visitor *v, const char *name,
  42                              void *opaque, Error **errp)
  43 {
  44     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  45     Error *local_err = NULL;
  46     uint64_t value;
  47
  48     if (memory_region_size(&backend->mr)) {
  49         error_setg(&local_err, "cannot change property value");
  50         goto out;
  51     }
  52
  53     visit_type_size(v, name, &value, &local_err);
  54     if (local_err) {
  55         goto out;
  56     }
  57     if (!value) {
  58         error_setg(&local_err, "Property '%s.%s' doesn't take value '%"
  59                    PRIu64 "'", object_get_typename(obj), name, value);
  60         goto out;
  61     }
  62     backend->size = value;
  63 out:
  64     error_propagate(errp, local_err);
  65 }
  66
  67 static void
  68 host_memory_backend_get_host_nodes(Object *obj, Visitor *v, const char *name,
  69                                    void *opaque, Error **errp)
  70 {
  71     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  72     uint16List *host_nodes = NULL;
  73     uint16List **node = &host_nodes;
  74     unsigned long value;
  75
  76     value = find_first_bit(backend->host_nodes, MAX_NODES);
  77     if (value == MAX_NODES) {
  78         return;
  79     }
  80
  81     *node = g_malloc0(sizeof(**node));
  82     (*node)->value = value;
  83     node = &(*node)->next;
  84
  85     do {
  86         value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
  87         if (value == MAX_NODES) {
  88             break;
  89         }
  90
  91         *node = g_malloc0(sizeof(**node));
  92         (*node)->value = value;
  93         node = &(*node)->next;
  94     } while (true);
  95
  96     visit_type_uint16List(v, name, &host_nodes, errp);
  97 }
  98
  99 static void
 100 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
 101                                    void *opaque, Error **errp)
 102 {
 103 #ifdef CONFIG_NUMA
 104     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 105     uint16List *l = NULL;
 106
 107     visit_type_uint16List(v, name, &l, errp);
 108
 109     while (l) {
 110         bitmap_set(backend->host_nodes, l->value, 1);
 111         l = l->next;
 112     }
 113 #else
 114     error_setg(errp, "NUMA node binding are not supported by this QEMU");
 115 #endif
 116 }
 117
 118 static int
 119 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
 120 {
 121     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 122     return backend->policy;
 123 }
 124
 125 static void
 126 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
 127 {
 128     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 129     backend->policy = policy;
 130
 131 #ifndef CONFIG_NUMA
 132     if (policy != HOST_MEM_POLICY_DEFAULT) {
 133         error_setg(errp, "NUMA policies are not supported by this QEMU");
 134     }
 135 #endif
 136 }
 137
 138 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
 139 {
 140     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 141
 142     return backend->merge;
 143 }
 144
 145 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
 146 {
 147     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 148
 149     if (!memory_region_size(&backend->mr)) {
 150         backend->merge = value;
 151         return;
 152     }
 153
 154     if (value != backend->merge) {
 155         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 156         uint64_t sz = memory_region_size(&backend->mr);
 157
 158         qemu_madvise(ptr, sz,
 159                      value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
 160         backend->merge = value;
 161     }
 162 }
 163
 164 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
 165 {
 166     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 167
 168     return backend->dump;
 169 }
 170
 171 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
 172 {
 173     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 174
 175     if (!memory_region_size(&backend->mr)) {
 176         backend->dump = value;
 177         return;
 178     }
 179
 180     if (value != backend->dump) {
 181         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 182         uint64_t sz = memory_region_size(&backend->mr);
 183
 184         qemu_madvise(ptr, sz,
 185                      value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
 186         backend->dump = value;
 187     }
 188 }
 189
 190 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
 191 {
 192     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 193
 194     return backend->prealloc || backend->force_prealloc;
 195 }
 196
 197 static void host_memory_backend_set_prealloc(Object *obj, bool value,
 198                                              Error **errp)
 199 {
 200     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 201
 202     if (backend->force_prealloc) {
 203         if (value) {
 204             error_setg(errp,
 205                        "remove -mem-prealloc to use the prealloc property");
 206             return;
 207         }
 208     }
 209
 210     if (!memory_region_size(&backend->mr)) {
 211         backend->prealloc = value;
 212         return;
 213     }
 214
 215     if (value && !backend->prealloc) {
 216         int fd = memory_region_get_fd(&backend->mr);
 217         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 218         uint64_t sz = memory_region_size(&backend->mr);
 219
 220         os_mem_prealloc(fd, ptr, sz);
 221         backend->prealloc = true;
 222     }
 223 }
 224
 225 static void host_memory_backend_init(Object *obj)
 226 {
 227     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 228     MachineState *machine = MACHINE(qdev_get_machine());
 229
 230     backend->merge = machine_mem_merge(machine);
 231     backend->dump = machine_dump_guest_core(machine);
 232     backend->prealloc = mem_prealloc;
 233
 234     object_property_add_bool(obj, "merge",
 235                         host_memory_backend_get_merge,
 236                         host_memory_backend_set_merge, NULL);
 237     object_property_add_bool(obj, "dump",
 238                         host_memory_backend_get_dump,
 239                         host_memory_backend_set_dump, NULL);
 240     object_property_add_bool(obj, "prealloc",
 241                         host_memory_backend_get_prealloc,
 242                         host_memory_backend_set_prealloc, NULL);
 243     object_property_add(obj, "size", "int",
 244                         host_memory_backend_get_size,
 245                         host_memory_backend_set_size, NULL, NULL, NULL);
 246     object_property_add(obj, "host-nodes", "int",
 247                         host_memory_backend_get_host_nodes,
 248                         host_memory_backend_set_host_nodes, NULL, NULL, NULL);
 249     object_property_add_enum(obj, "policy", "HostMemPolicy",
 250                              HostMemPolicy_lookup,
 251                              host_memory_backend_get_policy,
 252                              host_memory_backend_set_policy, NULL);
 253 }
 254
 255 MemoryRegion *
 256 host_memory_backend_get_memory(HostMemoryBackend *backend, Error **errp)
 257 {
 258     return memory_region_size(&backend->mr) ? &backend->mr : NULL;
 259 }
 260
 261 static void
 262 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 263 {
 264     HostMemoryBackend *backend = MEMORY_BACKEND(uc);
 265     HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
 266     Error *local_err = NULL;
 267     void *ptr;
 268     uint64_t sz;
 269
 270     if (bc->alloc) {
 271         bc->alloc(backend, &local_err);
 272         if (local_err) {
 273             error_propagate(errp, local_err);
 274             return;
 275         }
 276
 277         ptr = memory_region_get_ram_ptr(&backend->mr);
 278         sz = memory_region_size(&backend->mr);
 279
 280         if (backend->merge) {
 281             qemu_madvise(ptr, sz, QEMU_MADV_MERGEABLE);
 282         }
 283         if (!backend->dump) {
 284             qemu_madvise(ptr, sz, QEMU_MADV_DONTDUMP);
 285         }
 286 #ifdef CONFIG_NUMA
 287         unsigned long lastbit = find_last_bit(backend->host_nodes, MAX_NODES);
 288         /* lastbit == MAX_NODES means maxnode = 0 */
 289         unsigned long maxnode = (lastbit + 1) % (MAX_NODES + 1);
 290         /* ensure policy won't be ignored in case memory is preallocated
 291          * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
 292          * this doesn't catch hugepage case. */
 293         unsigned flags = MPOL_MF_STRICT | MPOL_MF_MOVE;
 294
 295         /* check for invalid host-nodes and policies and give more verbose
 296          * error messages than mbind(). */
 297         if (maxnode && backend->policy == MPOL_DEFAULT) {
 298             error_setg(errp, "host-nodes must be empty for policy default,"
 299                        " or you should explicitly specify a policy other"
 300                        " than default");
 301             return;
 302         } else if (maxnode == 0 && backend->policy != MPOL_DEFAULT) {
 303             error_setg(errp, "host-nodes must be set for policy %s",
 304                        HostMemPolicy_lookup[backend->policy]);
 305             return;
 306         }
 307
 308         /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
 309          * as argument to mbind() due to an old Linux bug (feature?) which
 310          * cuts off the last specified node. This means backend->host_nodes
 311          * must have MAX_NODES+1 bits available.
 312          */
 313         assert(sizeof(backend->host_nodes) >=
 314                BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
 315         assert(maxnode <= MAX_NODES);
 316         if (mbind(ptr, sz, backend->policy,
 317                   maxnode ? backend->host_nodes : NULL, maxnode + 1, flags)) {
 318             if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
 319                 error_setg_errno(errp, errno,
 320                                  "cannot bind memory to host NUMA nodes");
 321                 return;
 322             }
 323         }
 324 #endif
 325         /* Preallocate memory after the NUMA policy has been instantiated.
 326          * This is necessary to guarantee memory is allocated with
 327          * specified NUMA policy in place.
 328          */
 329         if (backend->prealloc) {
 330             os_mem_prealloc(memory_region_get_fd(&backend->mr), ptr, sz);
 331         }
 332     }
 333 }
 334
 335 static bool
 336 host_memory_backend_can_be_deleted(UserCreatable *uc, Error **errp)
 337 {
 338     MemoryRegion *mr;
 339
 340     mr = host_memory_backend_get_memory(MEMORY_BACKEND(uc), errp);
 341     if (memory_region_is_mapped(mr)) {
 342         return false;
 343     } else {
 344         return true;
 345     }
 346 }
 347
 348 static void
 349 host_memory_backend_class_init(ObjectClass *oc, void *data)
 350 {
 351     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 352
 353     ucc->complete = host_memory_backend_memory_complete;
 354     ucc->can_be_deleted = host_memory_backend_can_be_deleted;
 355 }
 356
 357 static const TypeInfo host_memory_backend_info = {
 358     .name = TYPE_MEMORY_BACKEND,
 359     .parent = TYPE_OBJECT,
 360     .abstract = true,
 361     .class_size = sizeof(HostMemoryBackendClass),
 362     .class_init = host_memory_backend_class_init,
 363     .instance_size = sizeof(HostMemoryBackend),
 364     .instance_init = host_memory_backend_init,
 365     .interfaces = (InterfaceInfo[]) {
 366         { TYPE_USER_CREATABLE },
 367         { }
 368     }
 369 };
 370
 371 static void register_types(void)
 372 {
 373     type_register_static(&host_memory_backend_info);
 374 }
 375
 376 type_init(register_types);