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