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 uint16List **host_memory_append_node(uint16List **node,
  68                                             unsigned long value)
  69 {
  70      *node = g_malloc0(sizeof(**node));
  71      (*node)->value = value;
  72      return &(*node)->next;
  73 }
  74
  75 static void
  76 host_memory_backend_get_host_nodes(Object *obj, Visitor *v, const char *name,
  77                                    void *opaque, Error **errp)
  78 {
  79     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
  80     uint16List *host_nodes = NULL;
  81     uint16List **node = &host_nodes;
  82     unsigned long value;
  83
  84     value = find_first_bit(backend->host_nodes, MAX_NODES);
  85
  86     node = host_memory_append_node(node, value);
  87
  88     if (value == MAX_NODES) {
  89         goto out;
  90     }
  91
  92     do {
  93         value = find_next_bit(backend->host_nodes, MAX_NODES, value + 1);
  94         if (value == MAX_NODES) {
  95             break;
  96         }
  97
  98         node = host_memory_append_node(node, value);
  99     } while (true);
 100
 101 out:
 102     visit_type_uint16List(v, name, &host_nodes, errp);
 103 }
 104
 105 static void
 106 host_memory_backend_set_host_nodes(Object *obj, Visitor *v, const char *name,
 107                                    void *opaque, Error **errp)
 108 {
 109 #ifdef CONFIG_NUMA
 110     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 111     uint16List *l = NULL;
 112
 113     visit_type_uint16List(v, name, &l, errp);
 114
 115     while (l) {
 116         bitmap_set(backend->host_nodes, l->value, 1);
 117         l = l->next;
 118     }
 119 #else
 120     error_setg(errp, "NUMA node binding are not supported by this QEMU");
 121 #endif
 122 }
 123
 124 static int
 125 host_memory_backend_get_policy(Object *obj, Error **errp G_GNUC_UNUSED)
 126 {
 127     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 128     return backend->policy;
 129 }
 130
 131 static void
 132 host_memory_backend_set_policy(Object *obj, int policy, Error **errp)
 133 {
 134     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 135     backend->policy = policy;
 136
 137 #ifndef CONFIG_NUMA
 138     if (policy != HOST_MEM_POLICY_DEFAULT) {
 139         error_setg(errp, "NUMA policies are not supported by this QEMU");
 140     }
 141 #endif
 142 }
 143
 144 static bool host_memory_backend_get_merge(Object *obj, Error **errp)
 145 {
 146     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 147
 148     return backend->merge;
 149 }
 150
 151 static void host_memory_backend_set_merge(Object *obj, bool value, Error **errp)
 152 {
 153     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 154
 155     if (!memory_region_size(&backend->mr)) {
 156         backend->merge = value;
 157         return;
 158     }
 159
 160     if (value != backend->merge) {
 161         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 162         uint64_t sz = memory_region_size(&backend->mr);
 163
 164         qemu_madvise(ptr, sz,
 165                      value ? QEMU_MADV_MERGEABLE : QEMU_MADV_UNMERGEABLE);
 166         backend->merge = value;
 167     }
 168 }
 169
 170 static bool host_memory_backend_get_dump(Object *obj, Error **errp)
 171 {
 172     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 173
 174     return backend->dump;
 175 }
 176
 177 static void host_memory_backend_set_dump(Object *obj, bool value, Error **errp)
 178 {
 179     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 180
 181     if (!memory_region_size(&backend->mr)) {
 182         backend->dump = value;
 183         return;
 184     }
 185
 186     if (value != backend->dump) {
 187         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 188         uint64_t sz = memory_region_size(&backend->mr);
 189
 190         qemu_madvise(ptr, sz,
 191                      value ? QEMU_MADV_DODUMP : QEMU_MADV_DONTDUMP);
 192         backend->dump = value;
 193     }
 194 }
 195
 196 static bool host_memory_backend_get_prealloc(Object *obj, Error **errp)
 197 {
 198     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 199
 200     return backend->prealloc || backend->force_prealloc;
 201 }
 202
 203 static void host_memory_backend_set_prealloc(Object *obj, bool value,
 204                                              Error **errp)
 205 {
 206     Error *local_err = NULL;
 207     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 208
 209     if (backend->force_prealloc) {
 210         if (value) {
 211             error_setg(errp,
 212                        "remove -mem-prealloc to use the prealloc property");
 213             return;
 214         }
 215     }
 216
 217     if (!memory_region_size(&backend->mr)) {
 218         backend->prealloc = value;
 219         return;
 220     }
 221
 222     if (value && !backend->prealloc) {
 223         int fd = memory_region_get_fd(&backend->mr);
 224         void *ptr = memory_region_get_ram_ptr(&backend->mr);
 225         uint64_t sz = memory_region_size(&backend->mr);
 226
 227         os_mem_prealloc(fd, ptr, sz, &local_err);
 228         if (local_err) {
 229             error_propagate(errp, local_err);
 230             return;
 231         }
 232         backend->prealloc = true;
 233     }
 234 }
 235
 236 static void host_memory_backend_init(Object *obj)
 237 {
 238     HostMemoryBackend *backend = MEMORY_BACKEND(obj);
 239     MachineState *machine = MACHINE(qdev_get_machine());
 240
 241     backend->merge = machine_mem_merge(machine);
 242     backend->dump = machine_dump_guest_core(machine);
 243     backend->prealloc = mem_prealloc;
 244 }
 245
 246 MemoryRegion *
 247 host_memory_backend_get_memory(HostMemoryBackend *backend, Error **errp)
 248 {
 249     return memory_region_size(&backend->mr) ? &backend->mr : NULL;
 250 }
 251
 252 void host_memory_backend_set_mapped(HostMemoryBackend *backend, bool mapped)
 253 {
 254     backend->is_mapped = mapped;
 255 }
 256
 257 bool host_memory_backend_is_mapped(HostMemoryBackend *backend)
 258 {
 259     return backend->is_mapped;
 260 }
 261
 262 static void
 263 host_memory_backend_memory_complete(UserCreatable *uc, Error **errp)
 264 {
 265     HostMemoryBackend *backend = MEMORY_BACKEND(uc);
 266     HostMemoryBackendClass *bc = MEMORY_BACKEND_GET_CLASS(uc);
 267     Error *local_err = NULL;
 268     void *ptr;
 269     uint64_t sz;
 270
 271     if (bc->alloc) {
 272         bc->alloc(backend, &local_err);
 273         if (local_err) {
 274             goto out;
 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                             &local_err);
 332             if (local_err) {
 333                 goto out;
 334             }
 335         }
 336     }
 337 out:
 338     error_propagate(errp, local_err);
 339 }
 340
 341 static bool
 342 host_memory_backend_can_be_deleted(UserCreatable *uc, Error **errp)
 343 {
 344     if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc))) {
 345         return false;
 346     } else {
 347         return true;
 348     }
 349 }
 350
 351 static void
 352 host_memory_backend_class_init(ObjectClass *oc, void *data)
 353 {
 354     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 355
 356     ucc->complete = host_memory_backend_memory_complete;
 357     ucc->can_be_deleted = host_memory_backend_can_be_deleted;
 358
 359     object_class_property_add_bool(oc, "merge",
 360         host_memory_backend_get_merge,
 361         host_memory_backend_set_merge, &error_abort);
 362     object_class_property_add_bool(oc, "dump",
 363         host_memory_backend_get_dump,
 364         host_memory_backend_set_dump, &error_abort);
 365     object_class_property_add_bool(oc, "prealloc",
 366         host_memory_backend_get_prealloc,
 367         host_memory_backend_set_prealloc, &error_abort);
 368     object_class_property_add(oc, "size", "int",
 369         host_memory_backend_get_size,
 370         host_memory_backend_set_size,
 371         NULL, NULL, &error_abort);
 372     object_class_property_add(oc, "host-nodes", "int",
 373         host_memory_backend_get_host_nodes,
 374         host_memory_backend_set_host_nodes,
 375         NULL, NULL, &error_abort);
 376     object_class_property_add_enum(oc, "policy", "HostMemPolicy",
 377         HostMemPolicy_lookup,
 378         host_memory_backend_get_policy,
 379         host_memory_backend_set_policy, &error_abort);
 380 }
 381
 382 static const TypeInfo host_memory_backend_info = {
 383     .name = TYPE_MEMORY_BACKEND,
 384     .parent = TYPE_OBJECT,
 385     .abstract = true,
 386     .class_size = sizeof(HostMemoryBackendClass),
 387     .class_init = host_memory_backend_class_init,
 388     .instance_size = sizeof(HostMemoryBackend),
 389     .instance_init = host_memory_backend_init,
 390     .interfaces = (InterfaceInfo[]) {
 391         { TYPE_USER_CREATABLE },
 392         { }
 393     }
 394 };
 395
 396 static void register_types(void)
 397 {
 398     type_register_static(&host_memory_backend_info);
 399 }
 400
 401 type_init(register_types);