numa.c

   1 /*
   2  * NUMA parameter parsing routines
   3  *
   4  * Copyright (c) 2014 Fujitsu Ltd.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "sysemu/sysemu.h"
  26 #include "exec/cpu-common.h"
  27 #include "qemu/bitmap.h"
  28 #include "qom/cpu.h"
  29 #include "qemu/error-report.h"
  30 #include "include/exec/cpu-common.h" /* for RAM_ADDR_FMT */
  31 #include "qapi-visit.h"
  32 #include "qapi/opts-visitor.h"
  33 #include "qapi/dealloc-visitor.h"
  34 #include "qapi/qmp/qerror.h"
  35 #include "hw/boards.h"
  36 #include "sysemu/hostmem.h"
  37 #include "qmp-commands.h"
  38
  39 QemuOptsList qemu_numa_opts = {
  40     .name = "numa",
  41     .implied_opt_name = "type",
  42     .head = QTAILQ_HEAD_INITIALIZER(qemu_numa_opts.head),
  43     .desc = { { 0 } } /* validated with OptsVisitor */
  44 };
  45
  46 static int have_memdevs = -1;
  47
  48 static void numa_node_parse(NumaNodeOptions *node, QemuOpts *opts, Error **errp)
  49 {
  50     uint16_t nodenr;
  51     uint16List *cpus = NULL;
  52
  53     if (node->has_nodeid) {
  54         nodenr = node->nodeid;
  55     } else {
  56         nodenr = nb_numa_nodes;
  57     }
  58
  59     if (nodenr >= MAX_NODES) {
  60         error_setg(errp, "Max number of NUMA nodes reached: %"
  61                    PRIu16 "\n", nodenr);
  62         return;
  63     }
  64
  65     for (cpus = node->cpus; cpus; cpus = cpus->next) {
  66         if (cpus->value > MAX_CPUMASK_BITS) {
  67             error_setg(errp, "CPU number %" PRIu16 " is bigger than %d",
  68                        cpus->value, MAX_CPUMASK_BITS);
  69             return;
  70         }
  71         bitmap_set(numa_info[nodenr].node_cpu, cpus->value, 1);
  72     }
  73
  74     if (node->has_mem && node->has_memdev) {
  75         error_setg(errp, "qemu: cannot specify both mem= and memdev=\n");
  76         return;
  77     }
  78
  79     if (have_memdevs == -1) {
  80         have_memdevs = node->has_memdev;
  81     }
  82     if (node->has_memdev != have_memdevs) {
  83         error_setg(errp, "qemu: memdev option must be specified for either "
  84                    "all or no nodes\n");
  85         return;
  86     }
  87
  88     if (node->has_mem) {
  89         uint64_t mem_size = node->mem;
  90         const char *mem_str = qemu_opt_get(opts, "mem");
  91         /* Fix up legacy suffix-less format */
  92         if (g_ascii_isdigit(mem_str[strlen(mem_str) - 1])) {
  93             mem_size <<= 20;
  94         }
  95         numa_info[nodenr].node_mem = mem_size;
  96     }
  97     if (node->has_memdev) {
  98         Object *o;
  99         o = object_resolve_path_type(node->memdev, TYPE_MEMORY_BACKEND, NULL);
 100         if (!o) {
 101             error_setg(errp, "memdev=%s is ambiguous", node->memdev);
 102             return;
 103         }
 104
 105         object_ref(o);
 106         numa_info[nodenr].node_mem = object_property_get_int(o, "size", NULL);
 107         numa_info[nodenr].node_memdev = MEMORY_BACKEND(o);
 108     }
 109     numa_info[nodenr].present = true;
 110     max_numa_nodeid = MAX(max_numa_nodeid, nodenr + 1);
 111 }
 112
 113 int numa_init_func(QemuOpts *opts, void *opaque)
 114 {
 115     NumaOptions *object = NULL;
 116     Error *err = NULL;
 117
 118     {
 119         OptsVisitor *ov = opts_visitor_new(opts);
 120         visit_type_NumaOptions(opts_get_visitor(ov), &object, NULL, &err);
 121         opts_visitor_cleanup(ov);
 122     }
 123
 124     if (err) {
 125         goto error;
 126     }
 127
 128     switch (object->kind) {
 129     case NUMA_OPTIONS_KIND_NODE:
 130         numa_node_parse(object->node, opts, &err);
 131         if (err) {
 132             goto error;
 133         }
 134         nb_numa_nodes++;
 135         break;
 136     default:
 137         abort();
 138     }
 139
 140     return 0;
 141
 142 error:
 143     qerror_report_err(err);
 144     error_free(err);
 145
 146     if (object) {
 147         QapiDeallocVisitor *dv = qapi_dealloc_visitor_new();
 148         visit_type_NumaOptions(qapi_dealloc_get_visitor(dv),
 149                                &object, NULL, NULL);
 150         qapi_dealloc_visitor_cleanup(dv);
 151     }
 152
 153     return -1;
 154 }
 155
 156 void set_numa_nodes(void)
 157 {
 158     if (nb_numa_nodes > 0) {
 159         uint64_t numa_total;
 160         int i;
 161
 162         if (nb_numa_nodes > MAX_NODES) {
 163             nb_numa_nodes = MAX_NODES;
 164         }
 165
 166         /* If no memory size is given for any node, assume the default case
 167          * and distribute the available memory equally across all nodes
 168          */
 169         for (i = 0; i < nb_numa_nodes; i++) {
 170             if (numa_info[i].node_mem != 0) {
 171                 break;
 172             }
 173         }
 174         if (i == nb_numa_nodes) {
 175             uint64_t usedmem = 0;
 176
 177             /* On Linux, each node's border has to be 8MB aligned,
 178              * the final node gets the rest.
 179              */
 180             for (i = 0; i < nb_numa_nodes - 1; i++) {
 181                 numa_info[i].node_mem = (ram_size / nb_numa_nodes) &
 182                                         ~((1 << 23UL) - 1);
 183                 usedmem += numa_info[i].node_mem;
 184             }
 185             numa_info[i].node_mem = ram_size - usedmem;
 186         }
 187
 188         numa_total = 0;
 189         for (i = 0; i < nb_numa_nodes; i++) {
 190             numa_total += numa_info[i].node_mem;
 191         }
 192         if (numa_total != ram_size) {
 193             error_report("total memory for NUMA nodes (%" PRIu64 ")"
 194                          " should equal RAM size (" RAM_ADDR_FMT ")",
 195                          numa_total, ram_size);
 196             exit(1);
 197         }
 198
 199         for (i = 0; i < nb_numa_nodes; i++) {
 200             if (!bitmap_empty(numa_info[i].node_cpu, MAX_CPUMASK_BITS)) {
 201                 break;
 202             }
 203         }
 204         /* assigning the VCPUs round-robin is easier to implement, guest OSes
 205          * must cope with this anyway, because there are BIOSes out there in
 206          * real machines which also use this scheme.
 207          */
 208         if (i == nb_numa_nodes) {
 209             for (i = 0; i < max_cpus; i++) {
 210                 set_bit(i, numa_info[i % nb_numa_nodes].node_cpu);
 211             }
 212         }
 213     }
 214 }
 215
 216 void set_numa_modes(void)
 217 {
 218     CPUState *cpu;
 219     int i;
 220
 221     CPU_FOREACH(cpu) {
 222         for (i = 0; i < nb_numa_nodes; i++) {
 223             if (test_bit(cpu->cpu_index, numa_info[i].node_cpu)) {
 224                 cpu->numa_node = i;
 225             }
 226         }
 227     }
 228 }
 229
 230 static void allocate_system_memory_nonnuma(MemoryRegion *mr, Object *owner,
 231                                            const char *name,
 232                                            uint64_t ram_size)
 233 {
 234     if (mem_path) {
 235 #ifdef __linux__
 236         Error *err = NULL;
 237         memory_region_init_ram_from_file(mr, owner, name, ram_size, false,
 238                                          mem_path, &err);
 239
 240         /* Legacy behavior: if allocation failed, fall back to
 241          * regular RAM allocation.
 242          */
 243         if (err) {
 244             qerror_report_err(err);
 245             error_free(err);
 246             memory_region_init_ram(mr, owner, name, ram_size);
 247         }
 248 #else
 249         fprintf(stderr, "-mem-path not supported on this host\n");
 250         exit(1);
 251 #endif
 252     } else {
 253         memory_region_init_ram(mr, owner, name, ram_size);
 254     }
 255     vmstate_register_ram_global(mr);
 256 }
 257
 258 void memory_region_allocate_system_memory(MemoryRegion *mr, Object *owner,
 259                                           const char *name,
 260                                           uint64_t ram_size)
 261 {
 262     uint64_t addr = 0;
 263     int i;
 264
 265     if (nb_numa_nodes == 0 || !have_memdevs) {
 266         allocate_system_memory_nonnuma(mr, owner, name, ram_size);
 267         return;
 268     }
 269
 270     memory_region_init(mr, owner, name, ram_size);
 271     for (i = 0; i < MAX_NODES; i++) {
 272         Error *local_err = NULL;
 273         uint64_t size = numa_info[i].node_mem;
 274         HostMemoryBackend *backend = numa_info[i].node_memdev;
 275         if (!backend) {
 276             continue;
 277         }
 278         MemoryRegion *seg = host_memory_backend_get_memory(backend, &local_err);
 279         if (local_err) {
 280             qerror_report_err(local_err);
 281             exit(1);
 282         }
 283
 284         memory_region_add_subregion(mr, addr, seg);
 285         vmstate_register_ram_global(seg);
 286         addr += size;
 287     }
 288 }
 289
 290 static int query_memdev(Object *obj, void *opaque)
 291 {
 292     MemdevList **list = opaque;
 293     Error *err = NULL;
 294
 295     if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
 296         MemdevList *m = g_malloc0(sizeof(*m));
 297
 298         m->value = g_malloc0(sizeof(*m->value));
 299
 300         m->value->size = object_property_get_int(obj, "size",
 301                                                  &err);
 302         if (err) {
 303             goto error;
 304         }
 305
 306         m->value->merge = object_property_get_bool(obj, "merge",
 307                                                    &err);
 308         if (err) {
 309             goto error;
 310         }
 311
 312         m->value->dump = object_property_get_bool(obj, "dump",
 313                                                   &err);
 314         if (err) {
 315             goto error;
 316         }
 317
 318         m->value->prealloc = object_property_get_bool(obj,
 319                                                       "prealloc", &err);
 320         if (err) {
 321             goto error;
 322         }
 323
 324         m->value->policy = object_property_get_enum(obj,
 325                                                     "policy",
 326                                                     HostMemPolicy_lookup,
 327                                                     &err);
 328         if (err) {
 329             goto error;
 330         }
 331
 332         object_property_get_uint16List(obj, "host-nodes",
 333                                        &m->value->host_nodes, &err);
 334         if (err) {
 335             goto error;
 336         }
 337
 338         m->next = *list;
 339         *list = m;
 340     }
 341
 342     return 0;
 343 error:
 344     return -1;
 345 }
 346
 347 MemdevList *qmp_query_memdev(Error **errp)
 348 {
 349     Object *obj;
 350     MemdevList *list = NULL, *m;
 351
 352     obj = object_resolve_path("/objects", NULL);
 353     if (obj == NULL) {
 354         return NULL;
 355     }
 356
 357     if (object_child_foreach(obj, query_memdev, &list) != 0) {
 358         goto error;
 359     }
 360
 361     return list;
 362
 363 error:
 364     while (list) {
 365         m = list;
 366         list = list->next;
 367         g_free(m->value);
 368         g_free(m);
 369     }
 370     return NULL;
 371 }