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 }
 110
 111 int numa_init_func(QemuOpts *opts, void *opaque)
 112 {
 113     NumaOptions *object = NULL;
 114     Error *err = NULL;
 115
 116     {
 117         OptsVisitor *ov = opts_visitor_new(opts);
 118         visit_type_NumaOptions(opts_get_visitor(ov), &object, NULL, &err);
 119         opts_visitor_cleanup(ov);
 120     }
 121
 122     if (err) {
 123         goto error;
 124     }
 125
 126     switch (object->kind) {
 127     case NUMA_OPTIONS_KIND_NODE:
 128         numa_node_parse(object->node, opts, &err);
 129         if (err) {
 130             goto error;
 131         }
 132         nb_numa_nodes++;
 133         break;
 134     default:
 135         abort();
 136     }
 137
 138     return 0;
 139
 140 error:
 141     qerror_report_err(err);
 142     error_free(err);
 143
 144     if (object) {
 145         QapiDeallocVisitor *dv = qapi_dealloc_visitor_new();
 146         visit_type_NumaOptions(qapi_dealloc_get_visitor(dv),
 147                                &object, NULL, NULL);
 148         qapi_dealloc_visitor_cleanup(dv);
 149     }
 150
 151     return -1;
 152 }
 153
 154 void set_numa_nodes(void)
 155 {
 156     if (nb_numa_nodes > 0) {
 157         uint64_t numa_total;
 158         int i;
 159
 160         if (nb_numa_nodes > MAX_NODES) {
 161             nb_numa_nodes = MAX_NODES;
 162         }
 163
 164         /* If no memory size if given for any node, assume the default case
 165          * and distribute the available memory equally across all nodes
 166          */
 167         for (i = 0; i < nb_numa_nodes; i++) {
 168             if (numa_info[i].node_mem != 0) {
 169                 break;
 170             }
 171         }
 172         if (i == nb_numa_nodes) {
 173             uint64_t usedmem = 0;
 174
 175             /* On Linux, the each node's border has to be 8MB aligned,
 176              * the final node gets the rest.
 177              */
 178             for (i = 0; i < nb_numa_nodes - 1; i++) {
 179                 numa_info[i].node_mem = (ram_size / nb_numa_nodes) &
 180                                         ~((1 << 23UL) - 1);
 181                 usedmem += numa_info[i].node_mem;
 182             }
 183             numa_info[i].node_mem = ram_size - usedmem;
 184         }
 185
 186         numa_total = 0;
 187         for (i = 0; i < nb_numa_nodes; i++) {
 188             numa_total += numa_info[i].node_mem;
 189         }
 190         if (numa_total != ram_size) {
 191             error_report("total memory for NUMA nodes (%" PRIu64 ")"
 192                          " should equal RAM size (" RAM_ADDR_FMT ")",
 193                          numa_total, ram_size);
 194             exit(1);
 195         }
 196
 197         for (i = 0; i < nb_numa_nodes; i++) {
 198             if (!bitmap_empty(numa_info[i].node_cpu, MAX_CPUMASK_BITS)) {
 199                 break;
 200             }
 201         }
 202         /* assigning the VCPUs round-robin is easier to implement, guest OSes
 203          * must cope with this anyway, because there are BIOSes out there in
 204          * real machines which also use this scheme.
 205          */
 206         if (i == nb_numa_nodes) {
 207             for (i = 0; i < max_cpus; i++) {
 208                 set_bit(i, numa_info[i % nb_numa_nodes].node_cpu);
 209             }
 210         }
 211     }
 212 }
 213
 214 void set_numa_modes(void)
 215 {
 216     CPUState *cpu;
 217     int i;
 218
 219     CPU_FOREACH(cpu) {
 220         for (i = 0; i < nb_numa_nodes; i++) {
 221             if (test_bit(cpu->cpu_index, numa_info[i].node_cpu)) {
 222                 cpu->numa_node = i;
 223             }
 224         }
 225     }
 226 }
 227
 228 static void allocate_system_memory_nonnuma(MemoryRegion *mr, Object *owner,
 229                                            const char *name,
 230                                            uint64_t ram_size)
 231 {
 232     if (mem_path) {
 233 #ifdef __linux__
 234         Error *err = NULL;
 235         memory_region_init_ram_from_file(mr, owner, name, ram_size, false,
 236                                          mem_path, &err);
 237
 238         /* Legacy behavior: if allocation failed, fall back to
 239          * regular RAM allocation.
 240          */
 241         if (err) {
 242             qerror_report_err(err);
 243             error_free(err);
 244             memory_region_init_ram(mr, owner, name, ram_size);
 245         }
 246 #else
 247         fprintf(stderr, "-mem-path not supported on this host\n");
 248         exit(1);
 249 #endif
 250     } else {
 251         memory_region_init_ram(mr, owner, name, ram_size);
 252     }
 253     vmstate_register_ram_global(mr);
 254 }
 255
 256 void memory_region_allocate_system_memory(MemoryRegion *mr, Object *owner,
 257                                           const char *name,
 258                                           uint64_t ram_size)
 259 {
 260     uint64_t addr = 0;
 261     int i;
 262
 263     if (nb_numa_nodes == 0 || !have_memdevs) {
 264         allocate_system_memory_nonnuma(mr, owner, name, ram_size);
 265         return;
 266     }
 267
 268     memory_region_init(mr, owner, name, ram_size);
 269     for (i = 0; i < MAX_NODES; i++) {
 270         Error *local_err = NULL;
 271         uint64_t size = numa_info[i].node_mem;
 272         HostMemoryBackend *backend = numa_info[i].node_memdev;
 273         if (!backend) {
 274             continue;
 275         }
 276         MemoryRegion *seg = host_memory_backend_get_memory(backend, &local_err);
 277         if (local_err) {
 278             qerror_report_err(local_err);
 279             exit(1);
 280         }
 281
 282         memory_region_add_subregion(mr, addr, seg);
 283         vmstate_register_ram_global(seg);
 284         addr += size;
 285     }
 286 }
 287
 288 static int query_memdev(Object *obj, void *opaque)
 289 {
 290     MemdevList **list = opaque;
 291     Error *err = NULL;
 292
 293     if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
 294         MemdevList *m = g_malloc0(sizeof(*m));
 295
 296         m->value = g_malloc0(sizeof(*m->value));
 297
 298         m->value->size = object_property_get_int(obj, "size",
 299                                                  &err);
 300         if (err) {
 301             goto error;
 302         }
 303
 304         m->value->merge = object_property_get_bool(obj, "merge",
 305                                                    &err);
 306         if (err) {
 307             goto error;
 308         }
 309
 310         m->value->dump = object_property_get_bool(obj, "dump",
 311                                                   &err);
 312         if (err) {
 313             goto error;
 314         }
 315
 316         m->value->prealloc = object_property_get_bool(obj,
 317                                                       "prealloc", &err);
 318         if (err) {
 319             goto error;
 320         }
 321
 322         m->value->policy = object_property_get_enum(obj,
 323                                                     "policy",
 324                                                     HostMemPolicy_lookup,
 325                                                     &err);
 326         if (err) {
 327             goto error;
 328         }
 329
 330         object_property_get_uint16List(obj, "host-nodes",
 331                                        &m->value->host_nodes, &err);
 332         if (err) {
 333             goto error;
 334         }
 335
 336         m->next = *list;
 337         *list = m;
 338     }
 339
 340     return 0;
 341 error:
 342     return -1;
 343 }
 344
 345 MemdevList *qmp_query_memdev(Error **errp)
 346 {
 347     Object *obj;
 348     MemdevList *list = NULL, *m;
 349
 350     obj = object_resolve_path("/objects", NULL);
 351     if (obj == NULL) {
 352         return NULL;
 353     }
 354
 355     if (object_child_foreach(obj, query_memdev, &list) != 0) {
 356         goto error;
 357     }
 358
 359     return list;
 360
 361 error:
 362     while (list) {
 363         m = list;
 364         list = list->next;
 365         g_free(m->value);
 366         g_free(m);
 367     }
 368     return NULL;
 369 }