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
  38 QemuOptsList qemu_numa_opts = {
  39     .name = "numa",
  40     .implied_opt_name = "type",
  41     .head = QTAILQ_HEAD_INITIALIZER(qemu_numa_opts.head),
  42     .desc = { { 0 } } /* validated with OptsVisitor */
  43 };
  44
  45 static int have_memdevs = -1;
  46
  47 static void numa_node_parse(NumaNodeOptions *node, QemuOpts *opts, Error **errp)
  48 {
  49     uint16_t nodenr;
  50     uint16List *cpus = NULL;
  51
  52     if (node->has_nodeid) {
  53         nodenr = node->nodeid;
  54     } else {
  55         nodenr = nb_numa_nodes;
  56     }
  57
  58     if (nodenr >= MAX_NODES) {
  59         error_setg(errp, "Max number of NUMA nodes reached: %"
  60                    PRIu16 "\n", nodenr);
  61         return;
  62     }
  63
  64     for (cpus = node->cpus; cpus; cpus = cpus->next) {
  65         if (cpus->value > MAX_CPUMASK_BITS) {
  66             error_setg(errp, "CPU number %" PRIu16 " is bigger than %d",
  67                        cpus->value, MAX_CPUMASK_BITS);
  68             return;
  69         }
  70         bitmap_set(numa_info[nodenr].node_cpu, cpus->value, 1);
  71     }
  72
  73     if (node->has_mem && node->has_memdev) {
  74         error_setg(errp, "qemu: cannot specify both mem= and memdev=\n");
  75         return;
  76     }
  77
  78     if (have_memdevs == -1) {
  79         have_memdevs = node->has_memdev;
  80     }
  81     if (node->has_memdev != have_memdevs) {
  82         error_setg(errp, "qemu: memdev option must be specified for either "
  83                    "all or no nodes\n");
  84         return;
  85     }
  86
  87     if (node->has_mem) {
  88         uint64_t mem_size = node->mem;
  89         const char *mem_str = qemu_opt_get(opts, "mem");
  90         /* Fix up legacy suffix-less format */
  91         if (g_ascii_isdigit(mem_str[strlen(mem_str) - 1])) {
  92             mem_size <<= 20;
  93         }
  94         numa_info[nodenr].node_mem = mem_size;
  95     }
  96     if (node->has_memdev) {
  97         Object *o;
  98         o = object_resolve_path_type(node->memdev, TYPE_MEMORY_BACKEND, NULL);
  99         if (!o) {
 100             error_setg(errp, "memdev=%s is ambiguous", node->memdev);
 101             return;
 102         }
 103
 104         object_ref(o);
 105         numa_info[nodenr].node_mem = object_property_get_int(o, "size", NULL);
 106         numa_info[nodenr].node_memdev = MEMORY_BACKEND(o);
 107     }
 108 }
 109
 110 int numa_init_func(QemuOpts *opts, void *opaque)
 111 {
 112     NumaOptions *object = NULL;
 113     Error *err = NULL;
 114
 115     {
 116         OptsVisitor *ov = opts_visitor_new(opts);
 117         visit_type_NumaOptions(opts_get_visitor(ov), &object, NULL, &err);
 118         opts_visitor_cleanup(ov);
 119     }
 120
 121     if (err) {
 122         goto error;
 123     }
 124
 125     switch (object->kind) {
 126     case NUMA_OPTIONS_KIND_NODE:
 127         numa_node_parse(object->node, opts, &err);
 128         if (err) {
 129             goto error;
 130         }
 131         nb_numa_nodes++;
 132         break;
 133     default:
 134         abort();
 135     }
 136
 137     return 0;
 138
 139 error:
 140     qerror_report_err(err);
 141     error_free(err);
 142
 143     if (object) {
 144         QapiDeallocVisitor *dv = qapi_dealloc_visitor_new();
 145         visit_type_NumaOptions(qapi_dealloc_get_visitor(dv),
 146                                &object, NULL, NULL);
 147         qapi_dealloc_visitor_cleanup(dv);
 148     }
 149
 150     return -1;
 151 }
 152
 153 void set_numa_nodes(void)
 154 {
 155     if (nb_numa_nodes > 0) {
 156         uint64_t numa_total;
 157         int i;
 158
 159         if (nb_numa_nodes > MAX_NODES) {
 160             nb_numa_nodes = MAX_NODES;
 161         }
 162
 163         /* If no memory size if given for any node, assume the default case
 164          * and distribute the available memory equally across all nodes
 165          */
 166         for (i = 0; i < nb_numa_nodes; i++) {
 167             if (numa_info[i].node_mem != 0) {
 168                 break;
 169             }
 170         }
 171         if (i == nb_numa_nodes) {
 172             uint64_t usedmem = 0;
 173
 174             /* On Linux, the each node's border has to be 8MB aligned,
 175              * the final node gets the rest.
 176              */
 177             for (i = 0; i < nb_numa_nodes - 1; i++) {
 178                 numa_info[i].node_mem = (ram_size / nb_numa_nodes) &
 179                                         ~((1 << 23UL) - 1);
 180                 usedmem += numa_info[i].node_mem;
 181             }
 182             numa_info[i].node_mem = ram_size - usedmem;
 183         }
 184
 185         numa_total = 0;
 186         for (i = 0; i < nb_numa_nodes; i++) {
 187             numa_total += numa_info[i].node_mem;
 188         }
 189         if (numa_total != ram_size) {
 190             error_report("total memory for NUMA nodes (%" PRIu64 ")"
 191                          " should equal RAM size (" RAM_ADDR_FMT ")",
 192                          numa_total, ram_size);
 193             exit(1);
 194         }
 195
 196         for (i = 0; i < nb_numa_nodes; i++) {
 197             if (!bitmap_empty(numa_info[i].node_cpu, MAX_CPUMASK_BITS)) {
 198                 break;
 199             }
 200         }
 201         /* assigning the VCPUs round-robin is easier to implement, guest OSes
 202          * must cope with this anyway, because there are BIOSes out there in
 203          * real machines which also use this scheme.
 204          */
 205         if (i == nb_numa_nodes) {
 206             for (i = 0; i < max_cpus; i++) {
 207                 set_bit(i, numa_info[i % nb_numa_nodes].node_cpu);
 208             }
 209         }
 210     }
 211 }
 212
 213 void set_numa_modes(void)
 214 {
 215     CPUState *cpu;
 216     int i;
 217
 218     CPU_FOREACH(cpu) {
 219         for (i = 0; i < nb_numa_nodes; i++) {
 220             if (test_bit(cpu->cpu_index, numa_info[i].node_cpu)) {
 221                 cpu->numa_node = i;
 222             }
 223         }
 224     }
 225 }
 226
 227 static void allocate_system_memory_nonnuma(MemoryRegion *mr, Object *owner,
 228                                            const char *name,
 229                                            uint64_t ram_size)
 230 {
 231     memory_region_init_ram(mr, owner, name, ram_size);
 232     vmstate_register_ram_global(mr);
 233 }
 234
 235 void memory_region_allocate_system_memory(MemoryRegion *mr, Object *owner,
 236                                           const char *name,
 237                                           uint64_t ram_size)
 238 {
 239     uint64_t addr = 0;
 240     int i;
 241
 242     if (nb_numa_nodes == 0 || !have_memdevs) {
 243         allocate_system_memory_nonnuma(mr, owner, name, ram_size);
 244         return;
 245     }
 246
 247     memory_region_init(mr, owner, name, ram_size);
 248     for (i = 0; i < MAX_NODES; i++) {
 249         Error *local_err = NULL;
 250         uint64_t size = numa_info[i].node_mem;
 251         HostMemoryBackend *backend = numa_info[i].node_memdev;
 252         if (!backend) {
 253             continue;
 254         }
 255         MemoryRegion *seg = host_memory_backend_get_memory(backend, &local_err);
 256         if (local_err) {
 257             qerror_report_err(local_err);
 258             exit(1);
 259         }
 260
 261         memory_region_add_subregion(mr, addr, seg);
 262         vmstate_register_ram_global(seg);
 263         addr += size;
 264     }
 265 }