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
  37 QemuOptsList qemu_numa_opts = {
  38     .name = "numa",
  39     .implied_opt_name = "type",
  40     .head = QTAILQ_HEAD_INITIALIZER(qemu_numa_opts.head),
  41     .desc = { { 0 } } /* validated with OptsVisitor */
  42 };
  43
  44 static void numa_node_parse(NumaNodeOptions *node, QemuOpts *opts, Error **errp)
  45 {
  46     uint16_t nodenr;
  47     uint16List *cpus = NULL;
  48
  49     if (node->has_nodeid) {
  50         nodenr = node->nodeid;
  51     } else {
  52         nodenr = nb_numa_nodes;
  53     }
  54
  55     if (nodenr >= MAX_NODES) {
  56         error_setg(errp, "Max number of NUMA nodes reached: %"
  57                    PRIu16 "\n", nodenr);
  58         return;
  59     }
  60
  61     for (cpus = node->cpus; cpus; cpus = cpus->next) {
  62         if (cpus->value > MAX_CPUMASK_BITS) {
  63             error_setg(errp, "CPU number %" PRIu16 " is bigger than %d",
  64                        cpus->value, MAX_CPUMASK_BITS);
  65             return;
  66         }
  67         bitmap_set(numa_info[nodenr].node_cpu, cpus->value, 1);
  68     }
  69
  70     if (node->has_mem) {
  71         uint64_t mem_size = node->mem;
  72         const char *mem_str = qemu_opt_get(opts, "mem");
  73         /* Fix up legacy suffix-less format */
  74         if (g_ascii_isdigit(mem_str[strlen(mem_str) - 1])) {
  75             mem_size <<= 20;
  76         }
  77         numa_info[nodenr].node_mem = mem_size;
  78     }
  79 }
  80
  81 int numa_init_func(QemuOpts *opts, void *opaque)
  82 {
  83     NumaOptions *object = NULL;
  84     Error *err = NULL;
  85
  86     {
  87         OptsVisitor *ov = opts_visitor_new(opts);
  88         visit_type_NumaOptions(opts_get_visitor(ov), &object, NULL, &err);
  89         opts_visitor_cleanup(ov);
  90     }
  91
  92     if (err) {
  93         goto error;
  94     }
  95
  96     switch (object->kind) {
  97     case NUMA_OPTIONS_KIND_NODE:
  98         numa_node_parse(object->node, opts, &err);
  99         if (err) {
 100             goto error;
 101         }
 102         nb_numa_nodes++;
 103         break;
 104     default:
 105         abort();
 106     }
 107
 108     return 0;
 109
 110 error:
 111     qerror_report_err(err);
 112     error_free(err);
 113
 114     if (object) {
 115         QapiDeallocVisitor *dv = qapi_dealloc_visitor_new();
 116         visit_type_NumaOptions(qapi_dealloc_get_visitor(dv),
 117                                &object, NULL, NULL);
 118         qapi_dealloc_visitor_cleanup(dv);
 119     }
 120
 121     return -1;
 122 }
 123
 124 void set_numa_nodes(void)
 125 {
 126     if (nb_numa_nodes > 0) {
 127         uint64_t numa_total;
 128         int i;
 129
 130         if (nb_numa_nodes > MAX_NODES) {
 131             nb_numa_nodes = MAX_NODES;
 132         }
 133
 134         /* If no memory size if given for any node, assume the default case
 135          * and distribute the available memory equally across all nodes
 136          */
 137         for (i = 0; i < nb_numa_nodes; i++) {
 138             if (numa_info[i].node_mem != 0) {
 139                 break;
 140             }
 141         }
 142         if (i == nb_numa_nodes) {
 143             uint64_t usedmem = 0;
 144
 145             /* On Linux, the each node's border has to be 8MB aligned,
 146              * the final node gets the rest.
 147              */
 148             for (i = 0; i < nb_numa_nodes - 1; i++) {
 149                 numa_info[i].node_mem = (ram_size / nb_numa_nodes) &
 150                                         ~((1 << 23UL) - 1);
 151                 usedmem += numa_info[i].node_mem;
 152             }
 153             numa_info[i].node_mem = ram_size - usedmem;
 154         }
 155
 156         numa_total = 0;
 157         for (i = 0; i < nb_numa_nodes; i++) {
 158             numa_total += numa_info[i].node_mem;
 159         }
 160         if (numa_total != ram_size) {
 161             error_report("total memory for NUMA nodes (%" PRIu64 ")"
 162                          " should equal RAM size (" RAM_ADDR_FMT ")",
 163                          numa_total, ram_size);
 164             exit(1);
 165         }
 166
 167         for (i = 0; i < nb_numa_nodes; i++) {
 168             if (!bitmap_empty(numa_info[i].node_cpu, MAX_CPUMASK_BITS)) {
 169                 break;
 170             }
 171         }
 172         /* assigning the VCPUs round-robin is easier to implement, guest OSes
 173          * must cope with this anyway, because there are BIOSes out there in
 174          * real machines which also use this scheme.
 175          */
 176         if (i == nb_numa_nodes) {
 177             for (i = 0; i < max_cpus; i++) {
 178                 set_bit(i, numa_info[i % nb_numa_nodes].node_cpu);
 179             }
 180         }
 181     }
 182 }
 183
 184 void set_numa_modes(void)
 185 {
 186     CPUState *cpu;
 187     int i;
 188
 189     CPU_FOREACH(cpu) {
 190         for (i = 0; i < nb_numa_nodes; i++) {
 191             if (test_bit(cpu->cpu_index, numa_info[i].node_cpu)) {
 192                 cpu->numa_node = i;
 193             }
 194         }
 195     }
 196 }
 197
 198 void memory_region_allocate_system_memory(MemoryRegion *mr, Object *owner,
 199                                           const char *name,
 200                                           uint64_t ram_size)
 201 {
 202     memory_region_init_ram(mr, owner, name, ram_size);
 203     vmstate_register_ram_global(mr);
 204 }