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