arch/arm64/kernel/topology.c

   1 /*
   2  * arch/arm64/kernel/topology.c
   3  *
   4  * Copyright (C) 2011,2013,2014 Linaro Limited.
   5  *
   6  * Based on the arm32 version written by Vincent Guittot in turn based on
   7  * arch/sh/kernel/topology.c
   8  *
   9  * This file is subject to the terms and conditions of the GNU General Public
  10  * License.  See the file "COPYING" in the main directory of this archive
  11  * for more details.
  12  */
  13
  14 #include <linux/cpu.h>
  15 #include <linux/cpumask.h>
  16 #include <linux/init.h>
  17 #include <linux/percpu.h>
  18 #include <linux/node.h>
  19 #include <linux/nodemask.h>
  20 #include <linux/of.h>
  21 #include <linux/sched.h>
  22
  23 #include <asm/cputype.h>
  24 #include <asm/topology.h>
  25
  26 static int __init get_cpu_for_node(struct device_node *node)
  27 {
  28         struct device_node *cpu_node;
  29         int cpu;
  30
  31         cpu_node = of_parse_phandle(node, "cpu", 0);
  32         if (!cpu_node)
  33                 return -1;
  34
  35         for_each_possible_cpu(cpu) {
  36                 if (of_get_cpu_node(cpu, NULL) == cpu_node) {
  37                         of_node_put(cpu_node);
  38                         return cpu;
  39                 }
  40         }
  41
  42         pr_crit("Unable to find CPU node for %s\n", cpu_node->full_name);
  43
  44         of_node_put(cpu_node);
  45         return -1;
  46 }
  47
  48 static int __init parse_core(struct device_node *core, int cluster_id,
  49                              int core_id)
  50 {
  51         char name[10];
  52         bool leaf = true;
  53         int i = 0;
  54         int cpu;
  55         struct device_node *t;
  56
  57         do {
  58                 snprintf(name, sizeof(name), "thread%d", i);
  59                 t = of_get_child_by_name(core, name);
  60                 if (t) {
  61                         leaf = false;
  62                         cpu = get_cpu_for_node(t);
  63                         if (cpu >= 0) {
  64                                 cpu_topology[cpu].cluster_id = cluster_id;
  65                                 cpu_topology[cpu].core_id = core_id;
  66                                 cpu_topology[cpu].thread_id = i;
  67                         } else {
  68                                 pr_err("%s: Can't get CPU for thread\n",
  69                                        t->full_name);
  70                                 of_node_put(t);
  71                                 return -EINVAL;
  72                         }
  73                         of_node_put(t);
  74                 }
  75                 i++;
  76         } while (t);
  77
  78         cpu = get_cpu_for_node(core);
  79         if (cpu >= 0) {
  80                 if (!leaf) {
  81                         pr_err("%s: Core has both threads and CPU\n",
  82                                core->full_name);
  83                         return -EINVAL;
  84                 }
  85
  86                 cpu_topology[cpu].cluster_id = cluster_id;
  87                 cpu_topology[cpu].core_id = core_id;
  88         } else if (leaf) {
  89                 pr_err("%s: Can't get CPU for leaf core\n", core->full_name);
  90                 return -EINVAL;
  91         }
  92
  93         return 0;
  94 }
  95
  96 static int __init parse_cluster(struct device_node *cluster, int depth)
  97 {
  98         char name[10];
  99         bool leaf = true;
 100         bool has_cores = false;
 101         struct device_node *c;
 102         static int cluster_id __initdata;
 103         int core_id = 0;
 104         int i, ret;
 105
 106         /*
 107          * First check for child clusters; we currently ignore any
 108          * information about the nesting of clusters and present the
 109          * scheduler with a flat list of them.
 110          */
 111         i = 0;
 112         do {
 113                 snprintf(name, sizeof(name), "cluster%d", i);
 114                 c = of_get_child_by_name(cluster, name);
 115                 if (c) {
 116                         leaf = false;
 117                         ret = parse_cluster(c, depth + 1);
 118                         of_node_put(c);
 119                         if (ret != 0)
 120                                 return ret;
 121                 }
 122                 i++;
 123         } while (c);
 124
 125         /* Now check for cores */
 126         i = 0;
 127         do {
 128                 snprintf(name, sizeof(name), "core%d", i);
 129                 c = of_get_child_by_name(cluster, name);
 130                 if (c) {
 131                         has_cores = true;
 132
 133                         if (depth == 0) {
 134                                 pr_err("%s: cpu-map children should be clusters\n",
 135                                        c->full_name);
 136                                 of_node_put(c);
 137                                 return -EINVAL;
 138                         }
 139
 140                         if (leaf) {
 141                                 ret = parse_core(c, cluster_id, core_id++);
 142                         } else {
 143                                 pr_err("%s: Non-leaf cluster with core %s\n",
 144                                        cluster->full_name, name);
 145                                 ret = -EINVAL;
 146                         }
 147
 148                         of_node_put(c);
 149                         if (ret != 0)
 150                                 return ret;
 151                 }
 152                 i++;
 153         } while (c);
 154
 155         if (leaf && !has_cores)
 156                 pr_warn("%s: empty cluster\n", cluster->full_name);
 157
 158         if (leaf)
 159                 cluster_id++;
 160
 161         return 0;
 162 }
 163
 164 static int __init parse_dt_topology(void)
 165 {
 166         struct device_node *cn, *map;
 167         int ret = 0;
 168         int cpu;
 169
 170         cn = of_find_node_by_path("/cpus");
 171         if (!cn) {
 172                 pr_err("No CPU information found in DT\n");
 173                 return 0;
 174         }
 175
 176         /*
 177          * When topology is provided cpu-map is essentially a root
 178          * cluster with restricted subnodes.
 179          */
 180         map = of_get_child_by_name(cn, "cpu-map");
 181         if (!map)
 182                 goto out;
 183
 184         ret = parse_cluster(map, 0);
 185         if (ret != 0)
 186                 goto out_map;
 187
 188         /*
 189          * Check that all cores are in the topology; the SMP code will
 190          * only mark cores described in the DT as possible.
 191          */
 192         for_each_possible_cpu(cpu)
 193                 if (cpu_topology[cpu].cluster_id == -1)
 194                         ret = -EINVAL;
 195
 196 out_map:
 197         of_node_put(map);
 198 out:
 199         of_node_put(cn);
 200         return ret;
 201 }
 202
 203 /*
 204  * cpu topology table
 205  */
 206 struct cpu_topology cpu_topology[NR_CPUS];
 207 EXPORT_SYMBOL_GPL(cpu_topology);
 208
 209 const struct cpumask *cpu_coregroup_mask(int cpu)
 210 {
 211         return &cpu_topology[cpu].core_sibling;
 212 }
 213
 214 static void update_siblings_masks(unsigned int cpuid)
 215 {
 216         struct cpu_topology *cpu_topo, *cpuid_topo = &cpu_topology[cpuid];
 217         int cpu;
 218
 219         /* update core and thread sibling masks */
 220         for_each_possible_cpu(cpu) {
 221                 cpu_topo = &cpu_topology[cpu];
 222
 223                 if (cpuid_topo->cluster_id != cpu_topo->cluster_id)
 224                         continue;
 225
 226                 cpumask_set_cpu(cpuid, &cpu_topo->core_sibling);
 227                 if (cpu != cpuid)
 228                         cpumask_set_cpu(cpu, &cpuid_topo->core_sibling);
 229
 230                 if (cpuid_topo->core_id != cpu_topo->core_id)
 231                         continue;
 232
 233                 cpumask_set_cpu(cpuid, &cpu_topo->thread_sibling);
 234                 if (cpu != cpuid)
 235                         cpumask_set_cpu(cpu, &cpuid_topo->thread_sibling);
 236         }
 237 }
 238
 239 void store_cpu_topology(unsigned int cpuid)
 240 {
 241         struct cpu_topology *cpuid_topo = &cpu_topology[cpuid];
 242         u64 mpidr;
 243
 244         if (cpuid_topo->cluster_id != -1)
 245                 goto topology_populated;
 246
 247         mpidr = read_cpuid_mpidr();
 248
 249         /* Uniprocessor systems can rely on default topology values */
 250         if (mpidr & MPIDR_UP_BITMASK)
 251                 return;
 252
 253         /* Create cpu topology mapping based on MPIDR. */
 254         if (mpidr & MPIDR_MT_BITMASK) {
 255                 /* Multiprocessor system : Multi-threads per core */
 256                 cpuid_topo->thread_id  = MPIDR_AFFINITY_LEVEL(mpidr, 0);
 257                 cpuid_topo->core_id    = MPIDR_AFFINITY_LEVEL(mpidr, 1);
 258                 cpuid_topo->cluster_id = MPIDR_AFFINITY_LEVEL(mpidr, 2) |
 259                                          MPIDR_AFFINITY_LEVEL(mpidr, 3) << 8;
 260         } else {
 261                 /* Multiprocessor system : Single-thread per core */
 262                 cpuid_topo->thread_id  = -1;
 263                 cpuid_topo->core_id    = MPIDR_AFFINITY_LEVEL(mpidr, 0);
 264                 cpuid_topo->cluster_id = MPIDR_AFFINITY_LEVEL(mpidr, 1) |
 265                                          MPIDR_AFFINITY_LEVEL(mpidr, 2) << 8 |
 266                                          MPIDR_AFFINITY_LEVEL(mpidr, 3) << 16;
 267         }
 268
 269         pr_debug("CPU%u: cluster %d core %d thread %d mpidr %#016llx\n",
 270                  cpuid, cpuid_topo->cluster_id, cpuid_topo->core_id,
 271                  cpuid_topo->thread_id, mpidr);
 272
 273 topology_populated:
 274         update_siblings_masks(cpuid);
 275 }
 276
 277 static void __init reset_cpu_topology(void)
 278 {
 279         unsigned int cpu;
 280
 281         for_each_possible_cpu(cpu) {
 282                 struct cpu_topology *cpu_topo = &cpu_topology[cpu];
 283
 284                 cpu_topo->thread_id = -1;
 285                 cpu_topo->core_id = 0;
 286                 cpu_topo->cluster_id = -1;
 287
 288                 cpumask_clear(&cpu_topo->core_sibling);
 289                 cpumask_set_cpu(cpu, &cpu_topo->core_sibling);
 290                 cpumask_clear(&cpu_topo->thread_sibling);
 291                 cpumask_set_cpu(cpu, &cpu_topo->thread_sibling);
 292         }
 293 }
 294
 295 void __init init_cpu_topology(void)
 296 {
 297         reset_cpu_topology();
 298
 299         /*
 300          * Discard anything that was parsed if we hit an error so we
 301          * don't use partial information.
 302          */
 303         if (of_have_populated_dt() && parse_dt_topology())
 304                 reset_cpu_topology();
 305 }