Merge branch 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[platform/adaptation/renesas_rcar/renesas_kernel.git] / arch / x86 / kernel / smpboot.c
index 3acaf51..433529e 100644 (file)
@@ -299,59 +299,90 @@ void __cpuinit smp_store_cpu_info(int id)
                identify_secondary_cpu(c);
 }
 
-static void __cpuinit link_thread_siblings(int cpu1, int cpu2)
+static bool __cpuinit
+topology_sane(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o, const char *name)
 {
-       cpumask_set_cpu(cpu1, cpu_sibling_mask(cpu2));
-       cpumask_set_cpu(cpu2, cpu_sibling_mask(cpu1));
-       cpumask_set_cpu(cpu1, cpu_core_mask(cpu2));
-       cpumask_set_cpu(cpu2, cpu_core_mask(cpu1));
-       cpumask_set_cpu(cpu1, cpu_llc_shared_mask(cpu2));
-       cpumask_set_cpu(cpu2, cpu_llc_shared_mask(cpu1));
+       int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
+
+       return !WARN_ONCE(cpu_to_node(cpu1) != cpu_to_node(cpu2),
+               "sched: CPU #%d's %s-sibling CPU #%d is not on the same node! "
+               "[node: %d != %d]. Ignoring dependency.\n",
+               cpu1, name, cpu2, cpu_to_node(cpu1), cpu_to_node(cpu2));
 }
 
+#define link_mask(_m, c1, c2)                                          \
+do {                                                                   \
+       cpumask_set_cpu((c1), cpu_##_m##_mask(c2));                     \
+       cpumask_set_cpu((c2), cpu_##_m##_mask(c1));                     \
+} while (0)
+
+static bool __cpuinit match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
+{
+       if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
+               int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
+
+               if (c->phys_proc_id == o->phys_proc_id &&
+                   per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2) &&
+                   c->compute_unit_id == o->compute_unit_id)
+                       return topology_sane(c, o, "smt");
+
+       } else if (c->phys_proc_id == o->phys_proc_id &&
+                  c->cpu_core_id == o->cpu_core_id) {
+               return topology_sane(c, o, "smt");
+       }
+
+       return false;
+}
+
+static bool __cpuinit match_llc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
+{
+       int cpu1 = c->cpu_index, cpu2 = o->cpu_index;
+
+       if (per_cpu(cpu_llc_id, cpu1) != BAD_APICID &&
+           per_cpu(cpu_llc_id, cpu1) == per_cpu(cpu_llc_id, cpu2))
+               return topology_sane(c, o, "llc");
+
+       return false;
+}
+
+static bool __cpuinit match_mc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
+{
+       if (c->phys_proc_id == o->phys_proc_id)
+               return topology_sane(c, o, "mc");
+
+       return false;
+}
 
 void __cpuinit set_cpu_sibling_map(int cpu)
 {
-       int i;
+       bool has_mc = boot_cpu_data.x86_max_cores > 1;
+       bool has_smt = smp_num_siblings > 1;
        struct cpuinfo_x86 *c = &cpu_data(cpu);
+       struct cpuinfo_x86 *o;
+       int i;
 
        cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
 
-       if (smp_num_siblings > 1) {
-               for_each_cpu(i, cpu_sibling_setup_mask) {
-                       struct cpuinfo_x86 *o = &cpu_data(i);
-
-                       if (cpu_has(c, X86_FEATURE_TOPOEXT)) {
-                               if (c->phys_proc_id == o->phys_proc_id &&
-                                   per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i) &&
-                                   c->compute_unit_id == o->compute_unit_id)
-                                       link_thread_siblings(cpu, i);
-                       } else if (c->phys_proc_id == o->phys_proc_id &&
-                                  c->cpu_core_id == o->cpu_core_id) {
-                               link_thread_siblings(cpu, i);
-                       }
-               }
-       } else {
+       if (!has_smt && !has_mc) {
                cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
-       }
-
-       cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
-
-       if (__this_cpu_read(cpu_info.x86_max_cores) == 1) {
-               cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
+               cpumask_set_cpu(cpu, cpu_llc_shared_mask(cpu));
+               cpumask_set_cpu(cpu, cpu_core_mask(cpu));
                c->booted_cores = 1;
                return;
        }
 
        for_each_cpu(i, cpu_sibling_setup_mask) {
-               if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
-                   per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
-                       cpumask_set_cpu(i, cpu_llc_shared_mask(cpu));
-                       cpumask_set_cpu(cpu, cpu_llc_shared_mask(i));
-               }
-               if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
-                       cpumask_set_cpu(i, cpu_core_mask(cpu));
-                       cpumask_set_cpu(cpu, cpu_core_mask(i));
+               o = &cpu_data(i);
+
+               if ((i == cpu) || (has_smt && match_smt(c, o)))
+                       link_mask(sibling, cpu, i);
+
+               if ((i == cpu) || (has_mc && match_llc(c, o)))
+                       link_mask(llc_shared, cpu, i);
+
+               if ((i == cpu) || (has_mc && match_mc(c, o))) {
+                       link_mask(core, cpu, i);
+
                        /*
                         *  Does this new cpu bringup a new core?
                         */
@@ -382,8 +413,7 @@ const struct cpumask *cpu_coregroup_mask(int cpu)
         * For perf, we return last level cache shared map.
         * And for power savings, we return cpu_core_map
         */
-       if ((sched_mc_power_savings || sched_smt_power_savings) &&
-           !(cpu_has(c, X86_FEATURE_AMD_DCM)))
+       if (!(cpu_has(c, X86_FEATURE_AMD_DCM)))
                return cpu_core_mask(cpu);
        else
                return cpu_llc_shared_mask(cpu);