summaryrefslogtreecommitdiff
path: root/arch/x86/mm/numa.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/mm/numa.c')
-rw-r--r--arch/x86/mm/numa.c212
1 files changed, 209 insertions, 3 deletions
diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index ebf6d7887a38..9559d360fde7 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -26,12 +26,50 @@ static __init int numa_setup(char *opt)
early_param("numa", numa_setup);
/*
- * Which logical CPUs are on which nodes
+ * apicid, cpu, node mappings
*/
+s16 __apicid_to_node[MAX_LOCAL_APIC] __cpuinitdata = {
+ [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
+};
+
cpumask_var_t node_to_cpumask_map[MAX_NUMNODES];
EXPORT_SYMBOL(node_to_cpumask_map);
/*
+ * Map cpu index to node index
+ */
+DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
+EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
+
+void __cpuinit numa_set_node(int cpu, int node)
+{
+ int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
+
+ /* early setting, no percpu area yet */
+ if (cpu_to_node_map) {
+ cpu_to_node_map[cpu] = node;
+ return;
+ }
+
+#ifdef CONFIG_DEBUG_PER_CPU_MAPS
+ if (cpu >= nr_cpu_ids || !cpu_possible(cpu)) {
+ printk(KERN_ERR "numa_set_node: invalid cpu# (%d)\n", cpu);
+ dump_stack();
+ return;
+ }
+#endif
+ per_cpu(x86_cpu_to_node_map, cpu) = node;
+
+ if (node != NUMA_NO_NODE)
+ set_cpu_numa_node(cpu, node);
+}
+
+void __cpuinit numa_clear_node(int cpu)
+{
+ numa_set_node(cpu, NUMA_NO_NODE);
+}
+
+/*
* Allocate node_to_cpumask_map based on number of available nodes
* Requires node_possible_map to be valid.
*
@@ -57,7 +95,174 @@ void __init setup_node_to_cpumask_map(void)
pr_debug("Node to cpumask map for %d nodes\n", nr_node_ids);
}
-#ifdef CONFIG_DEBUG_PER_CPU_MAPS
+/*
+ * There are unfortunately some poorly designed mainboards around that
+ * only connect memory to a single CPU. This breaks the 1:1 cpu->node
+ * mapping. To avoid this fill in the mapping for all possible CPUs,
+ * as the number of CPUs is not known yet. We round robin the existing
+ * nodes.
+ */
+void __init numa_init_array(void)
+{
+ int rr, i;
+
+ rr = first_node(node_online_map);
+ for (i = 0; i < nr_cpu_ids; i++) {
+ if (early_cpu_to_node(i) != NUMA_NO_NODE)
+ continue;
+ numa_set_node(i, rr);
+ rr = next_node(rr, node_online_map);
+ if (rr == MAX_NUMNODES)
+ rr = first_node(node_online_map);
+ }
+}
+
+static __init int find_near_online_node(int node)
+{
+ int n, val;
+ int min_val = INT_MAX;
+ int best_node = -1;
+
+ for_each_online_node(n) {
+ val = node_distance(node, n);
+
+ if (val < min_val) {
+ min_val = val;
+ best_node = n;
+ }
+ }
+
+ return best_node;
+}
+
+/*
+ * Setup early cpu_to_node.
+ *
+ * Populate cpu_to_node[] only if x86_cpu_to_apicid[],
+ * and apicid_to_node[] tables have valid entries for a CPU.
+ * This means we skip cpu_to_node[] initialisation for NUMA
+ * emulation and faking node case (when running a kernel compiled
+ * for NUMA on a non NUMA box), which is OK as cpu_to_node[]
+ * is already initialized in a round robin manner at numa_init_array,
+ * prior to this call, and this initialization is good enough
+ * for the fake NUMA cases.
+ *
+ * Called before the per_cpu areas are setup.
+ */
+void __init init_cpu_to_node(void)
+{
+ int cpu;
+ u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
+
+ BUG_ON(cpu_to_apicid == NULL);
+
+ for_each_possible_cpu(cpu) {
+ int node = numa_cpu_node(cpu);
+
+ if (node == NUMA_NO_NODE)
+ continue;
+ if (!node_online(node))
+ node = find_near_online_node(node);
+ numa_set_node(cpu, node);
+ }
+}
+
+#ifndef CONFIG_DEBUG_PER_CPU_MAPS
+
+# ifndef CONFIG_NUMA_EMU
+void __cpuinit numa_add_cpu(int cpu)
+{
+ cpumask_set_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
+}
+
+void __cpuinit numa_remove_cpu(int cpu)
+{
+ cpumask_clear_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
+}
+# endif /* !CONFIG_NUMA_EMU */
+
+#else /* !CONFIG_DEBUG_PER_CPU_MAPS */
+
+int __cpu_to_node(int cpu)
+{
+ if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
+ printk(KERN_WARNING
+ "cpu_to_node(%d): usage too early!\n", cpu);
+ dump_stack();
+ return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
+ }
+ return per_cpu(x86_cpu_to_node_map, cpu);
+}
+EXPORT_SYMBOL(__cpu_to_node);
+
+/*
+ * Same function as cpu_to_node() but used if called before the
+ * per_cpu areas are setup.
+ */
+int early_cpu_to_node(int cpu)
+{
+ if (early_per_cpu_ptr(x86_cpu_to_node_map))
+ return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
+
+ if (!cpu_possible(cpu)) {
+ printk(KERN_WARNING
+ "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
+ dump_stack();
+ return NUMA_NO_NODE;
+ }
+ return per_cpu(x86_cpu_to_node_map, cpu);
+}
+
+struct cpumask __cpuinit *debug_cpumask_set_cpu(int cpu, int enable)
+{
+ int node = early_cpu_to_node(cpu);
+ struct cpumask *mask;
+ char buf[64];
+
+ if (node == NUMA_NO_NODE) {
+ /* early_cpu_to_node() already emits a warning and trace */
+ return NULL;
+ }
+ mask = node_to_cpumask_map[node];
+ if (!mask) {
+ pr_err("node_to_cpumask_map[%i] NULL\n", node);
+ dump_stack();
+ return NULL;
+ }
+
+ cpulist_scnprintf(buf, sizeof(buf), mask);
+ printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
+ enable ? "numa_add_cpu" : "numa_remove_cpu",
+ cpu, node, buf);
+ return mask;
+}
+
+# ifndef CONFIG_NUMA_EMU
+static void __cpuinit numa_set_cpumask(int cpu, int enable)
+{
+ struct cpumask *mask;
+
+ mask = debug_cpumask_set_cpu(cpu, enable);
+ if (!mask)
+ return;
+
+ if (enable)
+ cpumask_set_cpu(cpu, mask);
+ else
+ cpumask_clear_cpu(cpu, mask);
+}
+
+void __cpuinit numa_add_cpu(int cpu)
+{
+ numa_set_cpumask(cpu, 1);
+}
+
+void __cpuinit numa_remove_cpu(int cpu)
+{
+ numa_set_cpumask(cpu, 0);
+}
+# endif /* !CONFIG_NUMA_EMU */
+
/*
* Returns a pointer to the bitmask of CPUs on Node 'node'.
*/
@@ -80,4 +285,5 @@ const struct cpumask *cpumask_of_node(int node)
return node_to_cpumask_map[node];
}
EXPORT_SYMBOL(cpumask_of_node);
-#endif
+
+#endif /* !CONFIG_DEBUG_PER_CPU_MAPS */