powerpc/numa: ensure per-cpu NUMA mappings are correct on topology update
authorNishanth Aravamudan <nacc@linux.vnet.ibm.com>
Sat, 18 Oct 2014 00:50:40 +0000 (17:50 -0700)
committerMichael Ellerman <mpe@ellerman.id.au>
Tue, 28 Oct 2014 22:41:22 +0000 (09:41 +1100)
We received a report of warning in kernel/sched/core.c where the sched
group was NULL on an LPAR after a topology update. This seems to occur
because after the topology update has moved the CPUs, cpu_to_node is
returning the old value still, which ends up breaking the consistency of
the NUMA topology in the per-cpu maps. Ensure that we update the per-cpu
fields when we re-map CPUs.

Signed-off-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/mm/numa.c

index ef45abf7319eb19533f45614dbec40efc18b68a1..b9d1dfdbe5bbaa5bdaad7446ec945aa909f2befd 100644 (file)
@@ -1509,11 +1509,14 @@ static int update_cpu_topology(void *data)
        cpu = smp_processor_id();
 
        for (update = data; update; update = update->next) {
+               int new_nid = update->new_nid;
                if (cpu != update->cpu)
                        continue;
 
                unmap_cpu_from_node(cpu);
-               map_cpu_to_node(cpu, update->new_nid);
+               map_cpu_to_node(cpu, new_nid);
+               set_cpu_numa_node(cpu, new_nid);
+               set_cpu_numa_mem(cpu, local_memory_node(new_nid));
                vdso_getcpu_init();
        }