Hi Srikar,
Thanks for figuring this out.
Srikar Dronamraju <sri...@linux.vnet.ibm.com> writes:
>
> Some of the per-CPU masks use cpu_cpu_mask as a filter to limit the search
> for related CPUs. On a dlpar add of a CPU, update cpu_cpu_mask before
> updating the per-CPU masks. This will ensure the cpu_cpu_mask is updated
> correctly before its used in setting the masks. Setting the numa_node will
> ensure that when cpu_cpu_mask() gets called, the correct node number is
> used. This code movement helped fix the above call trace.
>
>
> diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
> index 5a4d59a1070d..1a99d75679a8 100644
> --- a/arch/powerpc/kernel/smp.c
> +++ b/arch/powerpc/kernel/smp.c
> @@ -1521,6 +1521,9 @@ void start_secondary(void *unused)
>
> vdso_getcpu_init();
> #endif
> + set_numa_node(numa_cpu_lookup_table[cpu]);
> + set_numa_mem(local_memory_node(numa_cpu_lookup_table[cpu]));
> +
> /* Update topology CPU masks */
> add_cpu_to_masks(cpu);
>
> @@ -1539,9 +1542,6 @@ void start_secondary(void *unused)
> shared_caches = true;
> }
>
> - set_numa_node(numa_cpu_lookup_table[cpu]);
> - set_numa_mem(local_memory_node(numa_cpu_lookup_table[cpu]));
> -
Regardless of your change: at boot time, this set of calls to
set_numa_node() and set_numa_mem() is redundant, right? Because
smp_prepare_cpus() has:
for_each_possible_cpu(cpu) {
...
if (cpu_present(cpu)) {
set_cpu_numa_node(cpu, numa_cpu_lookup_table[cpu]);
set_cpu_numa_mem(cpu,
local_memory_node(numa_cpu_lookup_table[cpu]));
}
I would rather that, when onlining a CPU that happens to have been
dynamically added after boot, we enter start_secondary() with conditions
equivalent to those at boot time. Or as close to that as is practical.
So I'd suggest that pseries_add_processor() be made to update
these things when the CPUs are marked present, before onlining them.
