arm64: topology: re-introduce numa mask check for scheduler MC selection

Commit 37c3ec2d810f ("arm64: topology: divorce MC scheduling domain from
core_siblings") selected the smallest of LLC, socket siblings, and NUMA
node siblings to ensure that the sched domain we build for the MC layer
isn't larger than the DIE above it or it's shrunk to the socket or NUMA
node if LLC exist acrosis NUMA node/chiplets.

Commit acd32e52e4e0 ("arm64: topology: Avoid checking numa mask for
scheduler MC selection") reverted the NUMA siblings checks since the
CPU topology masks weren't updated on hotplug at that time.

This patch re-introduces numa mask check as the CPU and NUMA topology
is now updated in hotplug paths. Effectively, this patch does the
partial revert of commit acd32e52e4e0.

Cc: Catalin Marinas <[email protected]>
Cc: Will Deacon <[email protected]>
Tested-by: Ganapatrao Kulkarni <[email protected]>
Tested-by: Hanjun Guo <[email protected]>
Signed-off-by: Sudeep Holla <[email protected]>
Signed-off-by: Will Deacon <[email protected]>

diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c
index 03b0ed2480aa042d2172536d42a8a313f834cacb..0825c4a856e33da0d794f403fd2a6464543467c3 100644 (file)
--- a/arch/arm64/kernel/topology.c
+++ b/arch/arm64/kernel/topology.c
@@ -215,8 +215,13 @@ EXPORT_SYMBOL_GPL(cpu_topology);
 
 const struct cpumask *cpu_coregroup_mask(int cpu)
 {
-       const cpumask_t *core_mask = &cpu_topology[cpu].core_sibling;
+       const cpumask_t *core_mask = cpumask_of_node(cpu_to_node(cpu));
 
+       /* Find the smaller of NUMA, core or LLC siblings */
+       if (cpumask_subset(&cpu_topology[cpu].core_sibling, core_mask)) {
+               /* not numa in package, lets use the package siblings */
+               core_mask = &cpu_topology[cpu].core_sibling;
+       }
        if (cpu_topology[cpu].llc_id != -1) {
                if (cpumask_subset(&cpu_topology[cpu].llc_sibling, core_mask))
                        core_mask = &cpu_topology[cpu].llc_sibling;