]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
genirq/affinity: Consider that CPUs on nodes can be unbalanced
authorRei Yamamoto <yamamoto.rei@jp.fujitsu.com>
Thu, 31 Mar 2022 00:33:09 +0000 (09:33 +0900)
committerThomas Gleixner <tglx@linutronix.de>
Mon, 11 Apr 2022 07:58:03 +0000 (09:58 +0200)
If CPUs on a node are offline at boot time, the number of nodes is
different when building affinity masks for present cpus and when building
affinity masks for possible cpus. This causes the following problem:

In the case that the number of vectors is less than the number of nodes
there are cases where bits of masks for present cpus are overwritten when
building masks for possible cpus.

Fix this by excluding CPUs, which are not part of the current build mask
(present/possible).

[ tglx: Massaged changelog and added comment ]

Fixes: b82592199032 ("genirq/affinity: Spread IRQs to all available NUMA nodes")
Signed-off-by: Rei Yamamoto <yamamoto.rei@jp.fujitsu.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20220331003309.10891-1-yamamoto.rei@jp.fujitsu.com
kernel/irq/affinity.c

index f7ff8919dc9bbb1ab2a907c1b28a5605dc894251..fdf170404650f7e96559cc2df2d783618e5f93bc 100644 (file)
@@ -269,8 +269,9 @@ static int __irq_build_affinity_masks(unsigned int startvec,
         */
        if (numvecs <= nodes) {
                for_each_node_mask(n, nodemsk) {
-                       cpumask_or(&masks[curvec].mask, &masks[curvec].mask,
-                                  node_to_cpumask[n]);
+                       /* Ensure that only CPUs which are in both masks are set */
+                       cpumask_and(nmsk, cpu_mask, node_to_cpumask[n]);
+                       cpumask_or(&masks[curvec].mask, &masks[curvec].mask, nmsk);
                        if (++curvec == last_affv)
                                curvec = firstvec;
                }