On Tue, Jun 09, 2020 at 09:45:01PM -0400, Qian Cai wrote:
> When NUMA=n and nr_node_ids=2, in apply_wqattrs_prepare(), it has,
>
> for_each_node(node) {
> if (wq_calc_node_cpumask(...
>
> where it will trigger a booting warning,
>
> WARNING: workqueue cpumask: online intersect > possible intersect
>
> because it found 2 nodes and wq_numa_possible_cpumask[1] is an empty
> cpumask. NUMA=y has no such problem because node_possible_map will be
> initialized properly containing only node 0. Fix it by setting
> NODES_SHIFT=0 when NUMA=n.
>
> Fixes: 701dc81e7412 ("s390/mm: remove fake numa support")
> Signed-off-by: Qian Cai <cai@xxxxxx>
> ---
> arch/s390/Kconfig | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
Thanks! However I committed a different solution. Hope you don't mind:
>From dd3f1f08f2317768b35b2df3ff8285185df7e195 Mon Sep 17 00:00:00 2001
From: Heiko Carstens <heiko.carstens@xxxxxxxxxx>
Date: Wed, 10 Jun 2020 10:36:05 +0200
Subject: [PATCH] s390/numa: let NODES_SHIFT depend on NEED_MULTIPLE_NODES
Qian Cai reported:
---
When NUMA=n and nr_node_ids=2, in apply_wqattrs_prepare(), it has,
for_each_node(node) {
if (wq_calc_node_cpumask(...
where it will trigger a booting warning,
WARNING: workqueue cpumask: online intersect > possible intersect
because it found 2 nodes and wq_numa_possible_cpumask[1] is an empty
cpumask.
---
Let NODES_SHIFT depend on NEED_MULTIPLE_NODES like it is done
on other architectures in order to fix this.
Fixes: 701dc81e7412 ("s390/mm: remove fake numa support")
Reported-by: Qian Cai <cai@xxxxxx>
Signed-off-by: Heiko Carstens <heiko.carstens@xxxxxxxxxx>
---
arch/s390/Kconfig | 1 +
1 file changed, 1 insertion(+)
diff --git a/arch/s390/Kconfig b/arch/s390/Kconfig
index 2167bce993ff..ae01be202204 100644
--- a/arch/s390/Kconfig
+++ b/arch/s390/Kconfig
@@ -462,6 +462,7 @@ config NUMA
config NODES_SHIFT
int
+ depends on NEED_MULTIPLE_NODES
default "1"
config SCHED_SMT
--
2.17.1
