[4/4] powerpc/numa: Set fallback nodes for offline nodes

Commit Message

Srikar Dronamraju March 17, 2020, 1:17 p.m. UTC

Currently fallback nodes for offline nodes aren't set. Hence by default
node 0 ends up being the default node. However node 0 might be offline.

Fix this by explicitly setting fallback node. Ensure first_memory_node
is set before kernel does explicit setting of fallback node.

Cc: Andrew Morton <[e-mail]>
Cc: [e-mail]
Cc: Mel Gorman <[e-mail]>
Cc: Michael Ellerman <[e-mail]>
Cc: Sachin Sant <[e-mail]>
Cc: Michal Hocko <[e-mail]>
Cc: Christopher Lameter <[e-mail]>
Cc: [e-mail]
Cc: Joonsoo Kim <[e-mail]>
Cc: Kirill Tkhai <[e-mail]>
Cc: Vlastimil Babka <[e-mail]>
Cc: Srikar Dronamraju <[e-mail]>
Cc: Bharata B Rao <[e-mail]>

Reported-by: Sachin Sant <[e-mail]>
Tested-by: Sachin Sant <[e-mail]>
Signed-off-by: Srikar Dronamraju <[e-mail]>
---
 arch/powerpc/mm/numa.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

Comments

Bharata B Rao March 17, 2020, 2:22 p.m. UTC | #1

This patchset can also fix a related problem that I reported earlier at
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-March/206076.html
with an additional change, suggested by Srikar as shown below:

On Tue, Mar 17, 2020 at 06:47:53PM +0530, Srikar Dronamraju wrote:
> Currently fallback nodes for offline nodes aren't set. Hence by default
> node 0 ends up being the default node. However node 0 might be offline.
> 
> Fix this by explicitly setting fallback node. Ensure first_memory_node
> is set before kernel does explicit setting of fallback node.
> 
>  arch/powerpc/mm/numa.c | 11 ++++++++++-
>  1 file changed, 10 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
> index 281531340230..6e97ab6575cb 100644
> --- a/arch/powerpc/mm/numa.c
> +++ b/arch/powerpc/mm/numa.c
> @@ -827,7 +827,16 @@ void __init dump_numa_cpu_topology(void)
>  	if (!numa_enabled)
>  		return;
>  
> -	for_each_online_node(node) {
> +	for_each_node(node) {
> +		/*
> +		 * For all possible but not yet online nodes, ensure their
> +		 * node_numa_mem is set correctly so that kmalloc_node works
> +		 * for such nodes.
> +		 */
> +		if (!node_online(node)) {

Change the above line to like below:

+               if (!node_state(node, N_MEMORY)) {

Regards,
Bharata.

Srikar Dronamraju March 17, 2020, 2:29 p.m. UTC | #2

* Bharata B Rao <[e-mail]> [2020-03-17 19:52:32]:

Thanks Bharata.

> This patchset can also fix a related problem that I reported earlier at
> https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-March/206076.html
> with an additional change, suggested by Srikar as shown below:
> 
> >  
> > -	for_each_online_node(node) {
> > +	for_each_node(node) {
> > +		/*
> > +		 * For all possible but not yet online nodes, ensure their
> > +		 * node_numa_mem is set correctly so that kmalloc_node works
> > +		 * for such nodes.
> > +		 */
> > +		if (!node_online(node)) {
> 
> Change the above line to like below:
> 
> +               if (!node_state(node, N_MEMORY)) {
> 

Just to clarify, this is needed if we don't have
http://lore.kernel.org/lkml/[e-mail]/t/#u

Patch

diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
index 281531340230..6e97ab6575cb 100644
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -827,7 +827,16 @@  void __init dump_numa_cpu_topology(void)
 	if (!numa_enabled)
 		return;
 
-	for_each_online_node(node) {
+	for_each_node(node) {
+		/*
+		 * For all possible but not yet online nodes, ensure their
+		 * node_numa_mem is set correctly so that kmalloc_node works
+		 * for such nodes.
+		 */
+		if (!node_online(node)) {
+			reset_numa_mem(node);
+			continue;
+		}
 		pr_info("Node %d CPUs:", node);
 
 		count = 0;