Re: [PATCH] sched/topology: Correctly propagate NUMA flag to scheduling domains
From: Andrea Righi
Date: Mon Nov 11 2024 - 09:32:40 EST
Hi Peter,
On Mon, Nov 11, 2024 at 10:22:46AM +0100, Peter Zijlstra wrote:
> On Sat, Nov 09, 2024 at 03:56:28PM +0100, Andrea Righi wrote:
>
> > @@ -1974,6 +1981,7 @@ void sched_init_numa(int offline_node)
> > */
> > tl[i++] = (struct sched_domain_topology_level){
> > .mask = sd_numa_mask,
> > + .sd_flags = cpu_numa_flags,
> > .numa_level = 0,
> > SD_INIT_NAME(NODE)
> > };
>
> This doesn't seem right. This level is a single node, and IIRC we only
> expect SD_NUMA on cross-node domains.
Ah! This is the part that I was missing, thanks for clarifying it.
Basically I need to look at sd->groups of the SD_NUMA domain to figure
out the individual nodes.
Please ignore this patch then.
Thanks,
-Andrea