Re: [RFC][PATCH 5/6] x86/topo: Fix SNC topology mess

[Zhang Rui]

On Mon, 2026-03-02 at 17:43 +0100, Peter Zijlstra wrote:
> nt On Sat, Feb 28, 2026 at 03:35:26PM +0800, Chen, Yu C wrote:
> 
> > Here we check packages0 and 1, should we check all the packages?
> 
> Might as well I suppose.
> 
> Could you boot queue/x86/topo on an snc-3 machine to verify it
> doesn't
> explode with all the paranoia on?

Hi, Peter,

regarding slit_validate() in
https://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git/commit/?h=x86/topo&id=24ca94ac4b72803a7164b7ad84f06f0e9f0c49df

I suppose we want to use 
	WARN_ON_ONCE(!slit_cluster_symmetric(n))
rather than
	WARN_ON_ONCE(slit_cluster_symmetric(n));
right?

I tested the queue/x86/topo plus above change,

1. on GNR 4 sockets with SNC2, no difference about sched_domains in
/proc/schedstat compared with upstream 7.0-rc1, plus the below warning
is also gone
[   11.439633] ------------[ cut here ]------------
[   11.440491] sched: Expect only up to 2 packages for GNR or CWF, but
saw 4 packages when building sched domains.
[   11.440493] WARNING: arch/x86/kernel/smpboot.c:574 at
arch_sched_node_distance+0x133/0x140, CPU#0: swapper/0/1

2. on CWF 2 sockets with SNC3 and CWF 1 socket with SNC3, no difference
about sched_domains in /proc/schedstat compared with upstream 7.0-rc1

So

Tested-by: Zhang Rui <rui.zhang@xxxxxxxxx>

-rui