Re: [PATCH] arch_topology: Limit threads to one specific cluster

From: Gavin Shan
Date: Tue May 24 2022 - 05:05:41 EST


Hi Sudeep,

On 5/24/22 4:51 PM, Sudeep Holla wrote:
On Tue, May 24, 2022 at 04:12:12PM +0800, Gavin Shan wrote:
The sibling information for one particular CPU is updated after ACPI
PPTT table is parsed. struct cpu_topology::thread_sibling tracks the
the CPUs in same core. However, cluster isn't considered when it's
populated. In this case, multiple threads belonging to different
clusters can be put together through the sibling information. It
eventually leads to unexpected warning from sched subsystem.

For example, the following warning is observed in a VM where we have
2 sockets, 4 clusters, 8 cores and 16 threads and the CPU topology
is populated as below.

CPU Socket-ID Cluster-ID Core-ID Thread-ID
----------------------------------------------
0 0 0 0 0
1 0 0 0 1
2 0 0 1 0
3 0 0 1 1
4 0 1 0 0
5 0 1 0 1
6 0 1 1 0
7 0 1 1 1
8 1 0 0 0
9 1 0 0 1
10 1 0 1 0
11 1 0 1 1
12 1 1 0 0
13 1 1 0 1
14 1 1 1 0
15 1 1 1 1

[ 0.592181] CPU: All CPU(s) started at EL1
[ 0.593766] alternatives: patching kernel code
[ 0.595890] BUG: arch topology borken
[ 0.597210] the SMT domain not a subset of the CLS domain
[ 0.599286] child=0-1,4-5 sd=0-3

# cat /sys/devices/system/cpu/cpu0/topology/cluster_cpus_list
0-3
# cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list
0-1,4-5

This fixes the issue by limiting threads to one specific cluster.
With this applied, the unexpected warning disappears in the VM.


I have similar fix but as part of bigger series[1] to get DT support in
line with ACPI.


Your patch resolves the issue I have. So please ignore mine. Sorry
for the noise.

https://lore.kernel.org/lkml/20220518093325.2070336-3-sudeep.holla@xxxxxxx

# cat /sys/devices/system/cpu/cpu0/topology/cluster_cpus_list
0-3
# cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list
0-1

Thanks,
Gavin