Re: [tip: sched/core] sched: Add cluster scheduler level for x86

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Oct 20, 2021 at 08:12:51AM -0500, Tom Lendacky wrote:
> On 10/15/21 4:44 AM, tip-bot2 for Tim Chen wrote:
> > The following commit has been merged into the sched/core branch of tip:
> > 
> > Commit-ID:     66558b730f2533cc2bf2b74d51f5f80b81e2bad0
> > Gitweb:        https://git.kernel.org/tip/66558b730f2533cc2bf2b74d51f5f80b81e2bad0
> > Author:        Tim Chen <tim.c.chen@linux.intel.com>
> > AuthorDate:    Fri, 24 Sep 2021 20:51:04 +12:00
> > Committer:     Peter Zijlstra <peterz@infradead.org>
> > CommitterDate: Fri, 15 Oct 2021 11:25:16 +02:00
> > 
> > sched: Add cluster scheduler level for x86
> > 
> > There are x86 CPU architectures (e.g. Jacobsville) where L2 cahce is
> > shared among a cluster of cores instead of being exclusive to one
> > single core.
> > 
> > To prevent oversubscription of L2 cache, load should be balanced
> > between such L2 clusters, especially for tasks with no shared data.
> > On benchmark such as SPECrate mcf test, this change provides a boost
> > to performance especially on medium load system on Jacobsville.  on a
> > Jacobsville that has 24 Atom cores, arranged into 6 clusters of 4
> > cores each, the benchmark number is as follow:
> > 
> >   Improvement over baseline kernel for mcf_r
> >   copies		run time	base rate
> >   1		-0.1%		-0.2%
> >   6		25.1%		25.1%
> >   12		18.8%		19.0%
> >   24		0.3%		0.3%
> > 
> > So this looks pretty good. In terms of the system's task distribution,
> > some pretty bad clumping can be seen for the vanilla kernel without
> > the L2 cluster domain for the 6 and 12 copies case. With the extra
> > domain for cluster, the load does get evened out between the clusters.
> > 
> > Note this patch isn't an universal win as spreading isn't necessarily
> > a win, particually for those workload who can benefit from packing.
> > 
> > Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
> > Signed-off-by: Barry Song <song.bao.hua@hisilicon.com>
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > Link: https://lore.kernel.org/r/20210924085104.44806-4-21cnbao@gmail.com
> 
> I've bisected to this patch which now results in my EPYC systems issuing a
> lot of:
> 
> [    4.788480] BUG: arch topology borken
> [    4.789578]      the SMT domain not a subset of the CLS domain
> 
> messages (one for each CPU in the system).
> 
> I haven't had a chance to dig deeper and understand everything, does anyone
> have some quick insights/ideas?

Urgh, sorry about that. So this stuff uses cpu_l2c_id to build 'clusters',
basically CPUs that share L2, as a subset of the 'multi-core' topology,
which is all CPUs that share LLC (L3 typically).

Your EPYC seems to think the SMT group is not a strict subset of the L2.
The implication is that you have threads with different L2, which would
franky be quite 'exotic' if true :-)


If it does boot, what does something like:

  for i in /sys/devices/system/cpu/cpu*/topology/*{_id,_list}; do echo -n "${i}: " ; cat $i; done

produce?

I'll try and figure out how AMD sets l2c_id, that stuff is always a bit
of a maze.