[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 8 Aug 2018 01:19:42 -0700
From: Srikar Dronamraju <srikar@...ux.vnet.ibm.com>
To: Peter Zijlstra <peterz@...radead.org>
Cc: Ingo Molnar <mingo@...nel.org>,
LKML <linux-kernel@...r.kernel.org>,
Mel Gorman <mgorman@...hsingularity.net>,
Rik van Riel <riel@...riel.com>,
Thomas Gleixner <tglx@...utronix.de>,
Michael Ellerman <mpe@...erman.id.au>,
Heiko Carstens <heiko.carstens@...ibm.com>,
Suravee Suthikulpanit <suravee.suthikulpanit@....com>,
Andre Wild <wild@...ux.vnet.ibm.com>
Subject: Re: [PATCH] sched/topology: Use Identity node only if required
* Peter Zijlstra <peterz@...radead.org> [2018-08-08 09:58:41]:
> On Wed, Aug 08, 2018 at 12:39:31PM +0530, Srikar Dronamraju wrote:
> > - numa topology on 2 node systems no more marked as NUMA_DIRECT. After
> > this commit, it gets reported as NUMA_BACKPLANE. This is because
> > sched_domains_numa_level now equals 2 on 2 node systems.
> >
> > - Extra numa sched domain that gets added and degenerated on most
> > machines. The Identity node is only needed on very few systems.
> > Also all non-numa systems will end up populating
> > sched_domains_numa_distance and sched_domains_numa_masks tables.
> >
> > - On shared lpars like powerpc, this extra sched domain creation can
> > lead to repeated rcu stalls, sometimes even causing unresponsive
> > systems on boot. On such stalls, it was noticed that
> > init_sched_groups_capacity() (sg != sd->groups is always true).
>
> The idea was that if the topology level is redundant (as it often is);
> then the degenerate code would take it out.
>
> Why is that not working (right) and can we fix that instead?
>
All I have found is regular NUMA sched_domains use OVERLAP flag, which
inturn results in build_overlap_sched_groups(). The Identity node uses
build_sched_groups. Somehow build_sched_groups is unable to create the
group list. I still getting to see why that makes a difference.
I have tried with passing .flags = SDTL_OVERLAP to the identity node and
that works well.
However, I still think if majority of the cases the identity node is
going to be redundant, then we should use hint.
We could fix the numa topology to be NUMA_DIRECT for 2 node machines, by
checking if sched_domains_numa_levels == 2, but then I dont know what it
means for a system that has only NODE but not NUMA level.
i.e what should we say the numa topology type of a machine that has only
NODE but not NUMA sched_domain?
--
Thanks and Regards
Srikar
Powered by blists - more mailing lists