Path: csiph.com!news.mixmin.net!aioe.org!bofh.it!news.nic.it!robomod
From: Peter Zijlstra <peterz@infradead.org>
Newsgroups: linux.kernel
Subject: Re: [PATCH 1/2] x86/CPU/AMD: Present package as die instead of socket
Date: Wed, 05 Jul 2017 18:00:02 +0200
Message-ID: <tZUSB-8da-1@gated-at.bofh.it>
References: <tWWee-7HI-5@gated-at.bofh.it> <tWWee-7HI-7@gated-at.bofh.it> <tWWee-7HI-3@gated-at.bofh.it> <tWZ2q-1ch-11@gated-at.bofh.it>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: NeoMutt/20170113 (1.7.2)
Sender: robomod@news.nic.it
Approved: robomod@news.nic.it
Lines: 22
Organization: linux.* mail to news gateway
Xref: csiph.com linux.kernel:1681615

On Tue, Jun 27, 2017 at 08:07:10PM +0700, Suravee Suthikulpanit wrote:

> As I have described in the cover letter, this patch series changes how

Which you didn't send to me... thanks for that.

> kernel derives cpu "package" from package-as-socket to package-as-die in
> order to fix following issues on AMD family17h multi-die processor
> platforms:

> These issues are fixed when properly intepretes package as DIE.
> 
> For MCM Magny-Cours (family15h), we do not have this issue because the MC
> sched-domain has the same cpumask as the DIE sched-domain. This is not the
> same as for Zen cores (family17h) where the MC sched-domain is the CCX (2
> CCX per die)

Are you saying that your LLCs are smaller than your NUMA nodes ? You do
not in fact have a cache level that spans the memory controller?

Because that is indeed the assumption we have for Intel CoD/SnC and AMD
Magny-Cours.