Re: [PATCH v3] x86,sched: allow topologies where NUMA nodes share an LLC
From: Peter Zijlstra
Date: Thu Mar 29 2018 - 09:57:25 EST
On Thu, Mar 29, 2018 at 06:45:12AM -0700, Dave Hansen wrote:
> On 03/29/2018 06:16 AM, Thomas Gleixner wrote:
> >> This is OK at least on the hardware we are immediately concerned about
> >> because the LLC sharing happens at both the slice and at the package
> >> level, which are also NUMA boundaries.
> > So that addresses the scheduler interaction, but it still leaves the
> > information in the sysfs files unchanged. See cpu/intel_cacheinfo.c. There
> > are applications which use that information so it should be correct.
>
> Were you thinking of shared_cpu_list/map?
Yes, the fact that those are changed and the Changelog doesn't mention
them is a giant fail all by itself.
> The information in there is
> correct for core->off-package access. It is not correct for
> core->on-package access, unless that access is perfectly interleaved
> across both package "slices".
The fact that 'simple' measures like cache/cpu are now completely broken
is a problem.