Re: [patch 07/11] x86/perf/uncore: Track packages not per cpu data

[Thomas Gleixner]

Stephane,

On Thu, 18 Feb 2016, Stephane Eranian wrote:
> On Thu, Feb 18, 2016 at 12:13 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > Now clearly, BIOS can completely wreck things and indeed report too
> > small an apic_id range or whatever, and in this case we're up a creek
> > without a paddle.
> >
> > But I think you can check for that at boot and report errors/warns
> > whatever, because if you trigger this, your system is not really
> > 'correct' anyway.
> >
> The example I was worried about is as follows, take an Ivytown system
> (IVT) with 12 physical cores per package on a 2 socket system.
> 
> Thomas's  topology_max_packages()  macro would return 2, for 2 packages
> with CPU0..CPU47.
> 
> But let's assume that the BIOS does some weird mappings and that the
> id for socket0 is indeed 0
> but for socket1 it is 255. Then doing:
> 
>     pkg = topology_physical_package_id();
>     pmu->boxes[pkg];
> 
> Would crash because the boxes[] has space for 2 elements and not 256.

Right. It did not occur to me that this might be possible, but as I said in my
reply to Peter, we can simply create logical package ids and use them. That
makes us safe against BIOS tinkering.
 
> If we know this cannot happen, then the code is fine. If we are not
> sure, then I believe a check should be added
> and if a mismatch is found, then the uncore PMU init code should
> return an error. That's all I was trying to point
> out. I think Thomas' code is indeed a good simplification.

I'm glad you like it.

Thanks,

	tglx