linux-kernel - Re: [PATCH] x86/aperfmperf: Don't disable scheduler APERF/MPERF on bad samples

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <aTH6M8mU1Ods0F4t@tassilo>
Date: Thu, 4 Dec 2025 13:16:35 -0800
From: Andi Kleen <ak@...ux.intel.com>
To: Andrew Cooper <andrew.cooper3@...rix.com>
Cc: ggherdovich@...e.cz, linux-kernel@...r.kernel.org, peterz@...radead.org,
	rafael.j.wysocki@...el.com, x86@...nel.org
Subject: Re: [PATCH] x86/aperfmperf: Don't disable scheduler APERF/MPERF on
 bad samples

On Thu, Dec 04, 2025 at 08:02:09PM +0000, Andrew Cooper wrote:
> > However this can happen without any malfunction when there is
> > a long enough interruption between the two MSR reads, for
> > example due to an unlucky NMI or SMI or other system event
> > causing delays.
> 
> The list of problems is even longer under virt.  The vCPU can lose it's
> timeslice (-> arbitrary delay), or be scheduled onto a different CPU
> between the A and M reads (-> calculate the wrong ratio, and not in a
> way you could exclude it as a bad sample).

You could avoid that in the hypervisor by saving/restoring the counts (the
MSRs are writable), but it's likely rare and harmless enough that 
it's not worth the performance overhead.

-Andi