linux-kernel - Re: [PATCH] perf/x86/intel/ds: Fix the conversion from TSC to perf time

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] [day] [month] [year] [list]

Message-ID: <f661cad2-1299-932e-5845-d95ccd263d75@linux.intel.com>
Date:   Mon, 23 Jan 2023 16:34:59 -0500
From:   "Liang, Kan" <kan.liang@...ux.intel.com>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     mingo@...hat.com, linux-kernel@...r.kernel.org,
        namhyung@...nel.org, eranian@...gle.com, ak@...ux.intel.com,
        Thomas Gleixner <tglx@...utronix.de>
Subject: Re: [PATCH] perf/x86/intel/ds: Fix the conversion from TSC to perf
 time

On 2023-01-23 3:39 p.m., Peter Zijlstra wrote:
> On Mon, Jan 23, 2023 at 09:20:27AM -0800, kan.liang@...ux.intel.com wrote:
>> The TSC unstable case seems to be some corner cases (e.g., due to broken
>> BIOS). This patch doesn't support the conversion when the TSC is
>> unstable. The TSC in a PEBS record will be dropped and fallback to the
>> software perf time provided by the generic code.
> 
> :-(
> 
> You're saying there's modern systems (PEBS timestamps are fairly new)
> that trigger unstable TSC ?
> 
> What systems in specific have you observed this on -- we really need to
> name and shame them, this is fully unacceptable.

No, I didn't observe any systems which trigger unstable TSC.

Here, I tried to explain the reason why the patch didn't support the
conversion when the TSC is unstable. Because it's a corner case and very
unlikely to happen.
The "broken BIOS" case is quoted from the __sched_clock_work(), which
shows that the broken BIOS may be a reason of unstable TSC.

Sorry for the confusion caused.

Let me rephrase this.
"This patch doesn't support the conversion when the TSC is unstable.
Because the TSC unstable case is a corner case and very unlikely to
happen. Even if it happens, the TSC in a PEBS record will be dropped and
fall back to the inaccurate software perf time provided by the generic
code as perf does it on the previous platforms."

Thanks,
Kan