linux-kernel - Re: [REGRESSION] x86, perf: counter freezing breaks rr

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: <20181120194129.GC13936@tassilo.jf.intel.com>
Date:   Tue, 20 Nov 2018 11:41:29 -0800
From:   Andi Kleen <ak@...ux.intel.com>
To:     Kyle Huey <me@...ehuey.com>
Cc:     Kan Liang <kan.liang@...ux.intel.com>,
        "Peter Zijlstra (Intel)" <peterz@...radead.org>,
        Ingo Molnar <mingo@...nel.org>,
        Robert O'Callahan <robert@...llahan.org>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Arnaldo Carvalho de Melo <acme@...hat.com>,
        Jiri Olsa <jolsa@...hat.com>,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        Stephane Eranian <eranian@...gle.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Vince Weaver <vincent.weaver@...ne.edu>, acme@...nel.org,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [REGRESSION] x86, perf: counter freezing breaks rr

> rr, a userspace record and replay debugger[0], uses the PMU interrupt
> (PMI) to stop a program during replay to inject asynchronous events
> such as signals. With perf counter freezing enabled we are reliably
> seeing perf event overcounts during replay. This behavior is easily

It's hard to see how it could over count since the PMU freezes
earlier than the PMI with freezing. So it should count less.
Did you mean under count?

> Given that we're already at rc3, and that this renders rr unusable,
> we'd ask that counter freezing be disabled for the 4.20 release.

The boot option should be good enough for the release?

A reasonable future option would be to expose an option to disable it in
the perf_event. Then rr could set it.

-Andi