linux-kernel - Re: [PATCH 11/16] perf intel-pt: Add support for synthesizing callchains for regular events

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]

Date:   Sun, 19 Apr 2020 20:04:57 -0700
From:   Andi Kleen <ak@...ux.intel.com>
To:     Arnaldo Carvalho de Melo <arnaldo.melo@...il.com>
Cc:     Adrian Hunter <adrian.hunter@...el.com>,
        Jiri Olsa <jolsa@...hat.com>, linux-kernel@...r.kernel.org,
        Namhyung Kim <namhyung@...nel.org>,
        linux-perf-users@...r.kernel.org
Subject: Re: [PATCH 11/16] perf intel-pt: Add support for synthesizing
 callchains for regular events

> Andi? My concern is that if this is the optimal solution for a good
> subset of the machines out there, then we need to make it easy to use,
> even transparent, if possible and safe to take that path.

I'm not sure it's that great in the general case.  A PT call graph
would need a full PT recording from start to end.

The problem with full PT recording is that it doesn't really work for a
lot of workloads, because for anything doing enough computation the CPU
just generates too much data and you end up with a lot of gaps in the
trace when the perf record flushing cannot keep up.

Also even if it worked you might end up with far too much data that
will take a long time to process.

So I suspect it wouldn't work often enough to be generally useful.

For the leader sample case we just want to use some non PT method
to get the call graph.

-Andi