linux-kernel - Re: Ftrace vs perf user page fault statistics differences

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 for Android: free password hash cracker in your pocket

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

Date:   Wed, 14 Jun 2017 13:43:47 -0400
From:   Steven Rostedt <rostedt@...dmis.org>
To:     Will Hawkins <whh8b@...ginia.edu>
Cc:     Namhyung Kim <namhyung@...il.com>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: Ftrace vs perf user page fault statistics differences

On Wed, 14 Jun 2017 13:30:59 -0400
Will Hawkins <whh8b@...ginia.edu> wrote:

> > When perf profiles a program started by the same command line, it
> > disables the events by default and enables them during exec.  Please
> > see linux/tools/perf/util/evsel.c:perf_evsel__config().
> >
> > Thanks,
> > Namhyung  
> 
> Namhyung,
> 
> I think that this answers a very important question! Thanks for chiming in!

Yes. One difference between the design of ftrace and the design of
perf, is that, I avoided inserting call backs throughout the kernel.
Perf has a few function calls in the exec code. Just grep "perf" in
fs/exec.c. There's a few scattered around there. Causing a slight
overhead for when perf is not in use.

Hmm, I really should remove all perf injections and make them either a
tracepoint or generic jumplabels that anything may attach to. Then
ftrace could have the same features. And lttng for that matter.

-- Steve