linux-kernel - Re: [PATCH 03/10] perf report: Apply --percent-limit to callchains also

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:	Tue, 2 Feb 2016 23:35:13 +0900
From:	Namhyung Kim <namhyung@...nel.org>
To:	Arnaldo Carvalho de Melo <acme@...nel.org>
Cc:	Ingo Molnar <mingo@...nel.org>,
	Peter Zijlstra <peterz@...radead.org>,
	Jiri Olsa <jolsa@...nel.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Andi Kleen <andi@...stfloor.org>,
	David Ahern <dsahern@...il.com>,
	Frederic Weisbecker <fweisbec@...il.com>,
	Wang Nan <wangnan0@...wei.com>
Subject: Re: [PATCH 03/10] perf report: Apply --percent-limit to callchains
 also

On Tue, Feb 02, 2016 at 11:27:34AM -0300, Arnaldo Carvalho de Melo wrote:
> Em Tue, Feb 02, 2016 at 11:15:35PM +0900, Namhyung Kim escreveu:
> > On Tue, Feb 02, 2016 at 10:55:34AM -0300, Arnaldo Carvalho de Melo wrote:
> > > Em Tue, Feb 02, 2016 at 10:05:37PM +0900, Namhyung Kim escreveu:
> > > > Hi Arnaldo,
> > > > 
> > > > On Mon, Feb 01, 2016 at 05:19:36PM -0300, Arnaldo Carvalho de Melo wrote:
> > > > > Em Thu, Jan 28, 2016 at 12:40:50AM +0900, Namhyung Kim escreveu:
> > > > > > Currently --percent-limit option only works for hist entries.  However
> > > > > > it'd be better to have same effect to callchains as well
> > > > > 
> > > > > Documentation needs updating? It says:
> > > > > 
> > > > > --percent-limit::
> > > > >         Do not show entries which have an overhead under that percent.
> > > > >         (Default: 0).
> > > > 
> > > > Right.  Is it ok to you?
> > > > 
> > > > --percent-limit::
> > > >         Do not show entries and callchains which have an overhead under that
> > > >         percent.  (Default: 0).
> > > 
> > > Ok, but is the default zero?
> > > 
> > > That was what I was alluding to, as as soon as I applied the patch that
> > > made callchains honour this limit, about 60% of the entries in the
> > > particular perf.data file I was 'perf report'ing lost its '+' (callchain
> > > expansion) signs.
> > 
> > Yes, and this is what I want to say too. :)
> > 
> > The default value of percent limit is different for hist entry and
> > callchains.  For hist entry the default is 0, and for callchains it's
> > 0.5%.  But using --percent-limit option, we can set both at once from
> > now on.
> > 
> > Before this patchset, percent limit of callchains didn't checked.
> > Once applied, users can see callchains are disppeared like your case.
> >  This is just because 0.5% of the default limit for callchains is too
> > high IMHO - I guess it was originally set for 'fractal' mode which
> > calculates relative percents.  So I proposed to change the default in
> > the patch 10/10 but Andi thoughts 0.005% was too small.  We need to
> > choose other value like 0.05% ?
> 
> Unsure about the limit, but please fold the nice explanation you gave
> about the limits for hist entries and callchains and how the defaults
> are different, and how they can be set at once using --percent-limit.

Ok, will do.

> 
> Also, would it be overengineering to allow optionally set two limits,
> one for hists and one for callchains? I.e. like:
> 
>  --percent-limit 1,0.07
> 
> Then:
> 
>   --percent-limit 0.8
> 
> would be the same as:
> 
>   --percent-limit 0.8,0.8

Originally, the percent limit (threshold) for callchains can be set
using -g/--call-graph option.  What Andi asked is to set both limits
using a single option.  Not sure it's worth adding another way to
specify the callchain limit.

Thanks,
Namhyung