linux-kernel - Re: [PATCH] perf report: Display per-thread event counters

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:	Sat, 8 Aug 2009 14:14:56 +0200
From:	Ingo Molnar <mingo@...e.hu>
To:	Brice Goglin <Brice.Goglin@...ia.fr>,
	Frédéric Weisbecker <fweisbec@...il.com>,
	Mike Galbraith <efault@....de>,
	Arnaldo Carvalho de Melo <acme@...hat.com>
Cc:	Peter Zijlstra <a.p.zijlstra@...llo.nl>, paulus@...ba.org,
	LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] perf report: Display per-thread event counters

* Brice Goglin <Brice.Goglin@...ia.fr> wrote:

> Here's a better patch. I moved everything to utils/values.[ch] so 
> that we may reuse it in perf stat. [...]

Nice patch! I've applied it, you can find it in the latest -tip 
tree:

  http://people.redhat.com/mingo/tip.git/README

please send enhancements/fixes on top of this.

> [...] But I don't see yet where I am suppose to get something like 
> PERF_READ_EVENT in builtin-stat.c so I haven't touched it yet.

Yeah. 'perf stat' is not really getting events but is doing a 
read-out of the counter value(s) and constructs its 'read event' 
that way. So you wont find PERF_READ_EVENT in builtin-stat.c, you'll 
find:

                res = read(fd[cpu][counter], single_count, nv * sizeof(u64));

in read_counter(). The printout is then done in print_counter().

> We get something like this now:
> #  PID   TID  cache-misses  cache-references
>   4658  4659        495581           3238779
>   4658  4662        498246           3236823
>   4658  4663        499531           3243162
> 
> Then it'll be easy to add --pretty=raw to display a single line 
> per thread/event.

ok.

> By the way, -S was also used for --symbol... So I used -T/--thread 
> here.

Hm, indeed - and -s was taken for --sort. Maybe we could rename 
-S/--symbols to -y/--symbols - this too is an i think rarely used 
feature.

I think pure 'statistics' runs like you do will be a pretty popular 
workflow, so intuitive naming/placement of options is important.

> perf report: Add -T/--threads to display per-thread counter values
>     
> We get something like this now:
> #  PID   TID  cache-misses  cache-references
>   4658  4659        495581           3238779
>   4658  4662        498246           3236823
>   4658  4663        499531           3243162

Btw., another thing to do would be to allow the 'dual' recording of 
both the stat values (collected when threads exit) and regular 
samples that perf report deals with.

I.e. dont handle 'perf record -s' as an exclusive thing to regular 
'perf record', but instead have -s/--sample-type option that can 
have such combinations:

  -s stats
  -s samples
  -s call-graph

And any combination thereof, such as:

  -s stats,samples

The default would be '-s samples'. Right now call-graph recording is 
triggered via a separate option (-g/--call-graph) - but maybe it 
could be merged into a more generic -s/--sample option mechanism?

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/