Message-ID: <1329310113.2293.72.camel@twins>
Subject: Re: [PATCH] Perf: bug fix: distinguish between rename and exec
From: Peter Zijlstra <a.p.zijlstra@chello.nl>
To: Luigi Semenzato <semenzato@chromium.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>,
        Paul Mackerras <paulus@samba.org>, Ingo Molnar <mingo@elte.hu>,
        Arnaldo Carvalho de Melo <acme@ghostprotocols.net>,
        Andrew Morton <akpm@linux-foundation.org>,
        Vasiliy Kulikov <segoon@openwall.com>,
        Stephen Wilson <wilsons@start.ca>, Oleg Nesterov <oleg@redhat.com>,
        Tejun Heo <tj@kernel.org>,
        Paul Gortmaker <paul.gortmaker@windriver.com>,
        Andi Kleen <ak@linux.intel.com>,
        Lucas De Marchi <lucas.demarchi@profusion.mobi>,
        Greg Kroah-Hartman <gregkh@suse.de>,
        "Eric W. Biederman" <ebiederm@xmission.com>,
        "Rafael J. Wysocki" <rjw@sisk.pl>,
        Frederic Weisbecker <fweisbec@gmail.com>,
        David Ahern <dsahern@gmail.com>, Namhyung Kim <namhyung@gmail.com>,
        Robert Richter <robert.richter@amd.com>, linux-kernel@vger.kernel.org,
        sonnyrao@chromium.org, olofj@chromium.org, eranian@google.com
Date: Wed, 15 Feb 2012 13:48:33 +0100
In-Reply-To: <1329195360-10699-1-git-send-email-semenzato@chromium.org>
References: <1329195360-10699-1-git-send-email-semenzato@chromium.org>
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7BIT
Mime-Version: 1.0
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2066
Lines: 50

On Mon, 2012-02-13 at 20:56 -0800, Luigi Semenzato wrote:
> This makes it possible for "perf report" to distinguish between an exec and
> a rename (for instance from prctl(PR_SET_NAME)).  Currently a similar COMM
> record is produced for the two events.  Perf report assumes all COMM records
> are execs and discards the old mappings.  Without mappings, perf report
> can no longer correlate sampled IPs to the functions containing them,
> and collapses all samples into a single bucket.
> 
> This change maintains backward compatibility in both directions, i.e. old
> version of perf will continue to work on new perf.data files in the same
> way, and new versions of perf on old data files.
> 
> Another solution would be to not send COMM records for renames.  Although
> it seems reasonable, it might break existing setups.

Uhm, didn't you argue its already broken?

> +++ b/fs/exec.c
> @@ -1052,7 +1052,7 @@ char *get_task_comm(char *buf, struct task_struct *tsk)
>  }
>  EXPORT_SYMBOL_GPL(get_task_comm);
>  
> -void set_task_comm(struct task_struct *tsk, char *buf)
> +void set_task_comm(struct task_struct *tsk, char *buf, bool is_rename)
>  {
>  	task_lock(tsk);
>  
> @@ -1068,7 +1068,7 @@ void set_task_comm(struct task_struct *tsk, char *buf)
>  	wmb();
>  	strlcpy(tsk->comm, buf, sizeof(tsk->comm));
>  	task_unlock(tsk);
> -	perf_event_comm(tsk);
> +	perf_event_comm(tsk, is_rename);
>  }

I really dislike changing generic code purely for the purpose of
instrumentation like this. Better to pull perf_event_comm() out of here
if you want to change semantics.

Personally I couldn't care less about renames, I think they're a waste
of time, so I'm ok with the simple patch moving the perf_event_comm()
into setup_new_exec() and possibly renaming it to perf_event_exec().

Acme, do you care about renames?

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