Re: [PATCH v6 1/2] perf record: Propagate exit status of a command line workload

[Namhyung Kim <namhyung@kernel.org>]

Hi Jiri,

On Sun, 11 May 2014 15:30:18 +0200, Jiri Olsa wrote:
> On Fri, May 09, 2014 at 02:14:17PM +0900, Namhyung Kim wrote:
>> Currently perf record doesn't propagate the exit status of a workload
>> given by the command line.  But sometimes it'd useful if it's
>> propagated so that a monitoring script can handle errors
>> appropriately.
>> 
>> To do that, it got rid of exit handlers and run/call them directly in
>> the __cmd_record().  I don't see any reason why those are in a form of
>> exit handlers in the first place.  Also it cleaned up the resource
>> management code in record__exit().
>
> so in this last change we keep one exit handler, please update
> the changelog

Okay.

>
>> 
>> With this change, perf record returns the child exit status in case of
>> normal termination and send signal to itself when terminated by signal.
>> 
>
> SNIP
>
>>  
>>  out_delete_session:
>>  	perf_session__delete(session);
>> -	return err;
>> +	return status;
>
> we dont set status correctly before the 'goto out_delete_session'
> in the following condition:
>
>                 err = perf_evlist__prepare_workload(rec->evlist, &opts->target,
>                                                     argv, file->is_pipe,
>                                                     workload_exec_failed_signal);
>                 if (err < 0) {
>                         pr_err("Couldn't run the workload!\n");
>                         goto out_delete_session;
>                 }
>
> so we dont propagate status correctly in case the above function fails:
>
> [jolsa@krava perf]$ ulimit -n 6
> [jolsa@krava perf]$ ./perf record sleep 100
> failed to create 'go' pipe: Too many open files
> Couldn't run the workload!
> [jolsa@krava perf]$ echo $?
> 0
> [jolsa@krava perf]$

Right.  I'll send v7!

Thanks,
Namhyung