Re: [PATCH] perf trace: don't show comm/tid for regular syscalls for single-pid traces

[Namhyung Kim <namhyung@kernel.org>]

On Thu, Aug 21, 2025 at 12:58:32PM -0400, Anubhav Shelat wrote:
> Previously when attaching perf trace to a specific pid the output would
> look like this:
> 
>          ? (         ): vest/616160  ... [continued]: clock_nanosleep())                                  = 0
>      0.041 ( 0.013 ms): vest/616160 fstat(statbuf: 0xfffff2c60038)                                        = 0
>      0.073 ( 0.008 ms): vest/616160 read(buf: 0x3bede330, count: 4096)                                    = 17
>      0.090 (1000.095 ms): vest/616160 clock_nanosleep(rqtp: { .tv_sec: 1, .tv_nsec: 0 }, rmtp: 0xfffff2c601d8) = 0
>   1000.210 ( 0.136 ms): vest/616160 openat(dfd: CWD, filename: "/tmp/something.txt", flags: CREAT|TRUNC|WRONLY, mode: IRUGO|IWUGO) = 3
>   1000.361 ( 0.004 ms): vest/616160 fstat(fd: 3, statbuf: 0xfffff2c60088)                                 = 0
>   1000.375 ( 0.018 ms): vest/616160 write(fd: 3, buf: SOME TEST STRING\10, count: 17)                     = 17
>   1000.395 ( 0.072 ms): vest/616160 close(fd: 3)                                                          = 0
>   1000.476 (         ): vest/616160 exit_group()
> 
> which would cause an error in the perftool-testsuite since the comm/tid
> shouldn't be in trace lines for regular syscalls. With this change the
> output looks like this:

I'm not sure if it's the right test.  It's controlled by the
trace->multiple_threads condition which I believe should use
trace->opts.no_inherit condition instead of the attribute of the
first event.

Thanks,
Namhyung

> 
>          ? (         ): vest/612038  ... [continued]: clock_nanosleep())                                  = 0
>      0.039 ( 0.012 ms): fstat(statbuf: 0xfffff8e94cb8)                                        = 0
>      0.069 ( 0.008 ms): read(buf: 0x3b997330, count: 4096)                                    = 17
>      0.086 (1000.436 ms): clock_nanosleep(rqtp: { .tv_sec: 1, .tv_nsec: 0 }, rmtp: 0xfffff8e94e58) = 0
>   1000.545 ( 0.136 ms): openat(dfd: CWD, filename: "/tmp/something.txt", flags: CREAT|TRUNC|WRONLY, mode: IRUGO|IWUGO) = 3
>   1000.700 ( 0.006 ms): fstat(fd: 3, statbuf: 0xfffff8e94d08)                                 = 0
>   1000.720 ( 0.024 ms): write(fd: 3, buf: SOME TEST STRING\10, count: 17)                     = 17
>   1000.750 ( 0.073 ms): close(fd: 3)                                                          = 0
>   1000.833 (         ): vest/612038 exit_group()
> 
> Signed-off-by: Anubhav Shelat <ashelat@redhat.com>
> ---
>  tools/perf/builtin-trace.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
> index fe737b3ac6e6..4cc38ee6aad9 100644
> --- a/tools/perf/builtin-trace.c
> +++ b/tools/perf/builtin-trace.c
> @@ -2955,11 +2955,17 @@ static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
>  	if (trace->summary_only || (ret >= 0 && trace->failure_only))
>  		goto out;
>  
> -	trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
> -
>  	if (ttrace->entry_pending) {
> +		if (trace->opts.target.pid) {
> +			if (trace->show_tstamp)
> +				printed = trace__fprintf_tstamp(trace, ttrace->entry_time, trace->output);
> +			if (trace->show_duration)
> +				printed += fprintf_duration(duration, duration_calculated, trace->output);
> +		} else
> +			trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
>  		printed = fprintf(trace->output, "%s", ttrace->entry_str);
>  	} else {
> +		trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
>  		printed += fprintf(trace->output, " ... [");
>  		color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
>  		printed += 9;
> -- 
> 2.50.1
>