Re: [PATCH 3/3] perf test: fix "trace summary" test for musl-based systems

On Mon, Apr 27, 2026 at 09:57:44AM +0000, [email protected] wrote:
> Sashiko AI review found 1 potential issue(s):
> - [Medium] The `ps --help` test workload is brittle and will continue to fail on many musl-based systems that use BusyBox.
> --
> 
> commit 41e18297314ee059526c7fb6d14d02d61ba92d64
> Author: Martin Kaiser <[email protected]>
> 
> perf test: fix "trace summary" test for musl-based systems
> 
> This commit updates the trace summary test to use ps --help instead
> of true. The goal is to ensure the test workload triggers open, read,
> and close events from dynamic library loading on musl-based systems.
> 
> > diff --git a/tools/perf/tests/shell/trace_summary.sh b/tools/perf/tests/shell/trace_summary.sh
> > --- a/tools/perf/tests/shell/trace_summary.sh
> > +++ b/tools/perf/tests/shell/trace_summary.sh
> > @@ -14,7 +14,7 @@ OUTPUT=$(mktemp /tmp/perf_trace_test.XXXXX)
> >  
> >  test_perf_trace() {
> >      args=$1
> > -    workload="true"
> > +    workload="ps --help"
> 
> Will this workload still fail on musl-based systems that use BusyBox?
> 
> Many musl-based distributions, such as Alpine Linux or OpenWrt, frequently
> use BusyBox to provide standard utilities. BusyBox's ps is typically
> statically linked or links only to libc, without depending on libprocps.
> 
> Furthermore, running ps --help in BusyBox generally prints the usage string
> and exits without performing file I/O, which means no open, read, or close
> events will be generated.
> 
> Would a command that natively performs file I/O, such as cat /dev/null or
> dd if=/dev/null of=/dev/null, be more robust here?

Sounds right, it'd be nice if the workload doesn't print anything like
above.

Thanks,
Namhyung

> 
> >      search="^\s*(open|read|close).*[0-9]+%$"
> 
> -- 
> Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3