Re: [PATCH bpf-next 0/2] selftests/bpf: port test_ftrace.sh into test_progs

["Alexis Lothoré" <alexis.lothore@bootlin.com>]

Hi Alexei,

On Sat Oct 31, 2026 at 11:54 PM CET, Alexei Starovoitov wrote:
> On Fri, Oct 30, 2026 at 11:50 AM Alexis Lothoré (eBPF Foundation) <alexis.lothore@bootlin.com> wrote:
>> - commit 1 brings a few new helpers to manipulate ftrace from test_progs
>> - commit 2 performs the actual porting
> [...]
>>  5 files changed, 412 insertions(+), 46 deletions(-)
>
> This is quite heavy for a port of a 44 line script.
>
> Lets use a tracing instance. function_graph works in instances.
> An instance has its own set_ftrace_filter, current_tracer and ring
> buffer. rmdir undoes all of it.

I initially went with a ftrace instance, but I came back to global
ftrace because graph function filters (set_graph_function,
set_graph_notrace) are not handled at instance level, only at global
level. But maybe setting only set_ftrace_filter and set_ftrace_notrace
is enough for this test ? That would indeed simplify a lot the setup.

> What goes away:
>
>   series                          tracing instance
>   ------------------------------  ------------------------------
>   global tracefs files            files of the instance
>   save and restore, patch 1       rmdir
>   serial_                         not needed
>   all functions are patched       only the filtered functions

Are all functions really patched with the current series ? It is still
applying filters.

>   new progs                       progs/modify_return.c as is
>
> modify_return has fentry, fmod_ret and fexit on the same function
> and checks the return values.
>
> The removal of test_ftrace.sh stays.
>
> Instead of patches 1 and 2:
>
> diff --git a/tools/testing/selftests/bpf/prog_tests/modify_return.c b/tools/testing/selftests/bpf/prog_tests/modify_return.c
> --- a/tools/testing/selftests/bpf/prog_tests/modify_return.c
> +++ b/tools/testing/selftests/bpf/prog_tests/modify_return.c
> @@ -61,3 +61,28 @@ void test_modify_return(void)
>  		 0 /* want_side_effect */,
>  		 -EINVAL * 2 /* want_ret */);
>  }
> +
> +/*
> + * The same test while function_graph traces the same functions.
> + * A tracing instance has its own filter, tracer and ring buffer.
> + * rmdir() undoes all of it.
> + */
> +void test_modify_return_fgraph(void)
> +{
> +	char dir[128];
> +
> +	snprintf(dir, sizeof(dir), "%s/instances/bpf-XXXXXX",
> +		 access("/sys/kernel/tracing/trace", F_OK) ?
> +		 "/sys/kernel/debug/tracing" : "/sys/kernel/tracing");
> +	if (!ASSERT_OK_PTR(mkdtemp(dir), "mkdtemp"))
> +		return;
> +
> +	SYS(out, "echo 'bpf_modify_return_test*' > %s/set_ftrace_filter", dir);
> +	SYS(out, "echo function_graph > %s/current_tracer", dir);
> +
> +	test_modify_return();
> +
> +	SYS(out, "grep -q bpf_modify_return_test2 %s/trace", dir);
> +out:
> +	ASSERT_OK(rmdir(dir), "rmdir");
> +}
>
> pw-bot: cr





-- 
Alexis Lothoré, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com