latest ftrace test run results w/panand kernel patch

Hi Steve,

On Friday 20 February 2015 11:59 PM, Pratyush Anand wrote:
>> Here is another issue, which is related to ftrace kernel code. Although
>> it is generating a warning only, but causing a beaker stress test to
>> fail.
>>
>>
>> Run attached scripts (./ftrace_ftrace_enabled.sh &
>> ./ftrace_current_tracer.sh &) and we land into a situation after some
>> random iterations where ftrace_disable_ftrace_graph_caller is called two
>> times consecutively without ftrace_enable_ftrace_graph_caller in
>> between. Since ARM64 code validates a old instruction against
>> replaceable instruction, so it fails in this case and returns -EINVAL to
>> the generic ftrace code, which generates a warn.
>>
>>
>
> It seems to me that, there is a race between unregister_ftrace_graph and
> ftrace_enable_sysctl.

I analyzed this issue a bit more. This is my understanding:

When ftrace is enabled globally through proc interface, we must check if
ftrace_graph_active is set. If it is set, then we should also pass
FTRACE_START_FUNC_RET command to ftrace_run_update_code. Similarly, when
  ftrace is disabled globally through proc interface, we must check if
ftrace_graph_active is set. If it is set, then we should also pass
FTRACE_STOP_FUNC_RET command to ftrace_run_update_code.

Following patch resolves this issue. If you think it is right solution,
then may be I can send it to mainline.

Signed-off-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>