Re: [PATCH] time/sched_clock: mark sched_clock_read_begin as notrace

[Peter Zijlstra]

On Mon, Sep 28, 2020 at 06:49:52PM +0800, quanyang.wang@xxxxxxxxxxxxx wrote:
> From: Quanyang Wang <quanyang.wang@xxxxxxxxxxxxx>
> 
> Since sched_clock_read_begin is called by notrace function sched_clock,
> it shouldn't be traceable either, or else __ftrace_graph_caller will
> run into a dead loop on the path (arm for instance):
> 
>   ftrace_graph_caller
>     prepare_ftrace_return
>       function_graph_enter
>         ftrace_push_return_trace
>           trace_clock_local
>             sched_clock
>               sched_clock_read_begin
> 
> Fixes: 1b86abc1c645 ("sched_clock: Expose struct clock_read_data")
> Signed-off-by: Quanyang Wang <quanyang.wang@xxxxxxxxxxxxx>
> ---
>  kernel/time/sched_clock.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c
> index 1c03eec6ca9b..58459e1359d7 100644
> --- a/kernel/time/sched_clock.c
> +++ b/kernel/time/sched_clock.c
> @@ -68,7 +68,7 @@ static inline u64 notrace cyc_to_ns(u64 cyc, u32 mult, u32 shift)
>  	return (cyc * mult) >> shift;
>  }
>  
> -struct clock_read_data *sched_clock_read_begin(unsigned int *seq)
> +notrace struct clock_read_data *sched_clock_read_begin(unsigned int *seq)
>  {
>  	*seq = raw_read_seqcount_latch(&cd.seq);
>  	return cd.read_data + (*seq & 1);

At the very least sched_clock_read_retry() should also be marked such.

But Steve, how come x86 works? Our sched_clock() doesn't have notrace on
at all.