On Fri, 30 Oct 2020 10:13:50 +0100 (CET)
Miroslav Benes <[email protected]> wrote:
> how does this work in case of NMI? trace_get_context_bit() returns 0 (it
> does not change later in the patch set). "start" in
> trace_test_and_set_recursion() is 0 zero too as used later in the patch
> set by ftrace_test_recursion_trylock(). So trace_test_and_set_recursion()
> returns 0. That is perfectly sane, but then...
>
> > +static __always_inline void trace_clear_recursion(int bit)
> > +{
> > + unsigned int val = current->trace_recursion;
> > +
> > + if (!bit)
> > + return;
>
> ... the bit is not cleared here.
Yeah, I found that and fixed it yesterday, which discovered stack
overflow in perf:
https://lore.kernel.org/r/[email protected]
There's another bug that causes a false positive during interrupt
context transition. I have a fix for that too.
-- Steve
