Re: [PATCH] tracepoint: Fix CFI failures with tp_stub_func

[Mark Rutland <mark.rutland@arm.com>]

On Thu, Mar 23, 2023 at 12:26:50PM -0400, Steven Rostedt wrote:
> On Thu, 23 Mar 2023 08:53:21 -0400
> Steven Rostedt <rostedt@goodmis.org> wrote:
> 
> > > --- a/include/linux/tracepoint.h
> > > +++ b/include/linux/tracepoint.h
> > > @@ -303,6 +303,7 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
> > >  	__section("__tracepoints_strings") = #_name;			\
> > >  	extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name);	\
> > >  	int __traceiter_##_name(void *__data, proto);			\
> > > +	void __tracestub_##_name(void *, proto);			\
> > >  	struct tracepoint __tracepoint_##_name	__used			\
> > >  	__section("__tracepoints") = {					\
> > >  		.name = __tpstrtab_##_name,				\
> > > @@ -310,6 +311,7 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
> > >  		.static_call_key = &STATIC_CALL_KEY(tp_func_##_name),	\
> > >  		.static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \
> > >  		.iterator = &__traceiter_##_name,			\
> > > +		.stub = &__tracestub_##_name,				\
> > >  		.regfunc = _reg,					\
> > >  		.unregfunc = _unreg,					\
> > >  		.funcs = NULL };					\
> > > @@ -330,6 +332,9 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
> > >  		}							\
> > >  		return 0;						\
> > >  	}								\
> > > +	void __tracestub_##_name(void *__data, proto)			\
> > > +	{								\
> > > +	}								\  
> > 
> > I purposely did not do this because this adds over a thousand stub
> > functions! It adds one for *every* tracepoint (and that is a superset of
> > trace events).
> 
> And the commit that added this code:
> 
>   befe6d946551 ("tracepoint: Do not fail unregistering a probe due to memory failure")
> 
> Has this in the change log:
> 
>     [ Note, this version does use undefined compiler behavior (assuming that
>       a stub function with no parameters or return, can be called by a location
>       that thinks it has parameters but still no return value. Static calls
>       do the same thing, so this trick is not without precedent.
> 
>       There's another solution that uses RCU tricks and is more complex, but
>       can be an alternative if this solution becomes an issue.
> 
>       Link: https://lore.kernel.org/lkml/20210127170721.58bce7cc@gandalf.local.home/
>     ]

FWIW, I'd be happy with that approach too -- we just happened to race with our
last replies. :)

Mark.