On 15/04/20(Wed) 17:04, Martin Pieuchot wrote:
> On 15/04/20(Wed) 09:26, Martin Pieuchot wrote:
> > jca@ is currently building kernels with "-Wno-error=uninitialized" and
> > reported a warning in ART in SP builds:
> >
> > /usr/src/sys/net/art.c:256:10: warning: variable 'ndsr' is uninitialized
> > when used here [-Wuninitialized]
> > dsr = ndsr;
> > ^~~~
> > /usr/src/sys/net/art.c:221:2: note: variable 'ndsr' is declared here
> > struct srp_ref dsr, ndsr;
> > ^
> > warning generated.
> >
> > With a single CPU there's no need to save a reference and srp_leave()
> > just does nothing. In other words the warning above is harmless.
> >
> > The diff below prevents the false positive by turning srp_enter() into
> > a static inline function, I find that nicer rather than clutter the code
> > with #ifdef.
>
> Now without typo, spotted by jca@ thanks!
Any consensus about such warnings? Any ok or should I drop the diff?
> Index: sys/srp.h
> ===================================================================
> RCS file: /cvs/src/sys/sys/srp.h,v
> retrieving revision 1.14
> diff -u -p -r1.14 srp.h
> --- sys/srp.h 31 Mar 2019 14:03:40 -0000 1.14
> +++ sys/srp.h 15 Apr 2020 15:03:00 -0000
> @@ -96,11 +96,17 @@ void *srp_enter(struct srp_ref *, struc
> void *srp_follow(struct srp_ref *, struct srp *);
> void srp_leave(struct srp_ref *);
> #else /* MULTIPROCESSOR */
> +
> +static inline void *
> +srp_enter(struct srp_ref *sr, struct srp *srp)
> +{
> + return (srp->ref);
> +}
> +
> #define srp_swap(_srp, _v) srp_swap_locked((_srp), (_v))
> #define srp_update(_gc, _srp, _v) srp_update_locked((_gc), (_srp), (_v))
> #define srp_finalize(_v, _wchan) ((void)0)
> -#define srp_enter(_sr, _srp) ((_srp)->ref)
> -#define srp_follow(_sr, _srp) ((_srp)->ref)
> +#define srp_follow(_sr, _srp) srp_enter(_sr, _srp)
> #define srp_leave(_sr) do { } while (0)
> #endif /* MULTIPROCESSOR */
>
>
