On Wed 29-11-17 22:44:45, Tetsuo Handa wrote:
> Michal Hocko wrote:
> > On Sat 25-11-17 10:40:13, Tetsuo Handa wrote:
> > > Michal Hocko wrote:
> > > > On Fri 24-11-17 22:21:55, Tetsuo Handa wrote:
> > > > > Michal Hocko wrote:
> > > > > > > Since we can encourage register_shrinker() callers to check for failure
> > > > > > > by marking register_shrinker() as __must_check, unregister_shrinker()
> > > > > > > can stay silent.
> > > > > >
> > > > > > I am not sure __must_check is the right way. We already do get
> > > > > > allocation warning if the registration fails so silent unregister is
> > > > > > acceptable. Unchecked register_shrinker is a bug like any other
> > > > > > unchecked error path.
> > > > >
> > > > > I consider that __must_check is the simplest way to find all of
> > > > > unchecked register_shrinker bugs. Why not to encourage users to fix?
> > > >
> > > > because git grep doesn't require to patch the kernel and still provide
> > > > the information you want.
> > >
> > > I can't interpret this line. How git grep relevant?
> >
> > you do not have to compile to see who is checking the return value.
> > Seriously there is no need to overcomplicate this. Newly added shrinkers
> > know the function returns might fail so we just have to handle existing
> > users and there are not all that many of those.
>
> Newly added shrinker users are not always careful. See commit f2517eb76f1f2f7f
> ("android: binder: Add global lru shrinker to binder") for example.
> Unless we send __must_check change to linux.git, people won't notice it.
Crap code doesn't really warrant special handling. This is a failure of
reviewers...
Really, if we start abusing __must_check for non-fatal paths then it
will turn into an ignored class of warnings or find workarounds to
silent false positives.
Look, I will not lose more time discussing this borderline thing with
you even though obviously consider this the number one problem. But you
should really think more from a wider perspective rather than
obsessively focus into a unlikely corner case and beat the soul out of
it.
I really do appreciate you started fixing the remaining places of
course, that is the usual way to deal with these issues. But placing
__must_check around is just pointless. If Andrew thinks this is really
worth it, I will not object...
--
Michal Hocko
SUSE Labs
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
