Re: [PATCH] bio: gcc warning fix.

From: Al Viro
Date: Fri Jan 06 2006 - 14:55:25 EST


On Fri, Jan 06, 2006 at 12:33:56PM -0700, Zan Lynx wrote:
> On Fri, 2006-01-06 at 19:48 +0100, Jens Axboe wrote:
> > On Fri, Jan 06 2006, Jesper Juhl wrote:
> > > gcc is right to warn in the sense that it doesn't know if
> > > bvec_alloc_bs() will read or write into idx when its address is passed
> >
> > The function is right there, on top of bio_alloc_bioset(). It's even
> > inlined. gcc has absolutely no reason to complain.
>
> GCC complains because it is possible for that function to return without
> ever setting a value into idx. It's the "default" case in the switch.
> Of course, if that happens, idx will not be used and so it is not
> actually a problem.

If gcc would look at that code *after* it expands the call, it would
actually notice that everything's fine. The codepath leaving the
inlined block without setting idx would look like

bvl = NULL;
goto l1;
...
l1:
if (!bvl)
goto l2;
use idx
...
l2:
mempool_free(bio, bs->bio_pool);
bio = NULL;
goto out;

and after that exit collapses into jump directly to l2, we end up with
situation when every path to use of idx obviously going through l1 and,
before that, end of switch() in inlined block. All possible precursors
of that end of switch assign idx.

And yes, if you inline it manually gcc _will_ see that everything's OK.
Path that confuses it is
default in switch -> exit from bio_alloc_bs() -> l1 -> use of idx
and
return value will be NULL => we will go to l2
is what it doesn't notice when it inlines itself.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/