Re: [PATCH 1/2] Add thread_info_cache_init() to all archs

[Sam Ravnborg]

On Wed, May 21, 2008 at 11:41:47AM -0700, Andrew Morton wrote:
> On Wed, 21 May 2008 13:56:25 -0400 Benjamin Herrenschmidt <benh@xxxxxxxxxxxxxxxxxxx> wrote:
> 
> > 
> > On Fri, 2008-04-18 at 16:44 +1000, Benjamin Herrenschmidt wrote:
> > > > > so what
> > > > > about the patch below ?
> > > > 
> > > > I like it, but the compiler won't ;)
> > > > 
> > > > > If you're ok, I'll re-send with appropriate sob
> > > > > & adapted powerpc part.
> > > > 
> > > > Sure.
> > > > 
> > > > > +void __init __attribute__((weak) thread_info_cache_init(void)
> > > >
> > 
> > Back to this old subject...
> > 
> > I'm having reports that this is not working...
> > 
> > gcc is seeing the empty weak function and is optimizing it out
> > before it gets a chance to link to the arch provided one.
> > 
> > This would affect that and the other one next to it..
> > 
> > That seems pretty bad... it causes nasty crashes as we end up having no
> > idea what the compiler decided to generate... I suppose we could keep
> > the weak stubs out of the file where they are called but that sucks.
> > 
> > ie. This is some form of gcc 4.1.1
> > 
> > Is that a known problem ? A gcc issue ? Not sure what is expected from
> > those weak functions.
> 
> yup, gcc bug.  Discussed recently on lkml, "Subject: Re: huge gcc
> 4.1.{0,1} __weak problem".  I don't think anything ended up happening
> about it though.

It was discussed to add some run-time checks for this issue.
But the examples given were a bit fluffy so I never integrated anything
i kbuild to detect this.

As this is only a bug for const weak functions they could be made non-const
if they are seldomly used?

	Sam
--
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/