Re: (hacky) [PATCH] silence MODPOST section mismatch warnings

["Satyam Sharma" <satyam.sharma@gmail.com>]

Hi Sam,

On 5/11/07, Sam Ravnborg <sam@ravnborg.org> wrote:
> On Thu, May 10, 2007 at 09:54:27PM +0100, Russell King wrote:
> > On Thu, May 10, 2007 at 01:51:47PM -0700, David Miller wrote:
> > > From: Chris Wedgwood <cw@f00f.org>
> > > Date: Thu, 10 May 2007 13:34:18 -0700
> > >
> > > > MODPOST seems to be spewing bogus warnings.  It's not clear how best
> > > > to fix it so perhaps we should silence it for now?
> > >
> > > Most of them are legitimate, the only one that needs sorting
> > > is the mm/slab.c case and people are working on that.
> > >
> > > The rest are useful and I've been working to fix things up
> > > on sparc64 and the networking, and in fact I'm very happy
> > > about these notifications.
> > >
> > > Please don't apply a sledgehammer to this issue, thanks.
> >
> > I've not had one accurate one on ARM yet.
> You had one patch from me in latest submission to linus that
> was a clear bug.
>
> >
> > Here's another example:
> >
> > WARNING: init/built-in.o - Section mismatch: reference to .init.text:
> >  from .text between 'rest_init' (at offset 0x4c) and 'run_init_process'
> >
> > from init/main.c:
> >
> > static void noinline rest_init(void)
> >         __releases(kernel_lock)
> >
> > static void run_init_process(char *init_filename)
> >
> > Clearly, it just does _not_ work.
> As I have already explained to you this is a binutils issue
> that causes this false positive.
>
> The plan is to annotate functions that are not __init that
> they intentional reference a function or data in a init section.
> I just not there yet.

The thought makes me *extremely* uncomfortable. Better to _heed_
warnings, than _hide_ them, isn't it -- they only tell us what the
code says, after all. Personally, I'd much rather see a few warnings
on my screen (which I _know_ are false positives) than hide a genuine
potential issue only to crash later.

For bogus warnings caused by binutils issues, the solution clearly
lies in fixing _binutils_.

For the "genuine false positives" (where we *do* call .init.text from
.text, but somehow ensure that it doesn't happen after boot) -- I
don't really know, but a solution that solves the (kmem_cache_create
-> setup_cpu_cache + g_cpucache_up -> set_up_list3s) problem in
_logic_ sounds more elegant to me, than inventing something like
__nowarn_calls_init (?). The problem with stuff of the latter sort is
that they often tend to be abused later by everyone even where they
are avoidable / inapplicable, which opens up a can of worms.

Just my 2 paise,
Satyam
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/