sparse warnings

[Uwe Kleine-König]

On Thu, Nov 18, 2010 at 11:01:33AM +0100, Sascha Hauer wrote:
> On Thu, Nov 18, 2010 at 10:36:51AM +0100, Belisko Marek wrote:
> > Hi,
> > 
> > On Wed, Nov 17, 2010 at 10:11 AM, Sascha Hauer <s.hauer at pengutronix.de> wrote:
> > > Hi Marek,
> > >
> > > On Mon, Nov 15, 2010 at 03:20:47PM +0100, Belisko Marek wrote:
> > >> Hi,
> > >>
> > >> barebox compilation with C=1 produce a lot of sparse warnings.
> > >> Mainly concerning __iomem problems with readb() and similar functions.
> > >>
> > >> Make it sense to take care or just could be omitted?
> > >
> > > I think it makes sense to work on this. Then we can see the useful
> > > warnings buried under the __iomem warnings.
> > >
> > > I had the idea of adding a
> > >
> > > #define IOMEM(addr)        ((void __force __iomem *)(addr))
> > >
> > > and use it where appropriate.
> > Maybe stupid question but couldn't be __iomem mechanism removed completely?
> > Do we need to check for different address_space? In my opinion it
> > makes no sense in
> > barebox.
> 
> I have a better feeling letting it in. There may be no different address
> spaces on Arm, but there are for exmample on x86.
> You can simply do a #define __iomem in include/linux/compiler.h to
> silence these kinds of warnings temporarily if you are not interested.
> I agree that at least on Arm these warnings will not reveal any real
> bugs.
I can imagine you to get hard to find bugs if you interpret a pointer to
registers as normal pointer without volatile.  Something like:

	unsigned long *register = 0xabcdef04;
	while (*register & SOME_FLAG)
		/* nothing */

This might generate code that corresponds to

	if (*register & SOME_FLAG)
		while(1);

So I feel for letting it in, too.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |