Re: building kernel with gcc-4.3.x

[Nicolas Pitre <nico@xxxxxxx>]

On Thu, 3 Jul 2008, Russell King - ARM Linux wrote:

> On Wed, Jul 02, 2008 at 11:52:40PM +0100, Chris Steel wrote:
> > maybe not a gcc bug?
> > 
> > ------- Comment #4 from pinskia at gcc dot gnu dot org  2008-07-02 22:47 -------
> > "o" means any offsetable memory operand so GCC thinks you are accessing memory
> > that will always be valid.
> 
> Well, pld doesn't care whether the memory being accessed is available
> or not - it doesn't fault if the memory isn't mapped or is protected.

Right.

I think the problem has to do with some gcc optimizations that consider 
superfluous any tests to determine if a pointer is null if the same 
pointer has already been dereferenced.  So the "o" constraint makes gcc 
assume that the pointer was dereferenced and any test to find out the 
end of a linked list is optimized away.

> So, I think what the GCC people have just said is "we've removed the
> feature you were using to allow you to use this instruction."

Something like that.

Either a new constraint identifier needs to be created (and therefore 
making backward compatibility a problem), or we should simply use 
__builtin_prefetch() instead of our own open coded implementation.


Nicolas

-------------------------------------------------------------------
List admin: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm
FAQ:        http://www.arm.linux.org.uk/mailinglists/faq.php
Etiquette:  http://www.arm.linux.org.uk/mailinglists/etiquette.php