Re: [PATCH] x86: do not expose CONFIG_BSWAP to userspace

[Harvey Harrison <harvey.harrison@gmail.com>]

On Wed, 2009-01-28 at 14:15 -0800, H. Peter Anvin wrote:
> Harvey Harrison wrote:
> > I'm afraid my knowledge of gcc compiler flags for various models is
> > lacking, I used i486 as suggested, just wanted to make sure I understood
> > you corectly.
> 
> You did, but I misremembered... instead of having the __i386__, 
> __i486__, __i586__, __i686__ being an additive chain as would make 
> sense, gcc just has __i386__ plus whichever corresponds to the -march= 
> option.  I keep forgetting this because it's just so incredibly dumb.
> 
> Bloody hell.  This really f*cks thing up.

> What's worse, they seem to simply be adding new options, so at this 
> point you'd actually need something like:
> 
> # if defined(__i486__) || defined(__i586__) || defined(__i686__) || \
> 	defined(__core2__) || defined(__k8__) || defined(__amdfam10__)
> 
> Worse, there isn't any kind of macro that can be used to compare for a 
> negative (i.e. not i386).

Well, that's unfortunate, how about we just export the BSWAP version
unconditionally and hope pure i386 just goes away someday?

> 
> This obviously is screaming to be abstracted away into a header of its 
> own, but it really can't be done cleanly as far as I can tell because of 
> this particular piece of major gcc braindamage.
> 
> So, one ends up doing something like:
> 
> #ifdef __i486__
> # define __CPU_HAVE_BSWAP
> #endif
> #ifdef __i586__
> # define __CPU_HAVE_BSWAP
> #endif
> 
> ... and so on, and have to keep this up to date with the latest 
> inventions of the gcc people.  *Sob.*

Unpleasant indeed.  Is there a byteswap builtin in gcc?  At least AVR32
seems to use it, but perhaps it's not generally exposed...perhaps we
could ask the gcc-folk?

Harvey