Re: [mingo@elte.hu: [git pull] headers_check fixes]

[Linus Torvalds <torvalds@linux-foundation.org>]

On Tue, 27 Jan 2009, H. Peter Anvin wrote:

> Linus Torvalds wrote:
> > 
> > Do you think the "fix headers_check" patches spend lots of time analyzing
> > things? I bet no. They just try to make the warning go away, so you don't
> > actually end up with any more "coverage" anyway. Quite the reverse - instead
> > of having a simple rule ("CONFIG_xyz options simply do not exist in user
> > space"), you end up having ad-hoc hacks on a per-fix basis.
> > 
> 
> This is probably true.  I think we should add this as one more of the
> preprocessing rules which we really should just do, as well as automatic
> mangling of integer types.

Btw, the really scary thing is that I bet there are programs out there 
that "know" about kernel internals, and do things like

	#define CONFIG_SMP 1
	#define __KERNEL__ 1
	#include <asm/atomic.h>

in order to get the atomic helpers from the kernel, and using CONFIG_xyz 
markers to force the exact version they want.

And we will inevitably always end up breaking stuff like that. Nothing we 
can do about it - in the end, users can do infinitely odd things and know 
about our internals, and whatever changes we do will occasionally break 
some of the more incestuous code.

			Linus