"make allyesconfig"?

[Benjamin Herrenschmidt]

On Wed, 2004-04-28 at 03:12, Linus Torvalds wrote:
> One thing I've not done in a long time on my main machine is to try to
> make "allyesconfig" to get better coverage on stuff. And it just doesn't
> work on ppc64.
>
> Some of the stuff is pretty simple - I can make AGP compile (work? what's
> that?) by copying the ppc32 version of agp.h to include/asm-ppc64/ and fix
> a few other assumptions.
>
> And I've disabled a few drivers that won't build (the "pmac" IDE driver
> ends up missing "mediabay" defines, some ISDN drivers seem to be making
> nasty assumptions etc), but I was wondering if somebody has already done
> all the work - marking things broken on ppc64 in the Kconfig files so that
> "make allyesconfig" just does the right thing?

No, but I can do that, though it's weird that it tried to include the
media bay stuff, that should be allowed to build on a ppc64 config.

I'l have a look.

Also, p and i series are mutually exclusive at this point, so we would
in fact need a p-allyesconfig and an i-allyesconfig ;) But then, it's
known that iseries won't built with various things.

Finally, it's also expected that an allyesconfig kernel will not boot
a pmac due to the legacy drivers trying to tap random IO ports for
things that don't exist and crashing the box. I have no quick fix for
that at the moment but ugly bits of

#ifdef CONFIG_PPC
	if (_machine == _MACH_Pmac)
		return
#endif

In various places (this is what the distros are doing lately). I hope
that in 2.7, we'll define a neat way to "describe" the legacy device
setup from the arch code by creating struct device for them with some
way to express "ranges" of IO ports that can be probed. The arch would
then create that. It's one of the topic of the Kernel Summit this year.

Ben.


** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/