Re: 2.6.0-test9-bk16 ALi M5229 kernel boot error

[Ivan Kokshaysky <ink@jurassic.park.msu.ru>]

On Tue, Nov 11, 2003 at 10:58:45PM -0800, David S. Miller wrote:
> On Wed, 12 Nov 2003 04:31:34 +0000
> viro@parcelfarce.linux.theplanet.co.uk wrote:
> > Wrong fix, AFAICS.  Original condition is bogus, no arguments here.
> > However, the point is
> > 	"tweak our southbridge only if northbridge is known to be OK with that"
> > and not
> > 	"tweak southbridge only if it's ours"
> > 
> > IOW, proper check is || of those two.
> 
> I agree with Al's analysis, and this is the kind of logic needed on
> sparc64 boxes as well.

I'm not sure there was any logic at all, given extremely misleading
comments in the original code. That "south-bridge's enable bit" stands
for "enable input pins for 80-conductor cable detection" according
to my (rather sparse) docs, and I don't understand why the hell it has
anything to do with a northbridge.
Someone with a more complete ALi documentation ought to verify that...

> Indeed, blindly deref'ing 'isa_dev' here was pretty bogus :)

Perhaps the source of this bug was the fact that M5229 controllers
are always part of the southbridge chip and therefore respective
"isa_dev" must exist. However, PCI IDs are re-writable on newer
ALi chips, which was probably the case.

Ivan.