On 10/30/2011 18:34, Thomas Bogendoerfer wrote:
> no, but it will make live a lot easier, because address and interrupts don't
> need to be probed by the driver. Right now interrupts are on my todo, since
> there is some weirdness between guiness and fullhouse boxes...
It wouldn't be an SGI machine if it didn't implement something weird or just
> it still needs something to setup the PCI bus on the card and issue
> the probing. The problem with the Tulip Phobos cards is, that they
> messed up the endianess, so that none of the Linux Tulip drivers will
> work out of the box...
A.k.a., Tulip (and possibly ThunderLAN) assume little-endian, when we're
talking big-endian archs here. Interesting. Simple fix, as in defining a
few driver structures with little- and big-endian versions (if they're doing
something like packing bits or using bitfields)? Or is it more complex than
"The past tempts us, the present confuses us, the future frightens us. And
our lives slip away, moment by moment, lost in that vast, terrible in-between."
--Emperor Turhan, Centauri Republic