On Sat, Nov 25, 2006 at 08:19:05AM +1100, Benjamin Herrenschmidt wrote:

quoted

I haven't been following this very closely.  The problem is that you have 
a system with two different kinds of OHCI controllers, thus requiring two 
different and incompatible versions of the driver?

Sort-of. Basically, PowerPC can (and must in some case) build a single
kernel image that can be booted on a variety of platforms. There are at
least 3 platforms coming in now (this Efika is one of them) where that
includes a non-PCI OHCI or EHCI (big endian even) while the same kernel
must also contain the normal PCI OHCI/EHCI driver.

It's generally not a problem as the endian thingy is a runtime bit, and
in general, the whole code allows for this just fine, with, afaik, the
exception of the multiple module_init() bits...

quoted

If the only source of incompatibility is the module_init() and 
module_exit() routines, then your suggestion would be a simple fix.  
Ugly, yes, but I don't think anyone will really mind.

Is there anything else to prevent you from combining multiple HCIs into a
single driver?

Not that I know. But we could have a cleaner approach. The #include one
dates a bit... we could have for example an ohci-core.ko with the core
bits, and ochi-pci.ko, ohci-patform.ko etc... be separate modules linked
on the first and instanciating it.

I'll have a look and will come up with either solution.

Hi, ...

I am just wondering if i got dropped from the CCs or if there was no further
discussion on this.

The 2.6.19 kernel doesn't build when the efika patches are applied and the
ohci modules are built. We need a solution for this. Even disabling the patch
for ohci fails to build the ohci driver, but this could be a bad manipulation.

Is there any hint on the right way to go for this ? 

Friendly,

Sven Luther