Stefan Reinauer <[EMAIL PROTECTED]> writes:

> * Ronald G Minnich <[EMAIL PROTECTED]> [020703 15:58]:
> > Eric, for your PCI scheme, do you end up with an entry for every single
> > PCI device, or for classes of devices? I'm wondering how big linuxbios
> > will get to be if we have the whole PCI database in there.
> > 
> > I wonder how the commercial bioses handle this?
>  
> special handling is only needed for very rare pluggable devices. 
> it rather occurs for onboard bridges and such, which are specifically
> known when compiling.

Right.  There are lots of bridge devices that need a little special
tweaking. PCI-ISA bridges, northbridges etc.

> > can linuxbios make everything slow and let linux handle it?
> 
> Linux has a lot of fixups for devices anyways so this should not be a
> problem.
> I think keeping complexity out of LinuxBIOS in this case makes a lot of
> sense.

1) If it is motherboard specific Linux should not touch it.
2) I'm not advocating doing things in LinuxBIOS that we aren't
   doing right now.  Instead I'm advocating putting a driver model in
   front of it.  

This should reduce the number of hard codes and make LinuxBIOS more
maintainable.

The model is I know I have chips of types A, B, and C on the motherboard
so compile those in.

With things like 4G of ram I need to allocate all of the resources before
Linux starts up so I can be certain all pci resources will fit into the
hole provided for them.

Eric

Reply via email to