Re: [patch 2.5] 2-pass PCI probing, generic part

[Benjamin Herrenschmidt]

On Thu, 2003-01-09 at 23:16, Linus Torvalds wrote:
> On Fri, 10 Jan 2003, Ivan Kokshaysky wrote:
> >
> > Note that in most cases PCI-PCI bridges can be safely excluded from
> > pci_read_bases() simply because they have neither regular BARs nor
> > ROM BAR (even though PCI spec allows that).
> 
> This might be a good approach to take regardless - don't read pci-pci 
> bridge BAR (or host-bridge BAR's for that matter), simply because 
> 
>  (a) bridges are more "interesting" than regular devices, and disabling 
>      part of them might be a stupid thing.
>  (b) we're generally not really interested in the end result anyway

I completely agree. For example, I have already a bunch of cases where I
have to explicitely "hide" the host bridge from linux PCI layer as those
have BARs that mustn't be touched. A typical example is the 405gp
internal PCI host. It has a BAR that represents the view of system RAM
from bus mastering PCI devices. Of course, the kernel doesn't understand
that and tries to remap it away from 0 where it belongs ;)

Ben.

-- 
Benjamin Herrenschmidt <[e-mail]>