Re: cardbus reads from not exist function

[David Young <dyoung%pobox.com@localhost>]

On Mon, Jun 30, 2008 at 10:27:06PM +0200, Matthias Drochner wrote:
> 
> kiyohara%kk.iij4u.or.jp@localhost said:
> > +#if 0
> >         bcr |= PCI_BRIDGE_CONTROL_MABRT << PCI_BRIDGE_CONTROL_SHIFT;
> > +#endif 
> 
> Yes I think this makes most sense, at least for now.
> 
> jakllsch%kollasch.net@localhost said:
> > But maybe the error reporting should be enabled after the scan? 
> 
> Parity check etc should be OK. As I understand it it is just the master
> abort (which is kind of a timeout if the target function doesn't
> respond) which hurts. If the cardbus bridge translates this into
> a target abort it confuses the PCI framework of at least some
> architectures. (Istr some alphas check, but other ports rely on reading
> all-ones.)
> The problem with enabling after the scan is that "after the scan
> is before the scan" on cardbus. I'd say we can live without
> the MABRT bit set. Needs some testing of course.

Turning off master abort-forwarding is a step backwards.  A step forward
is to implement proper PCI exception handling.  That is not to say that
we should not take that backwards step for the short term, but let us
properly understand this bug, first.  Looks to me like there is a crash
in Cardbus_mapreg_unmap().  Previously, we had _map()'d the region.
Why the crash?

Dave

-- 
David Young             OJC Technologies
dyoung%ojctech.com@localhost      Urbana, IL * (217) 278-3933 ext 24