On Mon, Jul 6, 2009 at 12:09 AM, Blue Swirlblauwirbel@gmail.com wrote:
On 7/5/09, Igor Kovalenko igor.v.kovalenko@gmail.com wrote:
On Sun, Jul 5, 2009 at 10:32 PM, svn@openbios.org wrote: > Author: blueswirl > Date: 2009-07-05 20:32:50 +0200 (Sun, 05 Jul 2009) > New Revision: 505 > > Modified: > trunk/openbios-devel/drivers/pci.c > Log: > Don't try to configure non-existent BARs of bridge devices > > QEMU commit b7ee1603c16c1feb0d439d2ddf6cf824119d0aab improved the PCI bridge > device handling. PCI bridges use different header format from ordinary PCI > devices, especially BAR registers do not exist. OpenBIOS tried to configure > the bridges based on the non-BAR data which was assumed to be BARs.
I stumbled across this as well. My solution was inspired by linux code, please see attached sparc64-pci-bridge patch. The difference is actually small. Added value is dealing with ROM bars in other header type, also irq setup function and pci bar setup function are extracted.
This patch is against r504. In case this looks better here is my sign-off
Signed-off-by: igor.v.kovalenko@gmail.com
Yes, it looks like an improvement. Shouldn't the bridge check also check the subclass? Otherwise EBUS BARs would not handled like they should.
I think header type matters. Maybe it is just wrong for qemu ebus bridge.