libpciaccess bogus bridge data - commit 2bda5b73

[Jesse Barnes]

On Wed, 1 Dec 2010 18:54:22 -0800
Bryce Harrington <bryce at canonical.com> wrote:

> Jesse,
> 
> I have a question about change 2bda5b73 that you did to libpciaccess
> last year, to stop pci_device_get_bridge_buses() from looking at bus
> data if the bridge data was not read.
> 
> A ubuntu bug reporter says that this check is preventing his device from
> getting detected properly - see
> 
>  https://bugs.launchpad.net/ubuntu/+source/libpciaccess/+bug/681207
> 
> Looking at the code I see that for case 0x04 there's already a check for
> NULL, where it attempts to re-read the bridge info, so presumably there
> are at least some situations where NULL bridge.pci is expected?
> 
> What I'm wondering is if the check should be less strict in cases of
> multi-function cards, or if somewhere else in the code needs updated to
> read these devices properly.

The multifunction check is probably ok, but there should still be a
bridge above the device regardless, even if it's just a host bridge I
think, so we should figure out why that's not getting set...

-- 
Jesse Barnes, Intel Open Source Technology Center