Kernel PCI Scan Bug - No feedback 8 days after bug submittal (reposted with newlines)

[Jim Cornette]

Charles Butterfield wrote:
> I've had what seems like a fairly serious kernel bug related to PCI
> scanning open for over a week, with no hint that anybody has read it,
> triaged it, etc.  Some feedback would sure be nice.  I'm posting this
> on the email list, since I'm not really sure how else to get developers
> to notice the formal bug report.

Your bug does sound a bit of a tough problem. I guess posting to the 
list does expose your problem to a wider audience. The problem interests 
me. It also has me curious as to why /proc/bus/pci and the devices file 
have different numbering schemes. the processes are the same amount as 
the nodes or whatever is in the directory. The entries however seem to 
be in different base numbering schemes. (octal, hexadecimal or whatever 
the scheme is.) I don't know how it works at all.

> 
> History:
> 
>
> In the kernel used by FC6, there is a PCI scan/enumeration bug that
> manifests itself on some machines (like mine) as a mismatch between the
> contents of /proc/bus/pci/devices (e.g. 14 devices) and the nodes in
> /proc/bus/pci/xx/* (e.g. 15 devices).  Clearly this could cause numerous
> problems.  In my case it causes Xorg to fail to find my video card.
> YMMV

I had a one for one match in my devices file for each directory. The 
data is different than I would expect since the numbers except for zero 
do not seem to match one to one.

You have me stumped but curious. I have 10 devices under 00. One device 
under 01 (video card) and five devices under 02 (another video card 
being one of the elements)

>  
> This problem did not exist in the kernels used by FC5 (or at least did
> not manifest on my system).

Have you tried to install a kernel from the FC5 timefram to see if it 
works as in the past?

Sorry not a lot of help.

Jim

> 
> Regards
> -- Charlie B


-- 
I have a theory that it's impossible to prove anything, but I can't 
prove it.