Why devices ARE or NOT on bus 0 ?

[Eric W. Biederman]

Li-Ta Lo <ollie at lanl.gov> writes:

> Hello guys,
> 
> How does the enumeration decide if the devices in
> 8111/8131 are on bus 0 or other bus ? 

In LinuxBIOS only the cpus are on bus 0, at least
when we are done.

> Isn't that
> all these non-coherent devices are on PCIDEV(0,0,0),
> at power up ? How are they changed to their "real"
> PCIDEVFN ?

Not quite.  Their bridge decides their bus number.
The just start out at devfn 0 on which ever bus they
are assigned.

We do enumerate one of the chains statically before
the general resource allocator hits, and I think
we leave it at bus 0.

> We are having problems on ibm/e325 that these devices
> remain on bus 0 such that the current HT and PCI scan
> or resource allocation is totally fucked. The devices
> are on bus 0 device 1 or something, these devices are
> enumerated first than the northbridge so the northbridge
> does not get correct information about its "link" and
> can not set IO/MEM/BUS routing correctly (actually it
> refuse to set at all).

You might need to specify the bridges/busses in Config.lb.
I know the current code works and can handle this.

> Does anyone can help on this ?

Without something like a serial console trace so that the details
of what are going on are clear I don't know.

I have a hunch.  The default code for setting things up is
for pci devices.  In the opteron mainboard files we need to specify
non-default operations so we do the hypertransport scan.  One
of the things that does is to put everything back at devfn 0
if they have changed from someplace else.

amdk8_scan_root_bus very deliberately avoids all of the other
functions that are not cpus so if you are using it I don't see
how you would run into the problem you have described.

Eric