Re: [cobalt-developers] PCI issues with RAQ3

[Tim Hockin <thockin@xxxxxxxxxx>]

Dennis wrote:
 
> 1) With our card in the box, the eepro100 driver locks the system on probe
> using the C5 kernel or the C3 kernel (the original kernel in the box). With
> the rom kernel it doesnt.

Very odd, as they are essentially the same kernel.  Can you send me:
1) the kernel boot messages (serial port, from ROM to prompt) when you
boot normal
2) the same, but when you boot from ROM

> 2) with the ROM kernel an ioremap of our board's base[0] address returns a
> NULL when loading the module. The address (e800) that we are trying to map
> appears valid.

is 0xe800 an IO region or a memory-mapped IO region?  Is it hardwired,
or does PCI bios initialize it?
 
> I've noticed that board modules (include eepro100.o)  compiled on PCs dont
> load on cobalt machines without recompiling them so obviously there is
> something different. I'm going to have to hack the eepro100 driver to see
> why it hangs, but if you can fill me in on the bus issues it might help.

I'm not sure what you mean here - There isn't anything different - you
can diff the kernels and see that.  In fact,I compile my test kernels on
a PC, and load them onto a Raq3.  I've not tried doing eepro100.o, but
please tell me what you mean by "dont load" - error message? system
lock?  details... ;)
 
> Our cards are installed in hundreds of linux systems worldwide and there
> hasnt been one MB that they havent worked with. The hang occurs with no
> driver for the board present, so nothing *should* be accessing the card.

If there is a conflict with the eepro100 driver, we can find it.  I'm
curious as to why our PCI would be different on this card than any
other.  Have you tried setting the IO range to a different area?  

Lastly, There was a conflict that dealt with the Ethernet controller
being in MMIO mode, but we now turn off MMIO mode in BIOS.  Besides
that, it was affecting disk only.

Really to help I need more details, error messages if any exist, perhaps
the contents of /proc/pci, and the results of 

"for a in /proc/bus/pci/00/*; do echo $a; hexdump $a; done"

-- 
Tim Hockin
Software Engineer / OS Engineer
Cobalt Networks
thockin@xxxxxxxxxx