PCI bridges are not recognized by the kernel

[krokoziabla]

linux-3.8.13-gentoo
x86_64

Today I finally received my DFE-580TX multiport server network adapter and was looking forward to seeing it work. But so far I haven't succeeded in setting it up. In modern time it is a piece of antiques. But please don't blame me, I didn't have a free PCIe slot to insert a newer card.

But let's go back to the business. DFE-580TX simply consists of four DL10050C MAC/PHY microcontrollers connected to one Intel S21152BB PCI-to-PCI bridge. And my problem consists in that I cannot make the bridge work. Here is lspci printout:

[NeddySeagoon]

krokoziabla,

Help the kernel to find your buses. Add

[roarinelk]

Bridges like the ones on your card are normally transparent and don't need a driver to work.

Are you certain your pci card(s) (and pci slots on the board) are okay? When my 3c905 card died
it was detected as 16 pci devices by the kernel. To me your main problem seems to be a
failing PCI device which disrupts the others.

Experiment with the "pci" kernel parameter a bit. See the "pci=option" Section in linux/Documentation/kernel-parameters.txt
try at least "pci=conf1" conf2, bios, noacpi

[krokoziabla]

roarinelk & NeddySeagoon, thank you for your replies.

I'm glad to learn that the bridge is uncommanded, I do not need to bother about drivers. And it seems all the three bridges are.

I've done some tests of the kernel with pci parameter values you suggested and also some others. No effect.

Also I've made two tests against Windows. At first I put the board into another PC with Windows. The bridge was identified without a problem and I saw the four Ethernet adapters. So I can consider the board healthy. The next thing I did was taking the hard disk from the other PC and connecting it to my Gentoo PC. That way I wanted to see how Windows would identify the bridge in the original PCI topology. And to my grief the bridge wasn't correctly identified and I saw the same 96 false devices.

So it may be the problem of the compatibility between my motherboard and the network card although both of them comply with PCI 2.2 standard. BTW, the slot also seems to be healthy because before this network card another NIC worked quite well in it.

I've checked BIOS options and it looks like there is not that much to control behaviour of PCIe/PCI buses.

The last suspicious thing that worries me is the PCI topology of my system. Recently I've known that lspci can be run with -tv options.

[krokoziabla]

I have one more idea. It looks like the bridge is configured wrongly. If I understand rightly the configuration address space of a PCI device determines which memory and i/o address space regions are used by the device. It looks like the regions of S21152BB and iTE 8892 overlap. At least this would explain why there are 96 false devices and why they pretend to be devices from bus 03.

Does the kernel allow manual configuration of a PCI device? Probably I'll be able to configure the bridge myself if the kernel cannot do this automatically.
_________________
Vitaly

[roarinelk]

If possible, try the latest -git sources, and make a problem report to linux-kernel@vger.kernel.org and linux-pci@vger.kernel.org

Include full lspci and dmesg output, and if possible also a dump of all memory regions used by the 4 nics.

[krokoziabla]

roarinelk

Good, I'll follow your advise but maybe later. I want to get to the bottom of PCI configuration mechanism myself first.

So if somebody is interested in this topic I've just find out (from http://read.pudn.com/downloads13/doc/52518/PCI-to-PCI%20Bridge%20Architecture.pdf) that PCI-to-PCI bridges can be composed into a multi-level hierarchy. Thus my hardware configuration can be in theory compatible.

Let's see what I'll learn next.
_________________
Vitaly