SDK 1.7 fails to detect PCI-e devices on p5020

[darrenstevens]

Hi All,

I'm trying to boot a SDK1.7 kernel on my AmigaOne X5000, a p5020 based board, and while the kernel boots, it does not detect any PCI-e devices, not even the 5020 root ports.

I can log in and use the system over serial, and can see that other SOC devices (SATA/USB/ethernet) seem to working.

Some of the problem seems to be down to removal of the lines:

pcibios_scan_phb(hose);
pci_bus_add_devices(hose->bus);

from the end of fsl_add_bridge() function in arch/sysdev/fsl_pci.c

Adding them back partially fixes the problem, but X will still not start.

Anyone got any ideas?

Regards

Darren

[scottwood]

The removal of those lines relative to what tree?  Is there a kernel on which this worked before?

pcibios_scan_phb() and pci_bus_add_devices() are called from pcibios_init() in pci_64.c (or pci_32.c), not fsl_pci.c, and I don't see a previous kernel where that was different.  If you did mean pci_64.c (or pci_32.c), where do you see those lines being removed?

[darrenstevens]

The changes are relative to the 3.8.13-rt9 kernel based on the SDK1.5 that the board manufacturer provided us with. This kernel works mostly fine, in fact I am using Iceweasel on the board to post this.

I have pasted my dmesg here:

http://pastebin.com/wWVbGU3W

You can see the Radeon DRM driver is failing on a ring test, this happens with >3.5M ram enabled, one of the reason I want to move on to a newer kernel.

And here is the output from a default SDK1.7 kernel:

http://pastebin.com/aqzHmcPA

The supplied source does have an additional file to provide board initialisation, it's beginning to sound like more was patched than I thought.

FYI, this is the same board AlexPerez has been having trouble with.

Hope this makes things clearer.

Thanks

Darren

[scottwood]

Those two functions are not called in fsl_pci.c in SDK 1.5.  Could you post the diff between SDK 1.5 and the modified kernel you got from the manufacturer?  At least, files relevant to PCI and the board file.

[darrenstevens]

Scott,

Here is the patch we apply to the kernel (both SDK1.5 and SDK1.7)

Without this patch we see nothing at all when we run lspci - not even the processors root ports (0x1957 0x0421) so how is anyone else using PCI-e devices with this kernel without a similar patch? Is there something else going on?

The Cyrus board that we are running this has PCI-e port 0 connected directly to a PCI-e 16x connector, and a 4 lane to 4 lane and 4x1 lane switch. connected to port 1, so nothing unusual in its design.

Regards

Darren

[scottwood]

Is that the only change you've made to the SDK kernel (probably not, since this is a custom board)?  PCIe certainly works with an unmodified SDK on supported evaluation boards.  Do you have board code that may be doing something different from what corenet_generic.c is doing?