Without PCI_AUTO none of the pci devices will be configured properly,
unless the boot code sets them up. The pmon version you have does that,
I believe, but I'm not sure everyone can count on using the same boot
code. I think we fixed the problem you describe by skipping the system
controller (IT8172) config and leaving its default map. The small patch
which affects it8172_pci.c is in the source forge tree but I don't know
if it made it to oss, probably not since the interest in that board is
small in the community itself. In general, I think the IVR board
support in the sourceforge kernel is more up to date than oss.
On Tue, 2002-08-20 at 08:50, Tom Collins wrote:
> I'm not a subscriber to the list, so if anyone has questions or
> responses, please email me directly.
> I have a machine based on the Globespan IVR board, and just got done
> fighting through an upgrade from 2.4.2 to 2.4.19. In the process, I
> determined that (at least for my configuration), CONFIG_PCI_AUTO should
> not be set. When set, the I/O mappings for my devices (USB) were out of
> whack, and I was unable to use one of the on-board USB chips.
> Here's the configuration as it stands on my machine now:
> if [ "$CONFIG_MIPS_IVR" = "y" ]; then
> define_bool CONFIG_PCI y
> define_bool CONFIG_PC_KEYB y
> define_bool CONFIG_NEW_PCI y
> define_bool CONFIG_NONCOHERENT_IO y
> define_bool CONFIG_PCI_AUTO n
> define_bool CONFIG_IT8172_CIR y
> define_bool CONFIG_NEW_IRQ y
> define_bool CONFIG_NEW_TIME_C y
> I am still working on a problem with a PCI card on the unit though.
> It's a USB 2.0 interface which appears to the kernel as two OHCI (1.0)
> interfaces and one EHCI (2.0) interface. Right now, the kernel doesn't
> find the EHCI interface. If I find that changing other settings (like
> CONFIG_NEW_PCI) results in that interface showing up (without blowing up
> the others), I'll be sure to tell the list.
> Can someone here please make sure this change gets into the 2.4.20
> kernel? I haven't been using the 2.5 kernels because the last one I
> tried (2.5.25) wouldn't compile on my platform.
> Tom Collins