ioremap (still?) broken

[Paul Mackerras]

Troy Benjegerdes <hozer at drgw.net> wrote:

> The alteon driver does an 'ioremap' on the address provided by the kernel
> PCI services, and then expects to be able to access the PCI card's memory
> via accessing the memory returned using 'readl,writel'.

That should work.

> The first thing the driver does is check the card version by reading a
> particular register. By using appropriate printf's, I have found that the
> value returned is '0xffffffff'. (defininitely NOT the register it's
> looking for)

Possibly you may need to enable PCI memory or I/O space responses in
the PCI command register in the card's configuration space.

> Before I go on a wild chase again, could someone let me know whether the
> 2.2.12 kernel might have a better chance with this, or will I need to go

I just checked in a fix to ioremap in the vger 2.2 branch which fixes
the case where the mapping added by ioremap is the first in a 4MB
region.  You would be getting kernel DSI exceptions (page faults) if
you had hit that bug, though.

Paul.

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/