How to block pci config-reads during device self-test?

[Linas Vepstas]

On Thu, Jul 08, 2004 at 09:45:22AM -0700, Greg KH was heard to remark:
> On Thu, Jul 08, 2004 at 11:36:28AM -0500, Brian King wrote:
> > I've been doing some talking with various hardware folks to see if
> > there is a way to get this fixed and so far the answer I have been
> > getting is no. It is how the hardware works and there isn't much
> > that can be done about it in microcode.
>
> But this is limited only to a single PCI device that you have,
> correct?

I have a vagure recollection that this is not the first time this
has been seen, but that was a while ago, and I didn't understand
the issue then.

We don't know how many more devices are affected.  The pSeries
PCI hardware significantly clamps down on what is allowed on the
bus, all in the name of not corrupting the kernel.  Traditional
PCI bridges are a lot more permissive ... we actually don't know how
many more cards are out there that might go wild if touched
in unexpected ways.

However, note that most, (if not all?) of the devices on ppc64 are
now chasing or have recently chased rare bugs due to the EEH clampdown,
so this might be the tip of an iceburg.

> To add such a huge core change for only a single, broken device, that
> 99.99% of the Linux users and developers will never see is not a nice
> option.

I'll work with Brian to add something to arch/ppc64/kernel/pSeries_pci.c
that works around this problem.    There is a separate patch I'll mail
later today for dealing with EEH detection in the same codepath.

--linas

** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/